Analýza zvukových záznamů rotačních systémů pomocí algoritmů umělé inteligence

Abstract

This dissertation deals with the problem of quality control of rotating devices based on their sound expression. The scope of the thesis is to investigate the possibilities of using artificial neural network algorithm for products classification. Based on the analysis of real measurement samples, synthetic data has been generated in sufficient quantity to faithfully mimic real measurement conditions in manufacturing plants. The actual research was then performed on this data, with the main objective of finding the ideal transformation of audio recordings into image representations that are fed to the input of convolutional neural networks. A number of experiments, which focused on transforming audio signals into image representations with using of multiple transformation operations with different parameters, were performed in this work. Convolutional neural networks were then trained on the transformed data to classify the recordings according to the type of fault. The performances of the obtained classifiers were then verified on the test data, and their qualities were compared with each other to find the optimal representation of audio recordings for processing by convolutional neural networks.