Digitalization & Advanced Analytics Overview - FOR PRESENTATION ONLY June.2018 Modec do Brazil
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
Digitalization & Advanced Analytics Overview FOR PRESENTATION ONLY June.2018 Modec do Brazil
AGENDA
1. Quick introduction
2. What & Why is Advanced Analytics
3. What Modec is doing
2/33
AGENDA
1. Quick introduction
2. What & Why is Advanced Analytics
3. What Modec is doing
3/33
FPSO digitalization is in an important lever in MODEC group’s
2018-2020 mid-term business plan
Operational Organization & Expansion to
Excellence Talent new businesses
▪ Maximize life-cycle ▪ Maintain highest ▪ Expand to new related
profit compliance standards projects such as
FLGN and FSRWP
▪ Initiate FPSO ▪ Strengthen
digitalization governance and ▪ Capture new
corporate functions opportunities through
▪ Strenghten market
renewables (offshore
intelligence to over- ▪ Establish carrier
wind/ wave power)
come competitors development plan
and seabed mining
worldwide and ensure
talent retention
Modec 2018-2020 mid-term business plan
4/33
Digitalization opportunity: AI, RPA, Digital Twin…
Process design data Enhanced Project Management
Enhanced Cost Controlling/
Structural design data CAD
Decision Making
Enhanced Finance
Sensor data Process
Enhanced Legal
Advanced Analytics / AI Process (Legal Tech)
CMMS / ERP /
Historian data base
Purchasing
Optimization
Text / Report /
Scheduling data
Logistic
Optimization
Robotics Process Automation
Image, Video, Sound data Enhanced HR
Management (HR Tech)
Drone Production Optimization/
Failure Prediction
Satellite image / info Data Lake / Digital Twin Robotization
Weather / environmental data AR / VR
5/33
SOURCE: Icon created by Bakunetsu Kaito, Denis Klyuchnikov, Siraj C, Vectors Market, Creative Stall, Mahmudxon, Eugene Dobrik, Lero Keller, NOPIXEL, iconcheese, Wilson Joseph, Olivier Guin, Chameleon Design, Dinosoft Labs, Seb Cornelius from Noun Project
Digitalization opportunity in Modec: already ongoing at Brazil
Process design data Enhanced Project Management
Enhanced Cost Controlling/
Structural design data CAD
Decision Making
Enhanced Finance
Sensor data Process
Enhanced Legal
Advanced Analytics / AI Process (Legal Tech)
CMMS / ERP /
Historian data base
Purchasing
Optimization
Text / Report /
Scheduling data
Logistic
Optimization
Robotics Process Automation
Image, Video, Sound data Enhanced HR
Management (HR Tech)
Drone Production Optimization/
Failure Prediction
Satellite image / info Data Lake / Digital Twin Robotization
Weather / environmental data AR / VR
6/33
SOURCE: Icon created by Bakunetsu Kaito, Denis Klyuchnikov, Siraj C, Vectors Market, Creative Stall, Mahmudxon, Eugene Dobrik, Lero Keller, NOPIXEL, iconcheese, Wilson Joseph, Olivier Guin, Chameleon Design, Dinosoft Labs, Seb Cornelius from Noun Project
AGENDA
1. Quick introduction
2. What & Why is Advanced Analytics
3. What Modec is doing
7/33
Very rough
explanation
What is Advanced Analytics
8/33
Advanced Analytics (AA) ≒ ML + DL
9/33
SOURCE: AI & Machine Learning: The evolution, differences and connections (https://www.linkedin.com/pulse/ai-machine-learning-evolution-differences-connections-kapil-tandon/)
Concept of AA: “Butterfly Effect” but…
“Does the flap of a butterfly’s wings in Brazil set off a tornado in Texas?”
