基于动态统计滤波与深度学习的智能故障诊断方法

电流信号具有易采集、不易受环境噪声影响的优点,为难以通过振动传感器采集信号的特殊设备提供了可行的监测诊断思路,但电流信号也存在故障特征难以提取等问题。为此,将改进的动态统计滤波与深度卷积神经网络(DCNN)结合,提出一种基于电流信号进行机械设备智能故障诊断的方法。引入综合信息量指标(SIpq)优化滤波效果,基于改进的动态统计滤波方法,使不同状态信号间的特征差异最大化,以提高状态识别精度;通过交替堆叠特征图尺寸不变的卷积层与逐层递减的池化层,构建DCNN,提取电流信号中的高维故障特征。将动态统计滤波后的特征增强图像输入DCNN,识别故障类型。为验证方法有效性,以不平衡、不对中、松动3种故障为对象进行故障类型识别,分析结果表明,所提方法可有效识别故障类型,与传统的ANN、CNN等其他方法对比具有较好的识别精度。

Intelligent fault diagnosis method based on dynamic statistical filtering and deep learning

Electrical current signal possesses the characteristics of being easy to collect and not easily affected by environmental noise, which provides a feasible monitoring and diagnosis idea for the special equipment that is difficult to collect signals with vibration sensor. However, the electrical current signal also has the problems of being difficult to extract fault features. An intelligent fault diagnosis method for mechanical equipment based on electrical current signal is proposed, which combines the improved dynamic statistical filtering and deep convolution neural network (DCNN). In order to improve the accuracy of state recognition, the integrated information quantity index (SIpq) is introduced to optimize the filtering effect, which can maximize the difference of the features among different state signals based on dynamic statistical filtering. Through alternately stacking the convolution layer with invariant size of the feature map and the pooling layer with decreased the size of the feature map layer by layer, the DCNN is constructed to extract the high dimension fault features in the electrical current signal step by step. The feature enhanced image samples after dynamic statistical filtering are input into the DCNN to identify the fault type. In order to verify the effectiveness of the proposed method, take three kinds of faults inclnding unbalance, misalignment and looseness of rotating machinery as objects, the fault type identification was carried out. The analysis results show that the proposed method can effectively identify the fault type. Compared with other methods such as ANN and CNN, the proposed method has better recognition accuracy.