HRS
HRS means High Resolution System
We addressed and resolved the problem of rail corrugation with the data acquired by the Wave Partner II product produced by ADTS company and RVC1000. Driven by our innovative spirit, we wondered if there might be an alternative approach using different technologies.
Indeed, there is one. Inspired by published studies “Rail Corrugation Detection and Characterization Using Computer Vision” and “A cyber-enabled visual inspection system for rail corrugation” , it has emerged that high-resolution image processing can be used to detect the surfaces of rails where defects exist and even estimate the severity of these defects in terms of wavelength. Therefore, given the type of data required, it becomes evident that our product HRS meets the requirements and has been chosen to be applied in this context.
Using HRS instead of Wave Partner II would allow those who need to carry out lengthy diagnostics to shorten their acquisition times. In fact, the latter enables image acquisition at a resolution of 0.02mm per pixel at speeds of up to 100 km/h.
We opt for spectral analysis because it enhances robustness against variations in lighting and scale “Mel-cepstral methods for image feature extraction“. After selecting the portion of the rail to be analyzed, fundamental features in the frequency domain are extracted through Fourier transformation and then collected along with corresponding image portions.
Thanks to the expertise of RMT engineers, starting from the dataset described above, it was possible to leverage a neural network model trained as a multi-class classifier. Specifically, a ResNet18 “Deep Residual Learning for Image Recognition” was implemented, initialized with weights from ImageNet.
It’s called multi-class because the network is capable of recognizing when the railway track is affected by corrugation, when it has welds between various sections, and when there are other types of defects.
Below is a segment of data acquisition processed with our solution: