Service provider for industrial computed tomography

The use of artificial intelligence in NDT


The use of artificial intelligence (AI) ensures efficiency, cost savings and error prevention in all application areas. These effects can also be achieved in the field of non-destructive testing (NDT) by means of automatic evaluation procedures in which components are tested for defects.

Methods which automatically evaluate data from non-destructive testing reach their limits in the defect testing of components due to many special cases and correspondingly complex algorithms - in many cases the results do not show a clearly definable separation between a "permissible" (IO) and a "prohibited" (NIO) component. A manual follow-up check must therefore often be carried out afterwards. The time and cost savings aimed at by the use of an automatic evaluation are only partially achieved or completely nullified.

A typical example of this is the automatic evaluation of a test according to a limit sample catalogue. Because they are relatively easy to create, boundary sample catalogues are well suited for defining the desired quality. However, the requirements described in this way can hardly be implemented in an evaluation algorithm. For the quality assessment by means of modern AI, boundary sample catalogues are almost optimally suited. 

Machine learning is used in the NDT field for the detection of pores and burrs as well as for the evaluation of solder joints including a classification of components in IO/ NIO. It is also possible to quantify the size of disparities.

 

Lötstellenbewertung Bauteil IO

Solder joint evaluation - component IO

Lötstellenbewertung Bauteil NIO

Solder joint evaluation - component NIO

Your advantages with us

With AI, quick initial success can be achieved, but a high quality of assessment in the long term requires experience in the use of machine learning techniques. The provision of sufficient training information (= images) is particularly critical, because most of the time there are mostly IO images, but AI also needs NIO images for its learning process.
 
Microvista is regarded as a pioneer in the field of automated analysis and evaluation routines of CT scan data in the industrial sector and has developed reliable tools with which powerful solutions that are sufficiently robust for industrial use can be supplied even with few output data.
 

We offer you a fast, automated AI procedure that is highly robust against fluctuations in the measurement process.
We are able to teach our machine learning approach flexibly and adapt it to different customer requirements.

How do we proceed?

In a first conversation we ask you for an exact description of the problem, e.g. a limit sample catalogue or similar. On the basis of this information we can also tell you whether our AI procedure is suitable for your request.

We then discuss how we can train our AI for your application. If you already have component images, you can simply provide us with them.
 

If no pictures are available, it is best to send us sample parts from which we can create matching pictures.

The image material is used to build the learning and testing basis for our machine learning. If you do not have enough pictures, it is part of our know-how to deal with this problem. Then the AI learns how to evaluate your images and we then test what has been learned. This process is repeated until there are no more errors in the test phase - what sounds like a lengthy process is usually carried out by us in just one week.
 
The resulting AI module is now integrated into our internal software, which does everything from recording the component number and other master data to creating a customised test report.
 
 

For many people, machine learning is a process that seems too abstract and can therefore arouse distrust. However, it has been proven that results obtained with artificial intelligence are equivalent or even better than those obtained with manual testing, as the error rate is reduced. However, automatic procedures must also be subjected to random sampling and compared with reference data in order to guarantee the quality of the evaluation or to adjust it if necessary.