The design and development of various complex systems to support various sectors of the economy have surged recently. For these systems’ reliability and maximum operational efficiency, it is important to develop effective and robust reliability assessment and failure prediction methods for implementing optimal backups, maintenance, and repair schedules. Generally, for reparable systems, current reliability assessment methods are based on the principle of recurrence repair or failure times. With the increasing interest in the reliability assessment of reparable complex systems, several additional studies to enhance the reliability and operation of these systems under different conditions have also been reported.
The rate of occurrence of failures (ROCOF) is a critical consideration in dynamic reliability assessment. Typically, ROCOF follows well-defined deterministic functions over time and forms the basis of most failure prediction methods. However, it indicates the possible number of failures over a specified period that is highly susceptible to repair effects, uncertain environment conditions and inherent frailties common in most practical applications. These factors also result in random fluctuations of failure occurrence rates. Unfortunately, only a few reliability studies have focused on ROCOF with uncertainties, characterizing the uncertainties from unexplained variability. Additionally, the temporal uncertainties, common in the failure processes, have been overlooked despite their practical implications. This jeopardizes the chances of obtaining accurate modeling failure recurrence data leading to poor assessment performance.
To address the above issues, Dr. Peng Yizhen from ChongQing University, Dr. Wang Yu and Dr. Zi Yanyang from Xi’an Jiao Tong University, and Dr. Xie Jingsong from Central South University developed a novel adaptive stochastic recursive-filter-based dynamical model for accurate prediction of the dynamical failures in complex reparable systems under uncertainty conditions. Their main objective was to provide an alternative and effective approach for describing the failure counting process taking into account multiple uncertainties. Their research work is currently published in the journal, Reliability Engineering and System Safety.
In their approach, the authors adopted a state-space model framework comprising of the general nonhomogeneous Poisson process (NHPP) coupled with Brownian motion to describe the failure process under different uncertainty conditions. This model comprised a power law intensity function and AMSAA model to increase its adaptability and system failure prediction accuracy. Next, based on the Bayesian filter and EM algorithm, an adaptive stochastic inference method was derived to solve the complex statistical inference of the model, update its parameters and adaptively estimate its initial states. Finally, the feasibility of the proposed strategy was verified in the reliability prediction of the real gas pipeline system.
Results demonstrated the efficiency of the proposed approach in providing accurate predictions of the failure times in complex systems subject to different uncertainty conditions. Unlike the existing methods, introducing the Brownian process into the ROCOF accounted for the cumulative effects of the temporal uncertainties. Moreover, the model adaptively estimated the ROCOF distributions based on system evaluation. This was much better than relying on the training process, thus enhanced its overall performance. The improved performance in uncertain environments could be attributed to treating the ROCOF parameters as state variables to allow adaptive adjustment of the model parameters for tracking the temporal evaluations.
In summary, a novel doubly stochastic NHPP-based model, incorporating the AMSAA model and Brownian motion for failure prediction of complex reparable systems under temporal uncertainty conditions, was reported. The model proved reliable for describing failure process in stable operation and reliability growth stages of the reparable system. However, it was not fully appropriate for the entire life cycle of the system. Its application in the real natural gas pipeline compressor system proved a success, showcasing the feasibility of the model. The system is versatile and provides room for improving its limitation for effective field applications. In a statement to Avance in Engineering, the authors noted that the system is a potential candidate for improved reliability assessment in complex mechatronic systems.
Yizhen, P., Yu, W., Jingsong, X., & Yanyang, Z. (2020). Adaptive stochastic-filter-based failure prediction model for complex repairable systems under uncertainty conditions. Reliability Engineering & System Safety, 204, 107190.