The jour he journey i ney in r n railw ailway ay ana analytics ytics po power ered by ed by AI: AI: Towar ards ds railw ailway ay 4.0 4.0 Professor Diego Galar Lulea University of technology Head of Maintenance & Reliability, Tecnalia
Data driven models in railway is well trodden territory
But here be the dragons!!, approaches fail to scale
What analytics can be performed on railway?
Analytics and expectations also change
Types of data analytics
Descriptive analytics
Types of data analytics
Diagnostic analytics RELEVANT FEATURES DATA PREPARATION UNIFIED Feature Data DATA selection Reduction FORMAT Historical and live data DATA MINING BLOCK Optimal thresholds UPDATE MAINTENANCE PLAN PROCESS For features PREDICTIVE ADVANCED MAINTENANCE PREDICTION ESTIMATION PROGNOSTICS
Types of data analytics
Predictive analytics:RUL prediction Feature of item n crosses boundary in time t+dt Bearing mounted by Contractor 2 Bearing mounted by Contractor 1 Feature of item n crosses boundary in time t RUL considering two features
Types of data analytics
Types of data analytics
The way forward
Where analytics should be performed? 16
Edge agents versus cloud centralized
AI workflow @edge
Huge gap between data science and O&M
What can I see in my data? Forecasting Now casting 3) What will happen 1) What has happened in the future 2) What is happening 4) When will it happen
Domain knowledge and physics sometimes is not in the data
The method, let us twin reality
The twin as a service provider 29
The picture of Dorian Gray
Digital Twin: A virtual instance for services
Digital Twin Solution Architecture
Di Digital gital twin twin bas based ed on O on OT
Digital Di gital twin twin bas based ed on O on OT Internet eMaintenanc e Cloud Server Machine Maintenance On board Wireless System Analytics Data Information Knowledge
Di Digital gital twin twin bas based ed on O on OT
Wha hat a t abou bout IT t IT sys systems? tems?
Tax axono onomies mies and and ontologies ontologies 1 2 Rule-1 FailureMode(?x) ^ hasHappened(?x, true) ^ Device(?y) ^ happenedAt(?x, ?y) ^ FailureMode(?z) ^ theEndFffectIs(?z, ?x) ^ FailureMode(?a) ^ theHighEffectIs(?z, ?a)?theDirectFailureCauseIs(?x, ?a) ^ hasHappened(?a, true) 2 1
TRANSFORMA TRANSFORMATIVE TIVE MAI MAINTE NTENAN ANCE CE SOL SOLUTIO UTIONS NS Inte Integration & A tion & Applica pplication of tion of Tec echno hnologies logies IT OT
Digital Digital twin twin OT/IT T/IT inte integration tion OT IT
Digital twin Digital twin OT/IT inte T/IT integration tion Internet eMaintenanc e Cloud Server Technical services Truck scheduling On board Wireless System Machine Maintenance Analytics Data Information Knowledge
The he Way ay Forw orwar ard All Digital Data Computing Power Growth Context Engines Sensemaking Algorithms Time
Conte Context xt-aw awar are e Mainten Maintenanc ance e Decisi Decision on Suppo Support t Solution Solution Digital twin based on context Information Knowledge Context models models models Data Fusion Big Data Context Maintenanc e & Modelling & sensing & Data Integration Analysis adaptation
Let us b Let us be car e careful big eful bigger = smar ger = smarter? ter? • tolerate errors? • discover the long tail and corner cases? • more data, more error (e.g., semantic heterogeneity) • still need humans to ask right questions, lack of analytics
Bl Blac ack Sw k Swan Lo an Losse sses • Loss Distribution • Tail events are rare – very little data • Typically strong model assumptions
Da Data d ta driv riven en or mode or model based? l based?
Ev Evolution olution of of the the Pr Proce ocess ss Knowledge Capture Design & Validation of products Technological Advance Digital Mockup Digital twin 3D 2D Integration of Product Design and O&M 80s 2000 2016….. 90s
Hybrid Hybrid & C & Conte ontext Dri xt Driven en Se Services vices Physics Hybrid models Data of failure driven based Context Driven Services Context Awareness
Di Digital gital twin twin hyb hybrid rid IT ET OT
Di Digital gital twin twin hyb hybrid rid
Hybrid Digital T Hybrid Digital Twin win OFFLINE PROCESS: VIRTUAL COMMISSIONING Historical Cloud computing Diagnosis records Physics- Synthetic Pre- Feature Hybrid based model data processing extraction engine Maintenance Prognosis planning CONTEXT Maintenance action Physical CM Pre- Feature Failure Trend Risk Risk mitigation asset data processing extraction detection analysis assessment / actuator ONLINE PROCESS: OPERATIONAL LEVEL ANALYSIS 51
Application of railway twins O&M information Data loop Virtual commissioning services Design feedback loop Virtual assets Service/Repair Shop
Some hints
Concluding remarks • Digital twins and Hybrid models are needed for virtual commisioning to deliver O&M services • O&M based on Data driven solutions can lead to catastrophic failures • Life extension is not possible with big data analytics • Manufacturers must provide the integration of systems and data • Digital twin 4.0 will consider evoltionary models and normality dynamics
die diego go.galar@l .galar@ltu.se tu.se die diego go.galar@tecnalia. .galar@tecnalia.com com
Recommend
More recommend
