extruder automation with learning control 2009 m. pandit 3 extruder automation with learning control...

Extrusion 2009 Extrusion 2009 M. PanditM. Pandit1

Extruder Automation with Learning Control Extruder Automation with Learning Control

Extruder Automation with Learning ControlExtruder Automation with Learning Control

Madhukar Pandit, Madhukar Pandit, University of Kaiserslautern, 67653 Kaiserslautern, GermanyUniversity of Kaiserslautern, 67653 Kaiserslautern, Germany



AUTOMATION TASKS TO BE PERFORMED in EXTRUSIONAUTOMATION TASKS TO BE PERFORMED in EXTRUSION



Means availableSensors

Methods and Algorithms for control

Hardware and Methods for signal processing andcommunication

Tasks involved

Control of motion, temperature, force etc.

Archiving and data evaluation

Monitoring Visualisation of Process variables

Alarms and Interlocks

Goal

Raise quality, increase productivity

AUTOMATIONAUTOMATION TASKS TASKS -- General General --



Extruder

ram

billetpyrometer

profile

position-sensor

die

Ofen

hydraulics

PLC

PLC

billet billetbillet

pyrometer

Isothermal Isothermal –– IsorateIsorate ExtrusionExtrusion

For productivity:For productivity:Maximise extrusion rateMaximise extrusion rate

For product quality:For product quality:Extrude under prescribedExtrude under prescribed

conditions e.g. extrusion conditions e.g. extrusion

temperature and ratetemperature and rate

Goal:Goal: Increase productivity,Increase productivity,

enhance product qualityenhance product quality

To achieve both simultaneously:To achieve both simultaneously:Employ tight control ofEmploy tight control of

extrusion rate and extrusion rate and

temperaturetemperature

AUTOMATIONAUTOMATION TASKS in TASKS in EXTRUDERSEXTRUDERS



ContentsContents

11 INTRODUCTIONINTRODUCTION

22 AUTOMATION TASKS TO BE PERFORMED AUTOMATION TASKS TO BE PERFORMED

33 IMPLEMENTATION: Measurement and ControlIMPLEMENTATION: Measurement and Control

44 CONTROL: Modelling CONTROL: Modelling Iterative Learning ControlIterative Learning Control

55 A STATEA STATE--OFOF--THE ART AUTOMATION SYSTEMTHE ART AUTOMATION SYSTEM

66 CONCLUSIONS AND PERSPECTIVESCONCLUSIONS AND PERSPECTIVES



ExtrusionRam movement Control

2. AUTOMATION TASKS TO BE PERFORMED in EXTRUSION2. AUTOMATION TASKS TO BE PERFORMED in EXTRUSION

Billet HandlingBillet transportBillet loading Billet heating Control

Profile Handling Assess profile properties Image processingCool profilePuller movement Control



For optimal production:

Prescribe values of relevant process variables appropriately

Extrude such that the variables takeon the prescribed values

Process variables:

Billet temperature,

Extrusion force, Ram speed, Profile speed

Profile exit temperature,

Cooling of Profile




For choosing values appropriately

• Use experienceUse experience

•• Gather data, analyse it and store Gather data, analyse it and store optimal values in a dataoptimal values in a data--bank andbank andretrieve when requiredretrieve when required

To extrude such that values of the variables are

held in prescribed ranges• Display process variables and let the Display process variables and let the

operator control inputs manually operator control inputs manually

•• Use automatic controlUse automatic control

•• Use a combination of bothUse a combination of both

Data Acquisition Task

ControlControl TaskTask

VisualisationVisualisation TaskTask




Extrusion Extrusion Control TaskControl Task

Adjust billet temperature and ram speed such that the process Adjust billet temperature and ram speed such that the process

variables variables

Billet temperature, Billet temperature,

Profile exit temperature, Profile exit temperature,

Profile speed Profile speed

Max. extrusion forceMax. extrusion force

Cooling rateCooling rate

lie within the prescribed ranges and extrusion speed is maximumlie within the prescribed ranges and extrusion speed is maximum

3 IMPLEMENTATION: Measurement and Control 3 IMPLEMENTATION: Measurement and Control ……



2. Feed2. Feed--back controlled extrusion:back controlled extrusion:

Measure Measure process variables and adjust the input variables process variables and adjust the input variables

such that the desired runs are obtained.such that the desired runs are obtained.

