6/15/2017
1
LUCCA, ITALY, 2017-06-12
Future Challenges for Cyber-PhysicalSystems – An Industrial PerspectiveAlf Isaksson, ABB Corporate Research, Västerås, Sweden
Facts about ABB
Future Automation
MPC still the workhorse?
Integration of power and automation
Moving up the value chain
The modelling complexity
Future challenges
Conclusions
June 15, 2017 Slide 2
Outline
6/15/2017
2
”Power and productivity for a better world”
This is ABB
June 15, 2017 Slide 3
Utilities Industry Transport & Infrastructure
Corporate Research Centers
June 15, 2017 Slide 4
Close to major customers, universities andABB‘s business responsible units
Västerås SE
Raleigh US
Baden CHWindsor US
Ladenburg DE Krakow PL
Bangalore IN
Shanghai CNBeijing CN
6/15/2017
3
Facts about ABB
Future Automation
MPC still the workhorse?
Integration of power and automation
Moving up the value chain
The modelling complexity
Future challenges
Conclusions
June 15, 2017 Slide 5
Outline
What does that mean?
Industrial Digitalization
June 15, 2017 Slide 6
Product offering &business models
Distribution
Market channels
Customer contacts
Product development
Production & maintenance
Collaboration with sub-suppliers
Integration in energysystem
Production
R&D
Maintenance
Sales
Purchasing
Custom
ers
Distribution
Sub-supplier
Sub-supplier
Sub-supplier
Sub-supplier
Q, IT, etc
Partner PartnerPartner
Energy
6/15/2017
4
The Five Major Trends that Manufacturers Must Follow
Market Trends
June 15, 2017 Slide 7
Mobility
The hype
AnalyticsCloudComputing
BigData
InternetofThings
The Five Major Trends that Manufacturers Must Follow
Market Trends
June 15, 2017 Slide 8
Mobility
The hype
AnalyticsCloudComputing
BigData
InternetofThings
6/15/2017
5
Automation Network and Hierarchy
Today’s automation systems
June 15, 2017 Slide 9
Automation Network and Hierarchy
Today’s automation systems
June 15, 2017 Slide 10
ERP
(Level 4)
MES / CPM
(Level 3)
Supervisory control
(Level 2)
Regulatory control
(Level 1)
Process
(Level 0)
6/15/2017
6
What is happening next?
June 15, 2017 Slide 11
Upper levels moving to the cloud
ERP
(Level 4)
MES / CPM
(Level 3)
Supervisory control
(Level 2)
Regulatory control
(Level 1)
Process
(Level 0)
Trade-off between edge and cloud
Future automation system architecture
June 15, 2017 Slide 12
Real-time bus
On-site cloudplatform Global cloud
Devices
Firewall
6/15/2017
7
Optimizing control of Virtual Power Pools
OPTIMAX® PowerFit
June 15, 2017 Slide 13
Task
• Aggregate many small production units andtreat them like one big power plant
• Exploit multiple forms of energy (e.g. el andheat) and storages
Solution
• Build overall plant model (exploitingModelica multi-physics)
• Formulate optimizing control task asmathematical program
• Online optimization of set points and plantschedules
Digitalization enables the interconnection of power generation, consumption, storage andproduction
Facts about ABB
Future Automation
MPC still the workhorse?
Integration of power and automation
Moving up the value chain
The modelling complexity
Future challenges
Conclusions
June 15, 2017 Slide 14
Outline
6/15/2017
8
Model Predictive Control
June 15, 2017 Slide 15
Use model to
• Estimate where you are – stateestimation
• Optimize future controlsignals over a time horizon
• Repeat at next samplinginstant
• Shift horizon one step –receding horizon control
past future
k k+1 k+Ny
k+1k+2
k+1+Nu k+1+Ny
predictedcontrolled variable
u(k+n) manipulatedvariable
set-point
k+Nu
lndustry requirements vs available processing power
Model predictive control: advancing the frontiers
June 15, 2017 Slide 16
Pulp & Paper
Rolling Mills
Power Generation
Cement & Minerals
Electrical
Refining
Chemicals
6/15/2017
9
Digester APC Overview
Case study : Mondi Swiecie, Poland
June 15, 2017 Slide 17
Operator overview
APC display
Faceplates
Switch APC andSupervisory mode
Mid-point Kappa control: Results before & after APC implementation
Case study : Mondi Swiecie, Poland
June 15, 2017 Slide 18
Performance test run results Kappa variability reduced by 56.4%
Average Upper Lt Lower Lt Sigma % resultsBefore APC 88.86 89.00 85.00 3.68 36.8%After APC 87.93 89.00 85.00 1.61 71.4%Improvements 0.93 2.08 94%Before APC % of data within Limits 36.8%After APC % of Data within Limits 71.4%
Blow Kappa Test Run
56.4%
6/15/2017
10
Reduces Google Data Centre Cooling Bill by 40%
DeepMind AI for Data Center Cooling
June 15, 2017 https://deepmind.com/blog, 2016-06-20Slide 19
PUE = Power Usage Effectiveness
Facts about ABB
Future Automation
MPC still the workhorse?
