Date of Award

12-20-2018

Publication Type

Doctoral Thesis

Degree Name

Ph.D.

Department

Electrical and Computer Engineering

Keywords

Data Mining, Demagnetization Detection, Fault Diagnosis, Machine Learning, PMSM Motor Control, Signal Processing

Supervisor

Kar, Narayan

Rights

info:eu-repo/semantics/openAccess

Abstract

Monitoring and detecting PM flux linkage is important to maintain a stable permanent magnet synchronous motor (PMSM) operation. The key problems that need to be solved at this stage are to: 1) establish a demagnetization magnetic flux model that takes into account the influence of various nonlinear and complex factors to reveal the demagnetization mechanism; 2) explore the relationship between different factors and demagnetizing magnetic field, to detect the demagnetization in the early stage; and 3) propose post-demagnetization measures. This thesis investigates permanent magnet (PM) demagnetization detection for PMSM machines to achieve high-performance and reliable machine drive for practical industrial and consumer applications. In this thesis, theoretical analysis, numerical calculation as well as experimental investigations are carried out to systematically study the demagnetization detection mechanism and post-demagnetization measures for permanent magnet synchronous motors. At first a flux based acoustic noise model is proposed to analyze online PM demagnetization detection by using a back propagation neural network (BPNN) with acoustic noise data. In this method, the PM demagnetization is detected by means of comparing the measured acoustic signal of PMSM with an acoustic signal library of seven acoustical indicators. Then torque ripple is chosen for online PM demagnetization diagnosis by using continuous wavelet transforms (CWT) and Grey System Theory (GST). This model is able to reveal the relationship between torque variation and PM electromagnetic interferences. After demagnetization being detected, a current regulation strategy is proposed to minimize the torque ripples induced by PM demagnetization. Next, in order to compare the demagnetization detection accuracy, different data mining techniques, Vold-Kalman filtering order tracking (VKF-OT) and dynamic Bayesian network (DBN) based detection approach is applied to real-time PM flux monitoring through torque ripple again. VKF-OT is introduced to track the order of torque ripple of PMSM running in transient state. Lastly, the combination of acoustic noise and torque is investigated for demagnetization detection by using multi-sensor information fusion to improve the system redundancy and accuracy. Bayesian network based multi-sensor information fusion is then proposed to detect the demagnetization ratio from the extracted features. During the analysis of demagnetization detection methods, the proposed PM detection approaches both form torque ripple and acoustic noise are extensively evaluated on a laboratory PM machine drive system under different speeds, load conditions, and temperatures.

Download

Share

COinS