Last Updated: 10 Aug 2022
Note: Program Uses Hong Kong, China Standard Time and is 8 hours ahead of GMT (GMT+08:00)
Session Chair(s): Xiaoge ZHANG, The Hong Kong Polytechnic University
ICRMS22-F-0100 Research on Feature Selection Method Based on Bayesian Network and Importance MeasuresVIEW ABSTRACT
With the wide application of machine learning algorithms in various fields, feature selection becomes more and more important as a data preprocessing method which can not only solve the problem of dimension disaster, but also improve the generalization ability of algorithms. Based on this, the main work of this paper is as follows. Firstly, the importance measures and Bayesian network were combined to solve the problem that Bayesian network could not rank the importance of features. At the same time, a recursive feature elimination algorithm based on importance degree theory is proposed with importance degree as the screening index. Finally, the prognostic model of gallbladder cancer was established, which shows that the proposed algorithm has good performance.
ICRMS22-A-0045 Achieving High Accuracy with Few Run-to-failure Data for Deep Learning Based RUL Prediction: A Bayesian Deep Active Learning MethodVIEW ABSTRACT
Deep learning (DL) has been studied extensively for remaining useful life (RUL) prediction in the recent decade. Although with high precision and generalization ability, DL methods are greedy for run-to-failure data to guarantee their performance, and obtaining run-to-failure data is fairly expensive in many industrial applications. How to economically achieve high accuracy with few run-to-failure data becomes a critical and emergent issue. To this end, a Bayesian deep-active-learning-based method is proposed for RUL prediction, which goes beyond traditional passive learning and introduces a novel active learning perspective. A general framework equipped with Bayesian deep learning and active learning methods is constructed. Bayesian neural networks with Monte Carlo dropout inference are presented to predict RUL with uncertainty quantification for samples without run-to-failure labels. This prediction uncertainty is further used to develop an acquisition function for actively selecting target samples to obtain their run-to-failure labels. A recursive model training and active data selection mechanism is then constructed and leveraged to realize the proposed method for maintaining accuracy while reducing training set size. Two practical cases are presented, and experiment results demonstrate that 20 and 40 percentage run-to-failure data can be saved for the bearing and the battery RUL prediction, respectively.
ICRMS22-F-0115 AtFP: Attention-based Failure Predictor for Extreme-scale ComputingVIEW ABSTRACT
Extreme-scale computing is paving the way for unparalleled advances in scientific discovery and innovation. However, as systems scale, their propensity to failure increases significantly, making it difficult for long running applications that span a large number of computing nodes to make forward progress. Achieving high performance in extreme scale environments, while minimizing energy consumption, has emerged as a daunting challenge. Significant advances on how to deal with failure, both physical and logical, have been achieved, with varying degree of success. A key component of fault tolerance relies heavily on the ability of the scheme to predict failure accurately. Varies approaches, including intelligent methods, have been proposed to predict failures. In this paper, we propose an attention-based failure predictor (AtFP), which automatically extracts representative features from the raw event log data to predict failure. The results show that, using the same input and output layers, AtFP outperforms frequently used LSTM methods. The proposed model reduces the F1 score by 39% and the training time by 65%.
ICRMS22-F-0012 Deployable Lightweight ANN-Based Approach for Wind Turbine Fault DetectionVIEW ABSTRACT
Wind power is clean and renewable energy, which occupies an important position in the world’s energy. This paper proposes a multi-regional fault detection method for wind turbines, which makes full use of multiple sensor data. Specifically, a voting-based Artificial Neural Network (ANN) is constructed to achieve 96.5% detection accuracy, which is a light-weighted model with high operating efficiency. The robustness of this model is confirmed by various numerical experiments as well. Meanwhile, the detailed experiments of several other benchmark methods illustrate that our method has far higher accuracy. Furthermore, we complete the improvement of the model referring to grid search and the accuracy of the model is enhanced to 97.5%.
ICRMS22-F-0119 On-going Reliability Test Program and Test DesignVIEW ABSTRACT
On-going Reliability Test is one of the most important reliability activities in mass production phase to detect a possible out-of-control process that negatively impacts product reliability. ORT program definition, objectives, and benefits are discussed. The ORT test design method based on the operating characteristic curve, the controlled reliability, and the statistical risk is presented and discussed. Trigger limit for concluding an out-of-control process is defined and calculated with an illustration example.
Session Chair(s): Aibo ZHANG, City University of Hong Kong, Zhong LU, Nanjing University of Aeronautics and Astronautics
ICRMS22-F-0072 Reliability Analysis for Aero Engine Gear Considering Dependence of Multiple Failure Modes Based on CopulaVIEW ABSTRACT
A transmission gear commonly has multiple dependent failure modes. The reliability analysis methods that ignore the dependence effect often have substantial errors. This paper constructs the limit state functions of failure modes based on the stress-strength interference theory, characterizes the dependence among multiple failure modes with the help of Copula theory, transforms the multi-dimensional dependence problem into several two-dimensional Copulas based on Vine Copula, and uses square Euclidean distance to identify the optimal function. Using this method, a D-Vine Copula model is established for the spur gear in an aero engine accessory transmission system to carry out the gear reliability analysis considering the dependence of multiple failure modes, which demonstrates the rationality and effectiveness of the proposed method.
ICRMS22-F-0110 Failure Mode, Mechanism, and Effect Analysis of Visible Light Camera in an Optoelectronic PodVIEW ABSTRACT
Common photoelectric pods often have composite photoelectric sensing functions, such as visible light-sensing components, infrared night vision components, distance sensing components, etc. Among them, visible light sensing component is the most basic and widely used image sensing component. With the rapid development of modern optoelectronic technology, the reliability level of a visible light camera has become to be one of the most critical technical requirements in an optoelectronic pod, which greatly determine the success of a mission. For the entire equipment, the optoelectronic pod is like the eye of human beings. Once it fails, it may bring a devastating threat to the safety of the entire equipment and the performance of one task. This paper focuses on the reliability analysis of the visible light camera used in the optoelectronic pod. Through the analysis of its internal structure and performance indicators, the key and weak design links are obtained, which can provide important theoretical support for the reliability design of optoelectronics.
ICRMS22-F-0113 Utilisation of Fuzzy Fault Tree Analysis for Risk Assessment of Offshore Wind Farm ConstructionVIEW ABSTRACT
To prevent accidents in the construction stage of an offshore wind farm (OWF), this paper proposes the fuzzy fault tree analysis method to evaluate the safety of OWF. The kernel of the proposed method is to identify the risk factors leading to accidents in the construction stage of OWF using historic accident data and expert investigation, so as to establish the OWF fault tree in the construction stage. The basic event probability obtained by using fuzzy numbers instead of incomplete data is used to iteratively solve the risk level of the top event, and the Fussell-Vesely importance (FV-I) is employed to evaluate the contribution of each basic event. The result demonstrates that the fuzzy fault tree in the construction stage of OWF cannot only identify the risk factors but also give comprehensive evaluation opinions.
ICRMS22-A-0048 Application of Systems-theoretic Process Analysis to the Compressed Air Energy Storage SystemVIEW ABSTRACT
The development and utilization of renewable energy have been attracting increasing attention as a solution to the worldwide fossil energy crisis and environmental pollution issues. While the randomness and instability of renewable energies, such as wind and solar power, raise new challenges in the operation phase. Compressed air energy storage system (CAES) is proving technology to store electricity by compressing air to store energy and releasing it to produce electricity during the demand period. Due to the complexity of the system structure and operation process, how to keep the CAES operating safely and efficiently needs to be studied. As a relatively new technology, there are scarce studies regarding the safety and risk issues of CAES. Thus, it is expected to obtain more relevant information by leveraging STPA, which is a top-down method and based on the functional control diagram of the system. STPA method has been demonstrated with advantage in situations where there are many ‘unknown-unknowns’, or difficulties in the prediction of hazardous situations before they happen. The main purpose of this study is to apply the STPA to CAES to identify hazards and provide clues for practitioners to conduct the risk assessment.
ICRMS22-F-0065 A Research of a Modified Bearing Fault Diagnosis Model Based on Deformable Atrous Convolution and Squeeze-and-Excitation AggregationVIEW ABSTRACT
In order to accurately and intelligently identify the fault signal characteristics of bearing in modern rotating machinery industries, a comprehensive novel fault diagnosis model is introduced by using a few data points as input signals. In this paper, the main methods of the model are processing the two signal inputs, frequency-domain signal (FDS) and time-frequency graph (TFG), into a network built up with convolution (Conv) and deformable atrous Conv. Then, this network extracts their features respectively. After squeezing-and-excitation aggregation on the features, three types of outputs are obtained for tasks of faulty bearing position detection, fault type diagnosis, and damage size estimation. This method allows the fault diagnosis of different locations, fault types, and fault degrees to be completed simultaneously, and the state-of-Art effect can be achieved. Two bearing vibration signal datasets are used in the paper to evaluate the performance of the model, and experimental results prove an effective multi-task fault diagnosis ability on the three tasks.
Session Chair(s): Samira KEIVANPOUR, Polytechnique Montreal, Ibrahim KUCUKKOC, Balikesir University
ICRMS22-A-0026 Robot Reliability Path Optimization MethodVIEW ABSTRACT
A safe and smooth operating path is a prerequisite for mobile robots to accomplish tasks. Thus, a robot reliability path optimization method is proposed for fast and effective optimizing of path for mobile robots to travel in workspace, which combines neural network, genetic algorithm, reliability evaluation, and Bézier curve skillfully. Firstly, the data set of control points on the Bézier curve and the offsets of the path are constructed using the Latin hypercube sampling method. A surrogate model for the control point positions and offsets of path segments are established in the framework of neural network. Secondly, based on the surrogate model, a path reliability evaluation function is formulated by comprehensively considering multiple influential factors of mobile robot tracking accuracy, path reliability and smoothness. Finally, the genetic algorithm is introduced to detect the optimal control points to realize the selection of the optimal path of reliability in different environments. The result shows that the proposed robot reliability path optimization method exhibits excellent effectiveness and applicability. It demonstrates advantages of fast path planning, desirable path smoothness and high path reliability. Also, it can ensure the safety of mobile robot working along the planned path as availed by a pre-set criterion.
