Environmental Impact Reduction On

Flexible operation of Digital thermal

Plants By Using the Artificial
Intelligence
 Predictive Suite
by

AI for Flexible Operations

Presented at National Seminar on “Flexible
Operations of Coal-fired Power Plants”

Nov 1, 2019 at
 Flexibility - Thermal Power challenges
Aging
Assets

Asset
Part Load Stress and
Operations Failures

• Expected minimum loads at 25% of Base Load

• # of Ramp up and Ramp down cycles per day to increase
Always-On
Demand • Shorter response times to load cycles
Major issues reported In cyclic Operation

Boiler Turbine
•Boiler flameout •Damage to the turbines’ airfoils
•poor original design, •Resonance that led to failure in last-
•fabrication practices, stage blades
•fuel changes, •Cracks and deformation in the turbine
casing
•operation, maintenance,
•Rotor bow
•cycle chemistry
•Issues managing differential expansion
•Rubbing between radial/axial rotating
and stationary components
 Predictive Asset Analytics Solution to act
on Flexible Operation

We believe asset performance is driven by great

communication – between the asset and its owner.
With Pulse, your assets will reach out to you with
specific insights when they need your attention.
We leverage advanced analytics technologies and a wealth of available data to help run the industrial world with ZERO unplanned downtime,
highest possible performance and utmost safety.

CUSTOMER BENEFITS

Reduced Downtime and Improved Asset and Maintenance Workforce

Maintenance Costs Process Performance Empowerment
 - Technology and Features
Built on our proprietary AI-platform , Root Cause
Analysis
Predictive
Asset
Alarms and
Notifications
Diagnostics
Configuration
User
Management
our solution comes with deep asset knowledge to Monitoring

ensure the best decisions for each asset! Digital Twin

Processing and Analytics Engine Asset Knowledge Base

Fault Tree
Rule Engine Design Data
Evaluation

Performance Physics-based
and Risk Process Flow Diagram
Continuous and real time Analytics Assessment
Computation

Security
Data Historic process and
Prediction
Secure, Cloud-based data storage Validation performance data

Pattern Maintenance History and

Fingerprinting
Diagnostics and Recommendations recognition Activities

Equipment Library - System and Equipment Models

Mobile and Web Notifications, Dashboards

Collaborative Troubleshooting
Scalable Data Storage
Add custom analytic modules
ERP OPC CSV RDBMS
 - Overview
Pulse is purpose-built to apply advanced AI techniques to solve the most critical Industrial problems.

Technology Infrastructure Analytics Infrastructure

Integration – The first step to a robust industrial

AI solution is a secure integration with field assets
and data transfer to a remote platform. We
support industrial protocols (OPC, Modbus, Data Digital Twin – This forms the core of the platform
Files, RDBMS) and automation systems (DCS, PLC) and is solely responsible for efficiently converting
asset data into useful and timely insights. This
comprises the proprietary analytics framework as
Scalable Data Management – Based on a non- well as pre-built asset-specific models.
conventional database system designed for
handling voluminous data, this layer forms the
foundation to reliable and fast analysis. Also
functions as an asset/plant historian on the cloud. Business Applications – Analytics is only
beneficial when used by the right stakeholders to
make their decisions in a timely manner. This is
enabled by the applications layer which is
Security – Pulse takes every measure to ensure responsible for delivering insights on web/mobile
data integrity and security. This includes using
Firewall at the asset/plant side, read-only data
access from assets, encrypted data transfer to
cloud and user-restricted access to data from cloud
Digital Twin

The Digital Twin is the core engine that combines

Raw Data, Machine Learning and Asset Specific
knowledge to create insights for our customers.

CORE COMPONENTS

Equipment Library Asset Processing and Analytics

Knowledge Base Engine
 Digital Twin - Equipment Library
Benefit from our library of pre-built asset and system models which drastically reduces the time to
deploy AI for your assets and delivers value from Day 1.

