DETECTION OF TRAIN WEIGHT AND SPEED BASED ON ACCELERATION RESPONSE ON THE RAIL USING MACHINE LEARNING

• Main Author: Limanza, Jefferson
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/80687
• Institution: Institut Teknologi Bandung

The weight and speed of a train are crucial parameters that affect the condition of tracks and trains. Non-compliance with weight and speed regulations can impact the health of various components. Weight measurement is typically performed when the train is stationary which leada to a decrease in operational efficiency. Therefore, this research explores methods for measuring the weight and speed of a train based on the vertical acceleration of the track when the train traverses it, using machine learning. The data processed by machine learning comes from dynamic simulations using Universal Mechanism, with the analyzed objects being a passenger train and a flexible track as a multibody system. The flexible track is given geometrical irregularities to repres ent real-world conditions. The dynamic response of the track is represented by vertical track acceleration and time. Machine learning input features include the thirty highest values of the absolute vertical track acceleration and the corresponding time. Meanwhile, output features include the weight and speed of the train. Nine different regression learning methods are employed in developing weight and speed prediction models to find the method with the lowest root mean square error value. Furthermore, the characteristics of both models are analyzed based on root mean square error (RMSE), mean absolute error (MAE), and standard deviation (SD) values. The best method obtained is the random forest regression learning method for predicting both weight and speed values of the train based on the vertical track acceleration response when traversed by the train. The weight prediction model has an RMSE value of 0.83 ton, while the speed prediction model has an RMSE value of 0.92 km/jam.