Direct FeedDirect Feed--backback FeedFeed-- back over the cyclesback over the cycles

1. Simulated extrusion:1. Simulated extrusion:

Make calculations using simulation studies and calculate billet Make calculations using simulation studies and calculate billet

temperature and ram speedtemperature and ram speed ((Profile temperature not Profile temperature not

measured during extrusionmeasured during extrusion))

Methods of performing the Extrusion Control TaskMethods of performing the Extrusion Control Task

…… 3 IMPLEMENTATION: Measurement and Control3 IMPLEMENTATION: Measurement and Control



-- for measurement and control systems for temperature of for measurement and control systems for temperature of billet and profile exit with nonbillet and profile exit with non-- contact radiation pyrometers contact radiation pyrometers

NonNon-- contact temperature measurementcontact temperature measurement


0 50 100 150 200 250 300510

520

530

540

Time t in s

Ou

tput of ca

usal filte

r y

0 50 100 150 200 250 300510

520

530

540

Time t in sO

utp

ut of n

on-c

au

sal filter

y



Objectives: Isothermal Extrusion Control Isothermal Isorate Extrusion ControlRobustness of control

Control inputs: Extrusion velocityBillet temperature / - taperValve position of hydraulicsVelocity of billet/quench ring

hitherto: Feedforward control (simulated extrusion)Feedback control (on-line)

now: Employ control from cycle to cycle

Iterative Learning Control (ILR)

Strategy for Extrusion Control




Learning algorithm( Predictive feedLearning algorithm( Predictive feed-- forward adaptive feedforward adaptive feed-- back) :back) :

uk+1(l) = uk(l) + ∆∆∆∆uk+1(l) (Optimal ram speed / billet temp.)

∆∆∆∆uk+1(l) = F[ϑϑϑϑd(l), ϑϑϑϑk(l), ϑϑϑϑk+1(l), ϑϑϑϑΒ Β Β Β k(l), ϑϑϑϑΒ Β Β Β k+1(l), pk+1(l), vk(l),vk+1(l), ∆ϑ∆ϑ∆ϑ∆ϑk+1 ]

with:ϑϑϑϑd(l) : Desired run of exit temperature uk(l) : Input as a function of extruded length of cycle k

vk(l) : Ram velocity in current cycle k

ϑϑϑϑk(l) : Run of exit temperature in previous cycle k∆∆∆∆uk+1(l) : Increment of input calculated for cycle k+1

ϑϑϑϑΒ Β Β Β k(l) : Billet temperature in cycle kpk+1(l) : Extrusion force in current cycle

4 CONTROL: Modelling and Iterative Learning Control 4 CONTROL: Modelling and Iterative Learning Control ……

F[F[……] to be chosen (based on] to be chosen (based on modelmodel ) for fast convergence. ) for fast convergence.

With With exact modelexact model convergence to optimal input in 1 cycle !convergence to optimal input in 1 cycle !



Notation Notation

x : position of ram with respect to its starting point ,

t : time elapsed since the beginning of extrusion

v(x,t ) : velocity of the ram when it is at position x

ϑϑϑϑB(x,t): billet temperature

ϑϑϑϑP(x,t) : profile temperature

F(x,t) : extrusion force

ϑϑϑϑR(t) : temperature of the container

ϑϑϑϑdie(t) : temperature of the die

…… 4 CONTROL: Modelling4 CONTROL: Modelling



ModellingModelling ApproachesApproaches

�� Model Model basedbased on on exactexact physicalphysical lawslaws ((structurestructure and and parametersparameters) )

–– exactexact analysisanalysis leadsleads to partial differential to partial differential equationsequations ((p.d.ep.d.e.).)((unsuitableunsuitable forfor controlcontrol))

�� Lumpend Lumpend parameterparameter approximationsapproximations ((suitablesuitable forfor controlcontrol))

•• Model Model fromfrom physicsphysics nonlinearnonlinear o.d.eo.d.e. .

•• Model Model withwith formal formal structurestructure fromfrom inputinput outputoutput data (data (‘‘blackblack box box modelmodel‘‘))

-- ArtificialArtificial NeuralNeural NetworkNetwork (ANN)(ANN)

-- NonlinearNonlinear--RegressionRegression, , fuzzyfuzzy setssets etc. etc.

•• Model Model withwith structurestructure derivedderived fromfrom physicsphysics + + parametersparameters obtainedobtainedusingusing measurementsmeasurements –– ((‘‘greygrey boxbox‘‘ modelmodel ) ) ourour choicechoice !!




