Support vector regression-based grid localization method for acoustic emission sources from Chinese fir boards

Wood is an anisotropic composite material, whose variation can make it difficult to locate surface damage using non-destructive testing. In order to solve the problem of sound source localization on the surface of wood, this study used a first localization method combining grid-based feature mapping and machine learning. Chinese fir boards (Cunninghamia lanceolata) were divided into a grid and acoustic emission signals were generated through a pencil-lead break test. These signals were processed using wavelet packet decomposition (WPD) to create a database of energy feature vectors. Localization was then achieved by applying support vector regression (SVR), which compared the feature vectors from the experimental points with those in the database to determine the sound source location. The average absolute error of this localization method was 7.51 mm, the average relative error was 3.79%, and the positioning accuracy was 91.84%, which can effectively locate the sound source on the wood surface.