Real-Time Damage Detection in Fiber Lifting Ropes Using Lightweight Convolutional Neural Networks

TL;DR

This work tackles the risk and downtime caused by inspecting fiber lifting ropes by introducing a real-time, vision-based damage-detection system built on a lightweight CNN. An imaging setup with a three-camera circular array photographs rope surfaces during operation, and a new rope-image dataset is collected and labeled as normal or damaged. The authors design 16 CNN variants, select a top-performing model, and demonstrate high accuracy (≈96%), strong precision/recall balance, and an AUC around 99% with real-time inference (~33 fps) and a tiny memory footprint (~1.7 MB). Interpretability analyses using Grad-CAM and t-SNE validate that the model focuses on damage-relevant features and that the classes are well-separated, while acknowledging limitations such as surface-only detection and potential annotation noise, with future work including multi-class health states and regression scoring.

Abstract

The health and safety hazards posed by worn crane lifting ropes mandate periodic inspection for damage. This task is time-consuming, prone to human error, halts operation, and may result in the premature disposal of ropes. Therefore, we propose using efficient deep learning and computer vision methods to automate the process of detecting damaged ropes. Specifically, we present a vision-based system for detecting damage in synthetic fiber rope images using lightweight convolutional neural networks. We develop a camera-based apparatus to photograph the lifting rope's surface, while in operation, and capture the progressive wear-and-tear as well as the more significant degradation in the rope's health state. Experts from Konecranes annotate the collected images in accordance with the rope's condition; normal or damaged. Then, we pre-process the images, systematically design a deep learning model, evaluate its detection and prediction performance, analyze its computational complexity, and compare it with various other models. Experimental results show the proposed model outperforms other similar techniques with 96.5% accuracy, 94.8% precision, 98.3% recall, 96.5% F1-score, and 99.3% AUC. Besides, they demonstrate the model's real-time operation, low memory footprint, robustness to various environmental and operational conditions, and adequacy for deployment in industrial applications such as lifting, mooring, towing, climbing, and sailing.

Figures (11)

• Figure 1: The fiber rope crane used in this work. The photo is published with permission from Konecranes Crane_manual.
• Figure 2: The proposed vision-based damage detection system for synthetic fiber lifting ropes. The system is comprised of the following stages: (1) experimental setup with a three-camera circular array to capture rope images; (2) collection and annotation of the captured images; (3) preprocessing to enhance quality and down-sampling to reduce complexity; (4) data splitting into testing and training sets where the former is subdivided into 4-fold training and validation sets; (5) training/testing the proposed deep learning model; and (6) evaluating and analyzing the system's performance and computational complexity.
• Figure 3: Histograms illustrating the distribution of rope height ratio for the 876,847 images from the three ropes.
• Figure 4: Histogram equalization for an example rope image.
• Figure 5: Example images from the acquired dataset show significant variation in the severity and clarity of damages because of dirt and oil stains. The first row (a)-(e) shows damaged ropes while the second row (f)-(j) presents some healthy samples.