MAXIMIZE CONTROL & OPTIMIZATION ROI

Transcription

1 Perry Nordh, P.Eng. 19-June-2017 MAXIMIZE CONTROL & OPTIMIZATION ROI Profit Suite Honeywell Internal

2 Senior Leadership Perceptions 1 The biggest mistake we ve seen is the assumption lot of company leaders have that they have made such massive investments in IT that they must have optimization all over the place Senior VP Major Oil and Gas Company I was naive and thought we had a very well run and optimized business. Some room for improvement maybe, but not huge - or that would mean I wasn't doing my job well. Once we dug for data, we found out that the prize was much bigger than I thought.

3 Return on Investment: MPC and Optimization MPC Improves Process Stability and pushes the process to the most profitable operating point $ Profit Specification Limit Poor Control Improved Control Improved Profit by Changing Targets

4 PID control versus MPC 3 TC DV SP FC OP PV PID Control (feedback controller) Single Input / Single Output control Controls to a setpoint Does not handle interactions well Little/no optimization Customisation needed for complex control Corrects for current error only (requires an error before it responds) Like driving while looking in the rearview mirror! Advanced Regulatory Control (ARC) Typically control that goes beyond Basic PID control But typically done within the confines of the control system. Single Input / Single Output control Feedforward, Lead/Lag, Decoupling, etc. Little/no optimization Corrects for current error only

5 PID control versus MPC 4 CV 1 CV 2. CV n MPC. MV 1 MV 2 MV m... DV 1 DV 2 DV l Multivariable Predictive Control Application residing internal or external to DCS Contains a model of the process Controls within a range Multi-input / Multi-output control Little / no customization needed Optimization / constraint handling built-in Corrects for future errors Like driving looking through the windshield

6 Objective: More than just Value Generation 5 Even the best operator can NOT: - Monitor all operating envelops & change SP each minute to optimize - Manage disturbances (i.e. rate or composition changes) proactively Highest level of MPC CAN - Manage changing economics - Operate consistently & optimally - Manage disturbances and reduce operator changes - Improve stability - - Capture best operating practice - Stretch your existing infrastructure

7 6 Eliminate Process Variability Plants with Best Process Control Have Fewer Slowdowns and Shutdowns Correlation Between Uptime and MPC Implemented 2 Analyze constraints and operate within them Plant Plant Uptime 95% 65% 55% Reduces operator manipulation by 80% 15% 2. Birchfield, G., Solomon Associates Olefin Plant Reliability

8 7 Why don t we have MPC everywhere?

9 Agenda - Why don t we have MPC everywhere? - Addressing challenges with Profit Control and Optimization sustaining ROI Platforms now with Experion Profit Controller Design new tools to streamline and speed projects Adding process intelligence with UOP - Finding more ROI with Profit Suite Profit Optimization Profit Opportunities - What is Honeywell doing differently? Why? Proactive Monitoring and Honeywell Connected Plant

10 Sustainment Challenges 9 Operator Acceptance - Operating Philosophy - Operating Envelopes Operations conservatism Change Management Instrument reliability

11 How can Honeywell technology help? Technology inherently addresses the Deploy / Sustain challenges customers are facing it lends itself to minimizing lifecycle costs. Technology specifically built to reduce implementation effort and lifecycle burden of profit-generating control applications. Enables a lifecycle support program instead of providing a box of tools that the user must figure out how to configure/use. Every development of an extended technology or new tool has a goal to minimize the lifecycle cost.

12 Profit Controller Model-Based Predictive Control Past Future Inherently More Stable Minimum MV move no overcorrections Optimal response trajectories Less sensitive to model error Built in robustness and stability with On-line Dynamic matrix conditioning protection CV MV Optimal Response Setpoint Control Funnel Assumed Values Predicted Unforced Response Easier to Tune One-knob tuning No need for move suppression Independent CV by CV tuning (MVs too) >90% of applications work with default tuning T=0 Minimum Effort Move Easier to Commission and keep running Fully supports closed-loop identification Warm mode enables validation

