NATIONAL RESEARCH COUNCIL INSTITUTE OF INDUSTRIAL TECHNOLOGY AND AUTOMATION AUTOMATION: MODEL PREDICTIVE CONTROL IN MANUFACTURING PLANTS

Transcription

1 NATIONAL RESEARCH COUNCIL INSTITUTE OF INDUSTRIAL TECHNOLOGY AND AUTOMATION AUTOMATION: MODEL PREDICTIVE CONTROL IN MANUFACTURING PLANTS Andrea Cataldo

2 Andrea Cataldo ITIA CNR (Researcher) Institute of Industrial Technology and Automation National Research Council Italy ABB Corporate Research Italy (Industrial researcher) Foster Wheeler Italiana (Control engineer)

3 Outline What does Automation mean? Who is a Control Engineer? What does he do? Advanced Control System: a case study

4 What does Automation mean?

5 Who is a Control Engineer? What does he do? Mathematical Model Real Plant Analysis Modelling Control method + CONTROLLER synthesis

6 Who is a Control Engineer? What does he do? Simulation Mathematical Model Set point + Error CONTROLLER Control action - Feedback

7 Who is a Control Engineer? What does he do? Implementation Mathematical Model Real Plant Set point + Error CONTROLLER Control action - Feedback

8 Advanced Control System: a case study (Dynamic pallet routing in a manufacturing transport line) De-manufacturing pilot plant ITIA CNR, Via A. Corti 12, Milan (Italy)

9 Control Systems Design Control law (TBD) + Control Actuation - Measurement Task

10 Control Systems Design Control law (TBD) + Control Actuation - Measurement Task

11 System Modelling How to do that?

12 Buffer Zones and control Sequences Sequences The generic pallet transport module Buffer Zones

13 The Sequences technique implementation 36 different Sequences Finite State Machines Automata (Hybrid systems course Prof. Prandini) A. Cataldo, M. Taisch, B. Stahl, Modelling, simulation and evaluation of energy consumptions for a manufacturing production line, IECON Wien

14 The Sequences technique implementation SIMIO simulation platform The methodological aspect Transport line shop floor (Low level control implementation) C# Sw code IEC standard (Sequential Functional Chart) PLC Dynamic Discrete Event simulation model (SIMIO) (Isagraf - Rockwell)

15 The Buffer Zones model Sequence Buffer Zones model Buffer Zone

16 The transport line abstract description Sequence Transition Buffer Zone Buffer Zone

17 The High Level Control System design: main concepts Input: pallet Target Propositional calculus The transport line Mixed Logical Dynamic (MLD) model The transport line Model Predictive Controller (MPC) Output: Transition

18 From the manufacturing plant to the MPC design: main concepts Elements to be considered in order to build the MLD model Pallet movement (Target to be reached) Physycal system and control constraints Target distance Performance index and constraints

19 The MPC design: system modelling and MLD formulation Pallet movement (Target to be reached) Tpi ( k 1) Tpi ( k ) Tp (k ) u i j,i Tp (k ) u (k ) j I i,in i i, j ( k ), i 1,..,31 j I i,out Physycal system and control constraints X i1 ( k ) ( u i, j U n ( k ) U m ( k ) 1 ( k )) X i1 ( k 1) 0 j I i,in Target distance Tpi (k ) xi ( k ) f (Tpi (k )) Pallet distance x (k ) i (Linear function - H. Paul Williams, pag. 182; Sherali 2001) Performance index yi ( k ) Ci xi (k ) Pallet distance from the Target J (Q y yi ( k h )) (Qx xi ( k h )) Qu ui, j ( k h 1)) h 1 i 1 i 32 ( i, j ) I u RH

20 The de-manufacturing transport line: MILP formulation By re-arranging into the canonical form it comes: J min C ' x ' IP problem formulation s. t. A' x ' b' x ' [ x z' xb' ]T Z n min{ c ' x ' : A' x ' b', x ' Z n } x z' 0 x z' int. xb' {0,1} u(t ) vt* (0) Optimal solution (Receding horizon philosophy)

21 The MPC algorithm: Basic formulation RH J (Q y yi ( k h )) (Qx xi ( k h )) h 1 i 1 i 32 Deadlock if Receding Horizon (RH) 8

22 The MPC algorithm: Output quadratic term RH J (Q y yi (k h )) (Qx xi ( k h )) h 1 i 1 i 32 Deadlock if Receding Horizon (RH) 6 (Receding Horizon reduced by 2 steps)

23 The MPC algorithm: Integral control action RH J (Q y yi ( k h )) (Qx xi ( k h )) ( qni N i ( k h )) h 1 i 1 i 32 i 1 Deadlock if Receding Horizon (RH) 3 (Receding Horizon reduced by 5 steps)

24 The MPC algorithm: Off-limit zone RH J (Q y yi ( k h )) 2 (Qx xi ( k h )) ( qni N i ( k h )) k1 1U14 k1 1U 26 k1 1U 33 k 2 2U18 k3 3U 34 h 1 i 1 i 32 i 1 More efficient transport line control if (there is a pallet in M3 or N12) { then U10,12 (U18) is penalized in J }

25 MPC testing

26 MPC testing