A novel multi-scale CNN and attention mechanism method with multi-sensor signal for remaining useful life prediction. (July 2022)

• Record Type: Journal Article
• Authors: Xu, Xingwei; Li, Xiang; Ming, Weiwei; Chen, Ming
• Is Part Of: Computers & industrial engineering. Volume 169(2022)
• Publication Date: 2022-07
• Subjects: Remaining useful life -- Deep learning -- Tool wear prediction -- Turbofan RUL prediction; Engineering -- Data processing -- Periodicals; Industrial engineering -- Periodicals; 620.00285
• DOI: 10.1016/j.cie.2022.108204
• Languages: English
• ISSNs: 0360-8352

Abstract

Highlights: A novel pre-processing method is proposed for remaining useful life prediction. Multi-scale convolutional neural network and attention mechanism are developed. The experimental results verify the validity of the proposed method. Abstract: Remaining useful life prediction is crucial in smart manufacturing systems due to many advantages of early prognostics, i.e., downtime reduction, service time prolongation, ultimate work efficiency improvement, and cost-saving. However, the conventional methods highly depend on feature selection and extraction, the accuracy and generalization can not be guaranteed. Inspired by the development of deep learning, a new method is proposed. Firstly, a parallel one-dimensional convolutional neural network (CNN)and the pooling layer were developed to extract and fuse features from the multiple signals. The dilated convolution with the residual connection and attention mechanism were specially developed to deal with the features from the pooling layers. After that, the regression layer was designed to generate the remaining useful life (RUL). Moreover, to verify the prognostic performance of the proposed method, two experiments, including cutting tool wear prediction and turbofan engine RUL prediction, were conducted. The proposed model was compared with the related works, and the results showed that the new method was more robust and accurate than currently published methods.