FPGA Tabanlı Hata Tespit ve Sınıflandırma Sistemi Tasarımı

Abstract

Data-driven maintenance is defined as making inferences about the health status of electromechanical systems and producing maintenance plans using condition monitoring technology. Condition monitoring is the process of performing fault detection, fault classification and remaining useful life analysis of systems by using physical data, that is collected from electromechanical systems, with regression and classification models. In this regard, data acquisition, signal processing and machine learning subjects in the literature are utilized. Data-driven maintenance is widely used in the manufacturing, automotive, railway transportation, large construction equipment, construction and consumer electronics industries. However, the use of the data-driven maintenance approach in the Defense Industry is limited. Within the scope of this thesis study, vibration and current data were collected from a real Defense Industry product in different health conditions. The collected data was used in feature extraction and developing different machine learning models. The performance of Support Vector Machine (SVM), Decision Trees (DT), K-Nearest Neighbor (KNN), Random Forest (RF) and Linear Regression (LR) models with different features was examined. The SVM model that gave the most successful results was implemented in the FPGA and the performance of the FPGA design was tested. As a result of the thesis study, a data-driven maintenance study was carried out and applied on a real military mechanical system.