Integration of power and automation
Moving up the value chain
The modelling complxity
Future challenges
Conclusions
June 15, 2017 Slide 20
Outline
6/15/2017
11
End of Isolated SolutionsBalancing Between Control Systems
Energy availability and pricing(smart grids)
Process variations, e.g. quality,yield, disturbances (DCS)
Production Management(P&S, APC, Analytics, …)
Process control
Grid control
© ABB GroupJune 15, 2017 | Slide 21
Industrial demand-side management
Integration of scheduling and control
Coordination of production planning and energy managementIndustrial demand side management in pulp & paper
June 15, 2017 Shifting production to low cost timesSlide 22
§ Thermo-mechanical pulp (TMP)production is highly integratedwith other parts of paper plant
§ Most energy consumingproduction steps are moved tolow cost times
§ Paper output of plant is notreduced
Mechanical pulp production
0
20
40
60
80
t1 t6 t11
t16
t21
t26
t31
t36
t41
t46
t51
t56
t61
t66
t71
t76
t81
t86
t91
t96
t101
t106
t111
t116
t121
t126
t131
t136
t141
t146
t151
t156
t161
t166
Elec
tric
ityPr
ice
[€/M
Wh]
Time [hours]
Spot Market Hourly Prices (1st week, December 2014)
Reduced pulpproduction
024
t1 t6 t11
t16
t21
t26
t31
t36
t41
t46
t51
t56
t61
t66
t71
t76
t81
t86
t91
t96
t101
t106
t111
t116
t121
t126
t131
t136
t141
t146
t151
t156
t161
t166
Running Tasks with iDSM
6/15/2017
12
Evaluating market opportunities for thermo mechanical pulping (TMP) millsIndustrial demand side management in pulp & paper
June 15, 2017 Slide 23
§ Real world plant andproduction data of a Nordicpaper mill
§ Different scenarios evaluated
Case study with TMP mill
ScenarioEnergycost
Allowed pulpstorage levels
S0 No 20%-80%
S1 Yes 20%-80%
S2 Yes 5%-95%
050
100t1 t9 t17
t25
t33
t41
t49
t57
t65
t73
t81
t89
t97
t105
t113
t121
t129
t137
t145
t153
t161
Ele
ctri
city
Pri
ce[€
/MW
h]
Time [hours]
Spot Market Hourly Prices (1st week,December 2014)
050
100
t1 t9 t17
t25
t33
t41
t49
t57
t65
t73
t81
t89
t97
t105
t113
t121
t129
t137
t145
t153
t161
Ele
ctri
city
Pri
ce[€
/MW
h]
Time [hours]
Spot Market Hourly Prices (1st week,August 2014)
No. ofstarts Savings
S0 8 0%
S1 24 6%
S2 26 5%
No. ofstarts Savings
S0 7 0%
S1 34 4%
S2 33 4%
Controlling 48 MW at 1kHz sampling rate
NMPC for Load Commutated Inverters
June 15, 2017 Slide 24
LCIs play an important role inpowering electrically-drivencompressor stations. Enable LCIsto ride through partial loss of gridvoltage.
– Auto-generated NMPC algorithm(ACADO/qpOASES)
– Running at 1kHz on AC 800PEC
Solution running at a keyStatoil/GASSCO sites– Two out of six 41.2 MW compressor
strings for gas export at Kollsnes– Three 7.5 MW booster compressors at
Kårstø– First successful ride-through
November 2015
Goal
Solution
Results
6/15/2017
13
Facts about ABB
Future Automation
MPC still the workhorse?
Integration of power and automation
Moving up the value chain – Mining as example case
The modelling complexity
Future challenges
Conclusions
June 15, 2017 Slide 25
Outline
Mine - Processing - Port
Mining value chain
June 15, 2017 In co-operation with BolidenSlide 26
Automation touches every aspect of mining
Process optimization of the entire value chain in real-timeLowest level of automation is in the underground mine
Port
Mine
Processing
6/15/2017
14
Boliden/Garpenberg
Optimization of Flotation Process
June 15, 2017 Slide 27
• Boliden stated that the projectis impossible!
