Connor M. Byington†, Carl S. Byington*, Justin Zhang†, and Michael Hynson‡
Aircraft landing gear (LG) systems are subject to a wide range of stressors due to wear, fatigue, and impact driven mechanisms. Moreover, modern aircraft are often heavily reliant on avionics equipment to support anti-skid and related performance enhancements. As in many arenas, there is a strong desire to transition from overly preventative actions, scheduled part replacements, and unjustified removals towards performing maintenance upon evidence of need, i.e., Condition-based Maintenance (CBM). While maintenance cost drivers and opportunities for improvement are clearly evident, there are many sensor and observable limitations for landing gear prognostics and health management (PHM). To address these CBM challenges for landing gear systems and facilitate transition from scheduled maintenance to reliable implementation of diagnostics and prognostics, a team of developers from SJ Technologies, Inc. (SJ), PHM Design, LLC, and PavCon, LLC, have developed a novel usage-based methodology that incorporates reliability data, aircraft sensor data, and maintenance analysis.
This paper presents the evaluation of health-based algorithms, fault code reasoners, and ultimately a novel usage-based algorithm to address prognostic risk assessment for a critical anti-skid line replaceable unit (LRU) on the aircraft. This predictive analytics effort employed a range of PHM processing steps. The Team considered data from the Reliability and Maintainability Information System (REMIS) and sensor data from the aircraft onboard systems. The alpha version of this algorithm was developed and demonstrated using non-proprietary and readily available tools, namely Python, PyCharm, Anaconda, Javascript, and standard web browsers. The prognostic algorithm was designed to operate on the 1 Hz aircraft data and processes every second of valid and correct mode data within all the flight and ground operations. The Team processed approximately 57,000 flight and ground files with (about 1 Terabyte) of flight parametric and test point data from over 47 unique aircraft tails from 2019-2021. Using modified ETL (Extract, Transform and Load) processes the team extracted the relevant data sets from an AWS (Amazon Web Services) m5.4X environment using Scala and Python scripts. The subsequent files were organized into tail numbers, temporally named, as well as further sliced, diced, massaged, error corrected, searched, and processed by the algorithms with a range of Python developed scripts. Detailed process flow diagrams, logical steps, and algorithm elements are presented in the sections below.
The Team accomplished many significant, positive results with this usage-based prognostic algorithm including:
- Demonstrated very good predictive correlation for “believed faulted” cases with confirmed maintenance remove and replace (R2) events.
- Highly correlated alignment of high risk for high cycling usage considering all operations.
- Correctly assigned low damage risk for all tails with low flight and ground operations.
Several challenges and areas of improvement were also identified, namely that the algorithm:
- Underpredicts risk for low ground and flight usage conditions that witnessed an R2 (Remove and Replace) event. These are likely caused by other failure drivers, or “infant mortality” on the LRU.
Addressing this challenge ultimately relates to achieving some better level of health-based PHM indicators, which is currently being pursued. This algorithm evolution is ultimately intended to be accommodated using a hybrid fusion approach.
This Usage-based prognostic algorithm developed by the SJ/PHM/PavCon Team demonstrated a novel and unique solution to process full aircraft data sets, by tail number, to predict failure risk with usage surrogate drivers. This damage indication and risk metric can be utilized to trigger CBM+ maintenance codes for LRU inspection check as well as replacement for CBM causality (see Figure 5). The result is the achievement of reducing mission aborts and maintenance downtime as well as increasing fleet operational availability (Ao).
To read the full paper check out my Researchgate posting.