At the Novosibirsk State Technical University (NETI) proposed its own version for determining defects in industrial products. We are talking about a software package using a digital twin, which will help to identify even minor shortcomings of objects without destroying them.
Alexander Peshkov, a graduate student of the Department of Computer Engineering of the Faculty of Automation and Computer Engineering of NSTU-NETI, developed a software package in the form of a micro service application in Java, Python, Matlab for solving computed tomography problems using a digital twin in order to detect small defects in typical industrial products. To detect defects, the method of non-destructive testing is used — computed tomography with hard X-rays.
"Based on the difference in projection data for the reference sample of the product (digital twin) and for the test sample, it is possible to quickly diagnose the presence of a defect, its shape and location," said Alexander Peshkov.
In addition, a priori information from the digital twin about possible types of defects is already used to train a neural network that takes raw projection data as input and outputs the type of defect — crack, detachment, dent, curvature, the presence of internal voids.
The method of industrial flaw detection developed at NSTU-NETI differs from domestic and foreign analogues by a systematic approach. "Algorithms have been developed in all analogues, but there are no systems, and we offer a software package where you can integrate separately with each module if you want. We restore only the area of interest to us — this allows us to diagnose defects of very small thickness, "the scientist said.
In detail, this works as follows: a product model is generated (for example, an element of a building structure), taken as a reference (this is a digital twin), and a model of an instance of this product with a defect is also generated (for example, with a random crack), which is subjected to examination. The digital twin is used as a reference sample against which defects can be effectively detected.
According to Alexander Peshkov, currently there is no need to scan anything: the work is done with a scanning simulation. In the future, a module will be developed that will build a digital twin based on 3D drawings of the product and the results of its scanning. The operator responsible for quality control at the production site will receive an image with a defect and a conclusion of artificial intelligence about this defect, which will allow the employee to quickly and more accurately make a decision on the rejection of products.