What are Asset/System models?

We use Models as a proprietary way to digitally describe the Heat Exchanger Pulverizer Condenser

functioning of an asset or system, the relationship between

various data points within an asset or system and provide
deeper context to the higher-level machine learning algorithms Boiler Generator Inverter

to draw useful inferences.

Turbine Centrifugal Pump Gear Box

Pulse Equipment Library contains:

• Models of commonly used industrial equipment (50+)
• Models of common industrial processes Seals and Packings Valves Compressor
• Over 500 asset-specific fault diagnostics and
recommendations derived from prior known failures
• Reliability-centred maintenance strategy for given failure
Centrifugal Fan Industrial Fan Transformer

And more…
 Digital Twin - Knowledge Base
Two assets (e.g. Compressors) designed to the exact same specifications still have their own
individual characteristics. Knowledge Base is Pulse’s way to treat every single asset individually.

What is a Knowledge Base?

The Knowledge Base is an evolving document store of all information specific to an individual asset or
process. This is used by the higher level AI models to ‘learn’ and ‘develop’ deep knowledge of the
functioning of a specific asset or process. Pulse Knowledge Base contains:

Design Data – This is critical information that helps map the Asset model to
expected functionality of an asset

Process Flow – This is specifically useful to understand the physical

connectedness of various assets within a system

Maintenance Records – Knowledge of the existing maintenance process,

history of issues/breakdowns to understand criticality of asset

Historic Data – This is the most critical piece of information used by the
machine learning models to understand and benchmark an asset’s performance
 Digital Twin – Analytics Engine
For an Always-On asset/plant, it is critical to have a robust engine which is performing thousands of
analyses and applying hundreds of predictive models to generate timely and accurate insights.

How is data handled in Pulse?

The Processing and Analytics engine keeps a watch on an assets behaviour before triggering higher-level
diagnostics, root cause analysis and failure forecasting. It achieves this by orchestrating a series of
validations, computations, pattern detections and predictions for every single data point that the asset
generates before generating a single alert. This is done by the interplay between the following processors

Validation Prediction Root Cause

Auto detection Models for early Provide
of poor quality detection of diagnostics
or erratic data issues and through fault Risk Assessment
Fingerprinting and substitute Recognize Pattern forecast Physics-Based trees or perform Calculate a
Uses self- plausible data APR Algorithms Remaining Useful Models built automated RCA reliability index of
learning models values instead which mine for Life of Assets based on first using data every single asset
to understand relationships and principles and based on a
asset/process trace abnormal physics to help comprehensive
behaviour from patters and data validate actual check of all
historic data deviations asset performance parameters
Case Study – Boiler Stress evolution-
BTL advance warning with 95% accuracy
Customer Background:

• Coal-based 180 MW Power plant

Issue:
Frequent boiler tube leaks
• Unable to detect the onset of tube leak early enough to control the damage
• Cyclic (flexible) operations makes detection of signals more challenging
• Demand for Plant availability makes taking preventive shutdowns difficult
Steam profile changes
Analytics and Outcome:
• Studied patterns of data and changes in relationships between various tags
• Comparison of steam profile in a 15 days period leading upto a tube leak
• Able to detect minor changes in boiler operations close to 8 days before actual tube leak incident and
observe the progression of the issue

This abnormal region of Normal Operations

operations is observed from
10 days before a Tube Leak
compared to Steady
operations (May)

8 days before Tube Leak

 Feedwater profile change
8 days before Tube Leak

Normal Operations Tube Leak 1 Tube Leak 2 Tube Leak 3

FW temperature have started to drop almost 8 days before the

tube leak as indicated by the band of Yellow and Red which is
absent during normal operations of April and May.
 Flue Gas profile changes
8 days before Tube Leak

Normal Operations Tube Leak 1 Tube Leak 2 Tube Leak 3

Similar differences in FG Pressure observed (by 40-60 mmwc)

in the lead up to the tube leak event.

