Realizing the Full Value of MINING & METALS Industrial ... Valappil... · Jaleel Valappil, David...
Transcript of Realizing the Full Value of MINING & METALS Industrial ... Valappil... · Jaleel Valappil, David...
-
Jaleel Valappil, David MessersmithBechtel Oil, Gas and Chemicals
Kelly KnightBechtel Nuclear, Security and Environmental
Realizing the Full Value of Industrial Internet and Data Analytics in LNG Facilities
Level 4 - Bechtel Public
INFRASTRUCTURE
MINING & METALS
NUCLEAR, SECURITY & ENVIRONMENTAL
OIL, GAS & CHEMICALS
-
Introduction to Industrial Internet (IIoT)
Key Application Areas in LNG Operations
Process monitoring, Predictive maintenance, Process optimization
Leveraging Existing Assets
Benefits of EPC Phase Implementation
Case Study in Data Analytics
LNG plant amine system monitoring
Presentation Overview
2016 Bechtel | 2 Level 4 - Bechtel Public
-
Introduction to Industrial Internet/Data Analytics
2016 Bechtel | 3 Level 4 - Bechtel Public
Digital Transformation Enabled by
Cloud computing
Decreasing cost of sensors, storage, bandwidth
Big data analytics and Sensor technology
Integration of
Physical plant and Digital information
IT (ERP/CRM etc.) and OT (DCS/PLC etc.)
Cybersecurity
IT and OT security challenges
Analytics
Site 1 Site 2 Site 3 Office
-
Industrial Internet Domains/Solutions
Includes new ways of using existing technology
OEM, Automation provides and others
Sensors
Wireless sensors/Intelligent sensors
Vibration Monitoring
Virtual sensors
Big Data Analytics
Ability to deal with large datasets
Higher resolution/Storage
Better analytics Machine learning
Industrial Internet/Data Analytics
2016 Bechtel | 4 Level 4 - Bechtel Public
IIOT Functional Domains (IIC)
Business
Control
Op
era
tion
s
Info
rmati
on
Ap
pli
cati
on
Sense Actuation
-
Applications Related to LNG Facility Operation
Applications of Industrial Internet
2016 Bechtel | 5 Level 4 - Bechtel Public
Predictive
Maintenance
Complexity
Value
Anomaly/
Fault
Detection
Exploratory/
Statistical
Analysis
Process
Optimization
Data Analytics
-
Tracking Key Performance Indicators (KPI) for LNG
Efficiency, Production, Yield, Quality
Condition and Performance Monitoring of Equipments
Process Control Monitoring
Ensure loops are in auto and working
Abnormal Situation management/Anomaly or Fault Detection
Enhancement of safety
Reduce plant downtime
Early detection of anomalies Operator intervention
Fault diagnosis important
LNG Process/Equipment Monitoring
2016 Bechtel | 6
-
Approaches to Fault/Anomaly Detection
Numerous Application Areas in LNG
Amine system and dehydrators
Heavies/Scrub Columns
Plate and Frame and Other Exchangers
LNG Process/Equipment Monitoring - 2
2016 Bechtel | 7 Level 4 - Bechtel Public
Fundamental
Model BasedHistorical Data
Based
Supervised
(Known Fault
Samples)
Unsupervised
(Learn Online)
-
Continuous Monitoring and Prediction of Equipment Conditions
82% random failure pattern
Benefits Compared to Reactive/Preventive Maintenance (ARC)
Reduction in maintenance cost (50%)
Reduction in unexpected failures (50%)
Increase in MTBF (30%)
Increase in availability (30%)
Asset Management - Predictive Maintenance
2016 Bechtel | 8 Level 4 - Bechtel Public
Idle Plant
Time
LN
G P
rod
uctio
n
-
Asset Management - Predictive Maintenance - 2
2016 Bechtel | 9 Level 4 - Bechtel Public
Failure Latent Error and Enabling condition Methodologies for Predicting Failure
Engineered algorithms
Supervised learning for prognostics
Models for Remaining Useful Life (RUL)
Model for probability of failure in time
Unsupervised learning techniques
Limited failure history
Application Areas in LNG includes
Turbomachinery
Heat exchangers/Valves etc.
