55
Introduction Industrial Algorithms LLC. Jeff Kelly & Alkis Vazacopoulos July 18, 2013

Industrial Algorithms

Embed Size (px)

Citation preview

Page 1: Industrial Algorithms

Introduction Industrial Algorithms LLC. Jeff Kelly & Alkis Vazacopoulos

July 18, 2013

Page 2: Industrial Algorithms

Who we are?

– Jeff Kelly: 25-years of both production & process modeling &

optimization for planning, scheduling, control & estimation

(PSCE) problems in the process industries, worked in Shell,

Exxon, Honeywell, consulted for more than 30 companies.

– Alkis Vazacopoulos: 25-years of solving production planning

and scheduling in process, printing & publishing, consumers

goods, etc, worked for Dash Optimization, Fair Isaac, Verisk

and consulted for more than 100 companies.

7/19/2013 Copyright, Industrial Algorithms LLC

Page 3: Industrial Algorithms

Our Mission

• To provide efficient solutions to solve complex

APS (Advanced Planning and Scheduling)

problems.

• Our primary focus is to implement smaller

applications with large benefits versus installing

a large application with small benefits. To do

this, we help you identify your worst decision-

making bottlenecks and provide solutions

targeted to reducing their negative impact on

your bottom-line.

3

Page 4: Industrial Algorithms

4

What We Do

• iAL develops and markets iMPress, the

world’s leading software product for

flowsheet modeling and optimization.

• iAL provides in-house training for

customers along with complete software

support and consulting.

• iAL provides Industrial Modeling

Frameworks (iMf’s) for several problem

types.

Page 5: Industrial Algorithms

Our Mandate

• To provide advanced modeling and solving

tools for developing industrial applications in

the decision-making and data-mining areas.

• Our targets are:

– Operating companies in the process industries.

– Application software providers.

– Consulting service providers.

Page 6: Industrial Algorithms

Our Industrial Modeling

Frameworks

• Process industry business problems can be

complex hence an Industrial Modeling

Framework provides a pre-project or pre-

problem advantage.

• An iMf embeds Intellectual Property related

to the process’s flowsheet modeling as well

as its problem-solving methodology.

Page 7: Industrial Algorithms

What type of iMFs we have

developed

• Jet Fuel Supply Chain Design with Refinery and

multimodal transportation mode

• Maritime Shipping Supply Chain Design

• Real Time Blend Optimization

• Pipeline Scheduling Optimization

• Fast Moving Consumer goods – Planuling

Optimization

• Capital Investment & Facilities Location

• Advanced Production Accounting

• Advanced Process Monitoring

• Advanced Property Tracking/Tracing

7

Page 8: Industrial Algorithms

Our Business Model

• What do we license: iMPress, iMf, 3rd

Party Solvers.

• What is our pricing scheme: License

Fee, Support & Maintenance Fee.

• What are our license terms: Based on

Customer’s Needs i.e., Rental for a

specified period (months to years) or

Perpetual.

Page 9: Industrial Algorithms

9

iAL Services

• Application Support

– Free prototyping to get you started (for a

reasonable number of consulting hours).

– Full modeling and solving support.

• Consulting Services

Page 10: Industrial Algorithms

10

iAL Facts

• Founded 2012

• Offices in the US (New Jersey), Canada

(Toronto)

Page 11: Industrial Algorithms

11

Industrial Algorithms

Differentiators

• Modeling Focus

• Optimization Focus

• No Competition with

Consultants and OEMs

• Customer Support

• Solution of Hard Problems

• Flexible Licensing Terms

Page 12: Industrial Algorithms

Academic Collaboration &

Partnership

• Carnegie Mellon University

• University of Wisconsin

• Stevens Institute Of Tech

• Fairleigh Dickinson University

• George Washington University

Page 13: Industrial Algorithms

13

Academic Partnership Program

• Free Full-Edition iMPress licenses for degree-

awarding institutions for research and

teaching

Page 14: Industrial Algorithms

14

Working with Customers –

Current Projects

• Pipeline Optimization with DRA.

• Refinery Planning and Scheduling.

• Fast Moving Food Industry Planning and Scheduling.

• Jet Fuel Supply Chain.

• Beer Supply Chain Planning and Scheduling.

• Gasoline Blend Monitoring with ProSensus.

• Data Reconciliation Engine embedded in TUVienna STAN

Software.

Page 15: Industrial Algorithms

15

Development Directions

Performance

Ease of Use

Problem Types

Page 16: Industrial Algorithms

16

Development Directions

Performance

Ease of Use

Problem Types

Page 17: Industrial Algorithms

We solve problems that deal with the

following decisions:

Quantity

How much to produce?

What is the batch-size?

Quality

How to blend specific

products to satisfy certain

levels of quality?

Logic

What machines to use? How to sequence the jobs to

minimize setup costs?

Time

When to produce? How to respect past

decisions & future orders?

7/19/2013

17

Copyright, Industrial Algorithms LLC

Page 18: Industrial Algorithms

We solve these types of problems

Our system can model and solve problems which are a

mix of both planning & scheduling

decision-making.

