Two dimensional (2D) System Ideas for Industrial Processes

1
Two dimensional (2D) System Ideas for Industrial
Processes

• Peter Wellstead

2
Examples of Practical 2D Processes

• Paper making

• Plastic film extrusion
• Coating processes (adhesives on paper sheets)

• Steel rolling and continuous casting
• Spray actuation systems

3
Motivation Personal Experience

• 1970-72 real-time image processing for bubble
  chamber photographs
• 1975-85 self-tuning control
• 1980s self-tuning filters for 2D images
• 1980s modeling and control of polymer film
  extruders
• 1990s algorithms for practical 2D systems

4
Motivation Practical (e.g. Paper Making)

• Technical - product quality and plant
  flexibility

• Economics 1 reduction in waste produces a
  300,000 Euro saving per year per machine

• Environment - EU plant efficiency requirements

5
Motivation Research

• A generic class of 2-D dynamic systems - paper,
  plastic film, sheet forming, coating and
  converting

• Opportunity for innovation - 2-D concepts not
  previously used in sheet forming.

• Applications driven research - real 2-D systems
  as motivation for appropriate 2-D theory.

6
Idealised Plastic Film Extruder System Aspects
of the Control Problem
7
sensors
sensor signal processing
actuation
estimators
controllers
8
Control Issues Practical 2D Systems

• Models and identification
• Sensors and sensor signal processing
• Control

9
Models
10
Models
2-D models should describe the joint MD and CD
variations
11
Models

• Model for identification 2D-ARMAX

12
Models

• Two dimensional datastructures for sheet
  processes

This structure for 2-D control
13
Identification

• 2-D identification 2D-ARMAX estimation
• 2-D adaptive memory methods
• non causal model estimation methods
• edge effects

14
Identification

• non causal model estimation methods
• uses row-recursive methods for FIR 2-D filter
  implementation to generate prediction errors and
  simulate

15
Identification

• 2-D adaptive memory methods
• 2-D forgetting factors give selected weights to
  information from all directions

16
Identification

• Structure estimation
• Example shows a method for QP support size
  estimation

17
Control Issues Practical 2D Systems

• Models and identification
• Sensors and sensor signal processing
• Control

18
Sensors the requirement

• Extrusion line speeds move at 300m/min. Paper
  machines move at 1000m/min (30miles/h)
• Less than 0.002 of a paper roll is measured
• Need for increased density of measurement

19
Scanning gauge data collection
20
How do we get full sheet information?

• Hardware for Full Sheet Sensing
• sensor arrays

• Software for Full Sheet Sensing
• Generalised Sampling Theory

21
Distortion of sheet data using scanning gauges

• Collecting data along a zig-zag path scanning
  gauges are performing a 2-D SAMPLING PROCESS.

• 2-D spectral analysis shows that the two scans
  (left scan and right scan) collect sheet data in
  different ways.

22
Sampling theory reminder

• One dimension

time domain
t
0
T
2T
3T
4T
5T
6T
7T
frequency domain
Data spectrum
f
0
1/T
2/T
-1/T
-2/T
23
Spectra of scanninggauges

• The scans are NOT in the CD,
• Alternate scans are in opposite directions

• RESULT the two sets of spectra are distorted and
  in different ways

24
Scan averaging interpretation
In a basic scanner the results of adjacent scans
are averaged
25
Result of basic gauge signal processing
26
How to avoid distortion and get full sheet
information

• Use Generalised Sampling to reconstruct the MD
  signal.
• Get the full sheet information by assembling the
  reconstructed MD signals

27
Generalised Sampling

• By considering reconstruction along an MD line,
  the Generalised Sampling Theorem can be used to
  reconstruct the full 2-D sheet and double the
  bandwidth.

28
Signal processing interpretation
2T
Sampling along the MD as a generalised sampling
process
Signal processing block diagram
29
MD reconstruction results
Reconstruction of MD data using generalised
sampling
Reconstruction of MD using conventional signal
processing
30
Scan averaging interpretation
Generalised sampling reconstruction results in an
average of scans in both directions. The
weightings used at each CD point is different
31
Summary

• Conventional averaging of scanner data gives a
  distorted view of the sheet variations, and has
  an aliassing bandwidth of 1/2T.

• Generalised sampling reconstructs full sheet
  data by compensating for the scanning geometry.
  The bandwidth is DOUBLED to 1/T.

32
How do we get full sheet information?

• Hardware for Full Sheet Sensing
• sensor arrays.

• Software for Full Sheet Sensing
• use 2-D sampling theory find out how and under
  what conditions full sheet information can be
  reconstructed from scanning gauge data.

33
Multi-gauge scanning arrays

• Sensor signal processing doubles the scanning
  gauge bandwidth

• Arrays of gauges give an expensive solution if
  more bandwidth is required

• Scanning arrays are a scaleable solution to the
  bandwidth problem

34

Multi-gauge scanning array
35
Practical 2D Systems Scanning Sensor Array
Research System
36
Multi-gauge scanning arrays

• Calibration of sensors across the web/sheet done
  by special calibration transfer trick

• Only one expensive gauge is required

• Gauge technologies can be mixed (e.g. beta gauge
  and infrared)

• Generalised sampling is applicable to multiple
  gauges

37
Control Issues Practical 2D Systems

• Models and identification
• Sensors and sensor signal processing
• Control (Courtesy of Honeywell)

38
CD Profile Control Loop

• The pursuit of better paper quality has placed
  new demands on Cross Directional (CD) control
  systems
• smaller zone sizes
• faster response
• lower CD spreads

39
Symptoms of a tuning problem
40
Tuning just right!!!
smooth paper!
active, but not picketing
41
sensors
sensor signal processing
actuation
estimators
controllers
42
NEW DIRECTIONS 2D Scanning Actuators

• Consider Mass Deposition Processes
• eg spray painting
• Source of mass is spray gun that is moved over
  surface
• manipulation usually done by robot
• Aim to deposit specific mass profile on surface
• for most applications, desired profile is uniform
  (ie flat)

43
Scanning Actuators

• Given footprint of mass flow rate from gun
• What track should the gun follow over the surface
  to achieve desired mass profile?
• Scanning actuator is dual of scanning sensor

44
Raster Pattern
Part being Sprayed

• Results from 2D scanning theory tell you
• how close to put the tracks
• how far off edges you need to scan to avoid edge
  effects

Robot Path
45
More Complex Paths

• Generalised Scanning Theory also shows that this
  path is also valid
• Path is suitable for thermal spray processes
• aim to achieve specific temperature profile
• more difficult problem because heat flows

Robot Path
46
Example of 2D Spray Actuation

• SPRAY FORMATION OF METAL
• Spray forming of metal as an alternative to
  casting
• 2D generalised sampling ideas from sensing are
  DUALISED to get dual results for actuation.
• Metal is sprayed in a special pattern to optimise
  spray cast metal quality

47
Benefits of spray-forming

• Reduced cost
• Costs US 100Million to provide tooling for new
  car model
• Reduced time
• Takes gt18 months to produce tooling for big parts
  (bumpers, bonnets, door panels etc)
• Freeze design long before production

48
(No Transcript)
49
(No Transcript)
50
(No Transcript)
51
(No Transcript)
52
Typical sprayed steel flat
53
2D Spray Actuation

• Painting.
• Spray coating
• Metal deposition
• And many more

For example..
54
And 2D Sensing again Sub-sea profiling
55
Acknowledgements

• Greg Stuart of Honeywell Greg supplied the
  information and slides of his profile control
  system.
• Steven Duncan of Oxford University Steven
  supplied slides of his 2D actuation systems
• Final photograph from CropDusters