No abstract

The efficacy of fatigue life approximation methodologies for Landing Gear systems is studied and compared to the ongoing Structural Health Monitoring techniques being researched, which will forecast failures based on the system's specific life and withstanding abilities, ranging from creating a digital simulation model to applying neural network technologies, in order to simulate and approximate locations and levels of failure along the structure. Explainable Artificial Intelligence allows for the ease-of-integration of Deep Neural Network data into Predictive Maintenance, which is a procedure focused on the health of a system and its efficient upkeep via the use of sensorbased data. Test data from a flight includes a multitude of conditions and varying parameters such as the surface of the landing strip as well as the aircraft itself, requiring the use of Deep NeuralNetwork models for damage assessment and failure anticipation, where compliance to standards is a major question raised, as the EASA AI roadmap is followed, as well as the ICAO and FAA. This paper additionally discusses the challenges faced with respect to standardizing the Explainable AI methodologies and their parameters specifically for the case of Landing Gear.

This paper provides information on current certification of landing gear available for use in the aerospace industry. Moving forward, machine learning is part of structural health monitoring, which is being used by the aircraft industry. The non-deterministic nature of deep learning algorithms is regarded as a hurdle for certification and verification for use in the highly-regulated aerospace industry. This paper brings forth its regulation requirements and the emergence of standardisation efforts. To be able to validate machine learning for safety critical applications such as landing gear, the safe-life fatigue assessment needs to be certified such that the remaining useful life may be accurately predicted and trusted. A coverage of future certification for the usage of machine learning in safety-critical aerospace systems is provided, taking into consideration both the risk management and explainability for different end user categories involved in the certification process. Additionally, provisional use case scenarios are demonstrated, in which risk assessments and uncertainties are incorporated for the implementation of a proposed certification approach targeting offline machine learning models and their explainable usage for predicting the remaining useful life of landing gear systems based on the safe-life method.