13 Profit Optimization Suite Consistent Technology Multiple Platforms - MPC and Dynamic Nonlinear Real Time Optimization - Flexible Modeling environment - Unmatched operational awareness - Lowest lifecycle cost Profit Loop Experion Profit Controller (C300/ACE) Profit Suite - Profit Controller - Profit Sensor - Profit Online Modeler Profit Optimizer (DQP Multi-Unit Optimization) Profit Executive (SuperDQP Multi-Asset Optimization) Continuum of Control Solutions Single Variable Linear Control Multiple-Variable Non Linear Control and Optimization Seconds Hours/Days Profit CPM (Control Performance Monitor)

14 Experion Profit Controller (CEE C300 or ACE node) Experion Application Server (URT) Matrix vs Function / Benefit Value URT C300 ACE Redundancy (or Increased Robustness) Fast Execution Speed Reduced Infrastructure ( and therefore Cost) CV & MV Transformations Open & Closed Loop Step Testing (Profit Stepper) On-Line Matrix Conditioning One Knob Tuning Optimization (Across Multiple Units or Site-wide) * * Virtual Sensors (Profit Sensor Pro) * * Operator Guidance - Good Fit - Potential Fit - Poor Fit * - Requires EAS/URT based Supervisory Profit Controller

15 14 Profit Suite R440 Addressing challenges with Profit Control and Optimization sustaining ROI Design new tools to streamline and speed projects Adding process intelligence with UOP

16 Profit Suite R440 Final Scope Platform & Infrastructure Compliance - 64 bit only follows Experion specifications including support for Windows Server 2012, 2016, Windows 10 - Upgrading PHD to R321, UPS to R321 - Allow for custom install path Honeywell User Experience - CAB Prioritized enhancements (Stepper, Profit Suite Engineering Studio: back-build) - Profit Web Viewer - Profit Suite Operator Station: user configurable colors - MV Transformations - Models from historical data in Profit Suite Engineering Studio Process Intelligence: UOP Synergy - Hydroprocessing (hydrocracking) and Oleflex (C4) optimization Profit Optimizer Enhancements - Phase 1 APC Vendor Agnostic parameter mapping - Optimization engine interior point enhancements TSQP

17 Faster Setup with Profit Web Viewer 16 HTML5 based visualization using Internet Explorer or Chrome web browser

18 Minimize Plant Tests with History ID and Online Modeling 17 Automatic segment selection after removal of bad segments Checks for controllers in Manual Can handle large data as ID does not use all data (~ 6 months) Integrated with profit controller PSES infrastructure - Calculations - Transformations - Merge models - Build controller (xm, xs, xp) - Simulation Identify individual CVs in history ID nodes and assemble complete model in Merge model Use SEED models and go direct to online closed loop modeling

19 Manage Nonlinearities with MV Transformations 18 Nonlinear transformations - Extension of the nonlinear transformations to MVs - Allows Profit Controller to act on a transformed MV whereas the operator sees the real world values This is particularly useful when trying to linearize valve curves.

20 Manage Change with Auto Documentation & PSES Backbuild 19 Often running platforms are adjusted/modified - The variable combinations or logic block calculations which have been designed with PSES are adjusted by the user in URT - The changes can be done in the equation or connection or the value of coefficients - These calculation changes can now be read back into PSES to update the original calculations and be fully documented with Auto-Documentation in PSES

21 Adding Process Intelligence with UOP 20 Startup Your Facility Sooner Reach Target Production Faster Operate at Peak Performance UOP Critical Control Applications Experion Solution Suites Field Instrumentation/ Modular Design Profit Process Solution Suites UniSim Process Solution Suites DynAMo Process Solution Suites Modular Automation Connected Performance Services (CPS)

22 Adding Process Intelligence with UOP 21 What is it? - Pre-packaged MPC applications for key UOP units Best practice from HPS, Expert review and input from UOP, UOP toolkits as appropriate - MPC for UOP packages are currently available for: CCR Platforming, Oleflex What is the Value Proposition? - Best in class knowhow from both UOP and HPS - The inferential calculations enable better estimation of key variables that are either: Difficult to measure (e.g., tube wall temperature in furnaces) Difficult to measure online(e.g., octane or aromatics content)) - Allowing the customer to push their units to their constraints increasing throughput or yield - Empirical model gains of the MPC for UOP models can be kept up to date using the rigorous non-linear optimization models from UOP CPS/Optimization Advisor

23 22 Profit Suite R440 Finding more ROI with Profit Suite Profit Optimization Profit Opportunities

