anomalies detection with unisim design · flow transmitters & control valves anomalies the...

ANOMALIES DETECTION WITH

UNISIM DESIGNJOSEMARIA FERRER – (GERARD CAMPANYÀ, MANEL SERRA, JOSÉ-MARÍA NOUGUÉS)

SENIOR ADVISOR - INPROCESS

Honeywell Confidential - ©2019 by Honeywell International Inc. All rights reserved.

AGENDA

1

- Who we are

- Simulation: Historic perspective and the Digital Twin

- Process Plants break, physical laws don´t

- Inprocess Bad Actors application

- Conclusions


WHO WE ARE

2

Inprocess in brief

ACCOMPANY OUR CLIENTS IN THEIR SUCCESS IN

ACHIEVING SAFER, GREENER, MORE RELIABLE

AND MORE PROFITABLE INDUSTRIAL OPERATIONS

ESTABLISHED2006

BY FORMER ASPEN ENGINEERS

250+ MAN-YEAR

EXPERIENCE

45+SIMULATION ENGINEERS

Software & DCS agnostic

ProcessSimulation

Services

Process Simulation

• Process Simulation Training; Technology Courses; KIP -

Knowledge Improvement Program; Operator Training

• 300+ training courses conducted

KNOWLEDGE TRANSFER

• Commercial Simulation Software: HYSYS, UniSim, PetroSIM, VMGSim

• DCS independent.

• 50+ OTS commissioned

• Proprietary Instructor Station & communication layer

OPERATOR TRAINING SYSTEMS

PROPRIETARY APPLICATIONS

• ITOP: generic OTS for unit operations

• IIS – Inprocess Instructor Station

• IPSV; H2 Network Tool; PSA Simulator; Iflow; Bad Actors

ENGINEERING SIMULATION PROJECTS

• Steady State Simulations: What-If studies; KPIs; APC gains and

inferentials; Optimization; Refinery Reactor Modelling

• Dynamic Process Modelling: Compressors; Flare Systems; HIPPS; APC;

Flow Assurance

• 130+ simulation studies

Our services


SIMULATION:

HISTORIC PERSPECTIVE AND

THE DIGITAL TWIN

5


1930: Mechanical calculator

1934: Differential Analyzer

1939: Turing decrypter

1st Generation: Vaccum tube

1946: ENIAC

1952: IBM 701

2nd Gen.: Transistor

1959: IBM 1401

3rd Gen.: Integrated circuit

1964: IBM System/360

4th Gen.: Microprocessor

1971: Intel 4004

1977: VAX-11/780

1978: Intel 8086

1980: Sinclair ZX80

1982: Intel 80286, 386, 486

1993. Intel Pentium

2006. Intel Core line

2010. Intel Core i3, i5, i7, i9

1900: Planck´s Raditon Law

1908: Grüneisen´s Thermal L.

1913: Heisenberg principle

1923: Pauli´s Exclusion prin.

1925: Fermi-Dirac distribution

1949: EO Redlich-Kwong

1972: EO Soave R-K

1976: EO Peng-Robinson

1999: EO Elliot-Suresh-Donoh.

Chemistry, Mathematics, Physics

1600 1700 1800 1900 2000

1614: Napier Logarithms

1637: Descartes Cartesian geo

1662: Boyle´s Gas Law

1665: Calculus (Leibniz)

1669: Newton´s Method

1680: Algebraic logic Leibniz

1687: Newton´s Motion and

Cooling

1738: Bernoulli´s Law

1760: Lambert´s Law

1768: Euler´s Method

1785: Coulomb´s Law

1785: Laplace´s transform

1787: Charles´s Gas Law

1791: Richter´s reaction Law

1801: Dalton´s Law Partial P.

