An approach to Predictive Asset Management and predictive Maintenance using using Concept of Data Driven Modeling

1. A Practical Approach to Predictive
Asset Maintenance EHM through
System Identification (Data Driven
Modelling) and Machine Learning
Techniques
Mayur Dvivedi

2. Analytical Engine Frame Work

3. Overview of The Approach
• Failure Modes in Reliability
• System Observability
• System Identification
• Machine Learning in Reliability
• Comparative Data from Physical Models for Simulation
• Real World Data and Expert Analytics for Design and
Operations
• Building Scalability - Cloud Based Services for Remote
Monitoring
• Big Data Analytics – Anomaly Detection and Recommender
Systems
• Aggregate Model Approach
• Analytical Modules for Machines
Some important concepts to be understood

4. Failure Modes in Reliability Centric
Maintenance
• CBM enhances RCM for Engine Health
Monitoring
• Trend of Machine Wear Dynamics
• On Main Unit, Sub-unit, balance of Plant
• Prioritize for reliability by ranking failure
modes based on Risk Priority Number
• In RCM we select and monitor features that
track failure modes to optimize operations,
maintenance & design improvement

5. System Theory
• Real Life systems are Non Linear
• Can be linearised, simulated and desired inputs can be
scheduled for control
• Power Plant system has controllable units
• System Observability Concept for system identification
techniques for experimental modelling
• Essential to select Correct features or parameters,
variables assuring ‘Observability’
• Select Features from a combination of Process Variables
(speed, temp, pressure, flow, current, voltage etc) and RCM
/ CBM parameters for EHM (vibration-Frequency Based
Model, lube oil parameters, SBN, ABN, UWN, bearing
checker, hull potential, IC pressure signature, Motor current
analysis, Power quality, bearing checker value etc )

6. Systems Model
Rank Controllability Matrix : [B AB A^2B … A^n-1B]
Rank Observability Matrix : [C; CA; CA^2; …; CA^p-1]

7. Machine Learning in Reliability
• ML techniques can enhance reliability
• Algorithms for Anomaly detection and
recommender systems
• Near Real time analytics in different
operational modes
• Data from physical model and test bed /
installation commissioning can be effectively
used in algorithms for comparison and / or
validating anomaly detection

8. Real World Data and Expert Analytics
for Design and Operations
• Analytical tools for Unit, sub-unit, BOP
• Combine for Analytics of Aggregate
• Periodicity of observation recording and
flagging to be defined based on reliability
analysis
• Event based data logging
• Rules Based Expert System for Anomaly
Detection, Fault Flagging, Diagnosis and
Prognosis

9. Building Scalability - Cloud Based
Services for Remote Monitoring
• Cloud Based Expert - Remote Operations Center will be
the future of intelligence
• Incrementally define rules based learning
• Augmented Intelligence: Human Expert + Machine
Learning
• Enablers : Big Data Analytics – IOT, Machine Vision, DL /
ML tools, Anomaly Detection, Residual Useful Life and
Recommender Systems
• Build narrow AI – Analytical Engine
• Refine, make ‘Smart’ &
• AI e/e becomes highest ‘SME’

10. Aggregate Model Approach
• Key is to select the right features for each sub-unit to predict Health
• Model order reduction to 3 or 4 dominant states is useful
• Unit and Aggregate Equipment Health follow
• Dominant Parameters may vary based on operating mode / regime

11. Analytical Modules Score & Scale
EHM Score from analytical modules
– Sub-units & units to have GYOR scale
– Aggregate to have 0 to 100 score
– Visualization on Plant Configuration
– User defined alert notifications pushing to
location, remote, wireless user

12. How to Look at Machines for Developing
Analytical Modules in Different
Operating Modes
(Feature Selection)
• Selection of Output / measurable features based on
System Theory Models within valid operating mode /
regime / linear range
• Controllability leads to close loop control & automation
• Observability implies system identification &
predictability
• Useful to undertake Risk Based selection of features
• Modeling of Physical Machines and systems is easier
view laws of physics. Hence, majority features are valid
through the different regimes
• In Other models such as finance, the dominant
features largely vary in in different regimes / conditions

13. Analytical Modules for Gas Turbine
• List below is indicative only and also not in priority order of
dominance from failure modes
Starting Parameters (from start cycle parameters)
– Rectifier starter current
– Starting fuel time
– Maximum starting fuel pressure
– Lube oil pressure immediate at start & on idle
– Starting Cycle Time
– Hot Aborts
– Maximum exhaust gas temperature
Shut Down Parameters
– Run Down Time
– Time for Exhaust temperature drop

14. Analytical Modules for Gas Turbine
Running Parameters (from QAP reports)
– Unbalance of rotors and alternator
• Accelerometer
• Orbit analysis
– Misalignment of prime mover – drive
– Bearing shock Checker levels trend
– Drive gear vibration
– Speed-Torque-Load Vs Exhaust gas Temperature
characteristics

15. Analytical Modules for Gas Turbine
Running Parameters
– Compressor / Turbine bearing oil temperature trend
– Burner temperature Distribution
– Exhaust Temperature Slip
– Twin spool speed slip
– Lube Oil Pressure trend
– Fuel Injection Pressure trend
– Balance Chamber pressure trend
– Thrust Bearing vibration, thrust, oil temperature trend

16. Analytical Modules for Gas Turbine
Running Parameters
– Compressor / Turbine pressure ratio trend
– Surge Margin characteristics trend in acceleration
& SS
– Combustion pressures trend in acceleration, SS &
Dec
– Online LO monitor trend

17. – Spool Speeds / Vibration Signature
– Turbine Exhaust Temperature
– Compressor pressure
– Turbine Pressure
– Lube Oil Pressures trend
– Fuel Injection Pressure trend
– Balance Chamber pressure trend
– Thrust Bearing vibration, thrust, oil temperature
trend
Important & Probable GT Parameters
/ Features for Data Driven Modelling

18. Analytical Modules for HVAC
High Side
Starting and Stopping Parameters
Running Parameters – Low, Medium, High Duty
– Electrical drive current drawn starting and each
load 25, 50, 75, 100%, Motor CM parameters,
motor winding temp
– Mechanical- running speed, torque, vibration,
Pressure HP & LP, unbalance, misalignment,
bearing shock level, bearing temp, oil
temperature, DP pressure, poor HE efficiency

19. Analytical Modules for HVAC
Low Side
Starting and Stopping Parameters
Running Parameters – SAT, RAT, blower speed, valve
command, Delta T, CHWPI, CHWPO, FAD position, HE
efficiency
– Electrical drive current drawn starting and each load
25, 50, 75, 100%, Motor CM parameters, motor
winding temp
– Mechanical- running speed, torque, vibration,
unbalance, misalignment, bearing shock level, bearing
temp

20. …And many more to come