ICRMS22-F-0052 Reliability Analysis for Mixture Weibull Distribution with Progressively Censored Data Based on Stochastic Expectation-maximization MethodVIEW ABSTRACT
The mixture Weibull distribution is widely used in modeling lifetimes in reliability engineering. Due to the real-time maintenance and the replacement during operating environments, the field failure data are often progressively censored. The classical parameter estimation methods are not available for the lifetime data affected by both multiple failure modes and progressively censoring. This paper proposes an improved stochastic expectation-maximization (SEM) method based on the whale optimization algorithm (WOA). The method consists of two steps. The S-step aims to generate progressively censored data corresponding to each failure data by current parameters estimates. The M-step aims to maximize the Q function formed by the expanded data to obtain new parameters estimates. The WOA is used to optimize the Q function for optimal parameters estimates instead of the complex analytical process of maximization. A numerical example and a real-world dataset are carried out. The results demonstrate the accuracy and applicability of the proposed method in parameter estimations and data fitting.
ICRMS22-F-0026 GWO-LSSVM-based Strain Prediction and Importance Analysis for Full-scale Static Testing of Composite Wind Turbine BladesVIEW ABSTRACT
The strain measuring data in wind turbine blade (WTB) full-scale static testing is the basic for WTB mechanical property analysis, reliability assessment, design optimization, etc.; thus, accurate and sufficient strain data is essential. However, in the WTB full-scale static performance testing, the numbers of strain measuring positions are limited, thus the strain data for further analysis is insufficient. Since the strain response of the blade has a strong correlation with the applied load, measured positions, and is also directly influenced by the geometric non-linear of the blade structure, traditional numerical analysis methods based on physical simulation and mathematical models are difficult to meet the needs of accurate strain acquisition. Considering the significant advantages of Grey Wolf Optimizer-Least Squares Support Vector Machine (GWO-LSSVM) in dealing with multi-input parameters, non-linear fitting, etc., In this paper, a strain prediction method based on GWO-LSSVM for full-scale static testing of WTB is proposed in combination with full-scale static test data of a certain type of WTB. The accuracy and effectiveness of the proposed method are compared with traditional Least Squares Support Vector Machine (LSSVM) and Back Propagation Neural Network (BPNN). The study can provide more information for WTB reliability assessment and lifetime prediction.
ICRMS22-F-0021 Intelligent PHM Based Condition Monitoring in Nuclear Energy SystemsVIEW ABSTRACT
As the critical components age as time goes on, prognostics and health management (PHM) technology have been focused on nuclear energy systems. In this procedure, condition monitoring is vital for determining the turning point (TP) and ensuring an accurate assessment of system operations. This paper proposes a novel condition monitoring method combined with a deep time series-based feature analysis. Firstly, cross-correlation analysis is adopted to make a selection of the transient features related to the key metric. Second, the Piecewise Aggregate Approximation (PAA) is applied to perform further filtering on a time scale. Third, the deep convolutional neural network identifies the TP of a filtered feature in different accidents. Finally, the procedure is integrated with the designed intelligent PHM framework. The experiment results show the superiority of the proposed method. Besides, it has better model integration ability with the developed diagnosis and prediction module.
Session Chair(s): Ping JIANG, National University of Defense Technology, Xiaoge ZHANG, The Hong Kong Polytechnic University
ICRMS22-F-0082 Reliability Modeling of Bi-directional Circular Multi-State Sliding Window System for Sequential TasksVIEW ABSTRACT
This paper investigates a novel circular multi-state system model with bi-directional sliding window for sequential tasks. The multi-state elements (MEs) in our proposed system are mutually independent and circularly deployed. The functionality of our system depends on its ability to complete the pre-determined sequential tasks from any possible starting point with consecutive MEs in either clockwise or counterclockwise directions. Then, we employ the universal generating function technique in the model description of the multiple states of the system and its MEs and the system reliability evaluation. Finally, analytical examples are provided for demonstration and illustration.
ICRMS22-F-0043 Study on Standby Hydraulic of Single-heat-source Single-looped District Heating NetworkVIEW ABSTRACT
Since the concept of reliability was put forward, how to improve the reliability of district heating system to effectively save energy and ensure economic benefits has become a research hotspot. By analyzing the hydraulic performance of district heating network and the possible flow in each pipe under normal and failure scenarios, we proposed the standby hydraulic design framework. Then, the topology and standby hydraulic design mathematical models of typical single-heat-source single-looped district heating network were established. The variation of hydraulic performance was studied by changing the determination method of design flow, the number, and distribution of users on the mainline. Finally, several design rules which can effectively alleviate the hydraulic imbalance under failure scenarios were proposed. The results show that using the method of weighted summation of the possible flow and its probability in each pipe under various operation scenarios to design pipe diameter has the advantages of high hydraulic performance without greatly increasing investment.
ICRMS22-F-0050 A Specification for Phased-mission System with Redundancy Based on SysMLVIEW ABSTRACT
In engineering practice, phased mission systems (PMSs) may have phase redundancies. A failed task can be executed not only in the current phase, but also in its redundancy phases. Therefore, different from general PMSs, a PMS with phase redundancy (PMS-PR) has several mission execution sequences. Considering the complexity of the system, it is easy to make mistakes by manual modelling. Therefore, a certain tool is needed to support the modeling process. This paper proposes a specification model of PMS-PR based on system modeling language (SysML). Finally, a simplified PMS-PR is illustrated as an example to show the modeling process of the specification model.
ICRMS22-F-0051 A Multi-objective Optimization Based Safety Requirement Assignment for Aircraft Systems by Using NSGA-IIVIEW ABSTRACT
The assignment of safety requirements is an important task in the design and evaluation of complex aircraft systems. The safety assignment of development assurance level and failure probability for the items/functions can minimize the possibility of errors in the development process. The development assurance levels of the items/functions consisting of the system are taken as decision variables, the assignment principle of development assurance levels and the probability requirement of the top failure conditions are taken as constraints, and the minimization of the development cost and system weight is taken as the optimizing objective, the multi-objective safety requirement assignment model was established. Taking the vector composed of the development assurance levels of all items/functions as the individual chromosome, a method of solving the model based on non-dominated sorting genetic algorithm (NSGA-II) is proposed. Finally, an application instance is given based on a certain fly-by-wire system. The results show that the proposed method can reduce the dependence on designers' experiences or skills effectively.
Session Chair(s): Xiujie ZHAO, Tianjin University
ICRMS22-F-0111 A Selection Method of Effective Metamorphic RelationsVIEW ABSTRACT
Metamorphic testing can effectively alleviate the test oracle problem. However, the program under test needs to be executed one more time; it is particularly significant to select the metamorphic relation with the highly fault-detecting ability and improve the test cost-effectiveness. Due to the limitations of existing evaluation methods in engineering applications, a validity evaluation method for metamorphic relation is presented based on single factor and multi-factor mutation scores. It can guide their priority ranking and selection. Experiments show that this method can significantly reduce test costs.
ICRMS22-F-0117 A Dynamic Recognition Method of Metamorphic Relation IdentificationVIEW ABSTRACT
Metamorphic testing is one of the effective methods to alleviate the test oracle problem. Metamorphic relation is the core of metamorphic testing, and there is no effective automatic identification technology. This paper transforms the metamorphic relation recognition issue into a symbolic expression regression problem. The test input pairs are generated using the preset input pattern, and the output pattern is mined by gene expression programming. After reduction and verification, a valid metamorphic relation is obtained. Experiments show that this method can identify the relation obtained by static analysis and a more significant number. At the same time, compared with existing technologies, it has apparent advantages in effectiveness, reliability, and performance.
ICRMS22-F-0118 A Systematic Method for Identifying Safety-related Faults in Formal Specifications Using FTAVIEW ABSTRACT
The potential hazard in the formal specification of safety-critical systems is likely to cause the failure of the corresponding system that may lead to a catastrophic disaster. How to accurately identify the hazard-related faults in software is still a difficult problem. In this paper, we propose a systematic method for detecting potential hazards in formal specifications using fault tree analysis. Using this approach to a given formal specification, a fault tree will be constructed based on the structure of the specification. We discuss the rules for constructing fault tree analysis that are established based on various structures of specifications. A case study is conducted to demonstrate how the proposed approach works in practice.
Session Chair(s): Wenjie DONG, Nanjing University of Aeronautics and Astronautics, Shaomin WU, University of Kent
ICRMS22-A-0030 Condition-based Maintenance Policy for Systems Under Dynamic EnvironmentVIEW ABSTRACT
System degradation characteristics usually vary dramatically with environment, which brings challenges to the maintenance optimization upon such systems. This paper focuses on repairable systems whose degradation is modeled by a Wiener process that allows for the influence of dynamic environment. We investigate a condition-based maintenance (CBM) model aiming at minimizing the expected cost accordingly. Firstly, the evolution of environment is characterized by a Markov process and is accounted a covariate of the drift parameter in Wiener process. The choice between a corrective and preventive replacement rests on the periodic checks on the system and environment. Then, the CBM model is formulated in the framework of Markov decision process (MDP), meanwhile, the backward algorithm and value iteration algorithm are employed to gain the optimal maintenance policies over the finite and the infinite planning horizons respectively. Finally, we elucidate the proposed model in concrete terms along with a comprehensive sensitivity analysis.
ICRMS22-A-0031 Availability of a Load-sharing System Under the Heterogeneous Component SuppliersVIEW ABSTRACT
In a load sharing system, components are dependent with each other, because the load of a surviving component increases at the failure time of another component. Previous research has developed the reliability expression based on either the tampered failure rate model or the accelerated failure time model, without considering the repair process. In this presentation, we consider a load-sharing system composed of heterogeneous components that are partitioned into repairable modules. Then we provide a methodology to compute the system availability under the assumption that the component lifetimes follow exponential distributions whose parameters depend on the operating load, and the module repair times follow inverse Gaussian distributions. Using the flowgraph model, we derive the moment generating function of the entire system from which the system availability is obtained. A numerical example is presented to obtain the optimal system design for maximizing the system availability by comparing different system structures.
