Abstract
Induction motors are widely used in industry, making reliable and non-invasive condition monitoring significant for high-performance operations. Stator-current analysis is attractive because it is inexpensive and easy to implement. Conventional approaches, however, mainly target fault sidebands around the supply frequency, which often suffer from low signal-to-noise ratio (SNR) in practice, particularly under inverter-fed conditions where pulse width modulation (PWM) switching harmonics, winding non-idealities and fluctuating load torque, result in masking fault features. By contrast, higher-frequency components, such as those around the principal slot harmonic (PSH), are often less affected and may carry useful fault information. To address these challenges, this thesis approaches induction-motor condition monitoring through modulation signal bispectrum (MSB) analysis for stator-current feature extraction, and multiple coupled circuit (MCC) based modelling for fault-mechanism analysis and interpretation. The main contributions and findings are summarised as follows:(1) MSB interpretation and SNR-driven parameter optimisation
The MSB is interpreted as a fourth-order moment spectrum, and its noise-reduction behaviour is characterised using simulated modulation signals. An MSB-based SNR index is then used as the objective in a systematic grid search over record length, segment length and overlap, yielding an SNR-driven MSB configuration (i.e., the parameter set that maximises the proposed index within the investigated ranges). The resulting configuration is applied not only to supply-frequency sidebands but also to higher-frequency modulation components (e.g., around the PSH).
(2) Improved MCC modelling incorporates with PWM control voltage
The MCC model is extended to incorporate sinusoidal and trapezoidal winding distributions, open-loop PWM supply and calibrated parameters. The model and simulation have found that the PWM is the main sources leading to low SNR signals. A linear simplification of dynamic-eccentricity inductances is introduced to reduce computation while preserving key spectral characteristics, enabling efficient prediction of higher-frequency fault-related sidebands, on average, the computation time is reduced by 25% while the key resulting error remains within 1%.
(3) Experimental validation and current-based diagnostic performance
The proposed modelling and MSB coherent sideband energy (CSE) analysis are assessed experimentally under inverter-fed operation for typical faults including broken-rotor-bar, eccentricity and bearing defects. Within the investigated conditions, the MSB approach increases the SNR of higher-frequency sidebands by 7–9 dB compared with reference demodulation methods and shows consistent performance across the tested data lengths. The extended MCC model reproduces measured phase currents with less than 7% RMS error and predicts higher-frequency fault-related sidebands. Experimental MSB and CSE results indicate that broken-rotor-bar and eccentricity faults remain detectable via higher-frequency modulation components under inverter-fed conditions, whereas bearing faults with 1 mm defect on race ways produce weak and statistically insignificant diagnostic signatures, which is due mainly to the large size of the motor under investigation.
| Date of Award | 5 Feb 2026 |
|---|---|
| Original language | English |
| Supervisor | Fengshou Gu (Main Supervisor) & Andrew Ball (Co-Supervisor) |