Optimal multi-level classification and preventive maintenance policy for highly reliable products

Degradation modeling and classification of mixed populations using segmental continuous hidden Markov models

“…Thereby, 3 SCHMMs that are denoted by λ 1 , λ 2 , and λ 3 should be constructed here. The dataset for modeling learning is from our previous work . Since the classification policies are made on the basis of the degradation data and the Wiener process is often used for modeling degradation paths, the Wiener process is used to randomly generated the trained dataset, where the observation interval h = 25 and the number of inspections θ = 20.…”

“…Therefore, it is more extensive and representative to use the Wiener process to describe or generate degradation data. Using these data to training the SCHMMs has 2 advantages: (1) the data satisfy the Gaussian hypothesis of the distributions of observation and (2) the data are also used for classification in Chen et al We can compare the classification results based on the proposed method with the previous study intuitively. The threshold of failure D N is defined as 60% of the initial value.…”

“…However, more degradation data can be obtained during the test if more inspections have been made. In our latest paper, a classification method based on LDA is developed to make full use of the information for a more accurate classification and is proven to perform better than “one observation‐classification.” Nowadays, the machine learning techniques, such as HMM and ANN, are widely used for classification because of more and more data available. We adopted HMM‐based methods here since it is difficult to find the degradation law and observe the health state transition when using ANN‐based methods, Actually, we need this information to build the connection between mathematical models and the underlying physical degradation processes.…”

“…For fully using degradation data, Xu used linear discriminant analysis (LDA) to take the linear combination of the observed degradation data at each time as the basis for classification. Chen et al also used LDA to design policies for multilevel classification based on degradation data. However, all these methods mainly handle small samples.…”

A New Analytic Method of Multistate Single-component System Reliability Model

“…Wang et al [2] utilized the periodic preventive maintenance strategy to establish the reliability model for storage system and / kn system. Chen [3] adopted different preventive maintenance strategies for different levels of parts, and put forward the age-based combined multistage classification and preventive maintenance model. It may be simple and easy to perform the periodic or scheduled preventive maintenance, but it is often difficult to carry out the timely maintenance, so maintenance resources will be wasted.…”