High-quality fault samples are regarded as the foundation of data-driven high-speed train bearing fault diagnosis. An adaptive multi-timescale fault sample integration method based on long short-term memory network (...
详细信息
High-quality fault samples are regarded as the foundation of data-driven high-speed train bearing fault diagnosis. An adaptive multi-timescale fault sample integration method based on long short-term memory network (LSTM) is proposed to address the problem of insufficient multi-timescale fault samples during the degradation of bearing performance. First, we used particle swarm optimization (PSO) to extract multiple hyper-parameters from the LSTM network and constructed the PSO-LSTM prediction models, which can predict fault samples. Thereafter, we generated high-accuracy prediction PSO-LSTM models based on multiple timescales. Finally, multiple PSO-LSTM prediction models are integrated by using the Bagging algorithm. We compared the fault diagnostic rates of the integrated data model with the raw data using classic fault diagnosis algorithms. Experimental results indicated that the fault diagnostic rate of the integrated data is higher than that of the raw data.
暂无评论