We introduce nonlinear

optimization in large-scale

planning and scheduling

problems and

solve problems involving

quantity, logic & quality.

We properly manage

complexity in problems that

would normally be considered

as uncertainty by other

vendors.

We use data-mining

techniques to support the

solving of problems that

incorporate control,

feedback, and

maintainability issues.

7/19/2013

18

Copyright, Industrial Algorithms LLC

Page 19: Industrial Algorithms

19

Development Directions

Performance

Ease of Use

Problem Types

Page 20: Industrial Algorithms

How do we model the Superstructure?

Configure versus Code: Draw the flowsheet of connected industrial objects and

the sets, parameters, variables, constraints & derivatives are automatically created.

User, custom or adhoc sub-models can also be coded when required.

Unit-

Operation 1

Unit-

Operation 2

Port-State 1

Port-State 2 charge, batch & lot-sizing,

input-output yields,

stream flow bounding,

min/max run-lengths & cycle-times,

sequence-dependent setups,

certification delays,

density, composition & property limits,

nonlinear & discontinuous formulas,

economic, environmental & efficiency

objectives, etc.

Page 21: Industrial Algorithms

Why are we unique? • iMPress is flowsheet-based (i.e., a figurative

language).

– This means that the modeling is inherently network or superstructure “aware” with equipment-to-equipment, resource-to-resource, activity-to-activity, etc. as explicit language constructs or objects.

– It also means that all of the effort of generating the sparse A matrix in the LP, MILP and NLP is done automatically by automatically creating all of the sets, parameters, variables and constraints when the model is configured using our proprietary and comprehensive library of sub-models.

7/19/2013

21

Copyright, Industrial Algorithms LLC

Page 22: Industrial Algorithms

Why are we unique?

• iMPress is “shape-based” which is

different from other modeling systems:

– Algebraic modeling languages like GAMS,

AIMMS, AMPL, etc. are “set-based”.

– Applied engineering modeling languages

like ACM, gPROMS, APMonitor, NOVA-MS,

Modelica, etc. are “structure-based”.

– Array manipulation modeling languages like

Matlab, Mathematica, Octave, etc. are

“scalar-based”.

Page 23: Industrial Algorithms

Jet Fuel Supply Chain iMf

Page 24: Industrial Algorithms

How do you configure problems?

• Problems are configured either:

– Interfacing with our flat-file Industrial Modeling

Language (IML) or

– Interactively with our Industrial Programming

Language (IPL) using a programming language

such as C, C++, C#, Java, Python, etc.

7/19/2013 Copyright, Industrial Algorithms LLC

Page 25: Industrial Algorithms

25

Development Directions

Performance

Ease of Use

Problem Types

Page 26: Industrial Algorithms

Performance Issues

• Solve Large-Scale Problems.

• Take Advantage of MILP technologies

and multiprocessors.

• Efficient Memory Management.

• Strong Presolving.

• Nonlinear Technology that can handle

complex NLP problems.

• Support for decomposition/polylithic

modeling. 26

Page 27: Industrial Algorithms

What Math Programming and Solvers

we use?

Supply-chain planning and

scheduling optimization problems,

Logistics modeling and solving is

required utilizing Mixed-Integer

Linear Programming (MILP).

Production-chain planning and

scheduling optimization problems, both

Logistics and Quality optimization

models are solved using an integrated

and innovative combination of both

MILP and Nonlinear Programming

(NLP).

We currently have bindings to several linear and nonlinear

programming solvers such as

COINMP, GLPK, LPSOLVE, SCIP, CPLEX, GUROBI, XPRESS,

XPRESS-SLP, CONOPT, IPOPT,

KNITRO & IMPRESS-SLPQPE.

7/19/2013

27

Copyright, Industrial Algorithms LLC

Page 28: Industrial Algorithms

Real Time Blend Optimization iMf

28

Page 29: Industrial Algorithms

Fast Moving Consumer Goods iMf

Bulk-Line

Pack-Line

Sequence-

Dependent

Switchovers

Forecasted &

Firm Future

Demand Orders

Page 30: Industrial Algorithms

• Time Horizon: 60 time-periods w/ day periods.

• Continuous Variables = 10,000

• Binary Variables = 5,000

• Constraints = 20,000

• Time to First Good Solution = 10 to 30-seconds

• Time to Provably Optimal = 1 to 10-hours due to sequence-dependent switchovers.

• Solver: Tested with Xpress & Gurobi

7/19/2013 Copyright, Industrial Algorithms LLC

Fast Moving Consumer Goods iMf

Page 31: Industrial Algorithms

Cogeneration (Steam/Power) iMf

7/19/2013 Copyright, Industrial Algorithms LLC

• Time Horizon: 168 time-periods w/ hour

periods.

• Continuous Variables = 5,000

• Binary Variables = 1,000

• Constraints = 7,500

• Time to First Good Solution = 5 to 30-

seconds

• Time to Provably Optimal = 5 to 15-minutes

Page 32: Industrial Algorithms

Cogeneration (Steam/Power) iMf

Water

Pump

Page 33: Industrial Algorithms

• Time Horizon: 168 time-periods w/ hour

periods.

