脉频调制神经网络VLSI的设计及应用

Design and application of PFM neural network VLSI

中文关键词  神经网络  故障诊断  超大规模集成电路  脉频调制

英文关键词  neural network  fault diagnosis  very large scale integrated (VLSI) circuit  pulse frequency modulation

中文摘要

      本文提出了一种用于故障诊断识别的改进脉冲频率调制(PFM)VLSI神经网络电路,改进了传统的基于软件的机械故障诊断模式,发挥了神经网络超大规模集成电路(VLSI)的优势.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,而且该神经网络电路的突触权值不需要学习调整,降低了电路的复杂性.以此电路为基础,设计了进行主轴承噪声故障诊断的神经网络故障识别系统.将含有故障信息的原始噪声信号,经过前置信号处理分析、故障特征值提取和神经网络运算,得出VLSI电路输出端电容的电压——代表待识别信号与模板故障信号的“欧氏距离”,进而判断出故障的类别.经过仿真测试,基于硬件的诊断系统的识别性能接近于基于软件的系统.

英文摘要

      This paper presents an improved pulse frequency modulation (PFM) neural network VLSI circuit for fault diagnosis in order to improve the software - based fault diagnosis approach and to give full play to the merits of neural network VLSI circuit. By using the single-level perception network and the field effect transistor circuit, a new digital/analogue based synapse multiplier/adder is designed so that the threshold of the synapse need not be adjusted by learning and hence the circuit becomes less complicated. Based on this circuit, a neural network fault detection system is designed for the noise based fault diagnosis of main bearing. Through signal processing, extraction of fault feature, and neural network computation for the original noise signal containing fault information,the output capacitor voltage value of the VLSI circuit is derived. This value, representing the Euclidean distance between the template fault signal and the signal to be recognized is used to detect the fault. The simulation test shows that the recognition capability of the hardware-based diagnostic system is close to that of a software-based one.