An Optimized Deep Learning for Efficient Defect Inspection in Smart Industry with Fog Computing

With the development of the constantly expanding Internet of Things (IoT) devices and communications infrastructure, the majority of sensors have been used, resulting in a vast amount of data. One of the most frequent instances is the manufacturer’s inspection to precisely detect product flaws; which allows for adaptive optimization of the model’s parameters. This research proposes a novel Q-learning Reinforcement Learning (RL)-based Optimization Algorithm (ROA) for optimizing the Fog Computing based Convolutional Neural Network (FC-CNN) that can discover probable faulty items precisely, to develop an effective inspection system. Given the possibility of several assembly lines operating in a single plant, one of the primary issues is the real-time processing of this data. Subsequently, adjusting the CNN model’s hyperparameters significantly enhances its faults detection capabilities. The suggested FC-CNN-ROA transfers the burden from the central servers to the fog nodes, allowing it to handle very huge volumes of data. When compared with existing techniques such as You Only Look Once version4 combined with DL (YOLO v4-DL), MobileNet-V2, the proposed FC-CNNROA obtains a higher accuracy of $\mathbf{9 9. 9 7 \%}, \mathbf{9 9. 8 9 \%}, \mathbf{9 9. 7 5 \%}$, $\mathbf{9 9. 7 1 \%}$, and $\mathbf{9 9. 1 6 \%}$ for different sets of datasets such as 10,20, $30,40,50$ respectively.