Abstract:

A fault diagnosis method for the transmission system of lifting amusement rides was proposed to overcome the limitations of traditional fault diagnosis methods in handling complex signal features under specific operating conditions. First, vibration signals collected from the test bench were converted into images rich in time-frequency features using CWT to preserve detailed information in the time-frequency domain. Second, multilevel distributed fault-related features were extracted using ResNet. Third, a TFA module was introduced to selectively enhance time-frequency features, thereby improving the model's ability to capture critical information from complex time-frequency features. Finally, experiments were conducted on the test bench for the transmission system of lifting amusement rides to diagnose four types of faults at three locations. The results demonstrate that the fault recognition accuracies of the proposed method at the three locations reach 95.35%, 95.97%, and 99.15%, respectively. Substantial improvements in recognition metrics are achieved by the proposed method compared with multiple deep learning-based fault diagnosis models. The features extracted by the model are visualized after dimensionality reduction using t-distributed stochastic neighbor embedding (t-SNE), revealing an overall clustered distribution of samples under different operating conditions.

Key words: continuous wavelet transform (CWT), time-frequency attention (TFA), residual network (ResNet), transmission system of amusement ride, fault diagnosis