Developing a Hybrid Expert/Data-Driven Health Index for Railway Axleboxes Using Auto-Encoder Neural Networks

The railway axlebox is a critical safety component in rolling stock. It bears the weight of the train, and minimises the friction with the rotating axle. Its failure in service might cause derailment. Therefore, its maintenance is decisive. This article presents a methodology to create a standard Health Index that blends the hard threshold-based expert criteria developed at the workshop, with a record of actual vibration data, in order to increase the flexibility of its diagnostic and provide a soft granular feedback. The proposed method makes use of an auto-encoder neural network to accommodate two different approaches: one based on the principal component of a set of vibration features, and the other based on an anomaly detection scheme operating directly at the vibration signal level. The results of the analysis show that the former strategy provides a smooth uncertain index, while the latter yields a sharp sensitive index. Such different diagnosis designs are regarded as an opportunity to better fit the vision of the maintenance course of action.