Hierarchical Fault Diagnosis Method for Piezoresistive Pressure Sensor Based on GAF and CNN-SVM

Abstract

The requirements for reliability of piezoresistive pressure sensor in modern society are getting higher and higher because of wide application range of sensor and complex working environments. However, due to material properties and working principle (Wheatstone bridge), the failure caused by the degradation of different internal components in piezoresistive pressure sensors exhibits high degree of approximation, mainly reflected in the change in sensor output sensitivity coefficient and signal nonlinear distortion. This brings some challenges to the fault diagnosis of piezoresistive pressure sensor, including how to effectively extract features that can characterize different faults, how to accurately identify faults, and how to judge the severity of corresponding faults. To solve the above crucial problems, a hierarchical fault diagnosis (HFD) method based on Gramian angular fields (GAFs) and convolutional neural networks constructed with support vector machine (CNN-SVM) is presented in this article. First, the piezoresistive pressure sensor fault types are defined according to the functional area. Second, GAF is adopted to encode original output signal of the sensor, which can solve the problem of weak fault features in the original signal. Third, CNN is used to extract the features of Gramian angular summation field (GASF) images transformed by GAF. Moreover, SVM is used as the classifier connected to the flattening layer of CNN. Final, an HFD architecture is proposed to realize fault identification and fault severity judgment. The experimental results indicate the method in this article can effectively extract features, accurately identify different faults of the pressure sensor, and judge the severity of the fault. The accuracy of fault identification achieves 99.34% and the highest accuracy of severity judgment achieves 98%. It proves that the method proposed in this article is applicable and efficient for industrial application.