Imagery Dataset for Remaining Useful Life Estimation of Synthetic Fibre Ropes

This paper introduces a novel, publicly available image dataset comprising approximately 34,700 high-resolution images of eleven Dyneema synthetic fibre ropes subjected to cyclic fatigue loading, designed to support machine learning tasks for remaining useful life estimation and vision-based condition monitoring.

The Gist

Imagine you have a very strong, high-tech rope made of special fibers (like Dyneema). This rope is used for heavy lifting jobs, like hoisting wind turbines or moving giant loads on ships. Just like a rubber band that eventually snaps after being stretched and bent too many times, these ropes wear out over time. The big problem is that this wear-and-tear happens slowly and invisibly inside the rope, making it hard to know exactly when it's about to break.

This paper introduces a new "training library" for computers to learn how to predict when these ropes will fail. Here is the simple breakdown:

The Problem: Guessing the End of the Rope

Currently, if you want to know if a rope is safe, you have to stop work, look at it with your eyes, and guess. It's like trying to guess when a car tire is about to blow out just by looking at it once a month. It's risky and often wrong. The authors wanted to build a system where a camera could watch the rope and say, "You have about 500 more uses left before you break."

The Solution: A "Time-Lapse" Photo Album

To teach a computer to do this, the researchers needed a massive photo album showing the rope's entire life, from brand new to completely broken. They created a dataset containing about 34,700 high-resolution photos.

Think of it like a "time-lapse" video, but instead of a video, it's thousands of individual snapshots.

• The Actors: They used 11 different ropes.
• The Stress Test: They put these ropes on a machine that bends them back and forth over a wheel (like a pulley) thousands of times. This mimics the real-life bending they do on ships and cranes.
• The Pressure: They tested the ropes under different amounts of weight, from light loads (60 kN) to very heavy loads (280 kN).
• The Result: Some ropes lasted a long time (over 8,000 bends), while others under heavy stress broke quickly (in under 700 bends).

How They Took the Pictures

Every time the machine bent the rope a specific number of times (a "burst"), it stopped. Then, a high-speed camera took 10 photos of different spots along the rope's length.

Why 10 photos? Because damage isn't fair; it doesn't happen evenly. One spot on the rope might be fraying while the spot next to it looks perfect. Taking 10 photos ensures the computer sees the whole picture, not just one lucky spot.

The "Secret Sauce": The Labels

Every single photo in this dataset has a label attached to it. It's like a timestamp that says, "This photo was taken after 5,000 bends, and the rope broke at 8,000 bends."

This allows the computer to do simple math:

• Total Life: 8,000 bends
• Current Age: 5,000 bends
• Remaining Life: 3,000 bends

Because they have this math for every single photo, they can train artificial intelligence (AI) to look at a picture of a rope and calculate exactly how much "life" is left, even if the rope looks mostly fine to the human eye.

Why This Matters

Before this paper, there was no public collection of photos showing the entire life of these ropes from start to finish. Researchers had to build their own small tests, which took a long time and cost a lot of money.

Now, anyone can download this "photo album" and teach their AI to:

1. Spot the damage early.
2. Predict the future (how many bends are left).
3. Learn how different weights change how fast the rope wears out.

In short, this paper provides the "textbook" of images that computer scientists need to build smarter, safer systems that can tell us exactly when to replace a rope before it snaps, preventing accidents and saving money.

Technical Summary

Technical Summary: Imagery Dataset for Remaining Useful Life Estimation of Synthetic Fibre Ropes

Problem Statement
The safe operation of offshore cranes, wind turbine installation vessels, and heavy-load handling systems relies heavily on synthetic fibre ropes (SFRs), particularly those made from high-modulus polyethylene (HMPE) like Dyneema. While these ropes offer advantages in strength and weight over steel wire ropes, they are susceptible to progressive fatigue degradation during cyclic bending-over-sheave (BOS) operations. Current industry practices for monitoring rope health rely on periodic manual visual inspections and conservative "retire-on-time" policies, which are subjective and ill-suited for variable real-world loading conditions. Although data-driven condition monitoring is a promising alternative, the development of vision-based Remaining Useful Life (RUL) estimation algorithms has been hindered by the absence of publicly available image datasets that capture the complete degradation lifecycle of SFRs under controlled cyclic fatigue. Existing datasets, including the authors' previous work on defect classification, lack the temporal lifecycle information necessary for prognostics.

