Smart Hybrid Approach for Defect Detection Based on Analysis of System Entropy (DDetect)

Project no.: PP-91M/19

Project description:

Scientists and researchers from the Centre for Nonlinear Systems have experience in investigating models of real-world systems. Real data which was obtained experimentally and describes physical and physiological processes have been investigated during this project. Detection of non-typical or defect related measurements is one of the most important problems in engineering, medicine and other areas. In the first scenario the existence of a fault is identified and further degradation of the system is prevented while in the second – the illness is diagnosed. Constant increase in amounts of diagnostic and measurement data implies the need of automated and smart solutions for the detection of defects. Novel aspect of this project is the joining of artificial intelligence and methods of analysis of nonlinear dynamical systems. The data was analyzed not directly but by performing computations, during which representative two-dimensional digital images based on permutation entropy, Wada characteristics, time averaged geometric moiré were constructed, first. Thus the problem of detection of defects in signals was transformed to the problem of image classification and/or identification. This problem was dealt with methods of artificial intelligence during this project. Additionally, smart approach for detection of defects without prior knowledge of their existence was developed.

Project funding:

KTU R&D&I Fund


Project results:

2D digital images were generated based on permutation entropy as well as other techniques and were used for rolling bearing fault detection. The algorithm itself is novel, it transforms time series into 2D digital images and further processes them with neural networks combining elements of AI in decision making process.
Analysis of biomedical signals with the aim of early diagnostic of illnesses was also performed. The data of sudden cardiac arrest was analyzed, the change in ECG clusters were depicted. Visual evaluation alone enables an expert to notice a change in the extracted feature (which sometimes is not visible in the original time series) and in turn decide on preventative measures if needed.
The concept of defect detection was fully formulated, methodology was created, algorithms were created and coded in software and tested on real experimental data. 2 scientific papers were published (in journals Optics and Lasers in Engineering and Entropy), 2 presentations in international conferences (with printed conference material) were made.

Period of project implementation: 2019-04-01 - 2019-12-31

Head:
Mantas Landauskas

Duration:
2019 - 2019

Department:
Department of Mathematical Modelling, Faculty of Mathematics and Natural Sciences