1802: Henri´s Gas Law

1808: Gay-Lussac´s Law

1811: Avogadro´s Gas Law

1822: Fourier´s Heat Law

1823: F.T. Calculus (Cauchy)

1829: Graham´s Effusion Law

1831: Faraday´s Electrolysis

1840: Hess´s Enthalpy Law

1840: Poiseuille´s Flow Law

1850: Clausius´s Law Thermo.

1851: Stoke´s Viscosity Law

1852: Beer´s Absortion Law

1854: Boolean Algebra

1855: Fick´s Diffusion Laws

1864: Kopp´s Heat Cap. Law

1866: Maxwell´s Gas Viscosity

1869: Mendeleyev´s Periodic

1871: Coppet´s Freezing Point

1871: Boltzmann´s Distribut. L

1873: EO Van der Waals

1882: Raoult´s Vapor Pressure

1885: van´t Hoff´s Osmotic Pr.

1893: Sutherland´s Gas Visco.

1200: Abacus

1621: Slide Rule

1673: Leibniz’s Step Reckoner 1700

1801: Punched Cards

1822:Mechanical Computer (Babbage)

1879: Cash Register (Ritty)

Computers

This science has been there long time

ago, but we are the first generation of

people who has in our hands software

tools and desktop computers capable to

simulate dynamically entire plants.

Most of their applications are still in the early stages.

Our human brain is now the limiting factor

6

Honeywell Confidential - ©2019 by Honeywell International Inc. All rights reserved. 7

What is a Digital Twin?

It is the digital version of your running asset. It contains:• all the process layout and streams conditions (Compositions, Pressure, Temperature, Flow, etc);

• all the equipment geometric data (dimensions, elevation, tray sizing, sensor location, etc);

• all equipment manufacturer performance data (pump curves, compressor curves, heat exchanger rating data, etc);

• all actuated valves (valve pressure drop, sizing, characteristic, etc);

• and all instrumentation (control loops, PID tunning, instrument ranges, selectors, etc).

DIGITAL TWINS IN THE PROCESS INDUSTRIES

All this information is consolidated in a UniSim Dynamic Process Model


THE AUTOMATION PYRAMID WITH THE DIGITAL TWIN

The Process Plant

Valves & Instruments

Inferentials & Virtual Sensors

Basic Controls

APCARC

RTO, f(x), tables

Nothing

Advanced apps

The

Operator

• Better understanding of plant behaviour (SteadyState & Dynamics)

• Reproduce and study any plant operation issue

• Develop and tune Operating Procedures. HAZOP assistant.

• Discover faulty instruments. Automatic fault detection application.

• Improve basic control layout (this is the most important layer!)

• Checkout new DCS code. Checkout automatic sequences.

• Develop rigorous inferentials. Train AI data driven models.

• Develop&Test ARC. Deep Gain analysis and seed models for APC.

• SS model for RTO. Obtain simple optimization f(x) or lookup tables

• Equipment anomaly detection. Monitoring and look-ahead apps.

• Train operator on Start-up, Shutdown, Emergency & trip scenarios.

• Knowledge repository of plant incidents & best operation practices


PROCESS PLANTS BREAK,

PHYSICAL LAWS DON´T

9


Who did say this?

“Just a moment…Just a moment

I've just picked up a fault in the AE-35 unit.

It's going to go a hundred percent failure within 72 hours”

GETTING INSPIRATION

In that movie, HAL has the capability to predict failures of

equipment based in their working variables and how the unit

should ideally work.

Question is: Can we do the same for Process plants?

PD: Afterall, the AE-35 was not faulty, just the Hal´s mad plan to kill the crew!


• Process plants provide thousands of process values every second, most of them are shown in

the ICSS Operator displays and can trigger one or more alarms.

• Most of these alarms are a consequence of an anomaly in the process plant (warning: H/L Level,

Temp, etc) and can evolve later to trip an equipment or entire unit (HH/LL Level, Temp, etc).

• When plant production is affected by process issues (either trips of units or reduced throughput),

every operator and engineer start to build their own theory, sometimes obvious, sometimes not so

obvious.

