Detecting Defects in Sanitary Wares Using Deep Learning

Anomaly detection is an important task and has a wide range of applications, e.g. fraud, disease, and damage detection. Currently, the fault detection procedures in the industry of sanitary ware often rely on the senses of the test operators. One of the standard failure tests consists of hitting the sanitary ware with a small hammer and analyzing the sound emitted. Products with good quality emit a characteristic sound pattern. On the other hand, wares with defect emit sounds with slightly different patterns, and the goal is to determine when the emitted sound does not follow the typical pattern. There are two crucial problems in this procedure: its efficiency can be compromised due to fatigue and lack of attention of the human being performing the test, and the environment is noisy. This work proposes the development of an automatic system to aid the operator to identify defects in sanitary ware. The proposed system deploys a supervised deep learning technique. The trained models presented an average precision superior to 99%, and an execution time inferior to 0.018 s.