Convolutional Autoencoder-Based Flaw Detection for Steel Wire Ropes.

Area of Science:

• Engineering
• Materials Science
• Computer Science

Background:

• Visual perception methods offer potential for wire rope condition monitoring.
• Traditional methods struggle with noise (oil, dust, light) and limited flaw data.

Purpose of the Study:

• To develop a robust flaw detection method for wire ropes addressing noise and data scarcity.
• To improve the accuracy and efficiency of wire rope surface damage assessment.

Main Methods:

• A novel method combining a convolutional denoising autoencoder (CDAE) for noise reduction and feature learning.
• Integration with Isolation Forest (iForest) for effective flaw detection and isolation.
• Training CDAE with image reconstruction loss and fine-tuning with an iForest-based cost function.

Main Results:

• The CDAE-iForest method demonstrated superior performance in feature learning and flaw isolation.
• Effective detection of surface damage on real-world mine cableway hauling ropes.
• Outperformed other methods, especially when dealing with a small amount of flaw training data.

Conclusions:

• The proposed CDAE-iForest method provides a significant advancement in visual-based wire rope condition monitoring.
• It offers a practical solution for detecting wire rope flaws in challenging real-world conditions.
• The method is particularly advantageous when limited flaw data is available for training.