24 Truths of Real World Economic Optimization 23 No model exactly predicts plant behavior all the time - Many relationships aren t measured so can t be modeled - Feedback is used to capture missing relationships Optimization model requirements differ from control - Gain needs to be more accurate - Dynamics are less important - Non-linear process behavior needs to be accounted for All optimizers: - Use steady state models - Solve for steady state solution - Generate steady state optimal operating targets - Require a delivery mechanism to move process Value comes from moving the process - Generating a solution is not enough - Money is only made after solution is implemented

25 Introduction to Time-Sequenced Quadratic Programming (TSQP) Based on the dynamic QP formulation in Profit Optimizer and Profit Controller, Honeywell has developed a patented technology that extends this to dynamic nonlinear optimization Status of SS RTO and its implementation challenges - SS RTO has been in existence for more than 30 years - Two heterogeneous solutions MPC is dynamic, but RTO is steady state. Two sides have never been seamlessly integrated. - MPC executes typically once every minute, but steady state RTO cannot execute until the underlying process units all reach a steady state. When disturbances come in, which is when a re-optimization is needed the most, an RTO solution must however wait for a steady state before it can perform any optimization. Patented Time-Sequenced Quadratic Programing (TSQP) solves this dichotomy - It modifies the standard SQP algorithm (used in RTOs) into a TSQP algorithm; - Nonlinear optimization is formulated within an MPC framework, which eliminates the burden of having to wait for the steady-state, and enables the nonlinear optimization to be executed in tandem with the MPC controller(s) - The optimization result is implemented by the MPCs on a minute to minute basis.

26 TSQP combines the strengths of SS RTO and APC! Shared Variables MPC (data-driven) Models SS RTO (first-principle) Models MVs/DVs CVs Independent Variables Dependent Variables CVs for each unit Independent variables Dependent variables Some CVs between that impact profitability that represent process the units needle; capability MVs for each unit DVs for each unit (e.g., connections to and from other units) Unit Feed Rates (as MVs or DVs) Feed Rates always as optimizing variables Control-only Variables Transiently changing variables can be modeled well with the aid of dynamic bias-updating These variables must be included in order for the optimizer to guarantee the APC-feasibility of its solutions. Difficult to model accurately (these variables need to be dynamically bias-corrected) Often glossed over or not modeled -- Examples: flooding, valve OP, suction pressure, hydraulic or fluid dynamic constraints Comments: First-principle modeling approach is designed to model the process nonlinearities costeffectively (without having to deal with the underlying dynamics) Data-driven modeling approach is designed to model APC dynamics cost-effectively (without having to deal with the embedded nonlinearity) TSQP has an unique technical advantage of being able to combine the strength of both modeling approaches on-the-fly, possibly even within the iterations of TSQP algo.

27 Pursue MORE Profit without building another unit 26 Webinar series to further explain: Think Bigger: consider a larger scope and focus on the final product quality Keep it Real: dealing with the reality of nonlinearities and operational challenges Keep it Coming: Calculation and Sustainment of benefits The Math behind the Moves: the math that makes optimization feasible

28 27 Profit Opportunities

29 The APC View Blending Components Straight-run products Straight-Run and/or Blending APC model has a greater resolution but limited scope It captures essential unit or area operating constraints Optimizes unit production to possible detriment of plant

30 The Planner s View The model has a high level of abstraction It captures essential material and energy balances in holistic view of the refinery It leaves out many unnecessary or even obscuring details for the economic optimization Straight-Run and/or Blending Blending Components Straight-run products Feed y y y p p p n n Product

31 Real Plant Models/Predictions Can we bridge the gap between planning/scheduling and control? Not effective in dealing with model errors and unplanned events. Supply Side (upstream supply chain) Daily production average What Feed to buy? Monthly Planning Generate a month-plan based on an average model What products to sell? Customer Demand (downstream supply chain) Daily delivery average What Feed to use? Weekly Schedule What products to deliver? Feed Received Product to be Shipped A set of production activities (e.g., receipts & blending activities) Translate the plan into production activities to satisfy the plant logistics Control and Optimize to the targets but only on a best efforts basis Blending/shipping activities Optimizers Controllers More effective in dealing with model errors and unplanned events (as a control problem) month ahead days ahead hours ahead now hours ahead days ahead months ahead Raw Materials Execution Products

