Automated Single-Pole Double-Throw Toggle Switch Pin Inspection using Image Processing and Convolutional Neural Network Techniques

Abstract

The single-pole double-throw toggle switch bent pin inspection is an indispensable step in the switch production process. However, the traditional inspection process is conducted in manual work, and this may result in misunderstanding and reduces the manufacturing efficiency due to exhausted humans. To overcome these problems, the automated single pole-double throw toggle switch bent pin inspection method by using image processing and convolutional neural networks is proposed. Our proposed method can be achieved to inspect whether normal or abnormal bent pin of these toggle switches without sorting and no positioning arrangement. Our proposed method consists of five main steps: The first step, the HSV color segmentation is used for background extraction. Next step, the
morphological opening, and closing operation are applied for handling noise and holes that
conspicuously affect the system’s ability to identify the extracted object accurately. Next step, the
minimal enclosing rectangle and angle of rotation are calculated for identifying the positions of the
disorder toggle switch. Next step, the CNN is used for locating pins in order to extract only the pins
out from the binary switch image. In the final step, the average summation of the white pixel is
calculated for classifying normal and abnormal bent pins. The experimental results obtained by
our proposed method are statistically described as accurate compared to many different methods
based on the single pole-double throw toggle switch bent pin inspection. Results show that our
proposed method provides high accuracy than other comparative methods with a mean accuracy of 0.9940 from 6,270 images and uses mean time-consuming 0.3385 seconds.