Induction motors (IMs) are widely used as the primary driver in various industries, moreover with the rapid progress in electrification. To ensure safe and reliable operation of large volumes of motors, condition monitoring(CM) technologies have advanced greatly in recent years. However, due to the complexity of motors, existing CM technologies often suffer from insufficient accuracy in fault diagnosis, high costs and difficulties for system deployment and implementation. To address these limitations, this PhD research focuses on investigation and development of a novel monitoring system, based on emerging on-rotor sensing (ORS) technologies, that acquires data from a rotating shaft with wireless data transfer. In particular, the research has undergone four critical milestones to empower ORS systems: 1) Characterising the vibration responses of rotors by establishing a new model of rotor-bearing system with rotor flexibility, axial asymmetricity and nonlinear bearing excitations; 2) Advancing the ORS system, optimising the energy consumption and energy harvesting for long-term observation of the responses at either end of the rotor; 3) Evaluating the ORS system with the motors under different fault cases including bearing defects, broken rotor bar and rotor eccentricity accounting for over 60% breakdowns of IMs; 4) Developing a self-powered intelligent ORS by identifying and embedding key edging computing algorithms onto the system to obtain for the real-time monitoring of IMs.By achieving each milestone, the research had resulted in a number of important new findings and corresponding novel contributions. Characterising the vibration responses of IM rotors has found that the 4th and 5th bending modes of the rotor in the frequency up to 3000 Hz show significant differences in amplitudes, providing necessary information for localising and diagnosing the faulty bearings and rotor misalignment with just a single sensor node installed at either the drive end or non-drive end of the rotor. Moreover, modelling the system, including the flexibility and asymmetricity of the rotor along with nonlinear bearing excitation, provides accurate predictions of bearing faults at different severity, types and locations. Based on model predictions and thorough literature reviews along with low-power consumption requirements, it is identified that medium-size IMs produce vibrations within the range of ±16 g in a frequency band up to 3000 Hz, which can reflect key features of bearing and rotor-related faults. Therefore, a cost-effective design of ORS hardware was achieved by choosing electronic components with low cost, minimal power consumption, high resolution, and fast sampling rates. From numerous options, an ICM42688 MEMS accelerometer and the nRF52840 Cortex-M4F processor embedded with BLE 5.0 module were identified to be the core components of the system. Subsequently, the energy usage for raw data transfer is 50 % more than that of basic edging processing (FFT), but edging computing is preferred for real-time monitoring of a large system and reducing data storage and intensive processing in a cloud-based system. Furthermore, the rotor component-modulated ORS output signals relating to different bearing defects are also clarified for possible sensor installation errors and online deviation identification. The offline evaluation of ORS system based on industrial IMs has found that the prototype ORS system can capture data in real-time at a rate as high as 16 kHz at 18-bit data accuracy. The vibration responses obtained by ORS are very close to that of the model prediction in terms of magnitudes and frequency bands, such as high fidelity paving foundations for implementing digital twin-based monitoring. Importantly, it provides full diagnostics about bearing fault severity, location and components, which is achieved by popular FFT and Hilbert envelope spectra. Comparatively, conventional on-housing vibration signals cannot achieve these because of the low SNR signals due to complicated transmission path effects. During development of on-board intelligence onto ORS it has been identified that TKEO is a more efficient and sensitive algorithm for the edging computing to be realised in ORS sensor node. Testing has evidenced that the TKEO algorithm not only enables more accurate detection of bearing faults in the time domain, but also consumes fewer computing resources on the microcontroller, compared to the conventional Hilbert signal-based envelope.
| Date of Award | 12 Mar 2025 |
|---|
| Original language | English |
|---|
| Supervisor | Fengshou Gu (Main Supervisor) & Andrew Ball (Co-Supervisor) |
|---|