AI-Driven Predictive Quality Control for Real-Time Defect Detection and Process Optimization in Smart Manufacturing
Chidera Anastacia Ezeh
Department of Civil, Construction and Environmental Engineering, North Dakota State University, North Dakota, United States.
Ayomide Oladele
Department of Mechanical Engineering, South Dakota School of Mines and Technology, South Dakota, United States.
Ahmed Idowu Agbelejoye
Department of Mathematical Sciences, Osun State University, Osogbo, Nigeria.
Samson Anuoluwapo Adeniyi
Department of Statistics, Federal University of Agriculture, Abeokuta, Nigeria
Lawal Sulaimon Abiodun
Mechanical Engineering Department, Ladoke Akintola University of Technology, Ogbomoso, Oyo State Nigeria.
Gbuchie Chisom Vivian
Federal University of Technology Owerri Department of Electrical and Electronics Engineering Nigeria.
Confidence Adimchi Chinonyerem *
Abia State polytechnic, Abia, Nigeria.
*Author to whom correspondence should be addressed.
Abstract
Smart manufacturing requires quality-control systems capable of detecting defects and responding to process deviations in real time. This study develops an AI-driven predictive quality-control framework that integrates machine vision, sensor-based monitoring, sequential process modelling, and closed-loop multi-objective optimisation. The framework processes production images and sensor signals to classify surface defects, anticipate process anomalies, and adjust operating parameters, including machine speed, feed rate, temperature, and pressure. Evaluation was conducted using 15,000 surface-inspection images and 8,400 sensor-based process records collected from a controlled production facility. The convolutional neural network achieved a reported defect-classification accuracy of 97.8%, with precision, recall, and F1-score values of 96.9%, 96.3%, and 96.6%, respectively. The predictive component achieved 95.6% accuracy and identified process anomalies approximately 12–18 seconds before defect occurrence. Closed-loop optimisation was associated with reported reductions of 42.7% in defect rate, 18.3% in energy consumption, and 21.5% in production delay, together with a 16.8% improvement in overall throughput. The average processing latency was 103 milliseconds per image. These results indicate that integrating image-based inspection, sensor-driven prediction, and process optimisation can support timely quality-control decisions in a controlled smart-manufacturing environment, while further validation under variable real-world industrial conditions remains necessary.
Keywords: Predictive quality control, real-time defect detection, smart manufacturing, industrial internet of things, deep learning, process optimization, machine vision, industry 4.0