Image Analysis for Complex Data in Non Destructive Testing
This project will develop image analysis techniques for rail axle inspection as a visual inspection Non-destructive Testing (NDT) tool using microscope images.
The solution is to design a machine learning based artificial intelligent system that can automatically detect the pits and cracks on any microscope images from rail axle surface. It can give the detailed measurement of the numbers, locations and sizes of pits and cracks that are the very important information for industrial inspection that currently is done by human inspectors. This A. I. system will save the workloads of the workers and give accurate information about the conditions of the rail axle surface. A big image database will be collected from the surface of rail axle and pits and cracks are labelled by inspection experts following the industrial standard. These image data and their labels will be used for training advanced neural network models for pattern recognition. Advanced feature extraction methods will be used for capture the detailed information on the pits and cracks locally and globally. Feature selection methods will be used for reducing the size of feature vectors.
Both supervised and unsupervised machine learning techniques will be used and fused in the final system. The model will be further developed in a way that all the new images can be added for updating the artificial intelligent system continuously and the model can be improved further and further. Finally, optimisation methods will be used to simplify the algorithms and models for possible fast implementation.
The developed NDT tool can be used for fast surface inspection for rail axle surface and produce the measurements and numbers needed. It will reduce human eye inspection with better accuracy and reduced time. The developed NDT tool should be easily used for many similar applications as well.