Upload
others
View
2
Download
0
Embed Size (px)
Citation preview
NOVOTEL LONDON WEST • LONDON, UNITED KINGDOM • 2-4 APRIL 2019
DIOT: An Intelligent Predictive Maintenance
Strategy for Subsea Structure
Subrata Bhowmik
McDermott International
© M C E D e e p w a t e r D e v e l o p m e n t a n d G u l f Q u e s t L L C
Background
2
• Low oil price prediction ($55-$75)
• No increase in production rate (North Sea)
• Old and inefficient maintenance approach increase OPEX
• Operators need significant OPEX reduction
Possible Solution
• Inspection less often AND at fewer location ANDrapidly AND remotely AND improve Reliability
• Use modern technology IOT, Cloud Computing and Artificial Intelligence
© M C E D e e p w a t e r D e v e l o p m e n t a n d G u l f Q u e s t L L C
Sensor Data for Structural Health
3
Which is better for structural health assessment?
Visual Inspection
Sensor Data
Inspection
© M C E D e e p w a t e r D e v e l o p m e n t a n d G u l f Q u e s t L L C
Digital Intelligent Operational Twin (DIOT)
4
Digital Intelligent Operational Twin (DIOT)
Low Cost IoT Sensors
Low Cost Cloud Storage
Machine Learning Algorithm
Data Driven Model for Subsea Structure
Predictive Maintenance
Real-Time Data Monitoring
© M C E D e e p w a t e r D e v e l o p m e n t a n d G u l f Q u e s t L L C
Overview of Maintenance Strategy
5
Corrective or Maintenance
• Unscheduled interventions
• High downtime & cost
Planned or Preventive Maintenance
• Scheduled inspection
• Not preferable for subsea components
Predictive maintenance
• Indicate the possible Fault happening before it actual happen
• Reduction in unplanned downtime
• Reduction in inspection and maintenance time
• Reduction in OPEX
© M C E D e e p w a t e r D e v e l o p m e n t a n d G u l f Q u e s t L L C
Digital Intelligent Operational Twin (DIOT) Architecture
6
© M C E D e e p w a t e r D e v e l o p m e n t a n d G u l f Q u e s t L L C
Long-Short Term Memory (LSTM) Algorithm
7
Machine Learning Algorithm forTime Series Forecasting/ Prediction
Special kind of Recurrent Neural Network (RNN)
Save the current output and feedback to the input and predict the future output.
Capability to memorize the inputs due to its internal memory.
Capability to memorize information for long periods of time.
Long-Short Term Memory (LSTM) Algorithm
© M C E D e e p w a t e r D e v e l o p m e n t a n d G u l f Q u e s t L L C
Digital Subsea Field
8
Subsea Jumper
Subsea Jumper
Subsea Structure monitoring is required for
Xmas-Tree
Subsea Jumper
Pipeline Span
Subsea Wellhead
PLEM
© M C E D e e p w a t e r D e v e l o p m e n t a n d G u l f Q u e s t L L C
Case Study: Subsea Jumper
9
Sensors installed
Strain Gauge
Accelerometer
Pressure gauge
ADCPStrain gauge AccelerometerData
Logger
Cloud
Digital Model
Data Logger
Wireless data transmission Data are stored in Cloud(AWS)
Data Analytics on Cloud
Structural Assessment from Data Driven model
Predictive maintenance strategy
© M C E D e e p w a t e r D e v e l o p m e n t a n d G u l f Q u e s t L L C
DIOT Modelling Stages
10
StressEstimation
Machine LearningModel (Operational
Twin of Subsea Jumper)
Stage I: Data Preparation
Stage II: ML Model Building
Stage III: Stress Estimation from ML Model
StressRanges
Remaining Fatigue Life Calculation
Predictive Maintenance
Strategy
Stage IV: Remaining Fatigue Life Estimation
Stage V: Predictive Maintenance Strategy
© M C E D e e p w a t e r D e v e l o p m e n t a n d G u l f Q u e s t L L C
Results
11
Time(s)
0 50 100 150 200 250 300 350 400 450 -0.1
5
-0
.10
-
0.0
5
0
0.0
5
0
.10
0.1
5
Acc
ele
rati
on
(m
/s2
)
PredictedMeasured
Comparison between measured and predicted acceleration data
Comparison between FEM calculated and predicted remain fatigue life
© M C E D e e p w a t e r D e v e l o p m e n t a n d G u l f Q u e s t L L C
Summary
12
Reduction in ROV inspection frequency
Reduction in inspection time due to predictive maintenance schedule
Elimination of operational inefficiencies with cutting edge technologies
Iimprovement of useful remaining fatigue life using actual loading condition
Reduction in OPEX (Expected 30-50%)
Challenges
Sensor data transmission reliability in deepwater
Sensor data measurement rate
Model accuracy
Battery reliability
Majority of operators are still reluctant to AI
© M C E D e e p w a t e r D e v e l o p m e n t a n d G u l f Q u e s t L L C13
McDermott Digital Twin Journey
© M C E D e e p w a t e r D e v e l o p m e n t a n d G u l f Q u e s t L L C14