ICRMS22-F-0009 Robust Redundancy Allocation by the Min-max Regret Framework and Sequential Two-stage ApproachVIEW ABSTRACT
To effectively and quickly solve the robust redundancy allocation problem under the Min-Max regret framework, a sequential two-stage approach is proposed. The proposed metaheuristic approach includes two stages: the first stage is to solve an optimization problem with regard to various scenarios, and the second stage is to search for the optimal solution of decision variables. These two stages are executed sequentially and iteratively until the preset termination condition is satisfied. The listed example demonstrates the performance of the proposed approach.
Session Chair(s): Dequan ZHANG, Hebei University of Technology, Yiliu LIU, Norwegian University of Science and Technology
ICRMS22-A-0007 A Mixture Distribution for Evaluating the EVD of Engineering Structures Under Stochastic Seismic ExcitationsVIEW ABSTRACT
Evaluating the first-passage failure probability of nonlinear engineering structures under seismic excitations efficiently and accurately is a challenging problem. The first-passage failure probability can be estimated through the extreme value distribution (EVD) of the structures. The moment method is usually employed to construct the EVD by fitting the parameters of selected distribution model with statistical moments as constrains. However, it is hard to capture the real EVD of structures due to the estimation errors of high-order statistical moments and the inappropriate selections of distribution models. Hence, in this paper, a mixture distribution by combing the Gumbel distribution and inverse Gaussian distribution is proposed to reconstruct the EVD of structures and fractional moments are used for estimating unknown parameters of mixture distribution. The proposed mixture distribution is more flexible than a single distribution. A 3D reinforced concrete frame under stochastic seismic excitations are investigated to verify the efficiency and accuracy of the proposed mixture distribution.
ICRMS22-A-0036 Reliability Analysis and Error Compensation Method of Industrial Robot Positioning Accuracy Based on RBF Neural NetworkVIEW ABSTRACT
The positioning accuracy and high reliability are the key performance indicators of industrial robots to ensure its long-term stable operation in service. The actual motion position of the robotic end-effector may deviate from the target position due to the existence of uncertainties, resulting in the degradation of positioning accuracy of industrial robots. In this study, an efficient reliability analysis method for positioning accuracy of industrial robots is developed using the radial basis function (RBF) neural network. A hybrid learning algorithm for training the RBF neural network is proposed to construct the mapping relationship between the uncertain parameters and the position coordinates of the end-effector. Then, the positioning accuracy reliability analysis of industrial robots is performed in terms of single coordinate direction and single point through the improved RBF neural network. Following that, a spatial grid-based error field model is constructed to predict the actual position of end-effector in the workspace. A corresponding error compensation strategy is proposed to improve the positioning accuracy of industrial robots. Finally, the effectiveness and engineering practicability of the proposed reliability analysis and error compensation technique are verified through simulation-based and experimental methods.
ICRMS22-A-0059 Reliability of Risers of a Complex Offshore System Related to the Risk of Leakage Based on Fatigue Damage Assessment with Wave ScatteringVIEW ABSTRACT
The long-term reliability of the flowlines of an oil or gas platforms, FPSO, subject to the risk of leakage can be evaluated based on fatigue damage using wave-scatter. The study follows the discovery of large reserves of natural gas and the ongoing installation of floating liquefied natural gas in Area-4 of the Rovuma basin, Mozambique. The Mozambican coast has been the scene of extreme events, tropical storms and cyclic cyclones for over 10 years. The Rovuma basin characterized by sea currents caused by three large anticyclonic eddies. These environmental events can deteriorate risers at a higher rate, compromising system reliability and their availability. The fatigue analysis procedure is deterministic in essence. The risers long-term fatigue analysis is performed in time-domain using Monte Carlo simulation combined to lognormal distribution. Evaluating the limit state function aims to compute the probability of failure. The failure probability is used for the ultimate purpose of estimating the reliability of the production riser subsystem.
Session Chair(s): Weiwen PENG, Sun Yat-sen University, Zhiqiang CAI, Northwestern Polytechnical University
ICRMS22-A-0057 Early Prediction of Battery Lifetime Based on Statistical Health FeaturesVIEW ABSTRACT
The early prediction of lithium-ion battery lifetime exhibits timely feedback, short time-consuming and low cost. It is also crucial for safety and reliability of new energy vehicle. However, due to the nonlinear degradation path of battery capacity with a trivial variation in the early stage, it’s difficult to predict the lifetime with early degradation data. In this paper, we first analyze the change of voltage and temperature data during the cycle of charge and discharge and then design a series of health indicators. After using Box-Cox transformation to enhance their linear correlation with battery lifetime, we select two health indictors related to temperature and voltage with high correlation coefficients of -0.73 and - 0.90 respectively. After that, we apply Gaussian Process Regression on MIT data set, which consists of 124 commercial Lithium-ion batteries. Using only the first 100 cycles of data to predict lifetime, we get good result with a root-of-mean-square-error of 88.76 cycles, an average-absolute-error of 8.23% and a R2 of 0.90. The results verify the effectiveness and superiority of the method adopted in this paper.
ICRMS22-F-0114 Degradation Prognostics of Lithium-ion Batteries Based on Partial Features and Long Short-term Memory NetworkVIEW ABSTRACT
The accurate degradation prediction of Lithium-ion batteries is beneficial to the reliability and safety of battery-driven systems. In this paper, a long short-term memory network (LSTM) model is utilized to predict the capacity degradation trend using partial charge and discharge features of Lithium-ion batteries. Firstly, significant features are extracted from the original charge and discharge data. Then the Pearson correlation coefficient is adopted to filter the features with high correlation coefficients. Selected features are subsequently treated as the input of the prediction model. Finally, a LSTM model is developed and associated hyperparameters are established by Adam algorithm. The proposed method is validated by experimental results on the NASA battery dataset.
ICRMS22-F-0101 A Remaining Useful Life Prediction Framework for Aero-engine Using Information Entropy-based Criterion and PCA-RVMVIEW ABSTRACT
To deal with the challenge of feature selection and extraction in the remaining useful life (RUL) prediction for aero-engines, this paper proposes a framework using multi-sensors data, which involves three key components (i) an information entropy-based criterion for sensor selection, (ii) principal component analysis (PCA) for the construction of synthesized health index, and (iii) relevance vector machine (RVM)-based RUL prediction. The proposed method combines the PCA with RVM and improves the prediction accuracy by employing a novel entropy-based criterion for sensor selection. The effectiveness of this approach is demonstrated and validated with the turbofan engine released by NASA Research Center.
ICRMS22-F-0087 Dictionary Based Three-phase Driver MOSFET Cascading Fault DiagnosisVIEW ABSTRACT
MOSFET is widely used in motor driver because of its high switching frequency, but it often fails for many reasons which may results in driver faults. This paper presents a diagnosis method based on the output characteristic dictionary for open/short circuit mixed cascading faults of the three-phase driver. The small signal model is used to quantitatively analyze the fault influence, and a multi-value fault dictionary is constructed based on the time-frequency domain characteristics of online measurable signals such as output current and output voltage. Based on the numerical simulation method, the typical single faults and cascading faults are simulated and injected, which proves the effectiveness of this developed method.
ICRMS22-F-0030 Research on Circuit Fault Diagnosis Method Based on Multi-feature Information FusionVIEW ABSTRACT
Aiming at the shortcomings of low accuracy and efficiency of analog circuit diagnosis, a fault diagnosis method for analog circuits based on multi-feature information fusion is proposed. The method firstly combines statistical characteristics and Ensemble Empirical Mode Decomposition to extract the circuit fault features, and through the Principal Component Analysis method for dimensionality reduction, which provides a data preprocessing method for subsequent fault diagnosis. On this basis, in order to realize the accurate identification and classification of different fault modes, the information fusion method combining the Extreme Learning Machine and D-S evidence theory are introduced, and the ELM-D-S fault diagnosis method is proposed, which aims at fusing and diagnosing the fault signals of analog circuits from two aspects of current and voltage, so as to improve the accuracy of the fault diagnosis of the analog circuit. The simulation experiment results show that the information fusion method has higher diagnosis accuracy than the single information method, which shows the effectiveness and feasibility of the proposed method.
Session Chair(s): Man Ho LING, The Education University of Hong Kong, Niladri CHAKRABORTY, University of the Free State
ICRMS22-F-0094 Battery Remaining Useful Life Prediction Based on a Combination of ARMA and Degradation ModelVIEW ABSTRACT
The remaining useful life (RUL) of batteries is an important and helpful reference for battery management system. Since autoregressive moving average (ARMA) model is a relatively mature time series analysis method for prognostics, the long-term prediction results are not reliable due to dynamic noise and constantly cumulative system errors. In order to improve the accuracy of long-term prediction for battery RUL, a method combining ARMA and exponential degradation model is proposed in this paper. A case study using battery dataset from CALCE is performed to demonstrate the effectiveness of the proposed method, and the results show that the proposed method gives better prediction accuracy.
ICRMS22-F-0015 A Mode-based Approach for Lognormal Parameter Estimation on Heavily Censored DataVIEW ABSTRACT
The density function of the lognormal distribution is unimodal and its mode is always smaller than its median life. For the type-I censoring test, if the censoring time is not larger than the median life, the censoring degree of the dataset obtained in this way will be larger than 50% on average, implying that the dataset is heavily censored. In this case, the classical parameter estimation methods generally cannot provide stable estimates, but a relatively accurate estimate of the mode can be obtained. According to this argument, this paper proposes a mode-based approach for estimating the parameters of the lognormal distribution on heavily censored data. The proposed approach first uses the midpoint Kaplan-Meier estimator to augment the data; then uses the lognormal Q-Q plot to estimate the mode of the density function, from which the scale parameter can be expressed as a function of the shape parameter; and finally uses a single-parameter maximum likelihood method to estimate the shape parameter. Six datasets are analyzed to illustrate the proposed approach and its appropriateness.
