ANA's Predictive Maintenance Challenge: Replace Aircraft Parts Before They Break

All Nippon Airways Group started its air transport business in 1952 and celebrated its 70th anniversary in 2022. ANA Maintenance Center is working on predictive maintenance for aircraft maintenance. We detect failure signs of various aircraft components based on sensor data collected during flights, aiming to improve safety and on-time performance. To perform predictive maintenance, it is necessary to identify features to predict failures from sensor data. The ANA Maintenance Center leverages the deep domain knowledge of the maintenance crew and sensor data from more than 20 thousand flights around the world. The sensor data includes external factors, such as weather conditions, flight path, and number of passengers. In this talk, see how we used machine learning with MATLAB^® to test hypotheses and create machine learning models. We successfully identified robust features that enabled early failure detection of the cabin air compressor (CAC), one of the main air conditioning components. The preprocessing pipeline and the trained model were integrated into an existing system using MATLAB Compiler™ for daily inspection.

Highlights:

• Predictive maintenance for aircraft maintenance
• Hypothesis verification of failure-identifying features using domain knowledge and field data
• A case study on detecting the deterioration of aircraft components using a neural network