Methodology and Experimental Setup
To address this data gap, the authors constructed a novel imaging dataset through a controlled experimental campaign involving eleven Dyneema SK75/78 HMPE rope specimens (8 mm nominal diameter, 12-strand braided). The experiments were conducted on a sheave-bend test stand designed to replicate operational BOS loading.

• Loading Conditions: The specimens were subjected to cyclic fatigue at seven distinct axial load levels ranging from 60 kN to 280 kN. Testing continued until mechanical failure, defined as a sudden, complete loss of load-bearing capacity. Fatigue lifetimes varied inversely with load, ranging from 695 cycles (at 250 kN) to 8,340 cycles (at 60 kN).
• Data Acquisition: An imaging system comprising a Basler acA2000 high-speed camera (1920 × 1200 pixels) and a 12-megapixel lens was used. Illumination was provided by Aputure AL-MC RGBWW LED lights positioned at 45° to ensure uniform, shadow-free lighting.
• Sampling Strategy: Data collection occurred in "inspection bursts." After a fixed number of sheave cycles (10–20 cycles per burst, depending on load), the test stand paused, and ten images were captured at different cross-sectional positions along the rope. This spatial sampling strategy accounts for the non-uniform nature of fatigue damage along the rope length.
• Control and Synchronization: An NVIDIA Jetson Nano P3450 managed image capture, synchronized with the test stand's PLC via a handshake protocol to ensure precise timing and unambiguous association of images with cycle counts.
• Dataset Composition: The final dataset comprises approximately 34,700 high-resolution PNG images. Each image is annotated with the elapsed cycle count and the specific rope identity.

Key Contributions
The primary contribution of this work is the release of the first publicly available lifecycle imagery dataset specifically for HMPE ropes under cyclic fatigue. Key features include:

1. Complete Lifecycle Coverage: The dataset captures the visual trajectory from healthy conditions through progressive degradation to mechanical failure for all eleven specimens.
2. Load-Diverse Benchmark: By spanning seven load levels (60–280 kN), the dataset enables the study of load-conditioned degradation rates and the development of models that generalize across different tension levels.
3. Spatial and Temporal Resolution: The inclusion of ten spatially distributed images per inspection burst provides a richer training signal than single-view strategies, capturing the spatial variability of damage.
4. Direct RUL Labelling: Every image is paired with an elapsed cycle count, allowing for the direct computation of RUL ($RUL = N_{failure} - N_{elapsed}$). This supports both continuous regression tasks and discrete life-stage classification (early, mid, late).
5. Compatibility: The data format and structure are designed to be compatible with a wide range of deep learning architectures, including CNNs, Vision Transformers (ViT), and multimodal vision-language models.

Results and Data Characteristics
The paper does not present the results of specific machine learning models trained on the data; rather, the "results" are the dataset itself and its statistical properties. The data demonstrates a clear inverse relationship between applied load and fatigue life. The total image count of ~34,700 provides a statistically meaningful volume for training deep learning models. The dataset is organized into eleven folders (one per specimen), each containing metadata regarding load levels, rope speed, and cycle counts.

Significance and Claims
The authors position this dataset as a foundational benchmark resource for the research community. Its significance lies in:

• Enabling Automated RUL Prediction: It facilitates the development of vision-based systems capable of providing early warnings of impending failure, potentially reducing unplanned maintenance costs and preventing catastrophic safety incidents.
• Completing the Condition Monitoring Pipeline: When combined with the authors' previous defect classification dataset, this resource covers the full spectrum of condition monitoring (CM), from fault detection to remaining life prediction.
• Standardization: By providing a standardized, load-diverse, and fully annotated dataset, the authors aim to accelerate the development and fair comparison of prognostic health management (PHM) algorithms for SFRs, moving the field beyond the limitations of manual inspection and proprietary data.

The work is modest in its claims, focusing on the provision of the data infrastructure necessary to enable future research rather than proposing a specific algorithm or claiming immediate industrial deployment. The dataset is intended to serve as a tool for researchers to develop and validate their own vision-based prognostic models.