Fault Diagnosis of Wearable Temperature Sensors Based on Multi-scale Feature Extraction

Intelligent fault diagnosis is often used to improve the reliability, safety and stability of wearable sensors. In order to avoid the interference of expert experience to the fault diagnosis system, a multi-scale convolutional neural network model (MSCNN) is proposed. First, a novel MSCNN model is established to extract the deep features of the input data through convolution kernels of different sizes. Then, to improve the diagnostic efficiency of this method, a global average pooling layer is introduced to sub-sample the input features. Finally, the extracted sparse feature maps are input into the softmax classifier for fault classification. This method is applied to the diagnosis of wearable temperature sensor signals. The results show that this method has higher classification accuracy and faster efficiency than existing fault diagnosis methods such as artificial neural networks (ANN), traditional CNN, stackable automatic encoder (SAE) and support vector machines (SVM).