Air coupled ultrasonic detection of surface defects in food cans

In this paper, we describe an ultrasonic inspection system used for detection of surface defects in food cans. The system operates in the pulse-echo mode and analyses the 220 kHz ultrasonic signal backscattered by the object. The classification of samples into valid or defective is achieved with chi(2) statistics and the k nearest neighbour method, applied to features computed from the envelope of the ultrasonic echo. The performance of the system is demonstrated empirically in detection of the presence of the pull tab on the removable lid of easy-open food cans, in a production line. It is found that three factors limit the performance of the classification: the misalignment of the samples, their separation of the ultrasonic transducer, and the vibration of the conveyor belt. When these factors are controlled, classification success rates between 94% and 99% are achieved.