Fault Diagnosis of Rotating Machines Based on Combination of One‐Dimensional Convolutional Neural Network and Long Short‐Term Memory in Variable Working Conditions

Deep learning models, particularly one‐dimensional convolutional neural networks (1D CNNs), have shown great potential in the fault diagnosis of rotating machines. However, standard 1D CNNs face challenges in capturing long‐term dependencies in time series data. To overcome this, various studies have proposed hybrid models that combine 1D CNN with long short‐term memory (LSTM) networks. Despite these improvements, current 1D CNN‐LSTM methods still encounter issues such as low diagnostic accuracy, limited adaptability to changing conditions, poor generalization, and difficulty detecting compound faults. In response to these issues, this paper presents an improved hybrid fault diagnosis method based on a combination of 1D CNN and LSTM for rotating machines. In this approach, the 1D CNN extracts meaningful features from raw vibration data, which are then processed by the LSTM to model long‐term dependencies within the feature set. The experimental results show diagnosis accuracies of 99.57% for bearing data and 99.15% for the gearbox data, outperforming existing methods. These findings support the practical applicability of the proposed 1D CNN‐LSTM in real‐world industrial applications.