Deep Learning-Based Predictive Model for Defect Detection and Classification in Industry 4.0
Umesh Kumar Lilhore, Sarita Simaiya, Jasminder Kaur Sandhu, Naresh Kumar Trivedi, Atul Garg, Aditi Moudgil
Abstract
In the perspective of the Industry 4.0 (IR 4.0) model, the Deep Learning (DL) domain now has a significant impact on the production industry. The IR 4.0 model promotes intelligent sensors, systems, and devices to build intelligent industries that gather information regularly. DL method enables the development of implementable intelligence by analyzing the gathered information to boost production efficiency without dramatically changing the necessary materials. Component defects and discrepancies that impact component reliability are particularly massive in industrial processes. This research introduces a novel framework based on the VGG-16 with CNN model that creates the Intelligent Production learning center into an I4.0 production system. We describe the issue of recognizing tiny defects in an industrial inspection. The primary objective is to classify the pixel value correlating to a defect with a minimal level of false-positive results. Destructive Vs. non-destructive testing and classification procedures are mainly utilized for product quality assurance after production. Convolutional neural networks (CNN) based on machine learning (ML) methods are frequently utilized for this activity. Complex transfer learning (TL) strategies are examined in this research, which allows for the automatic detection and categorization of product defects in the manufacturing process employing industrial product samples. All the known performance metrics have been evaluated to measure and compare the model performance. The proposed VGG16 with CNN model has better outcomes for precision, recall, and accuracy as compared to exisitng CNN, VGG-16, EfficientNetB0, and Inception V3 methods.