ICRMS22-F-0007 Dynamics Modeling and Simulation of Multi-rotor UAV based on the Composite Wind Field ModelVIEW ABSTRACT
To study the influence of compound wind field disturbance on the stability of multi-rotor UAV(unmanned aerial vehicle), a dynamics modeling and simulation method for multi-rotor UAV under the compound wind field model is proposed. Firstly, based on the idea of equivalent modeling in practical engineering, the models of uniform wind, gust, turbulence and wind shear are constructed respectively. The numerical analysis and simulation research are then carried out. Second, the dynamic model of multi rotor UAV under the action of composite wind field is deduced, and the attitude control model is derived. Finally, the numerical analysis software is utilized to simulate and analyze the attitude response of the multi-rotor UAV under the condition of wind and no wind, and analyze its control performance under the interference of compound wind field. The simulation results show that the dynamic model of the multi-rotor UAV under the influence of the wind field can accurately reflect its dynamic performance under the action of the wind field, and the output attitude angle fluctuation is less than 2°. The obtained results show that the established wind field model can be effectively applied to the research on the flight performance of multi-rotor UAV under the interference of compound wind field.
ICRMS22-F-0010 Degradation Mechanism and Failure Analysis of Planar Type SiC MOSFETs Under Cyclic Stress of Surge CurrentVIEW ABSTRACT
The degradation behavior of planar silicon carbide metal oxide semiconductor field effect transistors (SiC MOSFETs) in terms of electrical characteristics is explored in this work under cyclic stress of surge current. Before and after surge current stress, planar-type SiC MOSFET is subjected to scanning electron microscope (SEM)-based failure analysis. During the surge current cyclic stress investigations, the accumulation of oxide trap charge and interface trap density leads to a drop in the threshold voltage(Vth) under surge current stress, according to the experimental data. The on-resistance(Rds(on)) displays the opposite tendency to the Vth during surge current cycling. The fact that it is a degradation phenomenon induced by the combined effect of Vth and package degradation caused by repetitive cycle thermal stress of the device is confirmed by testing. According to SEM studies, the major reason of SiC MOSFET failure following surge current cycling stress is that the surge current causes the Aluminum (Al) at the source terminal to melt and subsequently enter into the gate terminal, resulting in a gate-source short circuit and device failure.
ICRMS22-F-0092 A New Method for Deriving Reliability Qualification Test PlansVIEW ABSTRACT
A reliability qualification test (RQT) is used to determine whether a product meets pre-specified reliability requirements. For systems with high reliability, traditional RQTs are no longer preferred due to the test plans often require long test durations or pose high risks to both producers and consumers. To cope with this problem, this paper proposes a new method for RQT plan derivation that makes use of subsystem test data to derive system test plans, when both system and subsystems are following exponential distributions. Compared with conventional RQT plans that construct system test based on which decisions are made, the proposed method can derive a system test plan with much shorter test durations while keeping producer and consumer risks under control. The case study shows the validity of our proposed method.
Session Chair(s): Mimi ZHANG, Trinity College Dublin, Yiliu LIU, Norwegian University of Science and Technology
ICRMS22-A-0052 Fault Diagnosis of Mechanical Systems with Unlabeled Data Based on DBSCAN and Knowledge TransferVIEW ABSTRACT
The huge amount of operational data collected from mechanical systems has offered new opportunities and challenges in fault diagnosis. Traditional fault diagnosis methods assume different types of faults that happened in the system and the fault labels are given in advance. However, for machines that are new or just have been put into use, some faults may be unseen. Furthermore, the fault labels of each mechanical system are not easy to obtain in practice. To solve these problems, a novel fault diagnosis framework is proposed for a group of mechanical systems with unlabeled operational data. Firstly, for each mechanical system in the group, a distance-based clustering is applied to obtain the fault labels. Then, a summary of different fault types of this group of mechanical system is created. Finally, transfer learning techniques are used to transfer the knowledge of different types of faults into each mechanical system. In this way, the mechanical system can recognize unseen faults. The proposed framework is tested on a group of similar escalators with unlabeled vibration data from a metro corporation. Results demonstrate that the proposed method can generate data labels and achieve a good performance in fault diagnosis for unseen faults.
ICRMS22-A-0027 Reliability Modeling and B10 Life Analysis for Transmission Gear Based on Strength DegradationVIEW ABSTRACT
Aiming at the problem that the transmission gear of automobile transmission fails due to fatigue damage in the working stage, the working characteristics of the transmission gear are analyzed, and its reliability modeling and B10 life analysis method are studied. The fatigue failure of gears is caused by the interaction of strength degradation and accumulated damage, and the strength decreases with the change of the load and the increase of the number of stress cycles. Taking the working time of the transmission gear as the reliability measurement index, the influence of the accumulated damage on the strength degradation is analyzed, and a reliability model of the transmission gear considering the strength degradation is established on the basis of the stress-strength interference model, and its B10 reliability life calculation model is further established. Through the analysis and comparison of experimental data, the accuracy of the proposed reliability model is verified, and its B10 life can be calculated by using parameters such as stress and strength of the transmission gear, which can better guide the design, test, and use of the transmission gear.
ICRMS22-F-0096 A Knowledge Graph-based Link Prediction for Interpretable Maintenance Planning in Complex EquipmentVIEW ABSTRACT
Maintenance planning is a significant part of predictive maintenance, which involves task planning, resource scheduling, and prevention. Many data points will be collected during the monitoring and maintenance of sophisticated equipment thanks to the large-scale sensor systems installed in contemporary factories. As a result, with the help of collected maintenance data, maintenance plans may be more detailed and timelier. A knowledge graph (KG) has recently been proposed to manage massive and unorganized maintenance data semantically, enhancing data usage. Despite the fact that previous research had utilized KG for maintenance planning, they had only used semantic searching or graph structure-based algorithms and had not included the prediction of new links. To fill this gap, a maintenance-oriented KG is established firstly based on the well-defined ontology schema and accumulated maintenance data. Then, an Attention-Based Compressed Relational Graph Convolutional Network is proposed to find the potential solutions and explain the fault, specifically for the heterogeneous and sparse graph structure of maintenance-orient KG. A maintenance case of oil drilling equipment is carried out, which compares the proposed model with other cutting-edge models to demonstrate its effectiveness in link prediction.
ICRMS22-F-0067 Mission-oriented Prioritization Method for Health Management Objects of Complex EquipmentVIEW ABSTRACT
Under the conditions of systematic warfare, the diversified mission scenarios and the structure of complex equipment have led to problems such as difficulties in clarifying equipment health management objects and in implementing the allocation of support resources. To solve those problems under the influence of different mission profiles, this paper proposes a mission-oriented prioritization method for health management objects of complex equipment. Firstly, using the graph theory, function-dependent network models are constructed according to the composition of equipment system and mission sessions under the mission profile; Secondly, the PageRank algorithm is used to measure the importance of each system; Finally, combined with the characteristics of the mission profile, the priority list is derived by updating the PR values with expert knowledge, by which the key health management objects are determined. The proposed method is verified with a case study on the design of an equipment health management system. The results show that the method can obtain the priority of each sub-system of equipment efficiently, so as to provide appropriate objects for equipment health management and the distribution of support resources.
ICRMS22-A-0060 Image-based Health Monitoring Scheme for Electric-vehicle BatteryVIEW ABSTRACT
In recent years, the number of fire accidents caused by electric-vehicle battery has increased. The majority of the cases includes an abnormal operation due to high-speed charging. If the temperature of battery modules exceeds the proper range during the charge and discharge cycle, an exothermic phenomenon occurs which leads to thermal runaway as well as the degradation of battery modules. In this paper, we propose an image-based profile monitoring procedure for electric-vehicle battery. Using the thermal image streams measuring the charge-discharge cycling patterns, the propogation modeling in terms of time and space is conducted. Based on the spatio-temporal modeling result, dynamic time warping (DTW)-based distance is used to diagnose if the state of each image is in-control status. According to the application of thermal images over charge-discharge cycles, the proposed scheme can provide the real-time monitoring results by detecting the abnormal state during the operation.
Session Chair(s): Fakhri I. ALIFIN, Bandung Institute of Technology
ICRMS22-A-0015 Cost of Customized Extended Warranty with Flexible Purchase Date Via Bayesian ApproachVIEW ABSTRACT
Customized extended warranty with flexible purchase date can eliminate the traditional extended warranty that can only be purchased at the base warranty expires or with the product sold. Censoring of observational data due to flexible purchase date poses difffficulties for reliability estimation, which brings great risk for service providers on extended warranty cost estimation problem. In this paper we deduce the posterior distribution of the parameter in reliability model throughout bayesian approach incorporating non-informative prior and informative prior in accordance with the field failure data. As for whether and when to adopt PM, three customized extended warranty scenarios are presented and the corresponding analytical forms of cost and its confidence interval are given based on the update parameter. Numerical example results show that the implementation of posterior analysis can better reflect the reliability and extended warranty cost for a single product, and also reveal that the early intervention-type extended warranty behavior has more advantages in saving costs. By adopting the proposed customized extended warranty approach, customers with different performance of product could have satisfactory on extended warranty strategy of their product, and product providers also be beneficial from cost saving and competitive marketing strategy.
ICRMS22-F-0075 Warranty Cost Models for Products Protected by Lemon Laws Considering Mutual Failure InteractionVIEW ABSTRACT
In this study, we develop warranty cost models of a warranted product that is also under the protection of the lemon laws. The product is viewed as a multi-component system consisting of critical and non-critical components. The product is declared a lemon if the recurrent failures on either critical or non-critical components occur – i.e., the number of the failures reaches the threshold value, k (e.g., four failures). We consider that there is a mutual failure interaction (or two-way interaction) between critical and non-critical components. Hence, the number of failures of each component type is the sum of natural and induced failures. Whenever a lemon is declared, the manufacturer provides either (i) a refund or (ii) a replacement to the consumer. We obtain the warranty cost models for the two cases - (i) refund and (ii) replacement then find the optimal lemon law period that minimizes the expected warranty cost rate. The numerical examples are provided to illustrate the expected warranty cost and the optimal lemon law period.
ICRMS22-F-0103 Automobile MTBF Evaluation Method Based on Two-dimensional Warranty DataVIEW ABSTRACT
In order to accurately evaluate the reliability level of automobile based on warranty data, the MTBF calculation method considering the censored characteristics in warranty data and the influence of warranty period is proposed. Two types of censored problems in warranty data are analyzed, and a supplement method for censored mileage based on use intensity and automobile mileage distribution is constructed. Considering the influence of two-dimensional warranty period, the censored mileage is corrected and the calculation accuracy of MTBF is improved. Combined with the actual warranty data of an automobile enterprise, the accuracy and effectiveness of MTBF calculation method are verified and the proposed method can provide support for the evaluation and design of automobile reliability.