32 Planning Execution Gap Plan Execute Planner uses aggregate yield models to predict product makes for raw materials. Current plan may be infeasible for current operating conditions Planning model required to make complex decisions for which raw materials to purchase and which product to produce Today s yields may not match the planner s yield model resulting in different operation Detailed unit model is generally not required and can lead to deleterious effects for solver Mismatched operation can lead to imbalances in plant inventories Assumed planning constraints for unit performance may be too aggressive or too conservative compared to actual performance

33 Implementation of the plan today Optimizer -1 Optimizer -m MPC 1-1 MPC 1-2 MPC 1-3 MPC 1-n MPC 2-1 MPC w -1 MPC m-1 MPC m-p

34 Implementation of the plan tomorrow Returned values can include the actual versus predicted yields and proxy limits Profit Executive Manage intermediate and final: Inventory (volumes) Properties (quality) Timeline (just in time) Optimizer -1 Optimizer -m MPC 1-1 MPC 1-2 MPC 1-3 MPC 1-n MPC 2-1 MPC w -1 MPC m-1 MPC m-p Manage: Unit operation Limits Constraints

35 Profit Executive Components Planning Execution Optimizer What-If steady state analysis using current operating conditions Controller Dynamic operating plan moves provided to simulator Optimizer Generates optimal steady state operating plan based on actual costs Controller Moves plant toward optimal while accounting for immediate constraints Predictor Computes yield/property model biases using operating data APC Extensions Bounds optimization problem based on APC constraints

36 Profit Opportunities Using real operating data, Honeywell determined that the refiner could save million of dollars from: Feed cost reduction using cheaper crudes. Yield improvements optimal routing Better tradeoff between distillate and naphthas Reduce property giveaways Better coordination of production and blending Reduce overall energy and chemical usage Prolonging catalyst life and reduced hydrogen usage

37 Giveaway Estimation KBarrels/Month Monthly Diesel Production in a More Profitable Product Mix Product A Product B Product C Product D Product E Product F A more profitable product mix could have been produced in the study year KBarrels/Month Historical (Acutal) Monthly Diesel Production Month in the Study Year Over-Qualification: $65M/year ~$3/Barrel of Product The actual product mix was produced and sold in the study year The purpose for estimating the quality giveaway is to show: how less optimally the components were produced at a potentially higher cost. the potential room for reducing the component quality while still meeting the same demand.

38 37 Profit Suite Roadmap What is Honeywell doing differently? Why? Proactive Monitoring and Honeywell Connected Plant

39 Profit Control and Optimization Roadmap 38 Control Performance Monitor R600 Next generation Diagnostic workflow - Tune/Fix/Investigate (Expert Guidance) - Completely updated user centered reporting views Intuition platform - Coexistence with DynAMo - Improved security/user management CPM R601 - Enhanced cross correlation - Improved displays/trending - Revised Custom Templates - Runtime license update CPM R610 - Next generation MPC monitoring - Optimization monitoring - Integration with Intuition alerting (Pulse) Experion Profit Controller Profit Optimization Suite R440 - Optimization engine enhancements - UOP Oleflex and Hydroprocessing toolkits - MV Transforms - Profit Controller in CEE (C300/ACE) - Part of Experion R500, includes design and visualization - Profit Web Viewer - Models from Historical Data Profit Optimization Suite R441 - Enhancements for optimization - Phase 2 APC Agnostic Multi-Unit Dynamic optimization - Proxy Limits - Extend Profit Web Viewer - Profit Applications - Variable combinations naming & chaining Profit Optimization Suite R442 - Process Monitoring and data analytics - Extend thin client - Profit Applications - L4 - Runtime license update Profit R500 - Planning integration - Enhanced Optimization - Operator Guidance - Nonlinear constraints - Engineering AppStore - Process Monitoring/modeling with what-if and support capabilities - UOP toolkit enhancements

40 Profit Suite R500 Proposed Scope Platform & Infrastructure Compliance - OPC-UA Optimization Enhancements - Operator Guidance for optimization - Hessian updating nonlinear constraints - Profit Executive Master Controller UOP Synergy and toolkits - Hydroprocessing (hydrotreating) and Oleflex (C3, C3/C4) toolkits Engineering APP store - Downloadable starting points for projects/calculations Honeywell Connected Plant: Profit Performance Monitor - Digital Twin & Shadow Controller in the cloud What-if analysis Support/diagnostics - Enhance Process modeling/monitoring/machine learning

41 Leveraging Data to Drive Process Performance 40 Challenges Multiple, inconsistent data sources. Heavy dependence on spreadsheets and increasingly scarce skills. Little use of modern analytical tools as focus is on historical analysis, not prediction. Limited embedded intelligence and learning. Consequences Decisions are slow and not the best.

