NO FULL TEXT AVAILABLE. This thesis contains 3rd party copyright material. ----- More often than not, impaired mobility results in fewer opportunities for socialising, and pursuing personal goals and pursuits. This can produce depression, anxiety and social isolation for physically impaired people. As an aid to mobility, an electric-powered wheelchair can assist a disabled person in improving their quality of life. The safe control of a conventional powered wheelchair nevertheless requires a significant level of skill, attention, judgment and appropriate user response. Like an automobile user, a wheelchair user may also encounter an accident and injury.
To safely and comfortably accommodate a powered wheelchair user, there has been great advancement in recent decades in the development of autonomous and semiautonomous powered wheelchairs. However, the majority of this research focus on a supervisory control level related to developing navigation algorithms, shared-control approaches, and hand-free control strategies. Due to parameter uncertainties and external disturbances, an optimal performance and overall system robustness depends heavily on the low control level involved in precision motion control. Surprisingly, there is still very little research into control techniques for this level.
Although many advanced control techniques have been developed for the low level control, none of these techniques are expected to be optimal. This is largely because they have not treated the powered wheelchair as a multivariable system. Due to interactions between different inputs and outputs, a control design task for a multivariable system can also be complicated. Some very effective solutions for the inherent issues pertaining to multivariable systems involve decoupling techniques. These endeavour to simplify a multivariable control problem through reduction to a series of scalar control problems. However, there is still a scarcity of research regarding robustness under effects of uncertainty and external disturbance.
In this thesis, a combination of decoupling techniques and advanced control strategies leads to three new advanced multivariable control approaches. These approaches offer systematic solutions for problems of precision motion control in the low control level of a wheelchair system. New multivariable control approaches reduce a multivariable control problem into a series of scalar control problems, as such they have the advantage of involving less complexity and computation. Based on an identification framework, an approximate dynamic multivariable model is generated for a powered wheelchair to facilitate control implementation.
The first multivariable control approach is derived from a combination of a Triangularization technique and a Model Predictive Control strategy. This approach can guarantee an optimal performance for a linear dynamic multivariable system. Apart from providing a systematic solution for a linear multivariable system, this approach also handles control problems naturally, takes into account of actuator limitations, and allows operation closer to constraints. Real-time implementation indicates the effectiveness of this approach for the powered wheelchair.
The second multivariable control approach is derived from a combining a Triangularization technique and Optimal Neural Network Control strategy. This new approach guarantees the optimal performance for a dynamic multivariable system against parameter uncertainties. Experiment results related to path-following control of the powered wheelchair system reveal that, regardless of parameter uncertainty effects, this second multivariable control approach considerably improves system performance, robustness, and accuracy, in comparison with various popular control approaches. The results also demonstrate that coupling effects in the wheelchair dynamics are substantially reduced.
The third multivariable control approach is derived from the combination of Diagonalization technique and Robust Neuro-Sliding Mode Control strategy. This novel approach can effectively cope in real time with parameter uncertainties and external disturbances to achieve robustness and a desired performance for a multivariable system. This third approach fully reduces the coupling effects on a multivariable system, eliminates the chattering phenomena, and avoids plant Jacobian calculation. Furthermore, it can also achieve fast and global convergence with less computation. The effectiveness of this novel approach has been verified through real-time implementation of the powered wheelchair system. The results guarantee robustness and desired performance of the overall system, even with the compounding effect of parameter uncertainties and external disturbances.