• Engineers start to trend historical process data and talk with operators to try to diagnose what

has happened in the plant, understand the causes and think how to avoid it next time. Frequently

the root cause is not found and the trip is repeated after few days.

WHAT I HAVE SEEN IN CONTROL ROOMS

Can simulation models help here?


Inprocess Bad Actor is conceived to provide early warning of plant anomalies which could evolve to

a trip and provide additional information to the Operator/Engineer to diagnose it.

WHAT ALARMS WOULD PRODUCE HAL?

With the help of simulation, could we generate more advanced alarms like these?

• “HeatExchanger HX-001 is violating the energy balance”

• “Flow valve FCV-001 is operating outside characteristic, most likely fail in flowmeter”

• “Compressor C-001 has a lower efficiency than usual”

• “Colum T-001 will enter to flooding in 15 min”

• “Product quality will go off-spec in 30 min and the phone will ring”


REAL PLANT VS. IDEAL PLANT (UNISIM)

Equipment failure Real Plant Digital Twin (Unisim model)

Pump impeller erodedAccording to their

manufacturer test curves

Valve seat leakageAccording to their

manufacturer characteristic

Heat Exchanger tube leak

Or tube rupture or fouling

According to their

manufacturer datasheet

Instrument failure Exact measurement

Analyzer failure Exact measurement

Could a clever

application exploit

these differences to

determine the fault?

That is

Inprocess Bad Actor


INPROCESS BAD ACTOR

FOR ANOMALIES DETECTION

14

© Inprocess 15

Inprocess Bad Actors: functional blocks

RealtimeDatabase(ICSS OPC server or historian)

Graphical User Interface

INPUT

OUTPUT

Co

mm

un

icat

ion

laye

r

Dat

a va

lidat

ion

Detailed Equipment rating models

Steady-State models

Dynamic models

Data management

andmodels

execution

AnomalyDetectorEngine

Equipment and process rules

The application is designed to identify two types of anomalies:

Equipment anomalies:

- Flow transmitters & Control valves (FLOW-VALVE)

- Vessels

- Pumps

- Heat Exchangers

- Online analyzers

- Compressors

- Distillation columns

Process anomalies: When the individual equipment composing the process doesn´t identify an

anomaly but the overall process unit or subprocess behaves differently from their model. It

requires to customize the rules based on the specific process.

© Inprocess 16

Type of anomalies

© Inprocess 17

Flow transmitters & Control valves anomalies

In general, flow transmitters are the least reliable instruments in a plant. Most of them are orifice plates and

the causes of the anomalies can be multiple. Coriolis, vortex, etc, can also provide wrong measurements

when not working at design conditions (two phases presence).

Regardless of the type of flowmeter, the application is focused in finding anomalies in the operating point of

control valves with or without installed flowmeter.

For every control valve the application uses the installed

valve characteristic to monitor the actual operating point.

For valves with no flowmeter, the characteristic is inferred

from historical data run over the dynamic model.

For every valve, the application allows to adjust certain

parameters (tolerances, outside time, filter time, etc) to

eliminate the false positives.

Once anomaly is detected, then it is analysed by the app.

Valve saturated

Anomaly

OK

To discern that, the application is always running two digital twins in

the background, one governed with Flow SP (DTSP) and one

governed with valve OP (DTOP).

The one which is following better the plant when appears a FLOW-

VALVE anomaly will determine the guilty.

If DTOP follows better the plant: it is a flowmeter problem.

If DTSP follows better the plant: it is a valve issue (bypass open,

broken seat, etc)

© Inprocess 18

Flow transmitters & Control valves anomalies

The picture shows an real issue

with the flowmeter of the steam.

The DTOP fits better with the

plant data (Pressure & Level)

When a FLOW-VALVE anomaly appears, there could be two reasons: flowmeter problem or valve problem.

