Abstract
Reliable sensor data forms the core of critical process monitoring and control in modern manufacturing systems. From improving performance of machines in operation to efficient long-term maintenance, decisions based upon data, rather than human observation, are becoming more prevalent.One example, examined in this thesis, is vibration (free, forced, and self-excited) experienced on machine tools during operation. These vibrations can affect machined part accuracy (surface finish and roughness), machine and tool condition (wear and life), and machining efficiency. For the measurement of vibration, various high-cost state-of-the-art systems such as Integrated Electronics Piezo-Electric (IEPE) vibration sensors, laser vibrometers, and non-contact displacement transducers have been widely employed in industrial-manufacturing. Recent development of a variety of industrial-grade MicroElectroMechanical systems (MEMS)-based vibration sensors is proving to be a low-cost yet effective alternative to the aforementioned measurement systems; with an added possibility of being connected in numbers, essentially creating a network of sensors within the manufacturing sector to support smart machining and intelligent production.
The quality of any decision-making is governed by the choice of sensor and its location with respect to the effect of interest. This thesis develops a generalised framework for calculating and reducing uncertainty for vibration measurement, including both the sensor and its application. This was achieved by first identifying all sources of uncertainty through literature review, technical documentation, and experience of instrumentation design practice. A formalised method of evaluating the performance of those sensor characteristics that contribute to uncertainty was then produced. The framework recommends methods by which uncertainty can be reduced through calibration, compensation, or control.
The framework was tested by an in-depth case study which analysed industrial digital MEMS vibration sensors. The specific performance tests were devised according to the framework and conducted through bench-tests and on industrial CNC milling machine. This led to the development of an approach to reduce systematic and random errors, and therefore the overall uncertainty in vibration measurement. The systematic errors were mitigated using calibration and compensation routines, while the residual noise effects were mitigated using an implementation of Unscented Kalman Filtering (UKF).
The expanded uncertainty for the as-supplied MEMS vibration sensors was evaluated to be in the range of 1.56% to 11.38%, dependent on the sampling rate or relationship to natural frequency of experimental setup. Differences between the MEMS and reference IEPE in frequency measurements was recorded to be 27.93%. It was found that the sampling errors intrinsic to the MEMS sensor and data capture system were the dominant term for the spectral analysis, but that it could be reduced to within ±0.10% after correction. Similarly, the compensation of systematic baseline errors (bias, drift, sensitivity, temperature) led to an improvement of 90.36 %, while the noise reduction based on UKF led to a reduction of 59.10 % in residual noise for the investigated vibration sensor. Future work can lead to improvement and validation of the proposed framework to a broader class of vibration sensors and applications.
| Date of Award | 28 Sep 2022 |
|---|---|
| Original language | English |
| Supervisor | Naeem Mian (Main Supervisor), Simon Fletcher (Co-Supervisor) & Andrew Longstaff (Co-Supervisor) |