Accelerated Design And Parameter Tuning Of FPGA Based Predictive Controller

Model predictive control has been widely recognized by field of industrialprocess control as an optimization algorithm that can explicitly handlecomplicated constraints. However, MPC has some intrinsic difficulties in itsapplication. Firstly, the receding horizon optimization of MPC needs to solvea real-time optimization problem, whereas the complicity of on-linecomputation prohibits its application in embedded systems. Though recentlydeveloped Field Programmable Gate Array (FPGA) provides a good platformfor the on-line matrix computation of MPC by its unique parallel and effectiveassembly line structure, the on-line inversion of matrix nevertheless consumeslarge amount of time and resources. Secondly, it lacks quantified relationshipbetween the design parameter of MPC algorithm and the control performance.Hence, in real applications, many simulations are required for posteriorsetting of control parameters. For predictive controllers in embedded systems,this demands a joint simulation environment with the host computer todownload control parameters, which may be inconvenient for independentindustrial field applications. Moreover, highly experienced field techniciansare required for a satisfactory control performance.For the two problems above, this paper concerns the design and application ofMPC in FPGA and establishes the following researches:1) Design and implementation of FPGA based predictive controller with matrix inversion accomplished in the hardware. Concerning the problem offurther speeding up the computation of MPC in FPGA, we consider how toachieve effective computation of matrix inversion by using resources inFPGA. Noting that the matrices to be inversed in MPC are generally positivedefinite, we use a simple inversion algorithm for positive-definite matrices.This algorithm is implemented in FPGA using hardware resources. Moreover,this implementation is extended to quadratic programming, so that its solvingtime is remarkably reduced as compared with the software solving method.2) Design and implementation of FPGA based predictive controller for typicalindustrial objects. By theoretical analysis of MPC, control parameters of atypical kind of industrial object can be designed without simulation. Referringto the theoretical result, MPC are designed in FPGA for this object, and theprocesses of model identification, choosing of control parameters, and on-linereal-time control are completed.3) Case study of FPGA based predictive controller. The designed FPGA basedpredictive controller with matrix inversion accomplished in the hardware andthe FPGA based predictive controller for typical industrial objects are appliedseparately to the pressure-keeping process of an injection-molding machineand the combustion process of a circulating fluidized bed boiler. Effectivenessof the design of FPGA based predictive controller are verified by the controlperformances.