Aircraft Engine Audio Signal Analysis in Assisting Maintenance Inspections

• Published in: Journal of physics. Conference series Vol. 2218; no. 1; pp. 12002 - 12007
• Main Author: Sakai, Yukino
• Format: Journal Article
• Language: English
• Published: Bristol IOP Publishing 01.03.2022
• Subjects: Aircraft; Aircraft engines; Aircraft maintenance; Aircraft turbofan; Algorithms; Audio data; Audio signals; Combustion chambers; Commercial aircraft; Engine components; Engine inspection and maintenance; Fan blades; Fast Multi-Class SVM; High temperature; Image classification; Inspection; Machine learning; Physics; Signal analysis; Sound generation; Support vector machines; Turbofan engines
• DOI: 10.1088/1742-6596/2218/1/012002
• ISSN: 1742-6588

As the core component of modern commercial aircraft, turbofan engines have long been the center of focus in aircraft maintenance. Being subject to high temperatures and immense pressures causes problems for the engine components, such as the fan blades, as they are frequently burdened with the potential of overhaul and malfunction. Over many years, the industry has seen various methods of engine inspection and maintenance, ranging from manual inspection to computing large quantities of pre-existing data. Within, audio signal analysis has stood out as a productive, non-invasive method, with many alternate studies analyzing sound signals from components such as the combustion chamber. However, many of these methods, despite demonstrating good accuracy, are incredibly complex and require sophisticated apparatus. Therefore, this study begins by investigating the sound generation process of turbofan engines, especially how the features and form of the fan blade characterize its audio signals. This investigation proposes a solution that utilizes a fast Multi-Class Support Vector Machine (SVM) algorithm based on fan-blade-related audio signals from a perspective similar to the classification of music and images through supervised machine learning. Experimental results show that this fast Multi-Class SVM is more effective than traditional machine learning methods in its accuracy, F1-score, and other indicators.