Model-on-demand nonlinear estimation and model predictive control: Novel methodologies for process control and supply chain management

An integrated methodology for nonlinear system identification and control using the concept of Model-on-Demand (MoD) estimation is proposed. The methodology provides a structured framework that allows the user to translate a priori process information into a suitable identification experiment, enables judicious MoD user-decisions, and ultimately results in a Model-on-Demand Model Predictive Controller (MoDMPC). The major components of this methodology developed in this dissertation include the derivation of “plant-friendly” guidelines for multi-level Pseudo-Random Sequences (multi-level PRS) inputs, the application of new measures of nonlinear dynamical data quality, and enhancements of the MoDMPC formulation. With this methodology, the user is able to construct a high performance, nonlinear controller with less effort compared to methods which make use of global modeling techniques such as neural networks and semi-physical modeling. Various performance aspects of the methodology are demonstrated on a pilot scale brine-water mix tank, a simulated Rapid Thermal Processing (RTP) tool, and a non-adiabatic Continuous Stirred Tank Reactor (CSTR) simulation.; The available cost reduction associated with supply chain inefficiencies in most industries has been conservatively estimated to range in the billions of dollars. With the infrastructure now provided by the internet and today's e-business emphasis, in combination with the need for management systems which can deal efficiently with uncertainty and data inaccuracy, an opportunity exists to apply control and systems ideas in the operations management arena. MoD is shown to provide a viable method of estimating possible nonlinearities in the dynamics of semiconductor manufacturing lines and a Model Predictive Control (MPC) framework is presented for management of supply chains. The interpretation of the supply chain problem into process control terminology is examined and modifications to improve the customer service capability of the standard MPC algorithm are proposed. This work is supported with the extensive analysis of a two-node manufacturer-retailer supply chain whose results can be readily interpreted and conceptualized. Proof-of-concept results demonstrating the benefits of MPC on larger, more challenging networks are presented using a two-product, six-node, three-echelon demand network model developed by Intel Corporation. Results from this problem are used to demonstrate the ability of the MPC framework to deal with uncertainty, constraints, and a variety of information sharing configurations.