IE 361 Module 28. EFC and SPM: One Thing Control Charting is Not. Reading: Section 3.6 Statistical Methods for Quality Assurance

Transcription

1 IE 361 Module 28 EFC and SPM: One Thing Control Charting is Not Reading: Section 3.6 Statistical Methods for Quality Assurance ISU and Analytics Iowa LLC (ISU and Analytics Iowa LLC) IE 361 Module 28 1 / 11

2 EFC and SPC In this short module, we try to make explicit the difference between "engineering control" (as understood, for example, by EE s and ME s) and "SPC." Recall from Module 24 the basic insight that Shewhart "control charting" is about process watching for purposes of change detection as illustrated in the figure below repeated from that module. Figure: Statistical Process Monitoring (ISU and Analytics Iowa LLC) IE 361 Module 28 2 / 11

3 EFC and SPM Control Charts are NOT Process-Tweaking/Process-Guidance Tools Engineering Feedback Control is process guidance/on-line adjustment as represented in the figure below, where repeatedly an output Y is compared to a target value T and accordingly a change is ordered in some process variable X. Figure: Schematic of an Engineering Control Scheme (ISU and Analytics Iowa LLC) IE 361 Module 28 3 / 11

4 EFC and SPM Control Charts are NOT Process-Tweaking/Process-Guidance Tools The figures on panels 2 and 3 and the methodologies they represent are NOT the same. Automatic online process adjustment is not the same thing as process watching for purposes of change detection. SPM and EFC are NOT Competing Technologies both have their places both can be badly done both can contribute to variation reduction in an industrial process neither is a "weak version" of the other In many industrial applications, EFC creates the physical stability that SPM monitors. (ISU and Analytics Iowa LLC) IE 361 Module 28 4 / 11

5 EFC and SPM Control Charts are NOT Process-Tweaking/Process-Guidance Tools Here are some contrasts between EFC and SPM EFC is "automatic" (digital or mechanical) is compensation-oriented expects process "drift" is on-going "tweaking" is typically computer-controlled maintains optimal adjustment is tactical can exploit dynamic models SPM is often manual is detection-oriented expects "stability" triggers "rare" interventions typically alerts a human agent warns of "special cause" changes is strategic warns of departure from stability The technologies of EFC and SPM are quite different. SPM deals in control charts. EFC deals in feedback algorithms that prescribe exactly how adjustment variables are to be manipulated on the basis of process data. "PID" control algorithms are a common variety of these (see Section 3.6 of SQAME for a few details on PID control). (ISU and Analytics Iowa LLC) IE 361 Module 28 5 / 11

6 An EFC Example (Paper Making) From Section 3.6 of SQAME Prof. Jobe worked with the Miami U. Paper Science Department on the control of a paper-making machine. A schematic of that machine is below. Fundamentally, a pump feeds pulp onto a moving conveyor where the pulp dries to make paper. Figure: Schematic of a Paper-Making Process (ISU and Analytics Iowa LLC) IE 361 Module 28 6 / 11

7 An EFC Example (Paper Making) From Section 3.6 of SQAME Issues in developing an appropriate control algorithm included: pulp mix thickness WILL vary, but pump speed can be used to compensate, this is NOT an SPC problem! (it is an automated compensation problem), target paper density is 70 g/ m 2, 1 "tick" on pump dial changes density about.3 g/ m 2, and time delay and potential for over-compensation/oscillation are serious issues. To remove the time delay issue, a 5-minute sampling/adjustment interval was adopted. Through a series of experiments, an appropriate PID control algorithm was developed (see SQAME exposition and problems and the appendix to this module for details). (ISU and Analytics Iowa LLC) IE 361 Module 28 7 / 11

8 An EFC Example (Paper Making) From Section 3.6 of SQAME The following shows "uncontrolled"/baseline density data for 12 periods and data from 6 periods after the final algorithm was implemented and had been allowed to take effect. Figure: "Before" and "After" Paper Density To this point, we have an EFC success story. SPM now could have a role in monitoring for unexpected changes from this behavior. (ISU and Analytics Iowa LLC) IE 361 Module 28 8 / 11

9 Appendix: Some Details on the PID Controller for the Paper Machine A PID (proportional/integral/derivative) control algorithm for the paper making example is for X (t) = κ 1 E (t) + κ 2 E (t) + κ 3 2 E (t) Y (t) = measured paper density at time t X (t) = the knob position change (in "ticks") after observing Y (t) E (t) = the "error" at time t = T (t) Y (t) E (t) = E (t) E (t 1) and 2 E (t) = ( E (t)) = E (t) E (t 1) (ISU and Analytics Iowa LLC) IE 361 Module 28 9 / 11

10 Appendix: Some PID Control Details The "Integral" part of the algorithm, κ 2 E (t), reacts to deviation from target/offset, "Proportional" part of the algorithm, κ 1 E (t), reacts to changes in error (/level), and "Derivative" part of the algorithm, κ 3 2 E (t), reacts to curvature on plots of error. Jobe s final control algorithm was X (t) =.83 E (t) E (t) E (t) (ISU and Analytics Iowa LLC) IE 361 Module / 11

11 Appendix: Some PID Control Details The following table shows some hypothetical values for Y (t) and computation of corresponding values of X (t) for Jobe s control algorithm. t T (t) Y (t) E (t) E (t) 2 E (t) X (t) (ISU and Analytics Iowa LLC) IE 361 Module / 11