The profile temperature ϑϑϑϑP(t) depends on

- initial temperature in the deformation zone ϑϑϑϑI(t)

- the temperature rise ∆ϑ∆ϑ∆ϑ∆ϑD(t) :

dϑϑϑϑP(t)/dt = – ((((1/T2). ϑϑϑϑP(t) + (K2/T2).(ϑϑϑϑI(t)+ ∆ϑ∆ϑ∆ϑ∆ϑD(t)

K2 depends on ϑϑϑϑdie(t) , the ram velocity v(t ) force F(t),

ϑϑϑϑI(t) on temperature of billet ϑϑϑϑB(t), of receptacle ϑϑϑϑR(t) and v (t)

ϑϑϑϑI(t) = ϑϑϑϑB(t) + c1 ∫vRa(ττττ) dττττ + + + + c2(ϑϑϑϑB(t) – ϑϑϑϑR ).√t.

∆ϑ∆ϑ∆ϑ∆ϑD(t) in the deformation zone

d∆ϑ∆ϑ∆ϑ∆ϑD(t)/dt = – (1/T1). ∆ϑ∆ϑ∆ϑ∆ϑD(t) + (K1/T1)(c3.vk(t).exp(c4 .ϑϑϑϑI(t) ).

Relations Relations betweenbetween variables variables basedbased on on physicalphysical lawslaws

…… 4 CONTROL: Modelling 4 CONTROL: Modelling




Block Block diagrammediagramme of of thethe modelmodel



NonNon--linear models with Hammerstein structurelinear models with Hammerstein structure

Simplifying approximations for discretised model

a. The temperature change in the deformation zone is constant.

b. The recipient temperature ϑϑϑϑR is nearly equal to the billet

temperature ϑϑϑϑB .

c. The temperature of the profile ϑϑϑϑP follows changes of heat input instantaneously without lag.

d. A term is introduced to weight the influence of a ram speed

change v(n) depending on the billet temperature ϑϑϑϑB and ram

position LB(n).




ϑϑϑϑP(n) =kB.ϑϑϑϑB(n) + kI .LB(n)+ kv ·

ϑϑϑϑP(k) =ϑϑϑϑB(k) + ϑϑϑϑB*(k)

ϑϑϑϑB*(k) = ϑϑϑϑ0 + ∑g1(n).f(ϑϑϑϑB (k-n))

f(.) is a non-linear function approximated by

f(ϑϑϑϑB (k))= a1.ϑϑϑϑB(k) + a2.ϑϑϑϑB2(k) + a3.ϑϑϑϑB

3(k)

3ϑϑϑϑB(0)

ϑϑϑϑB(n) – klLB(n)· vRa(n)


EquationsEquations depictingdepicting thethe relationshipsrelationships in in discretediscrete formform




Block Block diagramdiagram withwith Hammerstein blockHammerstein block



ix••••

= fi (x1, x2, xn, u1,u2, …up,t) , i = 1, 2, …, nyk = gk (x1, x2, xn, u1,u2, …up,t), k= 1, 2, …, q

NonNon--linearlinear modelmodel in in generalgeneral statestate spacespace representationrepresentation

u1

u2

up

y1

y2

yq

.

...

…… 4 CONTROL: Modelling 4 CONTROL: Modelling

Black box Black box modelmodel in in statestate spacespace representationrepresentation



5 A STATE5 A STATE--OFOF--THE ART AUTOMATION SYSTEM THE ART AUTOMATION SYSTEM MoMASMoMAS®®

- Modular Measurement and Automation

Extruder press

Billet heating

PC with MoMAS-Software

Ram

BilletPyrometers

Furnace

Billet BilletBillet

Pyrometer

Commmunication

Link

PLC

Forcesensor

Positionsensor

Cooling



MOMASE vent control

Main-Task

A D-Conversio nM-5B-1/U

Pyrometer

VelocityACC ON-AGLinkR S-232C

ExtruderE xtruder PLC control

Optimisati onWatchdo g

Data acquisiti on Para meter dialog

MoMASMoMAS®®

Display at Display at

operatoroperator

consoleconsole

…… 5 A STATE5 A STATE--OFOF--THE ART AUTOMATION SYSTEM THE ART AUTOMATION SYSTEM MoMASMoMAS®®



Runs of die Runs of die exitexit temperaturetemperature and and ramram velocityvelocity




