Signal analysis approach to ultrasonic evaluation of diffusion bond quality

Solid state bonds like the diffusion bond are attractive techniques for joining dissimilar materials since they are not prone to the defects that occur with fusion welding. Ultrasonic methods can detect the presence of totally unbonded regions but have difficulty sensing poor bonded areas where the substrates are in intimate contact. Standard ultrasonic imaging is based on amplitude changes in the signal reflected from the bond interface. Unfortunately, amplitude alone is not sensitive to bond quality. We demonstrated that there is additional information in the ultrasonic signal that correlates with bond quality. In our approach, we interrogated a set of dissimilar diffusion bonded samples with broad band ultrasonic signals. The signals were digitally processed and the characteristics of the signals that corresponded to bond quality were determined. These characteristics or features were processed with pattern recognition algorithms to produce predictions of bond quality. The predicted bond quality was then compared with the destructive measurement to assess the classification capability of the ultrasonic technique.