The application verifies that the mass balance is respected taken into account the level variation,

pressure variation and the geometry of the vessel including level taps. This early detection can identify

issues before Low or High level alarms and prevent pumps or unit trips.

© Inprocess 19

Vessels anomalies

Level transmitter anomaly: When real level

transmitter is not following the DTIO.

Vessel leakage: When DTIO and DTOL are not

following the plant data.

Valve bypass open: When the outlet flows determine

by DTIL trigger a VALVE-FLOW anomaly.

Every vessels will define a subset dynamic model

and will run three digital twins of that model:

DTIL: Governed by Input flows and Level

DTOL: Governed by Output flows and Level

DTIO: Governed by Input and Output Flow

The application verifies that the manufacturer performance curves are respected in terms of hydraulic

and efficiency performance. Depending on the available instrumentation (pressure, flows, temps) the

kind of anomalies that can be detected will vary. If amperemeter is available the electric motor can be

also included in the scope.

© Inprocess 20

Pump anomalies

Hydraulic performance: When the DTC is not

delivering the pump outlet pressure. This can be caused

by pump cavitating.

Efficiency performance: When DTC is not following the

outlet temperatures. DTPT will provide actual

performance.

NPSH alert: When the DTC or DTPT approaches NPSH

limits.

Every pump will define a subset dynamic model

and will run two digital twins of that model:

DTC: Governed by pump curves

DTPT: Governed by pressure and temperatures

Anomaly

OK

The application verifies that the heat and material balances are respected taken into account the

pressure variation and the geometry of the exchanger or aircooler.

© Inprocess 21

Heat Exchangers anomalies

Sudden change in performance: when there is a

sudden change in the calculated heat transfer coefficient.

This could reveal a tube rupture, stream clogging, etc

Fouling factor alarms: Fouling factors are calculated

based in current performance and the system can be

configured to advise when it is needed to do a cleaning

Every exchanger is modelled individually with a rating modelling

tool (Unisim Heat Exchangers) using the design geometry,

process plant data from instrumentation and process data from

the Digital Twin model of the unit as needed.

The detection of anomalies is very dependent of the

instrumentation available around the exchanger.

© Inprocess 22

Inprocess Bad Actors: Setting Unisim models

Allows for multiple Steady-Stateor Dynamic Unisim cases

Windows services to manage Unisim models

OPC Data source or custom interface

How is the fouling factor calculated?▪ An objective function is defined▪ The function takes into account some key parameters as:

In/Out tubes temperature, In/Out shell temperature, Flows▪ Unsim Optimizer uses the fouling factor as the degree of

freedom.

Challenges & Solutions:▪ Exchanger energy balance:

• The balance didn’t match, the tubes required more energy than given by shell.

• Solution: Flow meter on shell side was bad calibrated. The asset recalibrated the sensor

© Inprocess 24

Customer case: Heat Exchangers

▪ Fouling Calculation has too much noise:• The plant data is noisy so It’s

necessary to do some treatment• Solution: Data reconciliation using

UniSim to recheck all plant values

• The value of the Inprocess Bad Actors application is increasing the plant uptime and avoiding

equipment damages by early identification of anomalies not currently detected by ICSS alarms.

• Inprocess is currently testing some modules of the application with customers who already own

dynamic process models of their plants, from previous DSS and/or OTS.

• The goal is to install 3-4 applications in different businesses (Oil&Gas, Refining, Chemical) and to

collect all the key requirements and anomaly casuistic.

• There are other areas of value of online Digital Twin that could be further explored to configure

other application modules (look-ahead, what-if, incident tracking)

© Inprocess 26

Takeaways

CONTACT:

JoseMaria Ferrer | Inprocess | Senior Advisor | +34 607 570 685 | [email protected] | www.inprocessgroup.com

mailto:[email protected]

http://www.inprocessgroup.com/

anomalies detection with unisim design · flow transmitters & control valves anomalies the...

Documents