🎯 Executive Summary
Detecting cracks and fractures in train wagon wheels from images during short stops — preventing rail accidents and cutting manual‑inspection cost.
💰 Estimated ROI: Several‑fold lower rail‑accident risk, savings on human inspectors, more useful wagon time in the network.
🏭 Target Industries: Rail transport, train‑fleet maintenance
⏱ Implementation Time: 8‑week POC, 4–6‑month station deployment
📊 Key KPIs: Crack‑detection rate, false‑negative rate, inspection time, accidents prevented
📋 Technical Project Details
This system is designed for continuous monitoring of wagon condition and automatic detection of wheel fractures or failures in real time. The main challenge was analyzing the large, continuous volume of data obtained from vibration and accelerometer sensors along the route, which had to be examined in real time so that potential damage is detected before an accident occurs.
Leveraging machine‑learning models and signal‑analysis algorithms, the system processes the vibration data collected from the sensors and identifies abnormal patterns that indicate a fracture or abnormal wear. This analysis is performed online and, when abnormal behavior is detected, alerts are sent to the maintenance unit so that the necessary actions are taken before a failure occurs.
In the development stage, the focus was on designing anomaly‑detection algorithms and extracting key features from the vibration signals. Hybrid models based on the Fourier Transform and Wavelet Analysis were used to accurately distinguish between ambient noise and abnormal vibrations caused by fractures. This combination raised detection accuracy to over 95%.
The analysis results are displayed graphically through a monitoring dashboard in the form of a health index for each wagon. This dashboard lets maintenance managers make preventive decisions and avoid sudden train stoppages.
Ultimately, the intelligent wheel‑fracture‑detection system is an effective step toward increasing rail‑fleet safety and reducing emergency‑repair costs. By continuously analyzing sensor data and predicting failures, this AI‑based approach paves the way from traditional maintenance to predictive maintenance and significantly increases operational productivity.