Predicting aircraft maintenance periods and failure counts through artificial intelligence techniques

Aviation industry is an area where a lot of knowledge and data are collected. In this area, it is typical to store a huge quantity of data, such as maintenance plans, failure notifications, repair orders, records of equipment's removal/installation, flight hours, number of sorties and active/inactive statuses. It is necessary to analyse these data to distinguish between the cases that come with false fault signals or false notifications and real faults. In addition, we need to extract meaningful data and eliminate unnecessary data and features. By writing programs for this purpose in the resource management system, firstly the MTBF and NFF values of the equipment that belong to the selected aircraft system were calculated, the maintenance periods specified in the manufacturer's catalogue of the materials were analysed in comparison with the calculated MTBF values. In addition, the success of the model was demonstrated by predicting the next failure day of the aircraft through the use of artificial intelligence/machine learning methods based on the aircraft activity reports obtained for a two-year period to help adjust the general maintenance periods. Within the scope of this dissertation, studies were carried out using artificial intelligence/machine learning and optimization algorithms to estimate the number of failures of aircraft equipment. For this purpose, a two-stage hybrid data preparation model was proposed to increase the success of the failure estimation. In the first stage, ReliefF, a feature selection method, was used to find the most effective and ineffective parameters through quality evaluation. In the second, a modified K-means algorithm was used to eliminate noisy or inconsistent data. The performance of the hybrid data preparation model in the maintenance dataset of the equipment was evaluated, using machine learning algorithms, including Artificial Neural Network (ANN), Multilayer Perceptron (MLP), Support Vector Regression (DVR) and Linear Regression (DR) and optimization algorithms, including Genetic Algorithm (GA), Differential Evolution Algorithm (DE) and Scatter Search (SS). The results indicated that the hybrid data preparation model increases the success in estimating the number of failures of the equipment.