A Study on Condition Monitoring for Permanent Magnet Synchronous Machines: Fault Diagnosis and Application to Optimal Scheduling of Microgrid

Abstract

This thesis proposes fault diagnosis methods for an interturn short circuit fault (ISCF) and demagnetization fault (DF) in permanent magnet synchronous machines (PMSMs). The ISCF and DF are frequently occurring faults in PMSMs and these directly reduce the efficiency and reliability of the machines. To diagnose these faults in an early stage before they get worse and spread to other faults, data-driven fault diagnosis methods are proposed. At first, a diagnostic method for an ISCF with an attention-based recurrent neural network is proposed. An encoder-decoder architecture with an attention mechanism estimates a fault indicator that is directly related to the severity of the ISCF, using three-phase currents and rotational speed as inputs. The attention mechanism helps the decoding process for accurate estimation and solves the long-term dependency problem. The proposed diagnostic method effectively diagnoses the ISCF in various operating conditions. Second, a fault diagnosis method that estimates the severity of an ISCF and DF in a PMSM using a self-attention-based severity estimation network (SASEN) is proposed. The SASEN receives the positive-sequence voltage and current, negative-sequence voltage and current as inputs, and outputs two fault indicators directly related to the severities of the ISCF and DF. The network diagnoses the ISCF and DF by estimating the severities of the faults through regression. In addition, the proposed method can diagnose the hybrid fault in which the ISCF and the DF occur simultaneously. The proposed diagnosis method diagnoses the faults under various load torques and fault conditions. Finally, a microgrid optimal scheduling strategy by incorporating the fault diagnosis is proposed. The faulty generator is gradually reduced in generation efficiency and capacity. Operating a microgrid without considering these impacts increases operating costs and reduces reliability. To overcome this problem, the performance degradation of the generator is analyzed and formulated using finite element analysis. The proposed optimal scheduling strategy reduces the total operating cost compared to the conventional methods by adaptively reflecting the increase in operating cost and the decrease in the capacity of the faulty generator. In addition, sudden shutdowns can be prevented by reflecting the remaining useful life of the generator.