42 Solution: Honeywell Connected Plant 41 3 Plant Availability & Optimization UOP Process Specialist OEM Partner Ecosystem Equipment Availability Utilizing Partnerships 4 2 Digital Twin of Processes & Assets Customer SME Analytics Based Insight at Enterprise Level 5 Optimize 1 Cyber Secured Data Stream Decisions & Actions 6 Monitor Troubleshoot What if? Planning Vectors Disparate Data Across Enterprise Process Data Asset Data Sensors ERP, Historian, etc. Connected Plant Plant and Enterprise Level Benchmark (future)

43 Annual Value $M Sustaining the Profit Opportunity Honeywell Profit CAMWeb Viewer 3 Process Engineer Profit Suite Indications on the operations monitoring dashboard show a reduction in financial performance AND lost opportunity Refinery Crude FCCU Operation Performance Long Term Performance Wee k Month Ja n Yea r Relative shift performance Day Feb Wee Metrics Select k Shift-A Shift-B Shift-C Shift-D Soft Sensor Performance LGO D86 90% Pnt. Jet Flash Pnt. Jet D86 90% Pnt. LGO Flash Pnt. Naphtha D86 90% Pnt. M ar Compare from: JAN-FEB-MAR-PRESENT Start Date Current APC Off Key Process Variables Feed GO Flowrate Naphtha Flowrate Energy Lost Opportunity Cost of Constraints Fuel Gas Pressure Chg. Heater COT Over flash flow Naphtha 95% Splitter reflux Limit 6 psi <benchmark Limit 5 F < benchmark Variance 1.5 x benchmark Soft sensor performance ± 4.2 f Valve in manual (long term) Profit Loss $K / yr Pending Action Notes Avg. Gap 17% 20% 33% 27% 01 Apr Apr 2017 View All S M T W T F S Equipment Limit Protection LE Compressor surge margin Chg. Heater O2 Chg. Heater Skin Temp. APC Off 01 Apr Apr Review drop in unit profitability and feedrate decrease with Tammi. Specific reference to why fuel gas pressure limit is a lost opportunity. 2. Review status of C103 (Splitter) reflux valve maintenance. 3. Review with shift B training with Tammi and Frank B. 4. Clarify COT upper limit with Rod. 20 View All

44 Drill Down From Lost Opportunity 43 Fired Heater Variables 01 Apr Apr 2017 S M T W T F S Fuel Gas Pressure Heater Outlet (COT) Preheat Heater Temp Inlet Temperature Profit Loss $K / yr Lost opportunity identified with the charge heater not firing as hard as expected, resulting in reduced plant capacity. Cost is $75k/year. F101 Fuel Gas (Burner) Pressure Limit Controller Reference Units High psig Low psig

45 Digital Twin 44 Types of Digital Twins Experience: Detailed first principle model for entire plant or complex across plant life Empirical: APC model generated from plant data or control models Equipment: Library - first principle; regressed; rulebased Augmented with process and control knowhow

46 Why Is Honeywell Different? 45 Data Analysis Results Not just a DIY platform and toolbox Embedded models. Embedded knowledge. Not just data scientists keen to have a go Leverages extensive experience along with state of the art analytics. Solve real-world problems to deliver improvements fast

47

48 What is the difference between CPS and APC? Strategic decisions - Open Loop - Rigorous UOP UniSim models (first principles) - Updates several times a day - UOP Process Expert consultation - Catalyst life versus throughput maximization CPS Operational decisions - Closed Loop - UOP Toolkits (correlations from UOP rigorous models) - Execution 10sec-5 mins (nominal ~1min) - Respond to instantaneous changes in raw materials, utilities, ambient, catalyst and equipment condition APC