ICRMS22-F-0064 Maintenance Decision System Design Based on Equipment Support Data SystemVIEW ABSTRACT
Equipment support plays an important role in the formation of equipment combat capability and the development of war. Meanwhile, Equipment support in the complex and changeable system combat requires more precision and intelligence. Combining the concept of enterprise data assets with equipment support, this paper proposes a framework for data system based on equipment support elements. This paper proposes the constructing approach for equipment support knowledge graph. And thoroughly, different methods of designing maintenance decision system are discussed. By analyzing the values and logical relationship among of equipment design data, usage data and support data, the capability of equipment health status evaluation and maintenance decision are improved. Finally, equipment support capability and combat effectiveness are elevated.
Session Chair(s): PIAO CHEN, Delft University of Technology (TU Delft), Weiwen PENG, Sun Yat-sen University
ICRMS22-F-0029 Model and Application of Intelligent Equipment Health Monitoring and Intelligent Warning Based on the Interaction of Virtual RealityVIEW ABSTRACT
The modeling of reliability of the complex equipment has many problems, such as multi-source reliability data, extremely unbalanced data distribution, uncertain information, weak model interpretation ability, and high misjudgment rate. This paper integrates cyber and physical system, deep learning and interpretable artificial intelligence to build virtual and real integration architecture for the health monitoring of marine equipment, to convert human "knowledge" into actual model and embed into deep learning network, and then proposes a migration health diagnosis method of large marine equipment lifting system based on the interaction of virtual reality. In addition, to transform the health warning problem of marine equipment into reinforcement learning problem of the continuous interaction between intelligent warning system and marine equipment, to establish the deep reinforcement learning model of the end-to-end mapping of "health state-warning strategy" for the intelligent warning decision of marine equipment. Finally, taking the marine equipment lifting system as an example, the application method of the model proposed above is introduced and its validity is verified.
ICRMS22-F-0081 An Improved FMEA Method Based on Performance Degradation Analysis for Offshore Wind Turbine Blade Trailing EdgeVIEW ABSTRACT
There are lots of performance degradation failure modes in offshore wind turbine blade trailing edge, such as stiffness degradation, strength reduction, et al. The occurrence and severity of such failure modes can’t be easily determined, and the traditional failure mode and effect analysis (FMEA) method is difficult to be applied. Considering the fact, an improved FMEA method based on performance degradation analysis is proposed for wind turbine blade trailing edge, in which the RPNs are calculated by new methods. Specifically, a performance parameter degradation model is established based on random processes, then the occurrence of each failure mode of that is assessed; The Pearson correlation coefficient of each failure mode and each performance parameter is calculated, then the calculation method of severity is given; Here the detection method of each performance degradation failure mode is assumed to be mature, thus the detections are considered equal. This research can provide reliability analysis methods for complex systems with performance degradation failure modes.
ICRMS22-A-0032 Reliability Assessment of Systems with Continuous and Discrete Degradation Units and Unilateral DependencyVIEW ABSTRACT
In this paper, a new reliability model for systems with continuous and discrete degradation units and unilateral dependency is developed. In this new model, the system is composed of a discrete degradation unit and a continuous degradation unit in series. The state transition dependency within the multi-state unit is characterized by copula functions. Considering the unilateral dependency between two units, that is, the degradation rate of the continuous degradation unit will gradually increase as the state of another unit decreases, a multi-stage gamma process is set up to describe the degradation process of the continuous degradation unit. Additionally, a Monte Carlo method is introduced to assess the system reliability. Subsequently, the likelihood function for synchronous inspection strategy is put forward to estimate model parameters by using the maximum likelihood method. The proposed method is demonstrated by an illustrative example.
ICRMS22-A-0039 Crack Modelling and Analysis with the Bayesian Lifetime Delayed Degradation ProcessVIEW ABSTRACT
Crack is one of the key failure modes for the critical structural components of aircraft. The corresponding fatigue phenomenon is presented as the crack, composed of sequential initiation and propagation phases. From the viewpoint of reliability analysis, the traditional fracture mechanics analysis cannot fully meet the assessment of remaining useful life. The purpose of this study is to evaluate the reliability and the remaining useful life of the delayed degradation products by applying the Bayesian-LDDP model to the real-time monitoring of crack degradation data. Both destructive testing and non-destructive testing cases are covered, and the Bayesian-LDDP model is used in the analysis of two practical cases. The crack inspection data of a transport aircraft and a center fuselage longeron are implemented on the proposed model for non-destructive tests and destructive tests respectively. Moreover, in terms of calculation, the Gibbs sampling is adopted for the Bayesian estimation of parameters, which greatly improves the efficiency. In addition, the results of further inference illustrate the MTTF of cracks as well as the prediction of the RUL of each sample in the optimal model. Keywords: Crack modelling, Degradation process, Delayed degradation, Gibbs sampling, Lifetime distribution, Prior information.
ICRMS22-F-0041 A Computationally Efficient Link Criticality Ranking with Perception Error and Route Overlapping for Road Transport NetworksVIEW ABSTRACT
In real-size congested road networks, ranking links with respect to their criticalities while capturing dependence of link travel costs on user behavior and travel demand is not an easy task. The existing measures either require multiple traffic assignments (i.e., full-scan) or make strong assumptions on users’ rationality while using a single traffic assignment. In this study, we propose an improved link criticality index that not only alleviates the computational burden associated with the full-scan network efficiency measures, but also accounts for travelers’ perception error and route overlap. The proposed link criticality index is based on the resulting stochastic user equilibrium (SUE) model with a specific discrete choice model. Particularly, we adopt the path-size logit (PSL) route choice model in the SUE model, in which a path-size factor resolves the route overlapping issue by adjusting the choice probabilities for routes with strong correlations. To solve the traffic assignment, we use a faster gradient projection algorithm based on Barzilai-Borwein step size determination scheme (GP-BB). Numerical experiments are conducted to verify and demonstrate the properties of the proposed link criticality measure with the loophole network.
Session Chair(s): Yifan ZHANG, City University of Hong Kong Shenzhen Research Institute, Zhizhong TAN, Northwestern Polytechnical University
ICRMS22-F-0018 Optimal Redundancy Strategy of Defense Resource Allocation for Power Grid System Based on Game Theory AnalysisVIEW ABSTRACT
The power grid system, as one of the most critical infrastructures, is confronted with multifarious attack threats, which can result in severe damages and enormous property losses. Hence, it is important to adopt appropriate redundancy strategy of defense resources allocation in an effective measure to ensure the power grid system runs stably and reliably. For investigating the influence of different redundancy strategies and determining the optimal strategy, we propose a sequential game model, in which the homogenous formulation is introduced to describe the action and belief for both attacker and defender, and the redundancy constraints are considered in the optimal strategy and boundary conditions. Finally, we designate the redundancy condition as the variable to investigate optimal redundancy strategy for defender, and find that the defense rewards decrease at first and then increase gradually with the redundancy coefficient rising. Consequently, it is recommended that the operator of power grid system set a reasonable redundancy coefficient to allocate defense resources for acquiring the extra defense utility.
ICRMS22-F-0036 Sensitivity Analysis-based Multi-modal Transportation Network Vulnerability Assessment with Weibit Choice ModelsVIEW ABSTRACT
This study proposes utility-based accessibility measures and a sensitivity analysis-based method for assessing the multi-modal transportation network vulnerability at different spatial levels. The proposed measures analyze the network vulnerability based on the expected travel disutility stemmed from the weibit choice model, thus are inherently suitable to assess the relative performance degradation and account for heterogeneous network scales. An advanced weibit-based combined modal split and traffic assignment (CMSTA) model is adopted to consider both mode choice and route choice behaviors in multi-modal transportation networks. To address the similarity issue in both choice dimensions, the accessibility at the modal split level is derived from the nested weibit model, while accessibility at the route choice level is obtained based on the path-size weibit model. Sensitivity analysis of the weibit-based CMSTA model is conducted for vulnerability assessment and critical link identification, which can reduce the computational burden of repetitively solving the CMSTA model required by the commonly used enumeration- or scenario-based approaches. Numerical experiments based on a simplified multi-modal network are conducted to verify and demonstrate the properties of the proposed sensitivity analysis-based multi-modal transportation network vulnerability analysis.
ICRMS22-F-0045 Resilience Modeling for a Single Component System Based on Markov ProcessesVIEW ABSTRACT
Modern systems are increasingly under the threaten of disruptive events like earthquakes, floods and storms. In engineering practice, multi-state models are often used to describe the behaviors of the system exposed to disruptive events. This article develops a resilience model for a single component system with multiple states in which the evolution of the performance level over time is described by a time-homogeneous Markov process. To characterize the different dimensions of the system resilience, four types of resilience metrics are proposed to describe the resistant, absorptive, recovery, and overall resilience, where each type consists of a metric that incorporates the time dimension and a metric that does not. The theory of aggregated stochastic processes is applied to derive explicit formulas for the four types of resilience metrics. They are used in a numerical example to illustrate the applicability of the proposed method in dealing with the situation when the state dimension is huge. The results show that the developed resilience model is able to comprehensively describe the resilience of single component systems threatened by disruptive events.
ICRMS22-F-0055 Accessibility and Vulnerability Analysis of the Integrated Road, High-speed Rail and Airline System in ChinaVIEW ABSTRACT
National-scale transportation systems are critical infrastructures to ensure the normal operation of the nation and offer essential services to modern societies. And they face a constant barrage of external stresses or threats that challenge their operation. This article analyzes the accessibility and vulnerability of the Chinese road, high-speed rail and airline systems. Firstly, it models the complementary relationship among road, high-speed rail and airline systems and takes them as an integrated transportation system (IRHAS). Then, the accessibility map from (to) the center of Beijing via IRHAS is displayed. Finally, modeling the extreme storm recently occurred in Zhengzhou as a reginal disruption, the vulnerability of the IRHAS is calculated. The findings in this paper provide guidance for city administrators to plan the road, high-speed rail, and airline systems as a whole.
