Deep convolutional neural network based broken magnet detection of PMSM using finite element analysis
Deep convolutional neural network based broken magnet detection of PMSM using finite element analysis
Dosyalar
Tarih
2023-02-13
Yazarlar
Matanagh, Amin Ghafouri
Süreli Yayın başlığı
Süreli Yayın ISSN
Cilt Başlığı
Yayınevi
Graduate School
Özet
Electrical machines are used more frequently, which raises the need for good validity and safety. Unanticipated failures of these devices, especially in mission-critical applications, can result in irreparable system failures. These failures in electrical machines could have dangerous effects on human life in medical robotics, aerospace, and the military. Adding additional modules to complex applications results in high cost, volume, and complexity. Artificial Intelligence (AI), Machine Learning (ML), and Deep Learning (DL) technologies are essential for continuously monitoring system performance and running tests on machines regularly to anticipate and prevent potential failures that could cause harm to the system or people. PMSMs are becoming increasingly popular for a variety of applications due to their ability to operate at both high and low speeds, their improved power density, their low rotor inertia which makes them easy to control, and their availability in different packages and sizes. These motors are being used in electric vehicles. In general, three types of electrical, mechanical, and magnetic faults can be classified according to the nature of the faults in motors. The class of electrical faults includes incorrect connection of the motor windings, grounding errors, short circuits of the stator phases, and open circuits of the entire phases. In contrast, mechanical faults such as shaft bending, bolt loss, bearing faults, and air gap eccentricity are all grouped together. In addition, excessive heat, imbalanced stator current, and fluctuating short-circuit currents can lead to the weakening of magnets, a type of magnetic malfunction. Different parameters of PMSM are thoroughly studied to understand electrical and mechanical faults. In this thesis, broken magnet faults and their corresponding reflectance on the variables of PMSM motors are investigated. The 3D simulation of the finite element machine approach is carried out for different shapes of cracks, and as a result, several characteristics are analyzed compared to the other variables. The winding current shows a significant difference by implementing various damages. In this research, the deep convolutional neural network (DCNN) was performed for broken magnet detection and classification in PMSM by using the data set obtained from simulation current curves. Ansys Electronics and Phyton design, implement cracks, train, validate, and test DCNN in PMSM motors. In this study, we have measured precision, recall, loss and calculated training and validation accuracies in prediction. As a result, 99.8% training accuracy and 98.9% validation accuracy were achieved with the DCNN model based on winding current data sets. Subsequently, the study proposed the integration of the best-performing DCNN models in the crack detection of PMSM motors.
Açıklama
Thesis (M.Sc.) -- İstanbul Technical University, Graduate School, 2023
Anahtar kelimeler
Fnite element analysis,
Sonlu eleman yöntemi,
Electric motors,
Elektrik motorları,
Neural network,
Sinir ağları,
Magnet motors,
Mıknatıs motorlar