These are early indicators of an upcoming leak event.

 Progression of issue
Day 1 Day 2 Day 3 Day 4

Relationship during
normal operations

Day 5 Day 6 Day 7 Day 8 (Tube Leak)

Gradual change in model relationship from 8 days before a Tube Leak

indicating significant change in Flue Gas profile as a possible early
indicator.
Outcome

The solution
• Real-time cause- effect Recognition helps monitor asset stress
• Solution inherently tuned for flexible operations and compares with best known historic operations
• Relationship changes between multiple tags and predictive models help accurately detect
component level degradation

Outcome
• Providing a predictive score (probability index) helps plant modify operations to reduce stress on
boiler or plan well for an unavoidable shutdown
• Real-time monitoring also enables operations at peak performance
Stressed out. Contour plots colour by maximum principal stress at the buck
stay junction location are shown here for both cases where the sliding joint
plate is present (left) and removed (right). Courtesy: Flownex SE
 Real-time Optimization for
Flexible Operations

We partner with a leading Australian engineering software firm with focus

on Real-time optimization of Thermal Power Plants in Flexible Operations
 ExactSpace + Synengco
• MoU for exchange of best practices between Australia and India in areas of AI, Asset Reliability and
Real-time Optimization
• Synengco Background:
• 20+ years of solutions for Thermal Power
• Specific focus on Flexibilization

• Real-time Optimization modules

• Throttle Valve RTO
• Combustion RTO
• Sootblowing RTO
• Condenser RTO
• GasPath RTO
• Cooling Water RTO
• Airheater RTO
• Dispatch RTO
 Real-time Optimization
• Real Time Optimization, in the hierarchy, is above the
APC and/or BPC layer.
Business planning(year)

• RTO computes the optimal operating conditions

Business Optimization(day/week)
from the plant economy point of view
Real Time Optimization(min,hour)
• RTO typically uses optimization engines.
Advanced Process Control (sec)
• RTO horizon exceeds that of APC; often
considers only steady state. Basic Process Control (msec)

• Scope of RTO may be local (the same process as Measurement / Instrumentation

controlled by APC), may include several units or
Plant / technology
may be plant-wide.
Integrated process
• RTO specifies targets (set-points, set-ranges and/or management

MV targets) for APC or basic process control

E.g. Sootblowing Optimization
Learned cause & effect Control Constraints Operator control points
• Impact of each SB on each • Min/max frequency • Soot blower sequence
boiler element’s TC • Maintaining control • Soot blower time
• Impact of each SB on back limits (spray flow, OPC write
end temperature emissions, etc.)
• etc.

Performance constraints
• Max back-end temperature

System efficiency When should each soot blower be

• Back-end temp & flows scheduled to maximise performance?
• Boiler element TCs
• L/R temperature balance
• Steam consumption

Thermodynamic performance
calculations

Instrumentation OPC read

Case Study
Customer Background:

• Australian Power Station – approx. 2800 MW

• Located in environmentally sensitive area
• Multiple coal sources with high ash content
• Facing transition due to renewables addition to grid

Solution:
• Deployed Real-time optimization modules to help improve plant performance
during flexible operations (

Outcome:
• 1.5-2% improvement in Heat Rate at full load
• 4% improvement in Heat Rate at part loads
• ROI of 3-12 months
• Improved plant capacity and reliability
AI in Gas Distribution

Optimization in Gas distribution Networks can be

extremely valuable as it involves expensive resources
operating in highly complex networks.

BENEFITS

Human and Proactive Maintenance Fraud and Theft

Environment Safety and Asset Availability Detection
 GAIL – Existing Data Setup
• 2 LPG Distribution Networks (Jamnagar, Vizag)
• 7 Natural Gas Distribution Networks
• Total Length of Network – 10,000+ KM
• Centralized Monitoring and Control using SCADA

Centralized Monitoring and SCADA

Architecture:
- Main Station at Noida (National Gas Management
Center)
- Backup Station at Jaipur, Triple Redundancy

Data Availability
- Operation Data for Billing/Allocation
- Maintenance related data
- All data available in historians

Central Control room setup for GAIL (source: Yokogawa)

 E.g. Compressor Station PdM
Compressor Stations are the most critical part of the Gas distribution network. It is important to
have the station and all its components working at top performance and availability at all times.