ICRMS22-A-0033 A Deep Reinforcement Learning Framework to Enhance Power Grids Resilience Under Natural HazardsVIEW ABSTRACT
Recently, the increased frequency of natural hazards brings power grids plenty of destructions on distribution components, e.g., towers and distribution lines. To enhance the power grids resilience and recovery the system performance, many optimization models have been proposed to obtain the planning of components post-disruption restoration. However, the typical optimization methods are difficult to be applied on large-scale power systems due to the high computational complexity, and unable to give emergency responses. This paper proposes a deep reinforcement learning (DRL) based framework to provide optimal repairing strategy for failed components under emergency disruptions. We embed an optimization-based resilience evaluation model into the deep Q-network (DQN) to reflect the rewards of actions during the restoration process. We also consider the uncertain repairing time of different components caused by varied repairing resources in the DQN. The proposed framework has been tested on IEEE 13-node system and IEEE 123-node system to showcase the effectiveness and adaptivity on large systems.
ICRMS22-A-0047 Simulated Scenario-based Reliability Assessment of Autonomous Driving SystemsVIEW ABSTRACT
Before the widely commercial use of autonomous driving vehicles, the reliability assessment of autonomous driving systems is an urgent need to address the safety-critical challenge. Though researchers have developed statistical models to predict the rates of accidents based on the prior knowledge of road testing, the models exist a large bias since the road testing miles are limited. Instead, we propose a simulated scenario-based method to assess the reliability of autonomous driving systems through exploring and generating all feasible driving scenarios. Based on the probabilistic outcomes achieved by perception module, the driving scenarios can be represented as a tuple of the state of ego car, the state of obstacles, and the state of roads. Combined with domain-specific constraints and generative models, we are able to generate valid driving scenarios with parameters of the states in specific ranges. For the safety-critical scenarios, we consider the hard braking, unsafe lane change, and collision cases to search out the corresponding sets of parameters. The reliability of the autonomous driving systems can be obtained by the statistical analysis of the safety-critical parameters among the valid range with a combination of our prior work on epistemic uncertainty and aleatory uncertainty of the perception module.
Session Chair(s): Aibo ZHANG, City University of Hong Kong, Huixing MENG, Beijing Institute of Technology
ICRMS22-A-0056 A Hybrid DNN-KF Model for Real-time SOC Estimation of Lithium-ion BatteriesVIEW ABSTRACT
Accurate state-of-charge (SOC) estimation of lithium iron phosphate (LFP) battery under different ambient temperatures is a long-standing problem in industry. In this paper, a hybrid model combining deep neural network and Kalman filter is proposed for SOC estimation of LFP battery under different ambient temperatures. After estimation via deep neural network, the estimated SOCs are further corrected using Kalman filter of high denoising capability. Data collected from dynamic stress test, US06 test and federal urban driving schedule under 25°C, 30°C, 40°C, and 50°C to verify the performance of proposed model, with the first two data as training set and the third data as testing set. Experimental results show that the proposed model can well meet the requirement of real-time estimation with mean absolute error within 2% and root mean square error within 2.4%. In addition, we also test the robustness of proposed model against different initial SOC values, and proves that the proposed model can well generalize the estimation to different initial values.
ICRMS22-A-0034 A Convolutional Autoencoder-based Fault Detection Method for Metro Railway TurnoutVIEW ABSTRACT
Railway turnout is one of the critical equipment of Switch&Crossing(S&C) Systems in railway, related to the train's safety and operation efficiency. This paper presents a convolutional autoencoder-based fault detection method for the metro railway turnout using the current curve data. Unlike existing methods, its data preprocessing considers human field inspection scenarios. This method does not require complex signal processing technology, nor does it destroy the spatial structure of the current curve. The proposed method comprises two modules. First, the one-dimensional original time-series signal is converted into a two-dimensional image by data preprocessing and 2D representation. Next, a binary classification model based on the convolutional autoencoder is developed to implement fault detection. The performance of our method is evaluated and tested on real-world operational current data in the metro stations. Experimental results show that the proposed method achieves a low error rate and is robust in practical engineering applications.
ICRMS22-F-0106 A Hybrid Model for Detecting Satellite Telemetry Data AnomaliesVIEW ABSTRACT
Anomaly detection of satellite telemetry data is of great significance for on-orbit satellite health monitoring. However, owing to the complex inherent characteristics, satellite telemetry data anomaly detection may be very difficult. To improve the accuracy of detecting satellite telemetry data anomalies, a novel hybrid model, called EEMD-SE-GWO-SVM, is proposed in this paper. In this model, ensemble empirical mode decomposition (EEMD) and sample entropy (SE) theory are firstly cooperated to extract critical features embedded in the raw telemetry data. Support vector machine (SVM) optimized by grey wolf optimizer (GWO) is then served as the predictive technique for the prediction of the reconstructed critical features. The anomaly can be finally detected by observing whether the residual error between the actual and predicted values exceeds a certain threshold. Experiment with the telemetry data of a real-world satellite is carried out and the results demonstrate that the proposed hybrid model outperforms comparison models in terms of anomaly detection accuracy.
ICRMS22-F-0097 A Hybrid Approach for Surface Roughness Prediction Based on Multi-domain Feature Fusion and Deep LearningVIEW ABSTRACT
The prediction of surface roughness in machining is of great influence on the assembly and reliability of precision equipment. Although the existing data-driven models consider both static and dynamic factors, the multi-domain features of dynamic factors are not effectively integrated, which results in unable to effectively capture the deterioration trend of surface roughness. This paper proposed a hybrid approach composed of a theoretical model and a data-driven model. Specifically, a novel deep network framework is designed to achieve the fusion of time-domain and time-frequency domain features. After that, the end-to-end prediction model of signal-to-surface roughness is established by the knowledge self-mining capability of deep learning. In addition, the transfer learning (TL) technique is also introduced to accelerate the training process of the deep learning network. The proposed approach is applied to surface quality inspection of the milling process and promising experimental results demonstrate the effectiveness of the proposed framework in practical engineering applications.
ICRMS22-F-0013 Research on Fault Prognostic of Photovoltaic System Based on LSTM-SAVIEW ABSTRACT
In the fault prognostic of photovoltaic systems, it is difficult to establish mathematical or physical models of complex components or systems. Therefore, this paper proposes a hybrid model of LSTM-SA, based on the principle of self-attention(SA) mechanism and long short-term memory (LSTM) neural network, combining the idea of self-attention and LSTM for timing problems processing capability to prognosticate faults of different equipment. Experimental verification of LSTM, LSTM-SA, BPNN and RNN models using the data of #102, #110 and #519 equipment respectively shows that the root mean square error (RMSE) of the model based on LSTM-SA is lower than that of the other three models in sunny days, indicating that the LSTM model with self-attention mechanism is optimized. Finally, the mixed model based on LSTM-SA is used to prognosticate the fault of different devices. The results are as follows: the fault of the #611 device at the 136th time point, and the fault of the #513 device at the 187th time point.
ICRMS22-F-0083 Sensitive Features Extraction of Wear Monitoring Signals Based on Wavelet Packet Energy SpectrumVIEW ABSTRACT
Tool condition monitoring is an essential issue in manufacturing process quality improvement, and there exist numerous sources of tool condition information. Force signals, vibration signals and acoustic emission signals are widely considered to be effective for identifying tool wear conditions, but the dilemma of redundant information is still hardly avoided. Therefore, to extract effective information of tool wear, this paper proposes a method to identify sensitive frequency band in the milling process based on wavelet packet energy spectrum. First, wavelet packet is proposed to decompose the vibration signal into multiple frequency bands. In addition, wavelet singular entropy is proposed to select appropriate decomposition parameters as well, so that weak vibration signals can be effectively extracted. Subsequently, the energy information is obtained from the decomposed frequency bands as characteristic parameters. Then identify the frequency bands sensitive to tool wear with Pearson correlation analysis. Finally, PHM2010 datasets are used to verify the feasibility and effectiveness of the proposed method, and the results demonstrate the applicability of the proposed method in practice for sensitive frequency band identification of tool wear.
Session Chair(s): JINFEN ZHANG, Wuhan University of Technology, Faqun QI, Wenzhou University
ICRMS22-F-0016 An Exploration into the Monitoring Methods for Key Parameters of Servo Drive Unit Based on Control Chart MethodVIEW ABSTRACT
In this study, the key parameters of servo drive units in CNC machine tools were explored by the online condition monitoring and reliability modeling during their long-term operation, in an attempt to eliminate the thorny problems in the reliability monitoring and evaluation of such servo drive units. Specifically, the model reference adaptive system algorithm was adopted to identify the online parameters of the motor by monitoring the stator resistance, stator inductance and flux linkage of the permanent magnet synchronous motor. Moreover, the percentage-based control chart was plotted to present the distribution of unknown parameters, with the intention of analyzing the reliability changing trend of the motor. Finally, the effectiveness of this method was validated via a case test. These findings contribute to the health monitoring and reliability evaluation of such servo drive units.
ICRMS22-F-0046 An Efficient Metamodel-based Method for Integrating Low-fidelity and High-fidelity Data on Reliability EvaluationVIEW ABSTRACT
Since the physical structure and mathematical models are more complex, reliability analysis in practical engineering can be expensive and difficult. A two-level multifidelity metamodel method for reliability analysis is introduced. Following the surrogate model in most of the relevant works, low-fidelity data and high-fidelity data are integrated by co-Kriging model. Besides, the co-Kriging model also provide an approximation for initial performance function. Bayesian method is adopted in model solution and a hybrid Markov chain Monte Carlo (MCMC) sampling algorithm is proposed. High-fidelity response of reliability performance function is estimated by the conditional distribution derivation based on Bayesian theory. Failure domain is identified by indicator function in sampling space that consists of samples derived from MCMC. Accordingly, failure probability estimations are obtained using Monte Carlo simulation (MCS). It is demonstrated through an illustrative example that the proposed method is valid and accurate.
