Cubic SVM neural classification algorithm for Self-excited Acoustical System
Krzysztof Lalik, Mateusz Kozek · 2020
This paper proposes a special classification system based on artificial neural networks. The algorithm was used to interpret the results for the Self-Excited Acoustical System (SAS) for ultrasonic stress measurement in elastic structures. The results obtained with the SAS system were transformed by Short-Timed Fourier Transform (STFT), and the resulting characterization images were used to train the artificial neural network using the Cubic SVM (Support Vector Machine) algorithm. Trained ANN was then used to classify materials on the basis of time-frequency characteristics. The article shows the principle of operation of the SAS system for materials such as stone, metal and composite material. The theoretical basis for usage of time-frequency transformations was presented, as well as the principle of operation of the Cubic SVM algorithm for classification tasks.