No description available.

No description available.

No description available.

We investigate the stabilization of unstable multidimensional partially observed single-station, multi-sensor (single-controller) and multi-controller (single-sensor) linear systems controlled over discrete noiseless channels under fixed-rate information constraints. Stability is achieved under communication requirements that are asymptotically tight in the limit of large sampling periods. Through the use of similarity transforms, sampling and random-time drift conditions we obtain a coding and control policy leading to the existence of a unique invariant distribution and finite second moment for the sampled state. We use a vector stabilization scheme in which all modes of the linear system visit a compact set together infinitely often. / Thesis (Master, Mathematics & Statistics) -- Queen's University, 2012-12-06 15:06:37.449

The control of object manipulation tasks involves the close interplay of predictive and reactive control mechanisms. For example, when lifting an object, people typically predict the weight based on object size and material as well as sensorimotor memory obtained from previous lifts of the object. When lifting objects with a precision grip, people increase vertical load force to a target level that slightly exceeds the predicted weight. When the object is heavier than expected, the mismatch between expected and actual tactile signals associated with lift-off triggers a corrective action within ~100 ms, that involves probing increases in load force that continue until the object is lifted. Here we investigated whether this correction action can be adaptively influenced by experience. Participants repeatedly lifted an object that was instrumented with force sensors to measure the forces applied by the fingertips, with weight that could be varied without the knowledge of the participant. In 80% of trials, the weight was set to 2 N and, in different blocks of 110 trials, the remaining 20 % of trials (2 trials randomly selected from each successive 10 trials) was set to either 4 or 6 N. We found that the rate of change of the reflexively triggered increase in load force that occurred in the 4 or 6 N trials, scaled with the additional weight. That is, following the initial increase in load force to ~2 N, the subsequent increase in load force was more rapid for the 6 N object than the 4 N object. In contrast, the onset time of the reactive increase in load force was independent of the additional weight. Finally, this adaptation of reactive load force control took place quickly and was evident after only a few lifts of the heavier weight. These results indicate that the reactive increases in load force that occur when a lifted object is heavier than expected can be adapted and tuned, to refine behavior. This further suggests that multiple predictions can be generated about object weight when lifting. / Thesis (Master, Neuroscience Studies) -- Queen's University, 2013-02-02 13:34:20.533

No description available.

This study considers the practical implementation of semi active suspension control on a test vehicle. The aim is to assess the viability of using simple models to describe the suspension state dynamics, both for simulation purposes and to enable feedback control. To this end, system identification techniques are employed to estimate model parameters, and the design of a suitable real-time observer is considered. The controller design itself, is not studied. The well known quarter-car approach is used to develop suspension models, and a simple time-domain method is presented for parameter identification. Simulated identification tests lead to the development of a new time-domain approach, based on an integrated form of the system differential equations. This is shown to have significant advantages over the direct identification method, under certain disturbance conditions. In a case study, the new method is applied to identify suspension parameters for the test vehicle, using data acquired during rig tests. Analysis of model errors then motivates a separate modelling exercise on the dampers in isolation. This yields a more complex nonlinear form of the model, which is finally validated against the simplest linear model, using measurements from the rig. For real-time state estimation, a linear Kalman Filter observer is developed. The observer's sensor requirement is examined, along with other parameters affecting the design, in a factorial experiment based on simulated and rig measured data. This allows an informed choice of the smallest sensor set that affords a high level of state estimation accuracy. The performance of the observer is also examined in the context of simulated closed-loop control. Finally, the design of an observer for the vehicle on the road is considered, and tests are carried out with the vehicle under semi-active control. Within the accuracy of state estimates derived from the available transducers, the observer performs well.

Emerging techniques of intelligent or learning control seem attractive for applications in manufacturing and robotics. It is however important to understand the capabilities of such control systems. In the past the inverted pendulum has been used as a test case. The thesis begins with an examination of whether the inverted pendulum or polecart balancing problem is a representative problem for experimentation for learning controllers for complex nonlinear systems. Results of previous research concerning the inverted pendulum problem are presented to show that this problem is not sufficiently testing. This thesis therefore concentrates on the control of the inverted pendulum with an additional degree of freedom as a testing demonstrator problem for learning control system experimentation. A flexible pole is used in place of a rigid one. The transverse displacement of the flexible pole adds a degree of freedom to the system. The dynamics of this new system are more complex as the system needs additional parameters to be defIned due to the pole's elastic deflection. This problem also has many of the signifIcant features associated with flexible robots with lightweight links as applied in manufacturing. Novel neural network and fuzzy control systems are presented that control such a system both in simulation and real time. A fuzzy-genetic approach is also demonstrated that allows the creation of fuzzy control systems without the use of extensive knowledge.

This work is concerned with the effects of Adaptive Differential Pulse Code Modulation (ADPCM) on digital data transmitted at 9.6 kbit/s. To facilitate the study, a system model was developed. The model simulates the complete communication system, including true bandpass simulation of the modulation and demodulation processes. The model was used to observe the behaviour of the ADPCM link and the distortion it produces when digital data are transmitted at 9.6 kbit/s. The observations were employed to devise two ADPCM distortion cancellation schemes. Scheme-1 employs a look-up-table. The cancellation is here determined by the last two detected symbols and the sampling phase. Scheme-2 operates without a memory. The cancellation is achieved by signal level compensation. The performances of these schemes, operating with different detectors, are compared with the performance of the system when a non-linear equaliser is applied to the digital data in the absence of ADPCM distortion. Tests were conducted using two channels: CH-1 which introduces mild distortion into the transmitted digital signal, and CH-2 which introduces significant distortion. The detectors used in the tests are a linear equaliser, a non-linear equaliser and a near-maximum-likelihood detector. Scheme-2, which is the simpler of the two schemes, when used with the near-maximum-likelihood detector has a significantly better performance than the other systems tested.

No description available.