Attention–Slope Entropy and Its Derivative-Driven Multiscale Version for Fault Diagnosis of Rail Transit Point Machines

Point machines play an important role in rail transit. However, they are prone to faults due to harsh working environment. This paper presents a vibration signal-based fault diagnosis method for point machines. First, considering the non-linear and non-stationary characteristics of the monitored signals, to achieve fast and effective extraction of non-linear features, a novel entropy named attention-slope entropy (ASE) is developed, which simultaneously considers the local extremum and their slope information. Then, regarding the issue of information loss in the classical coarse-graining process, multi-order derivatives-driven multi-scale algorithm is proposed, helping enrich the non-linearity information characterization. Finally, the effectiveness and superiority of the presented ASE and its derivative-driven multi-scale version are verified on logistic map and real dataset of rail transit point machines. The diagnosis accuracies of normal-reverse and reverse-normal switching directions reach 99.43% and 98.86%, respectively, demonstrating its effectiveness.