Jonathan Currie

Lightweight Model Predictive Control intended for embedded applications

Abstract The computational demands of Model predictive control (MPC) are well known, and due to its internal constrained optimiser, historically has been ill-suited for embedded controllers designed to tackle high-speed applications. This paper explores the options of developing a low-cost lightweight MPC controller destined for micro-controller or FPGA architectures for modest applications demanding reasonable controller horizons. An object based MPC development tool is introduced and applied to an experimental 4-tank level system to explore the performance of the algorithm.

Software Evolution Of A Hexapod Robot Walking Gait

This paper describes the process of evolving a Hexapod Robot walking gait within a simulated software environment. Initially a 3D mathematical model of the robot was created using Matlab, simulating full motion of each the robots six legs. Each leg has three degrees of freedom (DOF), allowing the robot to move in both lateral and rotational directions. The simulation allows the robots movements to be determined, based on a repeated sequence of static leg positions, and was used with a genetic algorithm (GA) to evolve walking gaits for the robot. The walking gait is described by a chromosome which is an 18times9 look up table (LUT) that lists the angular position of all eighteen servo motors over a sequence of nine discrete static positions, which will describe the walking gait. Each position in the LUT has twenty discrete states ranging from -45deg to +45deg, allowing a flexible range of achievable motions, while maintaining sensible evaluation limits. Successful evolution of these gaits was performed within 700 generations.

Rigorously modelling steam utility systems for mixed integer optimization

Given that industrial utility systems are essentially large energy converters, it is surprising that they are so often forgotten or ignored when optimizing plant performance. Significant operational savings are possible simply by redistributing steam generation and consumption, without adding extra equipment, and with minimal investment. However due to the discrete nature of a utility system where equipment can switched in and out of service, steam flows redistributed, and zero-flow conditions are normal, the optimizing of utility system requires a rigorous model based on thermodynamics and state-of-the-art numerical algorithms. This paper proposes a mixed integer modelling strategy to approximate a rigorous simulator model, combining regressions from literature, industrial experience and process specific knowledge resulting in a model suitable for optimization. Two case studies are presented to demonstrate the efficiency of the modelling design, a hypothetical three header model with cogeneration and a four header refinery utility system. Both systems are optimized using BONMIN in less than a quarter of a second on a standard desktop PC and result in substantial economic improvements.

Software integration for online dynamic simulation applications

Current process industries such as refineries and pharmaceutical industries have been facing increasing challenges with respect to productivity, reducing waste and energy consumption, as well as environmental and safety issues becomingly increasingly comprehensive. One way to address these issues is to utilize online dynamic simulation via process modelling and control software. While there are multiple process simulation packages available, such as VMGsim, HYSYS, Aspen etc., comprehensive control algorithms such as Model Predictive Control (MPC) or many Advanced Process Control (APC) algorithms which are available in the literature and third party packages cannot be easily and/or directly implemented in these packages. In this paper, we develop a software integration framework which allows Matlab to drive dynamic simulations built within a software simulator. The framework will open these high-fidelity process simulation packages for engineers and researchers to test, implement and tune advanced process control strategies, in both offline and online scenarios.

A systematic approach to modeling organic Rankine cycle systems for global optimization

The demand for organic Rankine cycle (ORC) systems to be efficient and economically competitive drives the need for a reliable and robust modeling approach that is suitable for optimization. However existing commercial simulation software is not typically tailored for optimization and they generally cannot guarantee a global optimum. This paper proposes a modeling approach to approximate a rigorous simulation model that is suitable for global optimization. This involves a combination of regression and thermodynamic analysis, in addition to integer programming techniques. Three different solvers, namely COBYLA, SCIP, and BARON, are used to optimize the ORC model and are compared against each other to demonstrate the prospect of achieving the global optimum using this approach. In addition, this paper also presents a technique to improve the model accuracy by using a piecewise fit to approximate the output characteristic of the ORC unit operations.