Projects

Bond Strength Detection Using ML


I spent a summer working in the Ultrasonic Nondestructive Evaluation Laboratory at UofT. An industry sponsored project, the goal was to develop a method of detecting bond strength between an adhesive and primer using ultrasound. My supervisor and I developed an SVM that could classify bond strength as weak or strong with high accuracy on one specific type of pipe.

An example cross-section of the pipe used in this research:

An ultrasonic pulse is emitted at the pipe and the echoes are recorded.

Peak detection was used to generate important features. We used FFT to eliminate high frequency noise.

A signal attenuation constant was part of our physical model for the adhesive. It had to be computed within a statistical confidence.

A decay constant for the primer echoes was fit to the signals obtained from samples of weak and strong bonds. A clear correlation is observed.