Detection of Railroad Anomalies using Machine Learning Approach

Abstract

Maintenance of assets owned by an organization or company is an activity that will never stop. From the time of implementation, maintenance will be more optimal if it is carried out before the asset is in a damaged condition or cannot operate. Pro-active repair model is proven to reduce 15-60% of operational costs. The existence of technology and computing models currently supports big data processing, both in the form of transactional data, historical data and statistical data. The asset maintenance cycle transformed into an autonomous and integrated system, will assist in the decision-making process. A machine learning approach that is supported by big data analysis is one solution that can realize the predictive maintenance process. To accurately predict the condition of critical components, it can be started with data collection, followed by detecting normal and abnormal behavior, and continued by training algorithms to make predictions. Detection of railroad anomalies is used as the initial process in the predictive maintenance of railroads. The process of detecting railroad anomalies can be done by comparing the lateral, longitudinal and vertical acceleration from the sensing results through the accelerometers on both sides of the train wheels. Differences will pay attention to the data acceleration draft rail geometry either angkatan or listringan. The results of rail anomaly detection will indicate the rail maintenance process that can be carried out immediately.