A new BRTCN model for predicting discharge status of WEDM based on acoustic emission. (July 2022)

• Record Type: Journal Article
• Authors: Yang, Xingxin; Liu, Changhong; Peng, Lingxi; Peng, Shaohu; Zhang, Yongjun; Xie, Ning; Zhong, Ray Y.
• Is Part Of: Journal of manufacturing systems. Volume 64(2022)
• Page Start: 409
• Page End: 423
• Publication Date: 2022-07
• Subjects: Wire electrical discharge machining (WEDM) -- Acoustic emission (AE) -- Deep learning -- Batch Relevance Temporal Convolution Neural Network (BRTCN) -- Batch relevance (BR); Manufacturing processes -- Periodicals; Production engineering -- Data processing -- Periodicals; Robots, Industrial -- Periodicals; Production, Technique de la -- Informatique -- Périodiques; Robots industriels -- Périodiques; Electronic journals; 670.42
• DOI: 10.1016/j.jmsy.2022.07.003
• Languages: English
• ISSNs: 0278-6125

Abstract

Abstract: Machine status detection is very critical to ensure the smoothness of processing workpieces in terms of efficiency and quality. In light of the significance of acoustic emission (AE) signals for the monitoring of wire electrical discharge machining (WEDM) processes and the strong feature extraction ability of deep learning, this paper proposes a physical propagation mechanism of AE and develops an effective deep learning dual-input model called batch relevance temporal convolution neural network (BRTCN) with a new labeling method after analyzing the collected signals. BRTCN, mainly composed of noise extractor, encoder, batch relevance and decoder, is applied to build the AE model that can accurately predict discharge status. Through comparative experiments, the encoder of BRTCN is capable of extracting the AE local sequence features, grabbing long dependencies and reducing computational costs. It is found that the noise extractor in BRTCN is essential for the AE model to converge stably. This paper innovatively detects the discharge status of WEDM based on dual channel AE signals and the proposed BRTCN model examines the relationship between AE and pulse time series with low computation and high accuracy. Highlights: Monitoring discharge status in WEDM using dual-channel acoustic emission sensor. A novel Batch relevance temporal convolution neural network (BRTCN) is proposed. Decoupling dual-channel AE signals used to establish relationship. A new data set labeling method is developed to train the model effectively.