一种基于BP神经网络的传感器故障诊断方案

A Fault Diagnosis Approach to Sensor Based on BP Neural Network

针对传感器故障,提出了一种BP网络和修正的Bayes分类算法(MB)的集成故障诊断方法.用BP神经网络建立传感器故障模型,对系统的状态和故障参数进行在线估计,再用修正的Bayes算法进行传感器故障的在线检测、分离和估计.对连续搅拌釜式反应器(CSTR)的仿真结果表明,该集成故障诊断方法能够对传感器故障进行快速准确的分离和估计,并对传感器故障具有容错性.

An ihtegrated fault diagnosis approach to sensor based on back propagation （BP） neural networks is presented in this paper. A BP neural network is used to estimate the state and fault parameters of the constructed model for sensor faults. The estimated fault parameters are processed by the improved Bayes algorithm to realize the sensor fault detection, isolation, and estimation on line. The simulation for continuous stirred tank reactor （CSTR） shows that the presented approach can isolate and estimate the muhriple sensor faults quickly and accurately and the integrated system is of tolerant ability to sensor faults.