Perpustakaan Digital ITB

In 2022, PT KAI (Persero) experienced a rise in passenger and freight services, leading to increased equipment failures due to longer operating hours. Maintenance is crucial for railway point machines (RPM), which significantly impact train safety. Currently, fault diagnosis is conducted manually due to the lack of a condition monitoring instrument for RPM equipment, resulting in increased labor expenses, delays, inefficiencies, and reduced accuracy. This study proposed a vibration-based fault diagnosis application using deep learning method. The study utilized ESP32 with ADXL345 accelerometer as a low- cost vibration sensor and a single channel convolution neural network (1D-CNN) as a simple deep learning algorithm. The calibration results showed that the ADXL345 has a reading error in magnitude value when monitoring vibrations above 230 Hz due to aliasing effect. However, recognizing patterns in the data is more crucial in fault diagnosis applications. The monitored vibration data were visualized in a GUI using LabVIEW application, providing intuitive information for users to access and read vibration data related to RPM equipment working conditions. The study investigated vibration signals in eight working conditions of RPM equipment, collecting 816 samples in three categories. The models were then tested in five scenarios of vibration data, with each model tuned to find the best architecture network in the DL algorithm. The results showed that the 1D-CNN implemented triaxial axis vibration data had significantly high accuracy performance with low variant for classifying RPM equipment working conditions with 97.58% accuracy. In addition, the 1D-CNN architecture in the proposed method has simpler configurations and acceptable computational cost.