基于隐Markov树的设备状态综合诊断模型

Synthetic machine condition diagnosis model based on hidden Markov tree

为了充分利用小波系数之间的统计相依性以更有效地诊断设备状态,提出了一种基于隐Markov树(HMT)的综合诊断模型。首先通过主成分分析将来自多个传感器的信号转换为主成分,求出各主成分对应的频谱,然后通过比较对已训练的各HMT模型的适应度,运用Bayes决策融合法则得到设备状态综合诊断决策。为了克服HMT模型存在的计算溢出困难,采用尺度变换对EM算法进行了改进。通过两个实例验证了该综合诊断模型具有较高的诊断准确率。

To fully exploit the statistical dependency of wavelet coefficients for more effective equipment condition diagnosis, a synthetic diagnosis model based on hidden Markov tree （HMT） is presented. The measurements from multiple sensors are converted into principal components by principal component analysis, and subsequently transformed to spectra. By comparing their likelihood fitness to the trained HMT models, and using Bayesian decision fusion, further diagnosis decision is made. To overcome the overflow difficulty existing in HMT model, a scaling algorithm is developed to improve expectation maximization （EM） algorithm. The high diagnosis accuracy of the model is illustrated by two cases.