ICRMS22-A-0049 Monitoring High-dimensional Heteroscedastic Processes Using Rank-based EWMA MethodsVIEW ABSTRACT
Various control charts are proposed to deal with the curse-of-dimensionality in high-dimensional process monitoring. The underlying dependency among variables is complicated and changes over time, which is called heteroscedasticity, also challenges the existing monitoring methods. We consider heteroscedasticity a common cause variability and propose an integrated monitoring scheme for location parameters in high-dimensional processes. Firstly, rank-based EWMA methods are designed to detect mean shifts in a small group of variables. A bootstrap algorithm determines the control limits by achieving a specified false alarm probability. Then, a post-signal diagnosis strategy is executed to cluster the shifted variables and estimate a time window for the change point. Simulation results show that the proposed methodology is robust to heteroscedasticity and sensitive to the small and moderate sparse mean shifts. It can efficiently identify out-of-control variables and the corresponding change points. A real-life example is provided to illustrate the proposed methodology. In this case study, sensor data is collected to ensure the system's safety and reliability. The monitoring of sensor data can help in preventative maintenance strategies.
ICRMS22-A-0058 Vibration Signals Detection of Early Faults Using LSTM-based Shewhart-type Control ChartsVIEW ABSTRACT
The deep learning method recurrent neural network and its variant have shown great success in processing sequences in recent years. However, this deep neural network has not aroused much concern in process control. We explore the ability of this neural network structure in monitoring process. This paper proposes a Shewhart-type control chart model based on long short-term memory (LSTM) prediction intervals. The proposed method is applied on a time series data of vibration, which confirms that the proposed method is an effective technique in process control. We will mainly focus on the vibration performance diagnosis of escalators in metro stations. Applying control chart techniques, the model can tell whether the health condition shows an early sign of a fault. By doing this, the maintenance scheme can be optimized when the early fault signals are triggered. We implement the model on vibration data of escalator components. The results validate that the proposed model manages to detect the fault signals from the potential information of vibration.
Session Chair(s): PIAO CHEN, Delft University of Technology (TU Delft), Yu LIU, University of Electronic Science and Technology of China
ICRMS22-F-0019 Research on Reliability Modeling of Bivariate Correlation Degradation Based on Nonlinear Wiener ProcessVIEW ABSTRACT
The performance of a product tends to degrade gradually with the use process, thus affecting its service life. Therefore, the research on the degradation process is helpful to grasp the residual life of products. Based on nonlinear Wiener process, the paper conducts a bivariate correlation degradation reliability modeling research. Firstly, the overall and partial correlations between two degenerate variables are calculated using the cosine similarity measure. Then, according to the calculation results and the characteristics of each copula function, select the appropriate copula function. Finally, a bivariate degradation model based on nonlinear Wiener process is established. Through the example analysis, the degradation fitting curve and degradation test results of bivariate degraded products are obtained by using the model, which has a good fitting degree and verifies the effectiveness of the model.
ICRMS22-F-0059 Operational Risk Modeling of CNC Machine Tool Considering Workpiece QualityVIEW ABSTRACT
With the advent of the era of big data and intelligent manufacturing, the structure of CNC machine tool has become more and more sophisticated and complex, and the risks of CNC machine tool during operation have become more and more diverse. Only by analyzing, evaluating and maintaining CNC machine operational risks activities to reduce its operational risk. Among the risks, substandard workpiece quality has gradually become the biggest risk in the operation of CNC machine tools, and this risk is usually ignored by people. Therefore, this paper proposes a modeling method for CNC machine tools based on operational risk that considers the quality of the workpiece. Firstly, define the operational risk system of CNC machine, and divide the operational risk of CNC machine into two parts: production risk and use risk. Secondly, the production risk and use risk of CNC machine are modeled separately，and the operational risk of CNC machine will be evaluated by Bayesian network for the CNC machine system. Finally, an example is given to illustrate the feasibility of the modeling method.
ICRMS22-F-0069 Photovoltaic Systems Degradation Modeling Under Dynamic Environmental ProfilesVIEW ABSTRACT
Reliability question of photovoltaic (PV) systems attracts significant attention due to performance degradation resulting from environment stresses. Different from monotone performance characteristics, such as wear and crack sizes, the power output of PV systems periodically degrades and is autocorrelated. Based on these considerations, we propose a semi-parametric model with autoregressive moving average (ARMA) model to extract the degradation rates of PV systems in this paper. Multivariate Bernstein bases with shape constraints are used to study interactions of time-varying covariates. The block coordinate descent method that jointly implements the quadratic optimization algorithm and nonlinear least square algorithm is used to estimate unknown parameters. Finally, different ground PV systems are provided to illustrate the effectiveness of the proposed method in practice.
ICRMS22-F-0104 Reliability Modeling and Analysis of Satellite-based Phased Array AntennaVIEW ABSTRACT
The degradation of satellite-based phased array antenna would affect the performance of the antenna, which in turn can prevent the signal from being transmitted properly. Therefore, satellite-based phased array antennas always require extremely high reliability. During the manufacturing and use stages, uncertainties are everywhere and performance degradation is inevitable. Both can threaten the reliability of satellite-based phased array antennas. However, existing studies mostly use voting models, which hardly reflect both points. In this paper, based on the principles of reliability science, a reliability model based on the performance margin is developed for satellite-based phased array antennas, considering the performance degradation and multiple sources of uncertainties. Firstly, the performance margin model for the degradation of the satellite-based phased array antenna is constructed based on its functional principle and the degradation processes of its critical components including phase shifters and power amplifiers. Then, uncertainties from different sources, including manufacturing tolerances and environmental temperature, are analyzed and each uncertainty is quantified by a probability distribution. Finally, a reliability model is developed. A numerical study of a satellite-based phased-array antenna is carried out. The results show the degradation processes of key performance parameters and reliability of the satellite-based phased array antenna during operation.
ICRMS22-F-0054 Reliability Analysis on Degraded Systems Subject to Coupled Effects of Zoned Shocks and Dynamic EnvironmentsVIEW ABSTRACT
In this paper, we focus on degraded systems suffer- ing from multiple dependent competing failure processes under coupled effects of zoned shocks and dynamic environments. A reliability model is developed for the system subject to gamma degradation and random shocks, in which the internal degradation rate is affected by a dynamic environment. Shocks are classified into four zones based on their magnitude, including the safety zone, probability damage zone, damage zone, and fatal zone. Explicit formulas for calculating reliability indexes of systems are derived. Meanwhile, simulation algorithms for calculating these reliability indexes are provided based on the Monte Carlo method, which is beneficial to the verification of the analytical method. Finally, a case study of micro-engine systems is given to illustrate the proposed models and obtained results, exhibiting that zoned shocks and dynamic environments have a significant effect on system performance.
Session Chair(s): Ping JIANG, National University of Defense Technology, Xiaoyue WANG, Beijing Technology and Business University
ICRMS22-F-0090 Reliability Classification Assessment of Logistics Service Capability Based on Universal Generating FunctionVIEW ABSTRACT
Reliability classification of service capability is essential for logistics enterprises to understand and assess their own ability level clearly, thereby making improvement more targeted. Few prior research has focused on reliability classification of logistics service capability. Furthermore, there are limitations in the research methods of logistics service capability. To fill these research gaps, this study establishes two models for reliability classification of logistics service capability based on two different criteria. An algorithm based on universal generating function (UGF) is proposed to obtain the reliability classification index of the models. Numerical illustrations based on a logistics integrator and several functional logistics enterprises are presented to demonstrate the applicability of models for reliability classification of logistics service capability.
ICRMS22-F-0093 Research on Smart Meter Prognostics and Health Management Model Based on Virtual MeasurementVIEW ABSTRACT
The smart meter metering anomaly analysis model can directly determine whether a metering fault occurs according to the metering data of a smart meter, and the model can connect the real-time sampling data of the smart meter to determine the metering fault. In this paper, a fault identification model of smart electricity meter is proposed. Through the example verification, the model can directly determine the metering fault according to the metering data of smart electricity meter, and accurately and truly identify the metering fault of smart electricity meter. At the same time, the accuracy of fault prediction reaches 88.7%, which can achieve good prediction effect. It can be seen from the results that the smart meter metering anomaly analysis model can directly determine whether a metering fault occurs according to the metering data of a smart meter, and the model can connect the real-time sampling data of the smart meter to determine the metering fault.
ICRMS22-F-0108 Reliability Evaluation Method of Motorized Spindle Based on Dual-source DataVIEW ABSTRACT
As one of the key functional components of the CNC machine tool, the motorized spindle seriously affects the reliability level of the CNC machine tool. At present, the researches on reliability evaluation methods of motorized spindles based on degradation data are still in the preliminary stage, and there are some problems such as single data source and difficult calculation, which affect the accuracy of reliability evaluation results. To solve these problems, this paper presents a novel reliability evaluation method of motorized spindles based on dual-source data. Taking the motorized spindle of the CNC machine tool as the research object, the degradation models based on nonlinear Wiener process are established according to the degradation data of laboratory reliability bench tests and field data respectively. In order to fuse the laboratory data and field data into dual-source data, a reliability model of the performance degradation of the motorized spindle is established based on the Copula function, which realizes accurate and efficient reliability evaluation of the motorized spindle. The effectiveness of the proposed method is verified by the degradation tests of the motorized spindles.
ICRMS22-A-0043 Reliability Evaluation of Consecutive-k-out-of-n: F Retrial Systems Subject to ShocksVIEW ABSTRACT
Based on the development trend of intelligent automation of products in the future, an intelligent automatic communication system consisting of a number of relay stations can be developed, in which the repair requests of failed relay stations can be sent to the repair station by the intelligent patrol robot with on-line monitoring function. This paper models the intelligent automatic telecommunication system as a repairable linear consecutive-k-out-of-n: F retrial system under Poisson shocks. The failure of system components can be caused by their intrinsic lifetimes or external shocks. When the repairman is busy, the repair requests will enter a retrial orbit (a virtual space), and then are sent at a certain retrial rate to the repairman in order to request repair. Based on the working time, repair time and retrial time of each component obey exponential distributions, the transition rates between system states are obtained. The formulas of several reliability indices are given by Markov process theory and Laplace transform. A numerical example is presented to illustrate the influence of various parameters on system reliability indices and compare the cost/benefit of the systems with and without retrial under Poisson shocks.