Components we can monitor

in a Compressor Station See our prior
Filters Gathering System Cooler Metering Environment Sensors
Compressor (Air Quality, Noise, etc)
success of
Predictive
Maintenance for a
Compressor OEM.

Benefits from Safety Considerations Compressor Maintenance

our AI solutions
- Sudden Pressure Some conditions that we can
drop detection predict and detect:
- Gas Leakage - Oil Leakage or clogging
- Compliance to - Rotor physical damage
norms (noise, air - Discharge piping issues
pollution, etc) - Damaged bearing, coupling
- Shaft Misalignment
- And many more…
1. Changing with the times. Network-level optimization can help utilities better understand the power landscape in real
time. Courtesy: GE Power
Why ExactSpace Low-Cost
50% lower than equivalent market
solutions with ZERO CAPEX

Customizable
We will use the same platform to add
analytics modules as per your need

Unified
We provide a single platform for all
monitoring and analytic needs

Value from Day 1

Pre-built Asset Models will deliver
value from day 1
 Plant Outcome Description
Prevent forced outages and unplanned plant
downtime
Asset
Reliability Reduce maintenance related costs due to early
detection of issues in all assets of the plant.

Data and event analysis by expert group in BHEL

Operate at best possible configuration during
flexible operations, thereby reducing Heat Rate of
the unit and cost of fuel
Flexible Recommendation to optimize the performance of
Operation system and equipment.
Optimization
Active combustion monitoring and reduced
Auxiliary consumption by performing
detailed specific power analysis

Collaborate with multiple stakeholders for

improved communication and troubleshooting

Easy access to 5 years of data from a single click

Workforce
Digitalization and Productivity Reduce/Eliminate manual effort to collect data
and digitize historic records for easy future access

Alarm and trip analysis report with

recommendation.
Spare Management
 New digital solutions additions to Pulse
Video Analysis for Rotary Equipment Environmental compliance from ESP monitoring

Data of rotary equipment from DCS is insufficient to
determine its health. To augment this, we now enable
periodic/continuous analysis of video+audio grabs
from bearings of rotary equipment to detect minute
changes in operations to avoid major damages.
Continuously monitor ESP fields and their operations
to maintain ideal stack opacity levels in the plant. Use
stack outlet images to correlate opacity with ESP
health and avoid excess stack emissions and stay
compliant with Environmental norms

Identify changes in bearing health Minimize stack emissions by

and operations from periodic video continuous monitoring and
grabs optimization of ESP health
 Case Studies
i. Steam Turbine Issue Prediction
ii. AI for Compressor OEM
iii. Boiler Degradation detection
iv. Turbine Issue Diagnostics
v. Theft Detection for Utility
Case Study 1 : Prediction and
Diagnostics for Turbine Issues
Case Study 1 – Predictive Turbine
Maintenance
Customer Background:

- Coal-based 30 MW plant
- Assets include AFBC, Turbines, 6 WHRBs
- 600+ tags monitored from DCS
- No convenient access to data for
troubleshooting – all data available only in files

Issue:
Turbine Vibration gradual increase
● Vibration levels gradually increased by 200% in
1 month
● No alarms in Vibration Monitoring system as
levels did not cross DCS threshold
● When issue was identified, regular Fault Tree Change in vibration trends and associated
diagnostics and SOPs didn’t help solve the predictive alerts generated by our system
problem
Case Study 1 – Predictive Turbine
Maintenance
ExactSpace Intervention:
● Early prediction of issue using Advanced Pattern
Recognition and other prediction models
● Millions of correlations performed on the data
(on both upstream and downstream processes)
to detect possible process induced variations
● Provided possible root causes and diagnostics
● Abrupt change in LP heater level was confirmed
as cause along with Operations team
● Vibrations reduced to normal levels after
adjusting LP Heater level

