Modeling Discrete Coating Degradation Events via Hawkes Processes

Matthew Repasky, Henry Shaowu Yuchi, Fritz Friedersdorf, Yao XiearXiv 2025 · DOI · arXiv

Abstract

Forecasting the degradation of coated materials has long been a topic of critical interest in engineering, as it has enormous implications for both system maintenance and sustainable material use. Material degradation is affected by many factors, including the history of corrosion and characteristics of the environment, which can be measured by high-frequency sensors. However, the high volume of data produced by such sensors can inhibit efficient modeling and prediction. To alleviate this issue, we propose novel metrics for representing material degradation, taking the form of discrete degradation events. These events maintain the statistical properties of continuous sensor readings, such as correlation with time to coating failure and coefficient of variation at failure, but are composed of orders of magnitude fewer measurements. To forecast future degradation of the coating system, a marked Hawkes process models the events. We use the forecast of degradation to predict a future time of failure, exhibiting superior performance to the approach based on direct modeling of galvanic corrosion using continuous sensor measurements. While such maintenance is typically done on a regular basis, degradation models can enable informed condition-based maintenance, reducing unnecessary excess maintenance and preventing unexpected failures.

Introduction

Protective coatings are the primary defense against corrosion, which costs the US Navy an estimated $8.6 billion annually. Current maintenance strategies often rely on fixed schedules, leading to significant material waste and environmental risks from toxic chemicals, regardless of the coating's actual condition. While high-frequency sensors, such as zero-resistance ammeters (ZRA), allow for real-time monitoring of galvanic corrosion current, the resulting "firehose" of data presents substantial modeling challenges. Furthermore, defining exactly when a coating has "failed" is subjective; visual inspections for blisters are infrequent and vary between observers.

To address these issues, this study utilized two primary datasets: a laboratory dataset consisting of 24 multi-sensor panels subjected to accelerated cyclic corrosion, and an outdoor dataset of 6 panels exposed to real-world elements in Daytona Beach, Florida. Sensors tracked galvanic corrosion current, relative humidity (RH), temperature, and surface conductance to provide a comprehensive view of the degradation process.

Methodology

The proposed framework transforms noisy, high-volume sensor data into a clear predictive signal through a three-stage pipeline.

Overview of the three-stage pipeline for failure prediction, from change point detection to Hawkes process modeling.

1. Data-Driven Failure Labeling

To establish a consistent "physical" failure point, the researchers employed a Window-Limited CUSUM (WLCUSUM) change-point detection algorithm. They identified that the "exponential tail" of the daily corrosion current cycle—the way current decays at the end of a cycle—serves as a highly sensitive indicator of internal coating health. This metric often signals failure before it becomes visually apparent to the naked eye.

Correlation between visual degradation (a), corrosion current (b), and the evolution of the exponential tail (c) used for failure labeling.

2. Discrete Event Extraction

Continuous sensor streams were distilled into three types of sparse events: Corrosion Events (current spikes), Environment Events (high humidity/contaminant windows), and Hybrid Events (corrosion spikes during adverse environmental windows). By reducing thousands of measurements to a few dozen discrete events, data volume was reduced by 2-3 orders of magnitude without sacrificing predictive accuracy.

3. Marked Hawkes Processes

The core innovation is the use of a marked Hawkes process to model these events. Originally designed for self-exciting phenomena like earthquakes, this process assumes that one degradation event (like a coating breach) increases the probability of subsequent events. The "mark" in this process represents the magnitude of the corrosion current, allowing the model to forecast the future trajectory of degradation.

Results

The discrete-event approach consistently outperformed traditional linear auto-regressive baselines, including OLS, Ridge, and LASSO regression.

  • Precision: In laboratory settings, the Hybrid Event model achieved a mean error of 1.11 weeks in predicting failure, compared to 1.17–1.31 weeks for traditional methods.
  • Confidence: The Hawkes models provided significantly narrower prediction windows (approximately 2.6 weeks wide) compared to baseline windows (over 4 weeks wide), indicating higher certainty.
  • Robustness: In chaotic outdoor environments, the Hawkes models achieved roughly half the error of linear models, demonstrating superior resilience to real-world environmental noise.

Conclusion

This research demonstrates that distilling continuous sensor data into discrete "degradation events" modeled via Hawkes processes provides a more accurate and efficient means of forecasting coating failure. While the current study is limited by a relatively small sample size of 30 panels, it paves the way for smarter, condition-based maintenance. Future work will explore integrating general weather databases to enable "sensor-less" modeling and testing the framework across a broader range of coating chemistries.

Summary updated

Abstract

Forecasting the degradation of coated materials has long been a topic of critical interest in engineering, as it has enormous implications for both system maintenance and sustainable material use. Material degradation is affected by many factors, including the history of corrosion and characteristics of the environment, which can be measured by high-frequency sensors. However, the high volume of data produced by such sensors can inhibit efficient modeling and prediction. To alleviate this issue, we propose novel metrics for representing material degradation, taking the form of discrete degradation events. These events maintain the statistical properties of continuous sensor readings, such as correlation with time to coating failure and coefficient of variation at failure, but are composed of orders of magnitude fewer measurements. To forecast future degradation of the coating system, a marked Hawkes process models the events. We use the forecast of degradation to predict a future time of failure, exhibiting superior performance to the approach based on direct modeling of galvanic corrosion using continuous sensor measurements. While such maintenance is typically done on a regular basis, degradation models can enable informed condition-based maintenance, reducing unnecessary excess maintenance and preventing unexpected failures.
BibTeX
@misc{repasky2025modeling,
  author = {Matthew Repasky and Henry Shaowu Yuchi and Fritz Friedersdorf and Yao Xie},
  title = {{Modeling Discrete Coating Degradation Events via Hawkes Processes}},
  year = {2025},
  eprint = {2504.09706},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2504.09706},
  howpublished = {arXiv},
}
Figure