Patsy Jammal
(Advisor: Prof. Mavris)

will defend a doctoral thesis entitled,

Digital Twin-Driven Condition Monitoring Approach for Aircraft Carbon Brakes

On

Wednesday, December 4 at 1:00 p.m. EST
Teams Meeting Link: Join the meeting now

Abstract

Wheels and brakes are among the largest contributors to aircraft component maintenance costs. Carbon brakes, in particular, are expensive to procure, and their current Time-Based Maintenance (TBM) approach—conducting maintenance at fixed intervals—often results in unnecessary inspections and increased aircraft downtime. Transitioning to Condition-Based Maintenance (CBM), which leverages the actual health state of the brakes, can significantly reduce maintenance costs and increase aircraft availability. CBM also enhances safety and prolongs carbon brake life by providing tailored insights to minimize wear and optimize brake performance.

CBM is enabled by Prognostics and Health Management (PHM) studies, which focus on diagnosing the current health of components and forecasting their future conditions and performance. Carbon brakes represent a significant revenue stream for manufacturers, who also gain access to operational data from their customers. By leveraging big data analytics on this information, innovative solutions can be developed to optimize brake maintenance. While traditional condition monitoring methods lacked real-time predictive capabilities, the emergence of Digital Twin (DT) technology, combined with the availability of multidomain data, has paved the way for a DT-driven approach. This approach employs advanced data-driven techniques, including Artificial Intelligence (AI) and Machine Learning (ML), to enhance predictive accuracy and maintenance efficiency.

Despite advancements, existing literature on carbon brake wear reveals several gaps. These include a limited understanding of how varying operational and environmental conditions influence carbon wear, minimal application of advanced machine ML techniques on high-dimensional datasets from real-world operations, and insufficient assessment and improvement of predictive model generalizability across domains (e.g., different aircraft types or route structures). To address these issues, this research aims to develop an optimized and generalizable data-driven methodology to predict carbon brake wear, incorporating aircraft-specific parameters, operational conditions, and environmental factors.

This research addresses the identified gaps by developing a rigorous and repeatable methodology that can also be applied to other wear-prone components. First, clustering techniques are employed to uncover brake wear patterns and the corresponding ranges of aircraft, operational, and environmental parameters associated with varying levels of wear severity. This analysis could support adjustments to duty cycle procedures for dynamometer brake performance testing, ensuring a more accurate reflection of real-world operations. Next, supervised ML algorithms are applied to create classification models that effectively assess the severity of carbon brake wear under diverse operational and environmental conditions. This capability facilitates the identification of flights with excessive brake wear, along with the most influential contributing factors.

The problem is further approached as a regression task to achieve more accurate and precise brake wear predictions, leveraging both traditional ML and advanced Deep Learning (DL) techniques. This approach enables benchmarking multiple algorithms to assess their suitability for DT modeling. Both regression and classification models depend on wear pin signals derived from aircraft flight data, which indicate the percentage of carbon brake disk thicknesses remaining. These signals are reported inconsistently, approximately every ten flights. To address this, the optimal frequency for collecting wear pin values is investigated to guide operators on how often to report brake thicknesses for aircraft without electronic wear pin sensors, enabling the development of supervised predictive models for such aircraft.

The generalizability of the predictive models is also evaluated to determine whether tailored models are required for different data segments (e.g., diverse aircraft types or route structures), or if a single model can achieve acceptable accuracy across the entire dataset. Lastly, Transfer Learning (TL) techniques are explored to enhance model performance across various data segments, ensuring robustness and adaptability in predictive capabilities.

Committee

• Prof. Dimitri N. Mavris – School of Aerospace Engineering (Advisor)
• Dr. Olivia J. Pinon Fischer – School of Aerospace Engineering
• Prof. Daniel Schrage – School of Aerospace Engineering
• Assoc. Prof. Brian German – School of Aerospace Engineering
• Dr. Gregory Wagner – Sr. Principal Engineer, Collins Aerospace