Time
Eq
uip
men
t C
on
dit
ion
Optimal
Broken
Cost to repair
Asset Condition Over Time
Predictive maintenance
detects problems early
Prognostics
Statistical
Machine
Learning
Physical
Model
-
LNG Process Optimization Current Status
Advanced Process Control (APC)
Supervisory Systems
Real Time Optimization (RTO) not Common
Opportunities from Industrial Internet
Supervisory optimization (Less frequent)
Smaller units/equipments that are currently ignored
Enterprise-wide optimization (Multiple facilities)
LNG Plant Optimization
2016 Bechtel | 10 Level 4 - Bechtel Public
Supervisory
Optimization
APC
DCS
-
Typical Application Areas in LNG Maximizing Plant Efficiency
Optimum Process Conditions MR compositions, Other operating parameters
Operating point for compressors to enhance efficiency
Optimal load distribution between compressors
5-15% power reduction reported in studies
12.9% (Wang et al., 2012)
4.5% (Alabdulkarem et al., 2011)
Energy Optimization for Utilities
LNG Plant Optimization - 2
2016 Bechtel | 11 Level 4 - Bechtel Public
-
Process Knowledge Key to Successful Applications
Selecting right variables
Selecting right objectives for optimization
Digital Plant Information
Digital repository of plant information
EPC project data for greenfield facilities
Existing Control Infrastructure
Working with DCS/PLC systems
Starting with supervisory levels and gradual move to base layer
Existing Assets for Data Analytics
2016 Bechtel | 12 Level 4 - Bechtel Public
-
LNG Simulation Models and Digital Twins
Steady State/Dynamic Models
Operator Training Simulator
Computational Fluid Dynamics Models
Digital Twins for LNG Plants
2016 Bechtel | 13
Level 4 - Bechtel Public
Selection of Variables/Causality
Model Based Optimization
Boosting Data Models
Existing Assets Applications
Process Model/OTS 3D Models Digital Data
IIoT
Equipment Model
-
Benefits of Early ImplementationPreparing sensors/IT infrastructure
Configuring models based on process/project information
Testing during startup and commissioning
Avoiding Retrofits after Commissioning
EPC Phase Implementation of IIoT
2016 Bechtel | 14 Level 4 - Bechtel Public
Operations
IT Infrastructure Sensor SelectionModel Scoping
StartupEPC
LNG PLANT EVOLUTION
Model RefinementImplementation
CommissioningModel DevelopmentSystem Testing
-
Monitoring of Amine System in LNG PlantAmine system performance important CO2 removal
Anomaly Detection Method used Identify abnormal operation considering group of variables
Statistical approach using data
Operating Data from Plant after StartupAbout 320 tags for amine system
Seven months of operation
Performed in Offline Mode
Case Study Anomaly Detection
2016 Bechtel | 15 Level 4 - Bechtel Public
LC
Acid gas to
Sulfur recovery
Absorber
Treated GasKO Drum
Amine SurgeTank
FG
L/R Amine Exchanger
Lean Amine Cooler
LC
Reflux Drum
Air CoolerCondenser
LC
LC
Regenerator
Flare
Antifoam
SRU
Flare
Antifoam
Rich Amine Flash Drum
LC
FC
FC
Filtration Pkg.
Inlet Feed
Separator
Reboiler
Wash Water
Optional
Wash Water
Mole Sieve Dryers
C3 Chiller
-
Big Data Analytics Framework/Tools
Splunk - Machine data database and indexer
Prelert Anomaly detection tool
Neo4j Graphical database
Design Data from Smartplant Instrumentation
Captures relationships among tags, equipment
Enables setup of monitoring system with
minimal effort
Anomaly Detection System Set-up
2016 Bechtel | 16
Level 4 - Bechtel Public
LNG Plant Data
Splunk
Prelert
Anomaly
Scores
Neo4j
SmartPlant
Instrumentation
-
Predictive Model Consists of Statistical Probability Distribution
Bayesian distribution modeling
Fit is determined by unsupervised machine learning
Distributions have different shapes (Not just bell curves)
Distributions used to determine what is unexpected (anomalies)
Anomaly Score Normalized Probability Score
Score measures level of unusualness
Output has no false positives or negatives
Anomaly Detection - Methodology
2016 Bechtel | 17 Level 4 - Bechtel Public
-
LC
Acid gas to
Sulfur recovery
Absorber
Treated GasKO Drum
Amine SurgeTank
FG
L/R Amine Exchanger
Lean Amine Cooler
LC
Reflux Drum
Air CoolerCondenser
LC
LC
Regenerator
Flare
Antifoam
SRU
Flare
Antifoam
Rich Amine Flash Drum
LC
FC
FC
Filtration Pkg.
Inlet Feed
Separator
Reboiler
Wash Water
Optional
Wash Water
Mole Sieve Dryers
C3 Chiller
Three Different Approaches
All amine tags used to calculate anomaly scores (unit-wide)
Individual equipment tags to calculate separate anomaly scores
Clustering based on Smartplant Instrumentation
Individual groups to calculate separate anomaly scores
Obtained from process understanding
Anomaly Detection Approaches
2016 Bechtel | 18 Level 4 - Bechtel Public
Amine
Inventory
Acid Gas
Extraction Acid Gas
Removal
Separator
Performance
-
Increase in CO2 to >100ppm Caused by Amine Regenerator Operation
Reflux failure leading to increase in regen column pressure
Case Study Scenario 1
2016 Bechtel | 19 Level 4 - Bechtel Public
CO2Composition
Anomaly
Score - Regenerator
Column
Pressure
-
Increase in CO2 to >100ppm Caused by Increase in Feed Gas CO2 Concentration
Column differential pressure increases
Case Study Scenario 2
2016 Bechtel | 20 Level 4 - Bechtel Public
CO2Composition
Anomaly
Score - Absorber
Column Diff.
Pressure
Feed Gas CO2
-
Ability to Quickly Evaluate the Root Cause
Equipment contribution to anomaly score
Further drill down to particular tag
Case Study Root Cause Analysis
2016 Bechtel | 21 Level 4 - Bechtel Public
Contributor to
Anomaly
-
Thank You
and
Questions
2016 Bechtel | 22 Level 4 - Bechtel Public