Condition Monitoring of Rolling Stock Supported by Artificial Intelligence Technique

The increasing use of condition monitoring of the railway infrastructure has led railway companies to take advantage of artificial intelligence (AI) technologies. The main goal of this research is to provide an unsupervised method for identifying railway wheel flats. The two-step procedure based on the automatic damage identification algorithm evaluates the acceleration on rails during the passage of traffic loads. The first step consists of assessing baseline responses from the rail to create a confidence boundary, while the second step involves evaluating damages according to their severity levels. The proposed procedure is based on a machine learning methodology and involves the following steps: (i) data acquisition from sensors, (ii) feature extraction from acquired responses using an AR (Auto Regressive) model, (iii) feature normalization using principal component analysis, (iv) data fusion and (v) unsupervised feature classification by implementing outlier and cluster analyses. This research shows that the proposed method is reliable and cost-effective and can successfully identify wheel flats considering different train speeds.