An adhesive bond strength prediction algorithm based on ultrasonic signal analysis

The problem of predicting adhesive bond defects for both surface preparation and undercure defects has been studied using an ultrasonic, experimental test bed system. This experimental test bed incorporates the ultrasonic and computer equipment necessary to acquire and process data from various types of adhesively bonded test specimens. The computer hardware and software have been developed to allow the design of reliable pattern recognition algorithms for the evaluation of surface preparation and bond cure. The specific problem studied is the inspection of the adhesive bond in an aluminium/aluminium step-lap joint whose strength could be affected by improper surface preparation or undercure. A set of 164 bond specimens was used to design an algorithm that is 91% reliable for separating the specimens into a good class, those bonds with no defects, or a week class, bonds with poor surface preparation on an undercured adhesive layer. A Fisher Linear Discriminant function was selected by the test bed as the best pattern recognition routine for this classification problem.