基于小波相关特征尺度熵的HSMM设备退化状态识别与故障预测方法

Equipment degradation state recognition and fault prognosis method based on wavelet correlation feature scale entropy and HSMM

隐半马尔可夫模型(HSMM)是隐马尔可夫模型(HMM)的一种扩展模型,是在已定义的HMM结构上加入了时间组成部分,克服了因马尔可夫链的假设造成HMM建模所具有的局限性,与HMM相比具有更好的建模能力和分析能力,而且可以直接用于预测。基于振动信号与语音信号的相似性,将HSMM引入机械设备退化状态识别与故障预测中,提出基于小波相关特征尺度熵(WCFSE)的HSMM设备退化状态识别与故障预测方法。首先将小波相关滤波法与信息熵理论相结合得到能敏感表征故障严重程度的WCFSE向量,并以此向量作为HSMM的输入进行训练,建立基于HSMM的设备运行状态分类器与故障预测模型,从而实现设备退化状态识别与故障预测。将其应用到滚动轴承的退化状态识别与故障预测中,验证了该方法的有效性。

Hidden semi-markov model （HSMM） is an expansion model of hidden markov model （HMM） and constructed by adding a temporal component into the well-defined HMM structure, which overcomes the modeling limitation of HMM due to the Markov chain assumption. Therefore, it improves the power in modeling and analysis, and further more, it can be directly used for prognosis. Based on the similarity between vibration and sound signals, HSMM is introduced to degradation state recognition and fault prognosis of machinery equipment, and a method of degradation state recognition and fault prognosis is proposed based on wavelet correlation feature scale entropy （WCVSE） and HSMM. Firstly, wavelet transform correlation filter and information entropy theory are combined to obtain the WCFSE vector, which can sensitively express the fault severity degree. Secondly, the WCFSE vector is inputted to the HSMM for training, and the running state classifier and fault prognosis model of the equipment are constructed based on HSMM to recognize equipment degradation states and prognosticate faults. The proposed method was applied to the degradation state recognition and fault prognosis of a roller bearing, and experiment results demonstrate that the method is effective.