Compact Convolutional Transformer for Bearing Remaining Useful Life Prediction

An accurate prediction of bearing remaining useful life (RUL) has become increasingly important for equipment maintenance with the development of monitoring technology and deep learning (DL). Although Transformers are currently the most commonly used unique learning algorithms for sequential data, concerns about their computational efficiency and cost exist. In this regard, Compact Convolutional Transformers (CCT) have emerged as a promising alternative that employs sequence pooling and replaces patch embedding with convolutional embedding to enhance computational efficiency while maintaining high prediction accuracy with smaller model sizes. This study proposes an RUL prediction modeling approach that utilizes the Continuous Wavelet Transform (CWT) to transform time-frequency domain features into images, subsequently fed into CCT to establish a highly accurate prediction model for the RUL of bearings. This study conducted experiments using the XJTU-SY rolling bearing dataset. The performance was evaluated in terms of root mean square error (RMSE) and maximum absolute error (MAE) by modifying the layer configuration and comparing with other state-of-the-art algorithms.