Cycle-to-cycle control of multiple input-multiple output manufacturing processes
Author(s)Rzepniewski, Adam K. (Adam Kamil), 1976-
Massachusetts Institute of Technology. Dept. of Mechanical Engineering.
David E. Hardt.
MetadataShow full item record
In-process closed-loop control of many manufacturing processes is impractical owing to the impossibility or the prohibitively high cost of placing sensors and actuators necessary for in-process control. Such processes are usually left to statistical process control methods, which only identify problems without specifying solutions. In this thesis, we look at a particular kind of manufacturing process control, cycle-to-cycle control. This type of control is similar to the better known run-by-run control. However, it is developed from a different point of view allowing easy analysis of the process' transient closed-loop behavior due to changes in the target value or to output disturbances. Both types are methods for using feedback to improve product quality for processes that are inaccessible within a single processing cycle but can be changed between cycles. Through rigorous redevelopment of the control equations, we show these methods are identical in their response to output disturbances, but different in their response to changes in the target specification. Next, we extend these SISO results to multiple input-multiple output processes. Gain selection, stability, and process variance amplification results are developed. Then, the limitation of imperfect knowledge of the plant model is imposed. This is consistent with manufacturing environments that require minimal cost and number of tests in determining a valid process model. The effects of this limitation on system performance and stability are discussed. To minimize the number of pre-production experiments, a generic, easily calibrated model is developed for processes with a regional-type coupling between the inputs and outputs, in which one input affects a region of outputs.(cont.) This model can be calibrated in just two experiments and is shown to be a good predictor of the output. However, it is determined that models for this class of process are ill- conditioned for even moderate numbers of inputs and outputs. Therefore, controller design methods that do not rely on direct plant gain inversion are sought and a representative set is selected: LQR, LQG, and H-infinity. Robust stability bounds are computed for each design and all results are experimentally verified on a 110 input- 10 output discrete-die sheet metal forming process, showing good agreement.
Thesis (Ph. D.)--Massachusetts Institute of Technology, Dept. of Mechanical Engineering, 2005."June 2005."Includes bibliographical references (leaves 199-200).
DepartmentMassachusetts Institute of Technology. Dept. of Mechanical Engineering.
Massachusetts Institute of Technology