Robotics-Based Automated Quality Inspection System Using Computer Vision and Machine Learning

Abstract

Conventional manual quality inspection in a manufacturing setting is intrinsically marred by a lack of subjective accuracy, human fatigue, and inconsistency of judgement that usually attracts inconsistencies, slower throughput and increased costs of operation. In order to overcome these difficulties, the study proposes an integrated Robotics-Based Automated Quality Inspection System with the help of computer vision and machine learning algorithms ensuring the high accuracy and real time detection of defects. The suggested demo incorporates an industrial 6-degree-of-freedom robotic arm that can accomplish fine manipulation and adapt fleet positioning of work parts with high-resolution imaging sensors and optimized illumination modules to guarantee the coherent visual data is obtained under diverse production conditions. The raw images are preprocessed and passed as an input of a convolutional neural network (CNN) model- fine-tuned over an EfficientNet backbone to detect and locate various defects, such as scratches, dents, misalignments, and surface contamination. The model was trained using a handcrafted dataset containing 12,000 labeled pictures aggregated using a variety of production environments and the category classification rate attained by the model was 98.2 percent, and a low rate of 2.1 percent false positives was experienced. Through system integration with robotic motion control it is possible to synchronize the inspection, overcoming the blind spot and being able to maintain throughput with the high speed conveyors. Experimental analysis revealed that the suggested system will decrease the amount of human effort put into operation by 65 percent and enhance the rate of inspection by 3.4x compared to the conventional manual method while guaranteeing greater consistency in detecting the defect. In addition to that, the modularization enables quick geometrical cleaning of the products and the nature of production and could, therefore, be used in electronics together with the automotive and packaging manufacturing industries. The work also not only confirms the practicality of combining more sophisticated deep learning methods with industrial robotics as a way to perform scalable, automated quality control, but also lays the groundwork towards improvements such as full 3D vision support and unsupervised detection of novel anomalies and flexible learning to adapt to the changing production environment, hence aligning with the Industry 4.0 goals of intelligent and autonomous manufacturing.