The analysis of the NDE measurement data and the subsequent interpretation to turn it into material quality decisions is often done manually by trained operators. Clearly, there is a strong interest within the Swedish aerospace manufacturing industry to decrease this human effort in data analysis in order to stay competitive. However, we predict that other manufacturing industries will have the similar needs due to digitalization, e.g. the ones relying on 3D printed metal parts (additive manufacturing).
A recent Vinnova Produktion 2030 funded project (Automatic inspection and other possibilities at the introduction of digital X-ray inspection in the manufacturing industry, Swerea KIMAB, GKN Aerospace, and Volvo Construction Equipment) found that scientific studies on X-ray inspection analysis in the last 15 years have aimed at a fully automatic methods, given an expected (fixed) input variability. However, in industry the unforeseen in terms of defects has to be considered during inspection of quality critical products; this was identified as the reason for the limited industrial implementation to date.
In this project we will instead explore semi-automatic approaches for data analysis rather than fully automatic ones. We will explore how an accurate and conservative estimation of confidences (how certain the algorithm is) combined with human operator interaction and active machine learning (how to effectively choose when to query the human) can solve this industrial problem.
Most of the automatic data analysis of X-ray images in industrial NDE applications can be divided into four steps: preprocessing, segmentation, feature extraction, and classification. In this project the focus is on the last steps.
NDE, as quality control in a production system, is an essential part of R&D driven high value manufacturing industries such as e.g. oil & gas and aerospace industry. The collected measurement data needs to be interpreted and translated to a decision in the production flow; the amount of such measurement data is already large and will grow even larger with the digitalization trend. To stay competitive, the amount of manual human interpretation work must therefore be reduced.
Forskningsmiljö / Institution
- Primus (KK-miljö)
- Institutionen för ingenjörsvetenskap
- Chalmers tekniska högskola