Detection, identification and classification of defects using ANN and a robotic manipulator of 2 G.L (Kohonen and MLP algorithms)

Abstract

The ultrasonic inspection technique had a sustained growth since the 80’s. It has several advantages, compared with the contact technique. A flexible and low cost solution is presented based on virtual instrumentation for the servomechanism (manipulator) control of the ultrasound inspection transducer in the immersion technique. The developed system uses a personal computer (PC), a Windows Operating System, Virtual Instrumentation Software, DAQ cards and a GPIB card. As a solution to detection, classification and evaluation of defects an Artificial Neuronal Networks technique is proposed. It consists of characterization and interpretation of acoustic signals (echoes) acquired by the immersion ultrasonic inspection technique. Two neuronal networks are proposed: Kohonen and Multilayer Perceptron (MLP). With this techniques non-linear complex processes can be modeled with great precision. The 2-degree of freedom manipulator control, the data acquisition and the net training have been carried out in a virtual instrument environment using LabVIEW and DataEngine.