Data-driven Approaches To Reliability Modeling And Assessment Of Complex Systems Under Dynamic Environments

With advanced industrial systems and military equipment designed towards larger scale,more complex,and intelligent,it is a challenging task to ensure their reliability in operational stages as they may possess highly integrated functions,improved performances,and complicated loads,and be exposed to severe environments.The failures of systems or equipment may lead to potential risk including purely economic loss and even severe safety accidents.To avoid the aforementioned potential risks,reliability of these systems must be accurately assessed and predicted in advances.On the one hand,engineering systems exhibit multi-state nature during their degradation or failure processes,leading to the high complexity of reliability modeling and assessment.On the other hand,most reported reliability models in literature were developed on the premise that systems operate in the prevailing environmental conditions which are either time-invariant or have no effect on failure processes.This assumption is not always held in real-world situations.In fact,engineering systems and equipment often operate their missions under time-varying environments which may also be stochastic in nature.The development of reliability models for engineering systems working under stochastic dynamic environments is an emerging area in the reliability field.The impetus comes because many available reliability models have been developed based on the simplified assumption of static environments.Ignorance of the variability and stochasticity or the effect of environmental conditions will result in inaccurate system reliability assessment.Thanks to the renovation of advanced sensors and development of condition monitoring techniques,inspection data reflecting systems' health status can be collected in the operational stage.Utilizing the inspection data can effectively track and predict the deterioration trend.This dissertation devotes to mathematically characterizing the stochastic degradation or failure behaviors of various engineering systems under stochastic dynamic environments,and then,developing data-driven methods to infer the unknown parameters in these new reliability models.The loading optimization problem for multi-state systems from a cumulative performance perspective will be also studied.The research results to be developed will provide a collection of new theoretical approaches and practical tools to ensure the operational reliability of engineering systems under dynamic environments.The primary research contributions and innovative outcomes are summarized as follows:(1)Development of reliability assessment and parameter estimation methods for systems with Weibull lifetimes under dynamic environments.The dynamic environment is characterized by a continuous-time Markov chain.Using the cumulative exposure principle,the stochastic time scale,resulting from the cumulative effect of the Markovian dynamic environment,is computed via a Markov reward model.Based on the above settings,the system reliability model under the Markovian dynamic environment is developed.Furthermore,the maximum likelihood estimates and confidence intervals for the reliability model parameters,including the transition rate matrix of the Markov chain,the reward rates of the Markov reward model,and the parameters of the baseline lifetime distribution,are obtained by utilizing the collected environment and lifetime data.The system reliability is then evaluated with the estimated parameters.The results show that the unknown reliability model parameters can be effectively estimated,and the proposed model with the consideration of the cumulative effect of the Markovian dynamic environment can provide a more accurate reliability estimate.(2)Development of reliability assessment and parameter estimation methods for Markov multi-state systems under dynamic environments and stochastic shocks.The deterioration behavior of the multi-state system is characterized by a nonprogressive continuous-time Markov chain,and the cumulative effect of the dynamic environment is characterized by the stochastic time scale.For systems working under dynamic environments and stochastic shocks,a distribution approximation algorithm is proposed to assess the system reliability.For systems working under only dynamic environments,two approximation algorithms,namely the distribution approximation algorithm and moment approximation algorithm,are developed to facilitate system reliability assessment.Comparative results show the moment approximation algorithm is superior to the distribution approximation algorithm.Furthermore,the maximum likelihood estimates and confidence intervals for the unknown reliability model parameters are obtained by utilizing the collected environment,shock,and lifetime data.The results show that the unknown reliability model parameters can be effectively estimated,and the proposed model can provide an accurate reliability estimate.(3)Development of reliability assessment and parameter estimation methods for semi-Markov multi-state systems under dynamic environments.The deterioration behavior of the multi-state system is characterized by a progressive semi-Markov model of which the sojourn time in each state follows the Weibull distribution,and the cumulative effect of the dynamic environment is characterized by the stochastic time scale.The full factorial numerical integration method and Gaussian quadrature are utilized to circumvent the multi-dimension integration in the system reliability assessment.Furthermore,parameter estimation is conducted for both the synchronous and asynchronous inspection cases.The results show that the unknown reliability model parameters can be effectively estimated,and the proposed model can provide an accurate reliability estimate.Moreover,the estimated parameters under the synchronous inspection case are more accurate than those under the asynchronous inspection case.(4)Development of a loading strategy for semi-Markov multi-state systems from a cumulative performance perspective.The cumulative performance and the corresponding mission success probability are formulated for both infinite and finite time horizons.The distribution of the cumulative performance of a system at failure or a particular time is evaluated using a set of multiple integrals.Correspondingly,two load optimization models are formulated to identify the optimal loading strategy for each state to achieve the maximum mission success probability.The results show that the proposed method can effectively evaluate the mission success probability from a cumulative performance perspective in a computationally efficient manner,and the optimal loading strategy can be determined by the proposed method for both infinite and finite time horizons.