Scalable Defective Tool Recognition and Classification System for Industry 4.0 Manufacturing Workshops

Abstract

In recent years, industrial automation and smart manufacturing have witnessed rapid growth, with global reports indicating that over 65% of workshops now rely on visual inspection systems to improve productivity and reduce operational errors, yet manual tool identification still contributes to nearly 30% of workflow inefficiencies and misclassification-related downtime. In this context, this work focuses on the automated recognition of industrial tools in workshop environments using intelligent vision-based techniques. The problem addressed in this study is the inaccuracy, time consumption, and subjectivity associated with traditional manual tool identification systems, where human operators rely on visual inspection and experience to classify tools. The research presents an intelligent image classification system that leverages deep learning and machine learning techniques for robust tool identification. The system utilizes InceptionResNetV2 (Inception Residual Network Version 2) as a pre-trained Convolutional Neural Network (CNN) for extracting high-level visual features from input images through transfer learning. These extracted feature vectors are then used to train multiple classifiers, including Decision Tree Classifier (DTC), K-Nearest Neighbors (KNN), and Perceptron Classifier (PC) as baseline models. The proposed approach introduces a hybrid ensemble model that combines a Deep Neural Network (DNN) for probability-based feature transformation with a Random Forest Classifier (RF) for final classification, forming a stacked learning architecture that enhances prediction accuracy and generalization. The system incorporates comprehensive evaluation metrics such as accuracy, precision, recall, F1-score, confusion matrix, and Receiver Operating Characteristic (ROC) curve with Area Under Curve (AUC) analysis to validate performance across multi-class industrial tool datasets. Additionally, a user-friendly Graphical User Interface (GUI) developed using Tkinter, along with role-based authentication via a MySQL database, enables seamless interaction for both administrators and end users. Experimental results demonstrate that the proposed hybrid model significantly outperforms traditional classifiers by effectively capturing complex patterns in tool images, thereby providing a scalable, efficient, and accurate solution for automated industrial tool recognition in real-world workshop environments.