- Philip Merilees, at the 139th meeting of the American Association for the Advancement of Science in 1972
“風が吹けば桶屋が儲かる”
10/37
Source: WikipediaConcept of AA: there could be many butterflies that leads to tornado
11/33Concept of AA: AA reveals the “important” butterflies
statistical / mathematical approach
12/33Concept of AA: example
Walmart事例
（実は都市伝説）
13/33Concept of AA: example
Source: www.amazon.co.jp
Source: https://aibo.sony.jp/
14/33Concept of AA: example
Source: リクルートにおける画像解析事例紹介 Source: Google, Tesla Source: https://www.jscore.co.jp/score/technology/
https://www.slideshare.net/recruitcojp/ss-
55725591
15/33Concept of AA: example
金融不正検知 オークションサービス 商品自動 迷惑メールフィルタ
認識 & 情報入力／不正出品
検知
サービス解約者予測・退職予測 旅館・ホテルの稼働率予測 & タクシー需要予測・配車支援
価格決定
スマートスピーカー（Amazon 医療画像（CT等）診断 機械翻訳
Echo, Google Home, (Google Translator等）
Apple HomePod, LINE
Clova）
16/33Why
17/33AA - different from convectional analytics
18/33
Source: http://livedoor.blogimg.jp/sts_114/imgs/8/d/8d3fa166.jpgComputing Power
19/33
Source: Time “The Year Man Becomes Immortal”Amount of data
250,000
(MB)
204800
200,000
150,000
100,000
50,000
5 10
0
1950 1960 1970 1980 1990 2000 2010 2020 Year
2030
20/33
Source: https://thenextweb.com/shareables/2011/12/26/this-is-what-a-5mb-hard-drive-looked-like-is-1956-required-a-forklift/
http://designbeep.com/2010/09/30/55-vintage-computer-ads-which-will-make-you-compare-today-and-past/
amazon.comEvolution of algorithm
21/33
SOURCE: https://pt.slideshare.net/roelofp/deep-learning-as-a-catdog-detectorAGENDA
1. Quick introduction
2. What & Why is Advanced Analytics
3. What Modec is doing
22/33Concept of AA: Advanced Analytics reveals the “important” butterflies
おさらい
statistical / mathematical approach
23/33Concept of AA: in our case, Butterflies = Pressure, Temp, Vibration…,
& Tornado = future Trip of the equipment
PI-2010A-03
Modecの場合
TI-2010A-01
Main A
Compressor
fails in 6hrs
statistical / mathematical approach
24/33
SOURCE: Icon created by Bakunetsu Kaito from Noun ProjectConcept of AA: moreover, it explorers not only raw but also
computational values that affect the trip
6hr StdDev PI-2010A-03
12hr Max TI-2010A-01
Main A
Compressor
fails in 6hrs
25/33
SOURCE: Icon created by Bakunetsu Kaito from Noun ProjectConcept of AA: we created a system to monitor important values to
predict failure in advance for FPSO MV22
6hr StdDev PI-2010A-03
12hr Max TI-2010A-01
Main A
Compressor
fails in 6hrs
26/33
SOURCE: Icon created by Bakunetsu Kaito from Noun ProjectWhere we are now
27/33Digitalization opportunity in MdB:
We are piloting failure prediction models
Implementation
Model development ▪ Fine tune models in the field
▪ Develop dashboards for field
Data collection ▪ Select key areas for model deployment
development based on ▪ Set IT infra-structure to host
criticality and operational
▪ Collect main operational data costs
models and dashboards
since 1st oil ▪ Have models implemen-ted
▪ Develop models, through with real time data
– Sensors registry advanced analytics, to
– Cost reports support improvements in ▪ Train ops team to use
– Daily / Monthly reports operational efficiency and new tools
– Lab reports maintenance spend
▪ Interview key operation team ▪ Evaluate the model
to understand vessel’s performance for PoC
challenges
We are here
We’ve already started this effort on MV22 and are preparing to roll out to the rest
of the fleet in 2018
28/33We’ve already completed the PoC on MV22, are now under
implementation phase (1/3)
Target issue / system on
MV22 Prediction scope Failuer warning time Model quality
Early signs of conditions 5 days R2: 0.92 (subsurface)
Scale Risk
leading to scale build up in advance R2: 0.73 (surface)
Water
Injection Trips or stops in seawater 24 hours Accuracy 99%
Pump injection pump in advance Precision 97%
Molecular Early signs of molecular sieve 5 days Accuracy 90%
Sieves dehydration failure in advance Precision 74%
Main A Trips or stops in main A 12 hours Accuracy 92%
compressor compressor in advance Precision 89%
Reinjection Trips or stops in gas 24 hours Accuracy 99%
compressor reinjection compressor in advance Precision 99%
29/33We’ve already completed the PoC on MV22, are now under
implementation phase (2/3)
Data storage
AA model
Dashboard
Satellite
(O3B)
Cloud
(Amazon Web Services)
CCR PC Equipment data
PI Server
MV22 Onshore 30/33We’ve already completed the PoC on MV22, are now under
implementation phase (3/3)
Seemingly first use
case for OSI / PI
Project
Subnet
VPN MSSQL on AWS Amazon
PI OLEDB
gateway Amazon Glue S3
Interface
RDS
Launched Dec. 2017
Early adoption use case
IAM
Data Scientist
On-Site Network
AWS SageMaker
Dashboards
Internet Amazon
Operators gateway EC2
Offshore AWS US-East-1
31/33Model will soon be implemented in real world
Dashboard
32/33Thank you for your attention
33/33You can also read
