Fault Diagnosis of TRD Machine Based on TimeGAN-LSTM-LightGBM

Le résumé fourni par la source

The construction of the TRD (Trench Cutting Remixing Deep Wall) machine faces many challenges, such as complex working conditions and numerous disturbance factors. Its fault conditions often manifest as a combination of multiple operational parameters. Due to the requirement that frequent faults are not allowed during construction, fault samples are typically difficult to obtain, which poses challenges for training fault diagnosis models. Additionally, if faults and their types cannot be diagnosed promptly and accurately, significant property losses and casualties may occur. To address the scarcity of fault samples for the TRD machine and leverage the temporal characteristics of faults, a fault diagnosis method based on the TimeGAN-LSTM-LightGBM model is proposed. First, the collected raw time-series dataset is cleaned and used to train the TimeGAN model with three types of fault samples for sample expansion. Next, a three-layer stacked LSTM network is employed to extract temporal features. Finally, the generated samples are used to train the LightGBM model for fault diagnosis and performance metric analysis. An experimental study was conducted using the operational data from a selected TRD machine over the past year, following the above steps. The results demonstrate that the proposed model exhibits good diagnostic performance and can detect a higher proportion of fault samples, thereby verifying the effectiveness of the diagnosis method.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.