• Continuous Variables = 5,000

• Binary Variables = 1,000

• Constraints = 7,500

• Time to First Good Solution = 5 to 30-seconds

• Time to Provably Optimal = 5 to 15-minutes.

• Solver: CPLEX

7/19/2013 Copyright, Industrial Algorithms LLC

Cogeneration (Steam/Power) iMf

Page 34: Industrial Algorithms

Power Generation iMf

• Three thermal-plants and two hydro-plants

with and without water storage.

• Three nodes or buses with voltage phase

angle inputs where each bus obeys

Kirchhoff’s current and voltage laws.

• One time-varying demand load located on

bus #3.

Page 35: Industrial Algorithms

Power Generation iMf

Page 36: Industrial Algorithms

Capital Investment/Facilities Location iMf

Expansion?

Installation?

Page 37: Industrial Algorithms

Maritime Industrial Shipping iMf

Inventory

Routing

Page 38: Industrial Algorithms

SubsTance flow ANalysis (STAN) iMf

• Large-scale data reconciliation and regression is performed to compute observability, redundancy and variability estimates.

• Substances are any material or meta/sub-material (concentrations) which need to be traced within the flowsheet or network to track their movements based on flow and composition measurements over time.

• STAN is a software development from TUVienna using iAL’s iMPress solver called SECQPE (successive equality-constrained QP engine).

7/19/2013 Copyright, Industrial Algorithms LLC

Page 39: Industrial Algorithms

7/19/2013 Copyright, Industrial Algorithms LLC

SubsTance flow ANalysis (STAN) iMf

Page 40: Industrial Algorithms

Other uses of IMPRESS …

• First-principles or rigorous process modeling to

manage difficult but high-valued bottlenecks.

• On-line process/production monitoring to compare

model predictions with plant actuals in real-time.

• Large-scale nonlinear optimization to solve

industrial scale problems where there is a large

portion of linear constraints and a smaller portion of

nonlinear constraints with multilinear cross-product

terms (x1*x2) using successive linear & quadratic

programming.

7/19/2013 Copyright, Industrial Algorithms LLC

Page 41: Industrial Algorithms

Linking iMPress with

IBM/ILOG ODME & Cplex (work

with DecisionBrain)

System Architecture

Page 42: Industrial Algorithms

ODME-IMPRESS-CPLEX

System Architecture

• A domain-specific data model was created in

ODME using the usual master-data and

transactional-data partitions.

• A mapping between iMPress’ data model and

ODME’s data model was established.

• Java code was written to export iMPress’ IML

file (Industrial Modeling Language).

• SWIG Java was used to create a Java Native

Inerface (JNI) to iMPress.

Page 43: Industrial Algorithms

ODME-IMPRESS-CPLEX

System Architecture

• Java code was written to call iMPress-CPLEX

using its API’s.

• Java code was written to access the solution(s)

from iMPress-CPLEX using its API’s and to

populate the ODME solution-data partition.

Page 44: Industrial Algorithms

ODME Screen Shots

Page 45: Industrial Algorithms

Data-Model in ODME

Page 46: Industrial Algorithms

Master-Data

Page 47: Industrial Algorithms

Transactional-Data

Page 48: Industrial Algorithms

Gantt Chart for Reference (Base)

Page 49: Industrial Algorithms

Trend Plots for Reference (Base)

Page 50: Industrial Algorithms

Demand Variability Scenario Data w/

Reference in ()

Page 51: Industrial Algorithms

Trend Plots for Demand Variability

Scenario w/ Reference

Page 52: Industrial Algorithms

Benefits • Perfectly fit your business model and decision processes

• Sophisticated optimization capabilities able to tackle complex,

non-linear and large-scale problems

• A solution that can be quickly adapted to new production

processes

• A user-friendly GUI to help planners driving refinery operational

excellence and analyzing refinery behavior

• What-if scenario analysis for confident decision-making

• See all your data and options in one place with drill-downs and

graphics

• Collaborate with other planners

• Powered by IBM ILOG CPLEX Optimizers

Page 53: Industrial Algorithms

How do we engage?

• We first consult to determine how we can

improve the profit and performance of the

problem as a whole.

• Then, depending on the benefit areas and

apparent bottlenecks, a tailored and

incremental solution is implemented which

focuses on both improving economics and

increasing efficiency whilst being

transparent and usable.

7/19/2013

53

Copyright, Industrial Algorithms LLC

Page 54: Industrial Algorithms

How do we engage?

• Using our Industrial Modeling Frameworks

(IMF): These are preconfigured solutions

that we can adopt to your specific

problems.

• We have IMFs in the following areas:

– Production Planning

– Plant Scheduling

– Pipeline & Marine Shipping

– Energy Management

7/19/2013

54

Copyright, Industrial Algorithms LLC

Page 55: Industrial Algorithms

For a demonstration of our IMFs

& IMPRESS, please Contact

• Alkis Vazacopoulos

• Industrial Algorithms LLC

• Mobile: 201-256-7323

[email protected]

7/19/2013

55

Copyright, Industrial Algorithms LLC