Value to customer:
● Prevention of possible breakdown - INR 60 High level of correlation and causality of
Vibration with LP Heater level
Lakhs
● Prevention of further mechanical damage to
turbine by early detection of issue
● Time to resolve issue reduced to 1 day
Case Study 2 : Early predictions of
issues in Compressors to optimize
maintenance visit and spare costs
(for an OEM)
Case Study 2 – Optimize
Maintenance visits and Spare costs
Customer Background:

- One of world’s largest manufacturers of

Compressors
- Hundreds of thousands of compressors are
installed worldwide

Issue:
High maintenance visit costs due to unforeseen
breakdowns in Compressors
● The OEM’s field assets at customer locations are Faults reported in the historical performance
under warranty and maintenance.
● Maintenance visits are based on pre-defined
schedule or breakdown service requirement
● Sometimes, assets with ‘Remaining Useful Life’
breakdown OR assets which undergo
maintenance are in good condition
● Threshold based rules were unable to accurately
alert maintenance teams of breakdown
Case Study 2 – Optimize
Maintenance visits and Spare costs
ExactSpace Analytics:
● Using individual compressor’s operating
parameters (load, pressure, temperature, etc) of
1000 assets over 1 year, we studied patterns
leading to historic faults/breakdowns
● Of all the data observed, changes in pattern in
3 parameters were statistically significant in
periods prior to breakdown but were not easily According to our models, Discharge temperature changes can help
predict breakdowns 2 months before the actual event
observable to the naked eye
● In all breakdown cases, our analysis is able to
detect these patterns 2 months prior to
breakdown

Value to customer:
● Real-time monitoring of parameters helps
accurately predict health of equipment
● Maintenance team can use this insight to
optimize their field visits and reduce spares cost
● Improved availability of the equipment  better Sump pressure is one of the parameters showing gradual decay in
customer experience the weks leading upto breakdown
Case Study 3 : Improving Boiler
performance by predicting
Economiser degradation
Case Study 3 – Boiler Performance
Improvement
Customer Background:

- 150 TPH Boiler fired by combination of fuel

- Steam from boiler used for Paper
manufacturing process as well as power

Issue:
Steaming capacity (Evaporation Ratio)
deterioration
● For similar coal quality and boiler output, the
boiler steaming performance deteriorated when
compared to past performance GCV of Fuel

● Due to varied operating profiles of the boiler

(output, coal quality, etc) and associated Performance in 2018 for similar input quality
process parameter changes, the cause for this was not in the expected range (highlighted
by red box)
could not be identified
Case Study 3 – Boiler Performance
Improvement
ExactSpace Analytics:
● While 100+ parameters of boiler where studied,
there were minor indications of degradation in
equipment like Economiser, Air Preheater and
ID fans as inferred from varying temperature
profiles of water and air in the boiler Higher temperatures at Economiser Exit –
● However, when downstream processes were Suggests deteriorating heat transfer
also analysed, there were more clear indications
of trouble in Turbine (vibration, wheelchamber
pressure, etc) which in turn coincided with a
drastic shift in silica levels in water
● Inference: Increased Silica levels in water leads
to deposition of layers in water walls of
Economiser leading to reduced heat exchange.

