A device health state assessment method based on convolutional neural network and an improved hierarchical Softmax strategy

In the light of the problems of complex feature extraction and low accuracy of fine-grained classification in traditional industrial equipment health state assessment, this paper proposes an equipment health state assessment method based on convolutional neural networks and an improved hierarchical Softmax strategy (hierarchical softmax convolutional neural network, HSCNN). By using convolutional neural networks (CNN) to learn feature representations from equipment health state data and mining the intrinsic features of the equipment state, and by introducing a hierarchical Softmax strategy for multilevel classification of the equipment health state, the original fine-grained classification task is converted into a hierarchical decision-making process. A Huffman tree is then used to solve the problem of the traditional hierarchical Softmax strategy, which may lead to the waste of computational resources and the degradation of the model performance, and realize the efficient and accurate assessment of the equipment health state. The proposed method was employed to assess the operational health status of large-scale coal mining equipment. Our new method of assessing the health status of equipment based on CNN and the improved hierarchical Softmax strategy has a higher accuracy than the traditional methods employing multilayer perceptron (MLP), recurrent neural networks (RNN), CNN, hierarchircal Softmax convolutional neural networks (CNN-LSoftmax), and long short-term memory networks (LSTM).