198 / 2024-09-01 20:36:04

Fault monitoring algorithm for trackside acoustic bearings based on LLM

Large Language Model (LLM), Doppler effect, microphone array, train bearing, wayside fault diagnosisIntroduction

Special Sessions > SS02 Artificial Intelligence in Condition Monitoring, Fault Diagnosis and Maintenance of Industrial Equipment

Abstract—With the launch of major strategic plans such as Made in China 2025, large-scale rotating machinery and equipment in the metallurgical industry have gradually transformed to intelligence. Strip is an essential material for automotive production, processing in the smart industry, and even in the aircraft and aerospace sectors, and bearings are the key equipment for plastic forming of strips. In order to realize the safe and efficient operation of strip rolling, it is particularly important to realize the status detection and fault diagnosis of rolling mill equipment. The vibration signals of different faulty bearings are obtained through the experimental rolling mill fault simulation test bench of the Engineering Center of Yanshan University, and the bearing fault identification under the condition of non-equilibrium training set is realized through various algorithms, focusing on the following contents: (1) Fault characterization and state evaluation of multi-row bearings in rolling mill based on multivariate multi-scale weighted arrangement entropy, and nonlinear dynamic fault characterization is carried out by multi-channel vibration signals in signal analysis according to the special working conditions of multi-row bearings in rolling mills, so as to realize the evaluation of bearing fault states. (2) Vibration signal processing and non-equilibrium dataset enhancement of adaptive multivariate variational modal decomposition and deep convolution generative adversarial networks. Modal decomposition is used to decompose and reconstruct the signal to remove invalid information from the signal. The generative adversarial network is used to augment the data of the non-equilibrium training set to improve the accuracy of the diagnostic model.