Value to customer:
● Insights from data analysis helped resolve the
water chemistry changes
● Resuled in Improved Turbine efficiency and
Evaporation ratio  0.8% reduction in coal
Change in water chemistry behind chemical deposition at
economiser leading to poor steam generation, and reduction in
evaporation ratio.
Case Study 4 : Prediction and
Diagnostics for Turbine Issues
Case Study 4 – Turbine Predictive
Maintenance
Customer Background:

- Coal-based 60 MW Power plant

- Assets include AFBC and Turbine
- 300+ tags monitored from DCS

Initial change in Turbine Vibrations level

Issue:
Turbine Vibration spikes and variation
● Vibration level suddenly increased by 30% on a
given date
● The levels started showing spikes and high
variability at the end of 30 days
● Spikes were very close to trip point of the
turbine unit
● Turbine OEM was called in to assess mechanical
damage but no mechanical issues were found
with frequency analysis
Sharp spikes and overall increase in vibration
level from Day 30 onwards
Case Study 4 – Turbine Predictive
Maintenance
ExactSpace Analytics:
● Our pattern recognition system observed two Turbine Vibrations
patterns whose changes worked in tandem to
create vibration level change and variability
● Control Oil Header Pressure – Highly correlated
with vibration and creates variability of vibration Control Oil Header Pressure
at higher levels of pressure
● Main oil temperature – Inversely affects
vibration and effects last upto 60 Hours after
change. Mainly causes change in vibration level
● We recommended decreasing pressure and Main Oil Temperature After Coolers
increasing temperature leading to immediate
decrease in variability of vibration and
decrease of level of vibration after 2 days

Value to customer:
● Prevention of possible breakdown - INR 1.15 Cr
● Could’ve avoided cost of OEM intervention
● Prevention of further mechanical damage
Vibration going back to normal after our
recommendation
Case Study 5 : Utilities Power theft
detection using pattern
recognition of 100,000+ meters
Case Study 5 – Utilities Power theft
detection
1
Customer Background: 2
- A mid-sized Energy meter manufacturer and Utility services provider
- Contracted with multiple Indian utilities to for Automated Meter reading

Issue:
Difficulty in accurately and automatically detecting theft of power from consumption patterns of
100,000+ consumers
● The customer receives continuous data from 100,000+ meters regularly
● Volume of data is massive and current architecture does not allow easy scaling
● Analysis of data is done manually leading to missing out lots of actual cases of Power theft

ExactSpace Solution:
Cloud-based scalable data platform and Advanced Pattern recognition for Theft detection
● Provided a cloud-based scalable data ingestion and long term data storage system
● Enabling accurate and automated detection of theft from across 100,000+ meters’ data by using models
which look for changes in consumption patterns for each consumer against patterns of its history,
industry, time of the day, day of the week, seasonality, etc
Our Team
Boben Anto C
Boben brings forth close to three decades of deep expertise in the Industrial (Process and Power) sector to be encapsulated
into the ExactSpace solution. Coming from a rich technical educational background, he spent the initial years of his career
with NTPC, India’s largest and leading power producer. He then moved on to work with international organizations like
STEAG and Uniper where he led India operations for Power solutions and services. A prominent expert in the Indian power
industry, he has authored numerous papers on topics covering efficiency improvement and advanced condition monitoring in
power plants.

Rahul Raghunathan
Rahul is responsible for the business strategy and customer development activities. Engineer and management graduate by
qualification, Rahul has been involved in data and analytics technologies right from the beginning of his career at various
capacities across industries. His most recent professional stints include global sales leadership roles at fast-growing IoT and
Analytics startups in India.

Arun Jose
Arun is responsible for Product strategy and engineering activities at Exactspace. Extremely passionate about engineering
and building solutions to create positive impact on a large scale, Arun has been involved in Analytics technologies from the
beginning of his career. Having implemented complex statistics-based models in the business of algorithmic trading, Arun
believes the value of this technology holds the maximum promise in the world of Industrial Manufacturing. His most recent
professional stint was as a Analytics Product leader at an IoT-focused organization where he has worked extensively with
manufacturing industries like Iron & Steel, Thermal Power and Cements.
Thank You!

Reach out to us for more details and solution demo.

Rahul Raghunathan | Director (Business)

rahul@exactspace.co
boben@exactspace.co
+91 99165 23232
+91 9051219801