Real-Time Defect Identification in Automotive Brake Calipers Using PCA-Optimized Feature Extraction and Machine...

Area of Science:

• Mechanical Engineering
• Materials Science
• Artificial Intelligence

Background:

• Automotive brake calipers are critical for Electric Parking Brake (EPB) systems.
• Manufacturing defects can compromise caliper performance, leading to braking issues.
• Real-time, non-contact defect detection is needed for quality control.

Purpose of the Study:

• Develop a non-contact automated impact-acoustic measurement system (AIAMS).
• Enable real-time detection of manufacturing defects in brake calipers.
• Improve the reliability and safety of EPB systems.

Main Methods:

• Implemented an automated impact hammer and microphone-based system.
• Utilized Fast Fourier Transform (FFT) and Principal Component Analysis (PCA) for feature extraction.
• Employed machine learning algorithms (SVM, KNN, DT) for defect classification.

Main Results:

• Shannon Entropy (SE) identified as the most discriminative feature via PCA.
• High classification accuracies achieved: SVM (99.2%/97.5%), KNN (98.8%/98.0%), DT (99.2%/96.5%).
• Field tests confirmed over 95% average defect classification accuracy.

Conclusions:

• The AIAMS is effective for real-time, non-contact quality control of brake calipers.
• The system enhances manufacturing defect detection for EPB components.
• Demonstrated practical applicability in industrial settings with high accuracy.