Bradley R. Holt

A Neural Network Structure for System Identification

Establishing a dynamic process model is the first step toward implementing a modern control algorithm. Because of the complexity of chemical processes, most models are identified, that is, determined from a known input/output sequence. Furthermore, models are usually linear and time invariant. This research focuses on the application of neural networks to the development of dynamic models. In particular, this paper presents a modification of the layered structure used most commonly with the Backward Error Propagation algorithm The modification is the addition of a set of weights connected directly from the input to the output layer, weights which contribute in a linear manner to the network output. This creates a number of advantageous compared to traditional structures, including initialization of network parameters based on process knowledge, additional insight to the leaning algorithm, and enhanced extrapolation outside of examples the learning data set.

A comparison of neural networks, non-linear biased regression and a genetic algorithm for dynamic model identification

Abstract A variety of non-linear modeling techniques were applied to a single input/single output dynamic model identification problem. Results of the tests show that the prediction error of an artificial neural network with direct linear feed through terms is nearly as good or better than the other methods when tested on new data. However, non-linear models with nearly equal and occasionally better performance can be developed (including the selection of the model form and order) with a genetic algorithm (GA) in far less computer time. The GA derived models have the additional advantage of being more parsimonious and can be reparameterized, if need be, extremely rapidly. The non-linear biased regression techniques tested typically had larger, though possibly acceptable, prediction errors. These model structures offer the advantage of low computational requirements and reproducibility, i.e. the same model is produced each time for a given data set.

Design of resilient processing plants. New characterization of the effect of RHP zeros

Abstract Right Half Plane (RHP) zeros restrict the achievable closed loop performance independent of controller design. A new characterization of all achievable closed loop setpoint/output transfer matrices is provided in terms of “zero-directions”. The zero directions also give some insight into what forms of partial decoupling are preferable.

An optimization approach to robust nonlinear control design

Abstract A new procedure of robust control design is presented for nonlinear systems with parametric uncertainty and disturbance. This nonlinear optimization-based approach can design robust nonlinear controllers of widely different structures and arbitrary complexity. By taking the worst-case design approach, a minimax optimization problem is formulated. A number of state-of-the-art optimization procedures are explored as possible candidates for minimaximization. It is determined that the tools of non-smooth analysis and optimization are the most useful for the efficient solution of practical size worst-case design problems. Included in the paper are the significant details of a special two-tier structured minimax programming algorithm implemented in this work. The overall algorithm is used in solving two robust nonlinear control design problems. The controller designs produced by the minimax optimization algorithm are discussed and the performance of the controllers is studied by simulations.

Neural network techniques for modeling sensor data

Some possible approaches to the use of neural networks for interpreting sensor, and particular spectral type, data are sketched. It is demonstrated how the structure of the neural network can be chosen to provide the capacity to fall back to the linear case when appropriate. An approach to the problem of underdetermined systems, based on adding random Gaussian noise to prevent the neural network from being locked into a local minimum, is presented.<<ETX>>

Scheduled Controllers for Robust, Non-Linear Control: Introduction and Application to the Shell Standard Control Problem

This paper presents a beginning philosophy for robust control using non-linear controllers. The motivation for the approach is the potential offered by a combination of variations on non-linear dynamic programming combined with scheduled or table-look up type controllers. The paper also analyzes and presents a solution to the Shell standard control problem using these ideas. Finally, it suggests that a steady state form of the IMC controller may provide a simple yet satisfactory control option for certain types of MIMO problems.

Control of Autoclave Processing of Polymeric Composites

Polymeric composite represent a relatively new and increasingly important material. A key step in one of the most important fabrication techniques involves curing the composite laminate under pressure and temperature cycles. The quality of the resulting component is highly dependent on this processing, a process which has traditionally been performed in an open loop manner. This paper provides a qualitative review of the production of polymeric composites components from prepreg. It examines the motivation and issues involved in applying feedback control to the processing of the components in an autoclave and qualitatively reviews the possible control strategies.