Detecting Machining Defects inside Engine Piston Chamber with Computer Vision and Machine Learning

Area of Science:

• Computer Vision
• Machine Learning
• Industrial Automation

Background:

• Traditional visual inspection for machining defects in automotive manufacturing is labor-intensive and prone to human error.
• The integration of machine learning and computer vision offers potential for automated quality control in industrial settings.
• Existing hardware in manufacturing plants can be repurposed for new automated inspection systems.

Purpose of the Study:

• To develop and implement a computer vision application for detecting machining defects in engine block piston chambers.
• To replace manual visual inspection with an automated system for enhanced efficiency and accuracy.
• To leverage existing hardware infrastructure and machine learning algorithms for a cost-effective solution.

Main Methods:

• Exploration of machine vision applications and machine learning in industrial robotics.
• Re-utilization of decommissioned hardware including cameras, Ethernet modules, and sensors for image acquisition.
• Implementation of a new processing system with a human-machine interface and production line communication, utilizing convolutional neural networks for defect detection.

Main Results:

• Demonstrated the efficiency of charged-coupled device (CCD) sensors in the manufacturing environment for image acquisition.
• Validated the robustness of convolutional neural networks (CNNs) within computer vision applications.
• Successfully implemented defect detection using thresholding and regions of interest techniques.

Conclusions:

• The implemented computer vision solution effectively detects machining defects, offering a viable alternative to manual inspection.
• Repurposing existing hardware is a practical approach for cost-effective automation in manufacturing.
• Machine learning algorithms, particularly CNNs, are robust and efficient for automated quality control tasks in industrial settings.