Modern train fleets have very demanding requirements in passenger safety, train service reliability and availability, comfort and life cycle costs. To reach these goals, maintenance intervals of more than thirty thousand kilometers besides serious failure-free objectives exceeding one and a half million kilometers are becoming a standard. This requires manufacturers to develop bold designs and to use advanced engineering tools for the Operations and Maintenance (O&M) of such trains. Condition Based Maintenance (CBM) solutions, using condition monitoring systems and advanced algorithms to detect commencing deterioration, may allow sufficient time for maintenance before serious failures can develop, which increases safety, reliability and availability while helping to reduce operating and maintenance expenses and the total cost of ownership.

This paper applies predictive analytics, big data processes and tools to design CBM Plans for train axle bearings, to increase both preventive maintenance (PM) intervals and dependability of the trains. The paper details how the machine learning predictive model is selected and how the model is trained with different data sets. Big data processes allow to test and accept a universal model per bearing position regardless the axle or train of the fleet, overcoming complexity that could be generated by the non-ergodicity of these assets. The originality of this work consists in the ability to identify bearing deterioration related anomalies, by an innovative modeling and prediction of axle bearing temperature using data analytics. Also, interpretation rules for early failure detection based on these advanced predictive analytics are compared to those already existing rules in the train on-board control monitoring system (TCMS) ensuring train’s safety. Conclusions of the work are related to the process followed and the validity of results.

现代火车车队对乘客安全，火车服务的可靠性和可用性，舒适性和生命周期成本有非常苛刻的要求。为了达到这些目标，除了严重的无故障目标超过一百五十万公里之外，超过三万公里的维护间隔也已成为标准。这就要求制造商开发大胆的设计并使用先进的工程工具来进行此类列车的运营和维护（O＆M）。基于状态的维护（CBM）解决方案，使用状态监视系统和高级算法来检测开始的恶化，可以留出足够的时间进行维护，以防止发生严重故障，从而提高安全性，可靠性和可用性，同时有助于减少运营和维护费用，并减少总支出。拥有成本。

本文运用预测分析，大数据流程和工具来设计火车车轴轴承的CBM计划，以增加火车的预防性维护（PM）间隔和可靠性。本文详细介绍了如何选择机器学习预测模型以及如何使用不同的数据集训练模型。大数据流程允许测试和接受每个轴承位置的通用模型，而无需考虑车队的轴或火车，从而克服了这些资产的非遍历性可能产生的复杂性。这项工作的独创性在于能够通过使用数据分析进行创新的建模和预测轴承温度来识别与轴承劣化相关的异常。也，将基于这些高级预测分析的早期故障检测解释规则与确保机车安全的列车车载控制监控系统（TCMS）中已存在的规则进行比较。工作的结论与所遵循的过程和结果的有效性有关。