Makine Öğrenmesi Kullanılarak Kestirimci Bakım

Abstract

Within the framework of the Industry 4.0 concept, the production processes undergo significant transformations. Less manpower use, reduced costs and increased efficiency are the main motivations for this transformation, with widespread automation. In this process, industrial facilities have started to be equipped with more industrial robots and auxiliary devices than ever before. The data transmitted by sensors measuring various values such as sound, temperature, fluidity and pressure, which are produced by these equipments on the production band, and their subsequent soundings, have increased greatly. These data provide opportunities for increasing the desired efficiency. Commonly used preventive maintenance approach is often criticized for the costs of maintenance that are considered unnecessary. This type of maintenance, where the current information of the system is not evaluated and is often repeated with short intervals for safety. In spite of this frequent maintenance, the possibility of a malfunction is not completely eliminated. Predictive maintenance techniques that estimate the need for maintenance by evaluating the sensor data monitoring the system are giving important promises in providing cost efficiency and high safety. The models created by training with the data obtained from the sensors can detect the degradation over time in the equipment and warn the operator about the need for maintenance before failure. In this thesis, the use of deep learning technique, which shows successful results in many areas in recent years in maintenance planning and its performance according to other machine learning methods are examined. The data set containing turbofan jet motor sensor measurements was used as application data set. The data set shows similar characteristics with the data that are likely to be produced in many maintenance scenarios.