CNN-Based ResNet-101: A High-Accuracy Quality Control Inspection Approach for Industrial Applications in Liquid Spreading on Skin-Mimic Substrates

Ensuring uniformity in liquid spreading on skin mimic substrate materials is crucial in quality control for automated industrial processes. This paper proposes a CNN-based method for visual inspection of liquid spreading on skin mimic substrate. In this approach, ResNet-101 is used to boost the performance of a designed liquid spreading system that integrates a machine vision subsystem and an UR10e robotic arm for spreading scanning and images capturing under different lighting conditions. Spreading is categorized into three classes: empty spread, full spread, and fault spread. A novel dataset is generated when the camera is assigned to capture images for training, validation and then testing purposes. The trained network effectively classifies the spreading status. Experimental results demonstrate that the ResNet-101 scheme achieved a training accuracy of 97.99%, outperforming classical machine learning and deep learning models by comparing accuracy and real testing of unseen samples. Real testing results proved that the system’s performance is enhanced with ResNet-101, which provides a promising novel quality control solution for automated coating and painting inspection task under complex industrial conditions and requirements