PFM神经网络VLSI电路的故障诊断应用

PFM neural network VLSI circuit for fault diagnosis

为了改变传统的基于软件的故障诊断模式,发挥神经网络超大规模集成电路(VLSI)的优势,提出了一种用于故障诊断的改进脉冲频率调制(PFM)模拟神经网络脉冲流VLSI电路。利用单层感知器网络、场效应管电路实现了一种新的数字模拟混合突触乘法/加法器电路。以此电路为基础,设计了进行主轴承磨损故障诊断的神经网络故障识别系统。用含有故障信息的噪声信号代替振动信号进行特征值提取,经过前置信号处理分析、故障特征值提取和神经网络运算,最后得出代表待诊断测试信号与标准故障模板之间"欧氏距离"的VLSI电路输出端电容的电压值。根据各个电压值,可以判断出故障类别。该电路具有较高的识别精度,可以实现实时在线的故障诊断。

This paper presents an improved PFM (pulse frequency modulation) analogue neural network/pulse stream VLSI circuit for fault diagnosis to replace software-based fault diagnostics. A single-level perceptron network and field effect transistor circuit were used to make a digital/analogue-based synapse multiplier/ adder. A neural network fault recognition system based on this circuit was designed for abrasion fault diagnosis of a main bearing. The characteristics were extracted from the noise signal including the fault information instead of using a vibration-based method. Signal processing was used to identify the fault characteristics with the neural network analysis used to calculate the output capacitor voltage of the VLSI circuit, which represents the Euclid Distance between the corresponding fault signal template and the test signal. The result was used to simultaneously identify the fault. Consequently, the circuit is very accurate and can realize real-time online fault diagnosis.