Enhancing Quality Inspection in Automotive Manufacturing through Deep Learning and Transfer Learning

Overview

Ensuring quality control and accurate defect detection are critical in the automotive manufacturing industry. This project introduces a deep learning-based approach leveraging transfer learning to automate defect detection process for Terminal Crimp Cross-Section.

---

Key Features

1. Utilizes pre-trained Convolutional Neural Networks (CNNs) such as DenseNet121, VGG19, and others.

2. Implements transfer learning to address limited labeled datasets.

3. Achieves high accuracy (>98%) in defect detection and classification.

4. Deploys models in a real-time production environment with a user-friendly GUI for the Quality team.

---

Repository Contents

• src: Scripts for preprocessing, training, and evaluating the CNN models.

• app: A user-friendly interface for real-time deployment in production environments.

• Documentation: Detailed descriptions of the methods, metrics, and experimental setup.

  Sample Data: Example datasets for terminal crimp cross-sections (restricted due to proprietary limitations).

---

Requirements

Hardware

• Machine 1: Ubuntu 22, 16 CPUs, 16 GB RAM, 100 GB Disk, Azure-based.

• Machine 2: Debian 11, 8 CPUs, 15 GB RAM, 140 GB Disk, On-premise server.

Software

• Python Libraries: TensorFlow, Keras, scikit-learn, seaborn, and other dependencies listed in requirements.txt.

---

Installation

Clone the repository:

git clone https://github.com/your-username/quality-inspection-automotive.git
cd quality-inspection-automotive

Install dependencies:

pip install -r requirements.txt

Launch the Streamlit application:

streamlit run your_script.py

---

Future Work

• Incorporating object detection models (R-CNN, YOLO) for precise defect localization.

• Expanding datasets to cover additional manufacturing applications.

• Enhancing model interpretability for better insights into defect classification.