Alf Isaksson Alexander Horch ABB Corporate Research Insert

Alf Isaksson, Alexander Horch ABB Corporate Research Insert image here Control Performance Monitoring © ABB Corporate Research Insert image here PROST Seminar 22 January 2002 Insert image here

© ABB Coprorate Research - 2 - Goal: detect and diagnose malfunctioning control loops

© ABB Coprorate Research - 3 - Bad control manifests itself as oscillation or

© ABB Coprorate Research - 4 - Methods needed to n detect oscillations n

© ABB Coprorate Research - 5 - Oscillation detection n Hägglund (1995). Consider areas between zero crossings (count if large enough). n Stattin and Forsman (1998). Based on same idea, easier to use. n Seborg and Miao (1999). Damping ratio of autocorrelation function.

© ABB Coprorate Research - 6 - Stattin index: Compare areas between zero crossings

Oscillation index © ABB Coprorate Research - 7 - 0 = no oscillation, 1 = perfect osc. 0. 88 0. 25 Controller re-tuned

Oscillation index trend plot © ABB Coprorate Research - 8 - index Valve IP converter replaced days

Major advantage: correlation analysis oscillation loop 2 © ABB Coprorate Research - 9 - 1 Conclusion: 0. 8 The loops interact. One of them is likely to cause both oscillations 0. 6 0. 4 0. 2 0 0 0. 2 0. 4 0. 6 0. 8 oscillation loop 1 1

Potential causes are. . . cycling load static friction F © ABB Coprorate Research - 10 - FC tight tuning

If the cause is stiction. . . © ABB Coprorate Research - 11 - process output control signal cross-correlation

If the cause is NOT stiction. . . © ABB Coprorate Research - 12 - process output control signal cross correlation

Stiction diagnosis n New method by Horch (1999) which utilizes that n when stiction in valve, process variable and control signal have odd cross-correlation ”not stiction” the signals are such that the cross-correlation is even (due to negative feedback) © ABB Coprorate Research - 13 - n when

Example: two coupled loops QC water Stiction Q © ABB Coprorate Research - 14 - pulp O. K. F FC

Example cont’d flow loop cross-corr. © ABB Coprorate Research - 15 - data concentration loop Diagnosis: stiction no stiction

Important assumptions 4 O Self-regulating process Oscillation detected Crosscorrelation method O. K. © ABB Coprorate Research - 16 - Integral action No compressible media

Example II: integrating plant © ABB Coprorate Research - 17 - two different level control loops stiction no stiction

CCF-method useless for integrating plants! level control loop Integration destroys the specific correlation in the stiction case. © ABB Coprorate Research - 18 - CCF is even, no matter if stiction or not. Re-calculation (differentiation) does not solve the problem

Idea! Look for discontinuities in the data! © ABB Coprorate Research - 19 - . . . ‘Second derivative is infinite’

1. ) Differentiate the process output! Y © ABB Coprorate Research - 20 - dy dt d 2 y dt 2 stiction no stiction

3 a. ) Histogram (ideally) © ABB Coprorate Research - 21 - d 2 y dt 2 stiction d 2 y dt 2 no stiction

3 b. ) Histogram (noise & filter) © ABB Coprorate Research - 22 - d 2 y dt 2 stiction d 2 y dt 2 no stiction

Level control with stiction d 2 y y(t) dt 2 © ABB Coprorate Research - 23 - stiction MSE: 0. 97 2. 01

Level control without stiction d 2 y y(t) dt 2 © ABB Coprorate Research - 24 - no stiction MSE: 1. 17 0. 46

Use Camel method also for self-regulating processes! Y Y dy dt © ABB Coprorate Research - 25 - Y’ d 2 y dt 2 stiction no stiction

Detect too large variance (too large 2 -sigma) Basic problem: 2σ © ABB Coprorate Research - 28 - -2σ Is this good or bad?

Performance index • Introduce a control performance measure: © ABB Coprorate Research - 29 - Current variance Ip = Theoretically opt variance n Possible to calculate denominator from normal operating data given knowledge of process time delay (deadtime). n Proposed by Harris (1989). n Modification presented in Horch and Isaksson (1999)

© ABB Coprorate Research - 30 - Modified Index: Before: 2. 11 After: 1.

Commercial tools / suppliers. . . ABB © ABB Coprorate Research - 31 -

LATTS – Loop Auditing and Tuning Tool Suite Part of ABB Industrial IT concept and uses the new Aspect Integrator Platform (AIP). © ABB Coprorate Research - 32 - Consists of three Aspects: n Process model identification n PID controller tuning n Loop auditing

© ABB Coprorate Research - 33 - Process Model Identification Aspect

© ABB Coprorate Research - 34 - PID Controller Tuning Aspect

Auditing Aspect n Computes 21 different quantities/indices. For example: n Control error standard deviation n Oscillation index n Stiction diagnosis (correlation) © ABB Coprorate Research - 35 - n Stiction diagnosis (histogram) n Modified Harris index

Auditing Aspect cont’d Combines these indices to test a number of hypotheses, such as © ABB Coprorate Research - 36 - n n Acceptable performance n Possible valve problem n Sluggish tuning The result is summarized in a report, either as a text file or in Internet Explorer

© ABB Coprorate Research - 37 - Auditing -- Index trend plots

© ABB Coprorate Research - 38 - Auditing -- Report

Conclusions n Methods exist for non-invasive n Oscillation detection n Stiction diagnosis n Minimum variance benchmark © ABB Coprorate Research - 39 - n New ABB Product LATTS under Beta testing right now. Product release approximately June 2002.

Future work (industrial as well as academic) detection and diagnosis of mill-wide oscillations distinction of linearly and non-linearly caused oscillations n performance assessment based on full process model (event-triggered estimation) n application of multivariable performance index n performance monitoring of MPC loops © ABB Coprorate Research - 40 - n n