Case Study

The Challenge: White Etching Cracks (WEC)

Wind turbine drivetrains operate under highly variable dynamic loads, leading to severe component stress. One of the most critical and cost-intensive failure modes facing the industry is the development of White Etching Cracks (WEC) in gearbox bearings.

WEC leads to premature, catastrophic bearing failure, often occurring at just 10% to 20% of the calculated design life. Because WEC initiates deep within the steel matrix of the bearing, it is virtually impossible to detect with conventional vibration sensors until irreversible macroscopic damage has occurred.

While mechanical stress is a known factor, the industry required a deeper understanding of the tribological and chemical influences—specifically the role of gearbox lubricants—that trigger WEC formation.

Machine Learning for Risk Classification

The PRONOWIS project aimed to move beyond traditional mechanical analysis by applying advanced data science to the chemical properties of lubricants.

To train and assess the predictive models, the research team utilized a robust dataset of approximately 700 high- and low-risk oil compositions. This dataset was provided by a leading global bearing manufacturer and compiled through rigorous physical testing and chemical simulations in collaboration with a specialized industrial analytics supplier.

Operating on anonymized chemical identities, the research applied a suite of machine learning techniques to uncover the hidden patterns linking a lubricant's chemical makeup to a high risk of WEC formation. This included Random Forests (RF) and Support Vector Machines (SVM), deployed alongside purpose-built Artificial Neural Networks (ANN) specifically designed for feature selection.

Objective: The goal was to isolate the specific chemical drivers behind the failures by mapping the complex, non-linear relationships between the elemental composition of a lubricant and its propensity to induce WEC.

The optimization of wind turbine reliability requires transitioning from reactive maintenance to predictive, data-driven condition monitoring. To address this complex industrial challenge, the Center for Wind Power Drives (CWD) at RWTH Aachen University conducted the PRONOWIS project (Project for the Integrated Optimization of Wind Turbine Generator Utilization through Innovative Sensor Systems), funded by the German Federal Ministry for Economic Affairs and Energy.

At Machinalytics, our advanced methodologies are rooted in this caliber of applied academic research. Dr.-Ing. Baher Azzam, data scientist and mechanical engineer, was a primary researcher on the PRONOWIS project. The following article details the published scientific results of this initiative, demonstrating how rigorous machine learning can solve previously unpredictable electromechanical failures.

Key Results: Isolating High-Risk Compounds

The PRONOWIS project successfully proved that data science could untangle the chemical root causes of a complex mechanical failure.

The developed Random Forest and SVM models were highly effective at classifying the WEC risk of a given lubricant solely based on the lubricant’s elemental composition. The predictive framework allowed accelerated inference compared to physical testing alone.

Out of the 21 constituting chemical elements present in the studied lubricants, the machine learning algorithms successfully isolated eight specific compounds as the primary statistical drivers that increase the risk of WEC formation.

Crucially, this data-driven research provided conclusions that were later confirmed by an independent laboratory team through physical experiments, specifically validating the top risk compound identified by the algorithms.

The PRONOWIS project demonstrates that true predictive maintenance goes beyond spotting dashboard anomalies—it requires domain-specific data science to uncover root causes at the chemical and physical level. By bridging complex datasets with advanced AI, the project established an accurate analytical model for understanding material failure.

At Machinalytics, this caliber of scientifically validated research forms our technical foundation. Whether optimizing wind turbines, analysing lubricant compositions, or complex production data, we leverage deep engineering expertise to deliver actionable insights that permanently reduce unplanned downtime.

To learn how our advanced predictive analytics can optimize your machinery's reliability, contact the Machinalytics team today.

Reference source: Azzam, B.; Harzendorf, F.; Schelenz, R.; Holweger, W.; Jacobs, G. Pattern Discovery in White Etching Crack Experimental Data Using Machine Learning Techniques. Appl. Sci. 2019, 9, 5502. https://doi.org/10.3390/app9245502

The Machinalytics Advantage

Contact Us

Have questions or want to chat about the project?