基于模型诊断技术的神经网络实现方法

针对基于模型的故障诊断流程中故障检测和故障识别两个关键问题,提出了一种基于神经网络的实现方法.首先利用BP神经网络进行参数估计,并结合系统模型进行故障检测;然后采用ART2神经网络进行数据聚类,并基于聚类结果进行系统故障识别;最后,设计实现了基于BP/ART2神经网络的故障诊断系统.基于BP神经网络的参数估计方法可以准确地估计诊断对象在不同状态下的参数,为故障检测提供有效依据;基于ART2神经网络的数据聚类不仅可以识别对象的已知故障类型,还可以识别出未知故障,对先验信息较少的系统进行故障识别更具有效性.通过永磁直流电机故障诊断案例的应用,证明方法能具有一定的工程实用性.

Realization of model-based fault diagnosis with artificial neural network

Based on artificial neural network, a method for fault detection and fault identification was proposed, which can be used during the model-based fault diagnosis (MBFD) process. Firstly, back-propagation (BP) neural network was presented for parameter estimation, combing with the system model, the faults can be detected. Then the adaptive resonance theory (ART2) neural network was used for data clustering, and based on the clustering results, the faults will be identified. Finally, a fault diagnostic system was developed based on BP&ART2 neural network. The presented BP-based parameters estimation method can estimate the parameters under different states accurately, which is important for the fault detection. ART2-based data clustering method can distinguish both the known and unknown fault modes, which is much efficient when diagnosing the system without enough priori information. A brushless direct current motor was selected as the application case, simulation result has proved that the presented method is available to estimate the motor parameters, detect and identify some typical faults, such as brushes position offset and armature open.