摘要
为了改变传统的基于软件的机械故障诊断模式,以及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识别的脉冲频率调制(PFM)模拟神经网络脉冲流VLSI电路。实现了一种脉冲流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,设计了进行主轴承磨损故障诊断的神经网络故障识别系统。利用含有故障信息的噪声信号代替传感器安装困难的基于振动信号的特征值提取,最后,根据代表待识别信号与标准故障模板之间欧氏距离的电路输出端电容电压值可以判断出故障类别。该电路具有较高的识别精度,可以实现噪声故障信号的实时在线识别。
?An improved Pulse Frequency Modulation (PFM) neural network VLSI circuit for fault diagnosis is presented in order to change the softwarebased fault diagnosis approach and facilitate the merits of neural network VLSI circuit. A new pulse stream digital/analogue based synapse multiplier/adder can be realized. The synapse weight values don't need learning, and it can also lessen the complexity of circuit. A neural network fault recognition system based on this circuit is designed for the abrasion fault diagnosis of main bearing. The noise signal including faults information, instead of vibrationbased method, is used to extract characteristic values. Finally, depending on each output capacitor voltage value of VLSI circuit that represents Euclid Distance between the corresponding standard fault template and test signal, the fault can be recognized. This circuit has a relatively high recognition precision and can realize a realtime fault diagnosis.
出处
《控制工程》
CSCD
2003年第4期289-292,共4页
Control Engineering of China
基金
国家自然科学基金资助项目(60274015)
国家863计划资助项目(2002AA412420)
