A review of rolling bearing fault diagnosis: data preprocessing and model optimization

Abstract Rolling bearing fault diagnosis is a critical process for ensuring the safe and efficient operation of rotating machinery. With the rapid advancements in artificial intelligence technologies, data-driven approaches have offered new perspectives for rolling bearing fault diagnosis and have significantly advanced progress in this field. However, most existing review articles primarily focus on specific models or methods, such as individual deep learning architectures, or particular signal processing techniques, but they often lack a systematic summary of optimization strategies throughout the diagnostic process. As a result, researchers are often limited to particular models or techniques when studying fault diagnosis methods, making it difficult to gain a comprehensive understanding of the role of various optimization strategies. This limitation hinders both method selection and innovation. To address this issue, this paper systematically reviews the research progress related to measurement optimization strategies for rolling bearing fault diagnosis. It conducts a comprehensive analysis from two perspectives: data preprocessing and model algorithm optimization. First, this paper highlights the measurement optimization strategies in the data preprocessing stage, including data acquisition, signal denoising, data augmentation, and feature extraction. This provides reliable support for establishing a high-quality data foundation. Subsequently, the latest advancements in model algorithm optimization strategies are thoroughly summarized, encompassing both machine learning and deep learning. Detailed analyses are conducted on the critical roles of hyperparameter tuning, network structure design, and training strategy optimization in enhancing model performance. Additionally, the potential of emerging technologies such as transfer learning and model integration techniques is discussed, focusing on their capacity to improve model generalization and adaptability to complex operating conditions. Finally, thelimitations of current research are examined, and future development directions are proposed based on practical application requirements. These findings provide a reference for optimizing measurement strategies in rolling bearing fault diagnosis.