Laser ultrasonic quantitative recognition based on wavelet packet fusion algorithm and SVM

With the development of industrialization and modern technology, laser ultrasonic technique is more and more used in aerospace, machinery and electronics, measurements, metallurgy chemical engineering, materials science, railway transportation, bridge engineering, etc. In order to maintain the excellent characteristics of new materials (such as thermal properties, mechanical properties, chemical properties and optical properties, material structure must be early diagnosed and monitored before properties change. Nondestructive Testing technology plays a great role in monitoring reliability of industry products. This paper proposes a new feature selection method based on wavelet packet algorithm, and applies SVM (support vector machine) for quantitative classification on the ultrasonic echo data generated by cracks in Laser ultrasonic experiment. By combining the nondestructive device and dimension reduction method in machine learning, this paper analyses the scatter plot of two cracks in 2d and the fitting surface in 3d and give the quantitative index for determining the performance of used methods.