• MPC + static nonlinearoptimization
• Optimize financial outputbased on dynamic pricemodels
• Increased throughput by 1-2percent units
Blast cycle expands the mine
Underground mine production
June 15, 2017 *Source: Final report Zepa, SMIFU Work package 1, Rock Tech Centre, 2011-12-15Slide 28
Production cycle Where automation can help
In reach is a significant production increase of 40-50% through automation
Source: 1 A pragmaticapproach to mineautomation, ABB MiningConference Cape Town,2013More than 50 operations in a harsh and high risk
environment whereof 10% is automated*
6/15/2017
15
Short term scheduling in closed loop
Automated scheduling - MineInsight
June 15, 2017 In co-operation with BolidenSlide 29
Real-time visualization of actual status and on-line replanning
Increased weekly blasting by 10% - Increased utilization of machines and facesA predictive production from the beginning of the value chain
Executor
Scheduler
Monthly production planPersonell and machine availabilityMachine characteristics and distances
Work list Progress list
Progress of workWork orders
Healthy working environment and Energy efficiency
ABB Smart ventilation
June 15, 2017 Slide 30
Ventilation where needed Real-time feedback control
Extended lifetime of existing infrastructureEnergy consumption reduction of 30-50% validated on site
6/15/2017
16
Increased safety and productivity
ABB Mine Location Intelligence™
June 15, 2017 Slide 31
3D visualization on a map Connection of people and machines
Enabled by mine wide wireless communication network through WLAN
Facts about ABB
Future Automation
MPC still the workhorse?
Integration of power and automation
Moving up the value chain
The modelling complexity
Future challenges
Conclusions
June 15, 2017 Slide 32
Outline
6/15/2017
17
Automatically generate models for control and optimization from CAD
Modelling vision – Automation of automation
June 15, 2017 Slide 33
CAD
Modelica model
Process graphics in 800xA
Commissioning using a (simulated) virtual reality
Virtual commissioning
June 15, 2017 Slide 34
Manufacturing: Mechanical objects up to cells, lines, incl. 2D or 3Dsimulation are coupled with automation systems (hardware or softwarein the loop)
Process automation more difficult due to lack of easily available processmodels. Currently piloting simulation models derived from P&I diagram tobe used for FAT.
6/15/2017
18
Demo cell at ABB in Gothenburg
Virtual commissioning
June 15, 2017 Slide 35
Finding intervals that are useful for modelling
Learning models from historic data
June 15, 2017 Slide 36
• Method available for systemidentification using input-output data
• Less than 5 % of normaloperating data useful foridentification
• Can (historic) data be usedalso for applications learningdecision models rather thanprocess models? For example
• Alarm management
• Production scheduling
• Supply-chain optimization
6/15/2017
19
Facts about ABB
Future Automation
MPC still the workhorse?
Integration of power and automation
Moving up the value chain
The modelling complexity
Future challenges
Conclusions
June 15, 2017 Slide 37
Outline
Systems Design: Open, Efficient and Easy to Engineer
Future Challenges
June 15, 2017 Slide 38
§ How to design future automation systems (ofsystems)– Easier usability of more complex system
(Smartphone)§ How to ensure seamless communication and access
to data– Can we agree on ONE set of standards?– How to make systems more open
§ How to make the design adaptive to facilitate thedynamic changes in the operating environment– Engineering 10 times more complex systems at
10 % of today’s effort§ Which functions become redundant / merged?
– Inheritance vs. innovativeness
6/15/2017
20
Operations: Less is More – Focus on the Essence
Future Challenges
June 15, 2017 Slide 39
§ Which decisions are relevant and necessary?– Are there redundant functions?
§ Role of an operator in the future– How to maximize his/her performance & quality
§ Create relevant information from data– Easy to get lost with all possibilities
§ How to ensure truly collaborative functions– Eliminate competition between local targets
§ What comes instead of the automation pyramid?– Do we end up in another form of hierarchy?
§ The 100 % available plant – meaning only plannedmaintenance stops
Modeling and Optimization: Dealing with Complexity
Future Challenges
June 15, 2017 Slide 40
§ How to model cross-topics (merge earlier separatedones)– How to link models to each other - global
awareness§ How to deal with nonlinearities / nonconvexities
– Do linearization schemes kill the performance§ How to optimize ever growing problems
– Are there true options to MI(N)LP?§ Collaborative optimization solutions ensuring best
performance in all situations§ Balance between Cloud and Edge computing
– What functions can and will have to be executedlocally and what can instead be moved to a localvs global (outside firewall) cloud?
6/15/2017
21
Facts about ABB
Future Automation
MPC still the workhorse?
Integration of power and automation
Moving up the value chain
The modelling complexity
Future challenges
Conclusions
June 15, 2017 Slide 41
Outline
Digitalization is Inevitable
Conclusions
June 15, 2017 Slide 42
Digitalization will have a tremendous impact on controland operations.Will be future business differentiator and growingacademic field.
We (may) see shifts from§ Proprietary to open architectures§ Multivariable control with few in-out to DeepMPC§ Isolated to integrated power and automation
systems§ Low level control to optimization of entire value
chain§ Manual to automatic generation of models§ Real to virtual commissioning
6/15/2017
22
The speaker is grateful to many colleagues, especially within ABB, for discussions and slides.
In particular I thank Iiro Harjunkoski, Guido Sand, Jan Nyqvist, Tomas Lagerberg, Rüdiger Franke,Abhijit Badwe, Lennart Merkert, Thomas Besselmann, Mehmet Mercangöz, Rainer Drath and MarioHoernicke.
Acknowledgements
June 15, 2017 Slide 43