基于行为的故障诊断系统跟踪自学习算法研究

Research on a Tracking Self-learning Algorithm for Behavior-based Fault Diagnosis System

针对现有绝大多数智能故障诊断系统自适应跟踪设备行为变化能力的不足,对基于行为的智能化故障诊断系统中模糊基函数网络的在线跟踪自学习算法进行了研究,提出了一种在线跟踪故障分类边界的自学习算法。该算法通过带有遗忘因子的样本均值和样本协方差矩阵保存样本所包含的故障可能性分布信息,并在此基础上产生新增样本的目标输出,用于训练FBF网络,以实现故障分类边界的在线跟踪。给出了带有遗忘因子的样本均值和样本协方差矩阵的递推算法,用以克服传统方法需要保存大量以往故障训练样本所带来的困难。理论研究和工程应用表明,在线跟踪故障分类边界的自学习算法可以有效地避免神经网络训练过程中的“突然遗忘”现象,是可行的。

Most of the existing inteiilgent fault diagnosis systems are lack of the ability to track the operation behavior of the machine, thus a on--line tracking self--learning algorithm for the fuzzy basis function （FBF） network was researched, which was used in the behavior-- based intelligent fault diagnosis system. Based on the previous fault possibility distribution kept within the sample mean and covariance matrix with forgetting factors, a strategy for constructing the target output of the new training sample set was proposed. With the new sample set the FBF network can be trained to track the variable fault clustering boundary. Meanwhile, a recursive algorithm for computing the sample mean and covariance matrix with forgetting factors was also given to overcome the difficulties of storing the vast old training samples. The results of the theoretical and practical applications show that the proposed online tracking self--learning algorithm is feasible, which can effectively avoid the phenomenon of ＂catastrophic forgetting＂ existing in neural network training.