Machine Learning for NDE Inspection

Aim and objectives of project:  

The main aim of this project is to tackle the automation of NDT ultrasonic data interpretation from Carbon Fibre Reinforced Polymers, with the use of AI technologies for defect detection and characterisation.  

What NDT problem I am working to solve:  

Carbon Fibre Reinforced Polymers (CFRPs) are increasingly used in both the civilian and military aerospace industries due to their physical properties such as high specific strength. However, the increased use of composite components has led to a significant requirement for Non-Destructive Testing (NDT) inspection during manufacturing, with ultrasonic testing (UT) being the most widely used. This testing is challenging to automate and can present a significant bottleneck for manufacturing at scale. The deployment of probes can be automated with robotic inspection however, this produces large amounts of data. In industry, this data is still primarily interpretated by skilled operators. Along with being a time intensive process, the requirement of a human operator also introduces potential for human error. By automating the interpretation of this data with Machine Learning we can reduce the burden on inspection for NDT operators and reduce the manufacturing bottleneck.  

Progress so far:  

So far research have been done into the best ways to generate synthetic training data, to overcome the sparsity of data available for ML training. High levels of detection accuracy of defects using Convolutional Neural Networks working with 2D and 3D ultrasonic data has been demonstrated on manufactured samples. Work is underway to include defect classification to turn this into a more challenging multiclass problem.