An end-to-end harmful object identification method for sizer crusher based on time series classification and deep learning. (April 2023)

• Record Type: Journal Article
• Authors: Bi, Yankun; Pan, Yongtai; Yu, Chao; Wang, Mengchao; Cui, Tongyu
• Is Part Of: Engineering applications of artificial intelligence. Volume 120(2023)
• Publication Date: 2023-04
• Subjects: Harmful object identification -- Time series classification -- End-to-end -- Deep learning -- Sizer; Engineering -- Data processing -- Periodicals; Artificial intelligence -- Periodicals; Expert systems (Computer science) -- Periodicals; Ingénierie -- Informatique -- Périodiques; Intelligence artificielle -- Périodiques; Systèmes experts (Informatique) -- Périodiques; Artificial intelligence; Engineering -- Data processing; Expert systems (Computer science); Periodicals; 620.00285
• DOI: 10.1016/j.engappai.2023.105883
• Languages: English
• ISSNs: 0952-1976

Abstract

Abstract: Nowadays, with a maximum annual capacity of 30 million tons in one large coal preparation plant, the corresponding belt speed can reach 7 m/s and the coal layer thickness will be more than 500 mm. These lead to harmful components, such as iron, gangue, and wood, being mixed into the coal and seriously damaging parts of sizer crushers (sizers) during the crushing process. The traditional method for harmful object identification and fault diagnosis is manual feature extraction (MFE), which has drawbacks such as dependence on experience, poor stability, and inflexibility. In this paper, an end-to-end (E2E) harmful object identification model was proposed for sizers based on time series classification (TSC) and deep learning. The model learned features directly from the one-dimensional multi-channel raw signals of sound pressure and vibration without MFE and has been tested on experimental and industrial datasets to validate its effectiveness and flexibility. Results showed that the E2E method proposed in this paper can reach a classification accuracy of 87.42% for feed including coal, iron, and wood. In particular, the identification ability for iron can be outstanding. The industrial testing indicated that the identification precision of iron could be 99.10%. To further improve the model's classification accuracy, the identification of mixture materials will be a key focus in the future.