Session Chair(s): Christophe BERENGUER, Université Grenoble Alpes, Bin LIU, University of Strathclyde
ICRMS22-F-0020 Mission Reliability Driven Risk-based Maintenance Approach of Multi-state Intelligent Manufacturing SystemVIEW ABSTRACT
Risk-based thinking can better characterize and quantify the defects in the operational process of the manufacturing system, so the probability and loss of defects can be reduced by considering maintenance from the perspective of risk. However, current studies about risk-based maintenance (RBM) for manufacturing systems are rare. Therefore, a risk modeling and maintenance method based on multi-state intelligent manufacturing system is proposed. First, a new definition of operational risk of manufacturing system is proposed by extending the conception of the mission reliability of manufacturing system, and the connotation of RBM is explained. Second, a manufacturing system operational risk model is built based on mission reliability. Third, a RBM framework is proposed to determine the priority of machine maintenance by evaluating the risk level of the machine. Finally, an example is given to illustrate the effectiveness of the proposed method.
ICRMS22-A-0016 Robust Markov Decision Process for Maintenance ModellingVIEW ABSTRACT
Markov decision processes (MDP) provide a powerful modelling tool for maintenance modelling. An implicit assumption of the existing research in maintenance literature is that the model parameters are precisely known. Yet when implemented to real problems, typically, the parameters are estimated, whether from data or experience. However, maintenance decisions can be sensitive to the uncertainty inherent in the parameter estimates. Therefore, we aim to explicitly build parameter uncertainty into a maintenance model to support development of a robust maintenance policy for a degrading system. We consider a multi-state system that degrades according to a stationary Markov chain. The underlying maintenance process is formulated into an MDP model. Different from the traditional approach to maintenance modelling that assumes precisely known parameters, we consider the case that the probability transition matrices are subject to uncertainty. A minmax optimisation model is formulated to account for the worst case scenario.
ICRMS22-F-0121 Joint Predictive Replacement Management and Spare Part Planning of High-speed Railway Bearing Under Bayesian FrameworkVIEW ABSTRACT
This paper proposes a joint Bayesian-driven replacement and spare parts provisioning optimization model for high-speed train bearings. The non-linear stochastic process with Brownian motion noise is employed to capture the non-steady health trends extracted from on-line vibration signals. The lifetime parameter acquisition is realized via the integration of offline estimation via Maximum likelihood estimation (MLE) and online updating via Bayesian inference. Then the lifetime distribution is calculated by dynamic space-time scale transform approach. The decision-making regarding replacement and spare part lead time are sequentially updated, and accordingly the cost model is formulated and optimized. The applicability of the model is demonstrated via a real-world case study on high-speed railway bearings.
ICRMS22-F-0068 Fuel Cell Stochastic Deterioration Modeling for Energy Management in a Multi-stack SystemVIEW ABSTRACT
Fuel cells are promising clean power sources which use hydrogen and oxygen to generate electricity. However, the limited durability that is decided by various degradation phenomena remains one of the main barriers hindering their commercialization. Fuel cell deterioration modeling contributes to model and reproducing fuel cell deterioration behaviors, thus serving as a key step to decreasing fuel cell system deterioration. Fuel cell deterioration behavior is characterized by two main features, namely, load-dependent and stack-to-stack deterioration heterogeneity. A Gamma process with random effect-based deterioration model is used to account for the above deterioration features of fuel cells. Different types of random effects are introduced to the studied Gamma process on its scale parameter, taken as a random variable following a gamma law. Fuel cell degradation trajectories are studied by the developed deterioration models using the Monte Carlo simulation method. The lifetime distributions of the proposed models are analyzed for investigating their deterioration behavior.
ICRMS22-F-0006 Optimal Ordering and Replacement Scheduling for a Deteriorating System Subject to ShocksVIEW ABSTRACT
This paper investigates a spare unit ordering and replacement policy for a deteriorating system with shocks in seeking for the optimal number of minimal repairs. The original system suffers from both deterioration and external shocks, in which the shocks are divided into two distinct categories with different time-dependent probabilities, including a non-fatal shock whose damage can be completely removed by a minimal repair and a fatal shock which can also break down the system. To be specific, the original system failure is subject to a competing failure process, where it occurs at the deteriorating failure time or at the first appearance epoch of a fatal shock. A spare unit with a constant delivery time is ordered emergently at system failure or preventively when the number of minimal repairs reaches a value, whichever occurs first, for the sake of minimizing the long run average cost rate in one renewal cycle. The optimal number of minimal repairs is theoretically demonstrated and an illustrative example is designed as a validation of the theoretical results, as well as sensitivity analyses of some key parameters.
ICRMS22-A-0005 Joint Optimization of Lot Sizing and Conditional Based Maintenance of a Production SystemVIEW ABSTRACT
The issues relevant to joint optimization of lot sizing and condition-based maintenance policy for a production system are widely investigated with help of the development of sensor technologies. In this study, we consider a two-stage deteriorating production system that produces a single product with constant production rate and demand rate. The system will experience the emergence of the crack and the propagation of the crack before failure. A general distribution and a Gamma process are utilized to describe the “crack initiation process” and “crack propagation process”. When the system fails, instead of production interruption, a proportion of nonconforming products which will cause extra cost will be produced. Under conditions with backlogging, the repair time of both preventive maintenance and corrective maintenance cannot be negligible. The lot-sizing and the condition-based maintenance threshold are optimized, where the long-run average cost rate is considered as the policy assessment function. We solve the problem in a semi-Markov process framework. A numerical example is given to illustrate the applicability of the proposed model. It may provide a theoretical reference in production and maintenance scheduling. Keywords: lot-sizing; condition-based maintenance; degradation modelling; semi-Markov process.
Session Chair(s): Olga FINK, EPFL, Lechang YANG, University of Science and Technology Beijing
ICRMS22-F-0116 Reliability Assessment Scheme for Intelligent Autonomous SystemVIEW ABSTRACT
As a product of intelligent technology integration, intelligent autonomous system (IAS) is central to future development. In order to analyze and evaluate the IAS reliability, artificial intelligence (AI) algorithms and the associated IAS functional structures are studied to propose a reliability assessment scheme for the IAS in this paper. Considering the task requirements and environmental conditions, the IAS can be divided into layers, and the reliability metrics of AI algorithms and the system reliability covering software and hardware are evaluated for each layer, respectively. Then, the reliability metrics of the IAS are calculated by weighting and fusing the reliability of AI algorithms and the system reliability of each level. Moreover, based on multi-source data fusion and uncertainty quantification techniques, the assessment method of the IAS reliability is introduced to provide technical guarantees for its reliable operation. In short, by analyzing the internal and external factors affecting the IAS reliability, the reliability assessment scheme of the IAS is established based on the IAS reliability metrics and the IAS evaluation method, to promote the development of reliability technology.
ICRMS22-F-0085 A Semi-supervised Deep Learning Model with Consistency Regularization of Augmented Samples for Imbalanced Fault DetectionVIEW ABSTRACT
With increasing requirements on reliability, maintainability and safety in modern ICT systems, fault detection, as an indispensable part of AIOps, has become essential in cloud computing or communication network environments. However, due to the lack of effective labels and class imbalance on faulty samples, fault detection performance based on the common classification model can't meet the system's operational requirements. Some recent approaches of SSL propose a consistency regularization loss to solve the problem of insufficient labels. However, these approaches are mainly for images based on artificial data augmentations but not feasible for all data types, and class-imbalance problem is not considered simultaneously. So, we propose a semi-supervised method for imbalanced fault detection with few labels, called SSLCR-IFD. In the method, we use a semi-supervised deep classifier based on consistency loss to solve the lack of labels, in which two sample augmentation methods based on clustering and GAN are used. Furthermore, a selective pseudo-labeling self-training strategy is proposed to solve the class-imbalance problem. Compared with the standard data augmentation, our methods alleviates the need for domain knowledge and can be used on multiple types of tasks. Finally, experiment results show that our method outperforms the baseline methods on two different AIOps tasks.
ICRMS22-A-0013 A Generic Physics-informed Neural Network-based Framework for Reliability Assessment of Multi-state SystemsVIEW ABSTRACT
We develop a generic physics-informed neural network (PINN)-based framework to assess the reliability of multi-state systems (MSSs). The proposed methodology consists of two major steps. Firstly, we recast the reliability assessment of MSS as a machine learning problem using the framework of PINN. A feedforward neural network with two individual loss groups is constructed to encode the initial condition and state transitions governed by ordinary differential equations in MSS. Next, we tackle the problem of high imbalance in the magnitude of the back-propagated gradients in PINN from a multi-task learning perspective. Particularly, we treat each element in the loss function as an individual task and adopt a gradient surgery approach named projecting conflicting gradients (PCGrad), where a task's gradient is projected onto the norm plane of any other task that has a conflicting gradient. The gradient projection operation significantly mitigates the detrimental effects caused by the gradient interference when training PINN. We showcase the PINN-based method for MSS reliability assessment in several different contexts. The results demonstrate that the proposed PINN-based framework shows remarkable performance in MSS reliability assessment, and the incorporation of PCGrad in PINN leads to substantial improvement in solution quality and convergence speed.
ICRMS22-F-0034 Data Augmentation to Improve the Performance of Ensemble Learning for System Failure Prediction with Limited ObservationsVIEW ABSTRACT
Ensemble learning has been widely used to improve the performance and robustness of machine learning algorithms on time series data. However, in real operational processes where the observed data is limited, it hinders the capability of ensemble learning algorithms. To address the challenge of limited observed data, this paper proposes a novel three-layer ensemble learning framework by use of data augmentation. Firstly, multiple classical time series augmentation methods are applied to increase the size of the data set. Subsequently, after pre-processing, these augmented data is trained by multiple basic learners with K-fold cross-validation as the first layer of the developed ensemble learning framework. The outputs of the first layer are integrated via LASSO to further improve the prediction performance, which serves as the second layer of the developed framework. Finally, the third-layer output is generated by averaging the prediction of the second layer and the output from an improved Long-Short Term Memory model that provides prediction based on the augmented data. A case study on a real wastewater treatment plant is used to illustrate the effectiveness of the proposed method.