Machine learning: approaches to predicting reliability and developing maintenance strategies

Current approaches to maintenance of rolling stock bogies are focused on compliance to wear limits as stipulated by OEM specifications. OEM recommendations are critical to providing an industry wide approach to safety and compliance. These are not operation specific and are often not the most cost-effective solutions. A system approach to reliability is an established approach that is applied in less complex systems where the relationships between components are well defined with historical data and predictable conditions. Extending this approach to more complex multi-variate systems where many relationships are not intuitively obvious or mathematically defined presents a challenge. Machine learning techniques have been applied to address such problems with examples in image recognition, tool wear prediction using multiple sensory inputs and estimating railway bogie wear using vibration inputs. [8,9,10] The aim of the study is to extend and adapt machine-learning techniques to the area of developing maintenance strategies for optimal business benefit with a specific focus on railway bogie maintenance. This study aims to present an insight into the variables, which includes bogie tracking condition affecting track side wear rate. A finding is that an in-depth study of each independent variable’s individual impact is a necessary step to efficient modelling. These include track geometry, operating and bogie component wear variables. Track side wear, curve radius, superelevation and track top variance have been found to be significant predictors of track side wear rate. These impact predictions are not consistent between the different rail tracks and are not exhaustive. Specifically, the impact of bogie performance requires inclusion. Combining these variables mathematically using statistical inference and convolutional theory with maximum likelihood estimators would establish a predictor for side wear rate for the specific operation. The paper finally discusses the relationship of area wear rate to side wear rate and the influences of grinding frequency and rail material type.