••measures and displays process variables profile exit measures and displays process variables profile exit temperature, extrusion force, ram velocity etc. during everytemperature, extrusion force, ram velocity etc. during everycycle cycle andand extrusion time per billet, idle times and the extrusion time per billet, idle times and the mean mean temperature over the cycles temperature over the cycles

••calculates the optimal process input functions and calculates the optimal process input functions and transfers this to the PLCtransfers this to the PLC

••determines best process inputs for an order, stores determines best process inputs for an order, stores them and retrieves them them and retrieves them autautomatically when neededomatically when needed

••provides the extruder managers and engineersprovides the extruder managers and engineerswith a data data base for monitoring and evaluation awith a data data base for monitoring and evaluation a

The functionsThe functions performed hitherto:performed hitherto:




Hardware and Software requiredHardware and Software required

-- PPyrometersyrometers forfor billetbillet and and forfor profileprofile exitexit

-- An Industrial PC with Windows operating system and An Industrial PC with Windows operating system and MoMASMoMAS

and OPC server software and OPC server software

-- A A sensor for measuring the position of the ramsensor for measuring the position of the ram

Prerequisite for the installationPrerequisite for the installation

availability ofavailability of

-- A hydraulically actuated ramA hydraulically actuated ram

-- A PLC (e.g. Siemens S5/S7, Allen Bradley 5/40 etc.)A PLC (e.g. Siemens S5/S7, Allen Bradley 5/40 etc.)

-- An inner velocity control loop and An inner velocity control loop and

-- Of advantage:Of advantage: An induction furnace for controlled taperAn induction furnace for controlled taper




Production Statistics of Extruder equiped with MoMAS MoMAS ®®

at SAPA Offenburg

No. of billetsextruded

Meanextrusion rate

RMS error of

exit temp. ϑϑϑϑ

Standard

deviation of ϑϑϑϑ

WithMoMAS 30 381 10,02 mm/s 12,1 °C 3.89 °C

WithoutMoMAS

26 180 9,29 mm/s (28,8 °C) 4,21 °C

Actual production statistics for the period

May – September 2001 of an extruder in Offenburg

Note Note increasedincreased meanmean extrusionextrusion rate of > 6%rate of > 6%




HowHow far far areare wewe withwith total total automationautomation ??????

AchievedAchieved !!!!!!

6 CONCLUSIONS AND PERSPECTIVES 6 CONCLUSIONS AND PERSPECTIVES ……



Commercial viabilityIs extruder plant working at its full capacity ?Is there a demand for increased product output ?Is a new market sector being envisaged ?

Technical feasibilityAre the hydraulics and ram position / speed control adequate ?Is the billet furnace temperature control in order ?Is the puller control adequate ?Is the PLC capable of handling the data traffic ?

Acceptance by the crewIs the crew capable of coping with the new system ?

…… 6 CONCLUSIONS AND PERSPECTIVES6 CONCLUSIONS AND PERSPECTIVES



ConclusionsConclusions

•• Contactless temperature measurement of extruded Contactless temperature measurement of extruded

aluminium with high accuracy is feasiblealuminium with high accuracy is feasible

•• Advanced extruder control based on exit temperature Advanced extruder control based on exit temperature

monitoring offers an useful tool for total automation monitoring offers an useful tool for total automation

•• Isothermal extrusion at constant extrusion rateIsothermal extrusion at constant extrusion rate has been has been

implemented employingimplemented employing MoMASMoMAS to achieve gains in to achieve gains in

productivity and qualityproductivity and quality

•• Improved models yield faster convergenceImproved models yield faster convergence

PerspectivesPerspectives

•• Fault detection and technical diagnosis using process modeFault detection and technical diagnosis using process model l

…… 6 CONCLUSIONS AND PERSPECTIVES6 CONCLUSIONS AND PERSPECTIVES



Whether automation based on temperature measurement and control is expedient seems to be a dogma…

If you think it is not necessary, you may be right....

or- more likely -

you may be left behind

��



INSTALLED SYSTEMSINSTALLED SYSTEMS

Presses which run with MoMAS MoMAS ®®

• 3 presses in Germany (SAPA , ALCANALCAN and

Research Centre for Extrusions, Berlin)

• 1 press in Sweden ( SAPA)

• 2 presses in Italy (ALEX and COMETAL)

• 2 presses in USA ( SAPA)

• 4 presses in Australia (CAPRAL)

• 2 presses in Russia (Minsk)

extruder automation with learning control 2009 m. pandit 3 extruder automation with learning control...

Documents