基于HDP-CHMM的机械设备性能退化评估
Performance Degradation Evaluation of Mechanical Equipment Based on HDP-CHMM

针对传统隐马尔可夫模型(hidden Markov model,简称HMM)状态数必须预先设定的不足，提出了一种基于分层狄利克雷过程-连续隐马尔可夫模型（hierarchical Dirichlet process-continuous hidden Markov model，简称HDP-CHMM）的机械设备性能退化评估方法。该方法利用分层狄利克雷模型的分层聚类原理，在狄利克雷过程(Dirichlet process，简称DP）模型的基础上进行扩展，利用多组关联数据实现了模型结构根据观测数据的自适应变化和动态调整，获得设备运行过程中的最优退化状态数，并结合连续隐马尔可夫模型(continuous hidden Markov model，简称CHMM)良好的分析和建模能力，获得设备退化状态转移路径，实现机械设备运行过程中的退化状态识别和性能评估。利用滚动轴承全寿命数据的多组特征值进行了应用研究，并与基于K-S检验算法的机械设备零部件性能退化评估方法进行了比较。结果表明，HDP-CHMM模型可以对轴承实际运行状态转移过程进行建模，有效识别轴承运行中的不同退化状态，为基于状态的设备维修提供了理论指导。
In the light of the deficiency of the traditional hidden Markov model (HMM), a new method to evaluate the performance degradation of mechanical equipment is proposed. It is based on the hierarchical Dirichlet process (HDP) and continuous hidden Markov model (CHMM). First, the self-adaptive structure of model changes dynamically with relative data for the optimal degradation state parameters during the operation. Then, the equipment degradation state transition path is obtained with the analysis and modeling based on CHMM to identify and evaluate the degradation. Finally, the life characteristic values of rolling bearings are used as the data to verify the proposed method.　The results are compared with those obtained by K-S algorithm. It shows that the proposed method of performance degradation assessment based on HDP-CHMM can be used to simulate equipment actual degradation process. It takes theoretical guidance for the maintenance of the equipment based on the state.