基于深度学习的模拟电路故障诊断方法

Analog Circuit Fault Diagnosis Method Based on Deep Learning

针对模拟电路的故障诊断问题,提出了一种基于深度学习的故障诊断方法。首先测量模拟电路各个故障类别的脉冲响应数据,随后应用深度学习中深度信念网络方法进行特征提取,最后将提取的特征用于建立基于极端学习机的故障诊断模型,从而对模拟电路的各个故障类别进行区分。通过四运放双二阶高通滤波器电路的故障诊断实验对提出的故障诊断方法进行了验证。通过对比实验表明,提出的基于深度信念网络的故障特征提取方法明显优于传统的基于小波分析的故障特征提取方法,有助于提高模拟电路故障诊断正确率。

A fault diagnosis method based on deep learning is proposed for analog circuit fault diagnosis.Analog circuit impulse response signals are measured firstly,and then deep belief network method is used to extract features from the signals.Finally,an extreme learning machine based diagnosis model is constructed based on extracted features to identify different fault classes.Four-op-amp biquad highpass filter circuit fault diagnosis is performed to test the proposed fault diagnosis method.Meanwhile,the comparison result reflects that the proposed DBN based features extraction is superior to the traditional the traditional wavelet analysis based features extraction method,which is helpful in improving the analog circuit fault diagnosis accuracy.