Research On Machinery Diagnosis Based On Deteriorating Hidden Markov Process

Equipment maintenance management directly affect a manufacturing system's reliability and economic efficiency, so it gains much attention in business management and becomes an important way for enterprises to reduce production costs and gain competitive advantage. Meanwhile, with the rapid development of science and technology, equipment structure and fault mechanism grow more and more complex, based on previous research on maintenance planning such as failure-based maintenance and time-based maintenance, condition-based maintenance(CBM) has been studied by more and more researchers, which ensure that ensure that enterprises can carry out active maintenance, and can reduce and avoid the accident occurred in the production process. As a key support technology for the realization of CBM, the research of machinery diagnosis is also of great value and significance. Hence, how to evaluate machine condition accurately becomes a vital issue to support condition-based maintenance. Considering machine degradation in the production process, a diagnosis model based on recession characteristics is proposed to realize the scientific description of the diagnosis model considering the equipment operation.This paper studies a single machine system. Based on the available research on intelligent equipment diagnosis technologies, a comprehensive analysis is given, which summarizes the advantages and disadvantages of existing research methods. Accordingly, this paper focus on the Hidden Markov Models(HMM)-based machinery diagnosis. Firstly, the basic theory of hidden Markov model is studied, and then based on the theoretical support, the research framework of machinery diagnosis based on HMM considering equipment degradation is constructed. According to the research framework, the following work are finished in this paper:(1) Determine the classification method of equipment condition. In this paper, the key performance parameters of equipment or its important components are choose as the equipment performance indicators, build a unified equipment health index value by normalizing the monitoring data. And then the different status levels are divided, which can be described qualitatively.(2) Optimize the traditional HMM training algorithm. Based on the traditional expectation maximization(EM) training algorithm, heuristic training algorithm(SAEM) is proposed to solve the problem of local optimum and initial value selection sensitivity in the traditional training process by introducing simulated annealing algorithm(SA) because of its probability's global convergence.(3) Construct HMM considering equipment degradation. Define the aging factor under two forms of constant and multiplier, respectively, identify the effects of different aging factors on the state transition probability, and then construct new expression form of the state transition matrix; to estimate the optimal aging factor, establish the likelihood function, and use numerical analysis method with a iterative algorithm; finally update the state transition matrix iteratively with the estimated value of aging factor to realize the evaluation of the equipment condition.The three research contents above are connected together to form a systematic equipment diagnosis process. Equipment state classification determine the different states corresponding to the observation sequence, which is the input of degraded hidden Markov model for diagnosis; what follows is optimizing the traditional EM training algorithm, and training the model with the SAEM algorithm; then on the basis of the training results, construct hidden Markov model considering equipment degradation property for equipment condition assessment by integrating aging factor to state transition matrix. Finally, a numerical examples is given to demonstrate the effectiveness and feasibility of the above research content.