Vibration Analysis to Detect Anomalies on Railway Track Using Unsupervised Machine Learning

Efficient railway maintenance is vital for a well-functioning transportation system. Indonesian Law No. 23 of 2007 mandates adherence to railway infrastructure maintenance standards carried out by qualified personnel. Damaged rails lead to disruptive vibrations, necessitating rail vibration detectors for assessment. This study employs smartphone-linked accelerometers to gather vibration data from miniature rails, simulating eight rail conditions, including normal and abnormal scenarios, using Phyphox. The research aims to develop a clustering approach for effective damage detection across diverse railway conditions. By utilizing K-Means Clustering and manual statistical analyses, distinct vibration patterns corresponding to different damage levels are identified. Machine learning experiments reveal optimal clustering with data variations up to three, as higher variations yield multiclass misclassification errors. This study demonstrates K-Means Clustering's efficacy in categorizing rail damage patterns and emphasizes limiting data variations to enhance accuracy.