| Title |
Inspection System for Defective Products of a Automotive Blower Motor Using CNN Ensemble Learning |
| Authors |
전병주(Byoung Ju Jeon) ; 김동헌(Dong Hun Kim) |
| DOI |
https://doi.org/10.5370/KIEE.2025.74.10.1710 |
| Keywords |
CNN; Ensemble learning; blower motor; defective detection; inspection system |
| Abstract |
When producing a blower motor used in automobile seats, various problems such as non-payment, over-payment, and damage due to overheating occur during the automated soldering process. These problems can reduce the durability of the motor over time, which can cause problems in long-term use. In order to detect defects during the process, machine vision-based defect detection methods are effective in standardized work but have limitations in unstructured defect detection. To solve this problem, this study proposes a defect detection system for an automobile blower motor using deep learning. The performance of the CNN model based on the Inception, Xception, and VGG19 architecture was evaluated and the performance was compared by applying soft voting and stacking, which are ensemble techniques that combine the three models. In addition, the generalization of the model was strengthened by applying the data augmentation technique to the image data. As a result, the proposed ensemble model showed high accuracy and consistent defect detection performance compared to a single CNN model. |