BUP43 - CT-INR

Computed tomography (CT) can be used to precisely measure components using X-rays. In addition to statements such as length, angle or similar about the component, internal structures such as pores can also be analyzed without destroying the component. CT therefore has a number of advantages as a measurement technique. However, the complex physical measurement principles of CT systems require extensive expert knowledge for their operation. This makes the entire measurement process very time-consuming and often leads to less than optimal measurement results. 

Aim

The aim of this project is to develop a simulator for industrial computer tomographs that can be used to design measurements. Current simulators do not meet the required simulation accuracy and computing time, which is why machine learning is to be used for simulation. For this purpose, approaches from the area of "Implicit Neural Representations" are used, which have proven themselves for related problems from signal processing.

Approach

• Development of a real data set 

• Development of a suitable supervised learning architecture using expert knowledge and approaches from AutoML

• Validation of the results and comparison with existing simulators 

Benefits

Added value for production/production engineers: The advantages of industrial computed tomography as a measurement technology cannot currently be fully exploited. The reason for this is the complex physical interactions that can currently only be addressed with comprehensive expert knowledge. This complexity can be reduced with fast and accurate simulation of these systems. Industrial computed tomography can thus be enabled for widespread use in quality assurance.