基于HDP-HMM的机械设备故障预测方法研究

针对隐马尔科夫模型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-隐马尔科夫模型(HDP-HMM)的机械设备故障预测方法。该算法通过构造HDP作为HMM参数的先验分布,利用HDP分层共享和自动聚类的优点,实现了模型结构动态更新,获得设备运行过程中的隐状态数;基于HDP-HMM所建立的退化状态动态转移关系,确定设备早期故障点和功能故障点,实现设备的健康等级评估和故障预测。利用美国USFI/UCR智能维护系统中心提供的滚动轴承全寿命数据进行了应用研究。结果表明,针对多观测序列,HDP-HMM能有效实现组合聚类,识别结果不依赖于算法初始参数的选择,具有较强的鲁棒性;与基于K-S检验的退化评估算法比较表明,HDP-HMM更能有效描述设备实际退化过程。

A prognostic method of mechanical equipment based on HDP-HMM

Aimed at the deficiency of the Hidden Markov model whose hidden states must be determined in advance, a prognostics method of mechanical equipment based on the Hierarchical Dirichlet Process-Hidden Markov model (HDP-HMM) was proposed. By constructing HDP as the prior distribution of HMM, the structure of HMM was dynamically adjusted and the state number during the operation of the equipment degradation was obtained according to hierarchical sharing and automatic clustering of HDP. Based on the dynamic transition state relationship established by HDP-HMM, the early failure point and functional failure point of the equipment were determined, and the health grade evaluation and prognostics of the equipment were realized. The application of life data of rolling bearings provided by the USFI/UCR intelligent maintenance system center was studied. The results show that HDP-HMM can effectively achieve the combination clustering for multiple observation sequences and the recognition results do not depend on the choice of initial parameters of the algorithm which has strong robustness. Compared with the K-S test algorithm of degradation assessment, HDP-HMM can describe the actual degradation process of the equipment more effectively.