Study on Classification of Waveforms Using Machine Learning

Abstract

Concrete structures such as buildings and bridges that were intensively constructed during the high-growth period are now 50 years old, and the progression of age-related deterioration has become a major problem. This trend is becoming stronger with each passing year. Therefore, there is a need for a method to understand the state of deterioration of these structures and to maintain and manage them in a rational and effective manner. In this study, we examined a method for estimating the presence or absence of damage and the state of deterioration by classification, using wood as the test specimen, applying wavelet transform to the vibration waveform obtained by striking it, and training the obtained wavelet image with a convolutional neural network (CNN). As a result of performing a cut-out process of the vibration waveform to increase accuracy, the inference accuracy was improved from about 50% to about 80%.