Defect Classification of Adhesively Bonded Joints Using Pulse-Echo Ultrasonic Testing in Automotive Industries

Amid all nondestructive testing (NDT) methods Ultrasound is considered the most practically feasible modality for quality assessment and detection of defects in automobile industry. Pattern recognition of the ultrasonic signals gives us important information about the interrogated object. This information includes size, geometric shape and location of the defect zone. However, this would not be straightforward to extract this information from the backscattered echoes due to the overlapping signals and also the presence of noise. Here in this study, we suggest a new method for classification of different defects in inspection of adhesively bonded joint. At the first step of this method, the problem of parameter estimation of the reflected echoes is defined in a Maximum Likelihood Estimation (MLE) framework. Then a space alternating generalized Expectation Maximization (SAGE) algorithm is implemented to solve the MLE problem. At the next step, a feature called decay rate of reverberating echoes is defined to serve for the classification. Decay rate can be calculated using amplitudes of reverberant echoes estimated by SAGE algorithm. Final step would be Bayesian classification of defects based on the calculated decay rate feature. By applying the proposed method, void-disbond, poor adhesion, and presence of wrong materials such as grease and water in the front interface of the joints with thickness of 0.5 mm could be detected and classified. To validate the accuracy of the classification procedure ten-fold cross validation is applied on the constructed dataset and the average accuracy of 94.3% is obtained.