FAULT DETECTION ON RAILWAYS TURNOUT SYSTEM POINT MACHINE SIMULATOR USING INTERNET OF THINGS AND TINY MACHINE LEARNING TO SUPPORT PREDICTIVE MAINTENANCE

• Main Author: Alfarezi Pratama, Zastra
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/68740
• Institution: Institut Teknologi Bandung

Signaling equipment in the railway industry has a vital role in ensuring trains' safety and smooth operation. A turnout system is a train signaling equipment that plays a critical role in train operations. Problems such as delays, train collisions, and train wheels out from the track are problems caused by fault or failure of the turnout system. The point machine is one of the critical components and is often the cause of operational failure in the turnout system. Therefore, a predictive maintenance (PdM) strategy is needed to detect faults, reduce equipment downtime and reduce equipment maintenance costs while knowing the useful life of equipment as an early warning, especially on point machines. PdM can be combined with the Internet of Things (IoT), but the use of IoT in PdM has other impacts, such as obtaining, storing, maintaining, and analyzing data requires costs that increase as the number of data increases. In addition, latency, payload, and data transfer rate are challenges for PdM implementation using IoT on point machines. Therefore, an approach is needed to balance the speed of data transmission and storage and provide up-to-date information about the status of the point machine. In addition, PdM is a big concept with steps that must be done before making predictions or making decisions, one of which is the fault detection phase. In this study, the fault detection phase was simulated on the point machine using a DC motor by combining IoT technology and Tiny Machine Learning (TinyML) to find a balance between storage efficiency and the data collected while providing fast information. The simulation results show that using IoT and TinyML can reduce 60.9% of data storage on the Thingsboard platform while maintaining a transfer rate of ? 1 second.