from digital twin to predictive maintenance...in pycon.de & pydata berlin, 07.-09.10.2019, 2019....
TRANSCRIPT
From Digital Twin toPredictive Maintenance
AI Monday Leipzig
Andreas HantschLeipzig, 25th November 2019
CLOUD&HEAT Technologies GmbH. We build the most energy e�cient data centers. Worldwide. https://cloudandheat.com 1
Challenge
Fig.: Jones (2018) [1], Data: Andrae & Edler (2015) [2]CLOUD&HEAT Technologies GmbH. We build the most energy e�cient data centers. Worldwide. https://cloudandheat.com 2
Liquid-cooled hardware: Micro data centre (MDC)
• on the market since 2014• 20 own projects with approx. 150 MDCheatingI 16 single-family houseI 3 multi-family house (>200 flats)I 1 kindergarten
• Customer projects, such as Innogy SE
CLOUD&HEAT Technologies GmbH. We build the most energy e�cient data centers. Worldwide. https://cloudandheat.com 3
Liquid-cooled hardware: Data centre container (DCC)
CLOUD&HEAT Technologies GmbH. We build the most energy e�cient data centers. Worldwide. https://cloudandheat.com 4
Distributed cloud operation
CLOUD&HEAT Technologies GmbH. We build the most energy e�cient data centers. Worldwide. https://cloudandheat.com 5
Concept
However
• Much more complex system• Water in the data centre• Few data at the beginning of operation
Solution, based on earlier ideas of, e.g., [3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13]• Predictive maintenance system• Both model-based and data-based approaches for normal system’sbehaviour
• Machine learning for anomaly detection
CLOUD&HEAT Technologies GmbH. We build the most energy e�cient data centers. Worldwide. https://cloudandheat.com 6
Concept
However
• Much more complex system• Water in the data centre• Few data at the beginning of operation
Solution, based on earlier ideas of, e.g., [3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13]• Predictive maintenance system• Both model-based and data-based approaches for normal system’sbehaviour
• Machine learning for anomaly detection
CLOUD&HEAT Technologies GmbH. We build the most energy e�cient data centers. Worldwide. https://cloudandheat.com 6
Concept Predictive Maintenance Unit
Va
lid
ati
on
Start-up routine
Retrieving data via API
Pre-processing
Compare data with standard
Digital Twin
model-based
Machine Learning
data-based
Determine differences
Anomaly detection
algorithm
Post-processing
Shut-down routine
Tra
inin
g
Tim
e l
oo
p
CLOUD&HEAT Technologies GmbH. We build the most energy e�cient data centers. Worldwide. https://cloudandheat.com 7
Concept: First prototype – Thermohydraulic system
Demo in jupyter notebooks
CLOUD&HEAT Technologies GmbH. We build the most energy e�cient data centers. Worldwide. https://cloudandheat.com 8
Summary & Outlook
Summary
• Advantages due to combination of model and data approach• Digital twin already there (employed as FMU)• Automatic data analysis• Machine learning for subsequent model adaption
Outlook
• Improvement of anomaly detection• Set-up of online learning
CLOUD&HEAT Technologies GmbH. We build the most energy e�cient data centers. Worldwide. https://cloudandheat.com 9
Summary & Outlook
Summary
• Advantages due to combination of model and data approach• Digital twin already there (employed as FMU)• Automatic data analysis• Machine learning for subsequent model adaption
Outlook
• Improvement of anomaly detection• Set-up of online learning
CLOUD&HEAT Technologies GmbH. We build the most energy e�cient data centers. Worldwide. https://cloudandheat.com 9
Contact
Andreas HantschCLOUD&HEAT Technologies GmbH
Königsbrücker Straße 9601099 Dresden
https://cloudandheat.com
CLOUD&HEAT Technologies GmbH. We build the most energy e�cient data centers. Worldwide. https://cloudandheat.com 10
References I
[1] Nicolas Jones.How to stop data centres from gobbling up the world’s electricity.Nature, 561:163–166, 2018.
[2] Anders S. G. Andrae and Tomas Edler.On global electricity usage of communication technology: Trends to2030.Challenges, 6(1):117–157, 2015.
[3] R.A. Adey and D. Sriram.Applications of artificial intelligence to engineering problems.(OSTI ID: 5113597), 1 1987.
CLOUD&HEAT Technologies GmbH. We build the most energy e�cient data centers. Worldwide. https://cloudandheat.com 11
References II
[4] Ying Guo, Davood Dehestani, Jiaming Li, Josh Wall, Sam West, andSteven Su.Intelligent outlier detection for hvac system fault detection.Technical report, CSIRO ICT Centre, Sydney, Australia, 2012.
[5] Muhammad Waseem Ahmad, Monjur Mourshed, Baris Yuce, andYacine Rezgui.Computational intelligence techniques for HVAC systems: A review.Building Simulation, 9(4):359–398, mar 2016.
[6] Michael Wetter.A modelica-based model library for building energy and controlsystems.In Eleventh International IBPSA Conference, Glasgow, Scotland, July27-30, 2009, pages 652–659, 2009.
CLOUD&HEAT Technologies GmbH. We build the most energy e�cient data centers. Worldwide. https://cloudandheat.com 12
References III
[7] Jens Bastian, Christoph Clauß, Susann Wolf, and Peter Schneider.Master for co-simulation using fmi.In Modelica Conference, 2011.
[8] Felix Bünning, Roozbeh Sangi, and Dirk Müller.A modelica library for the agent-based control of building energysystems.Applied Energy, 193:52–59, may 2017.
[9] Andreas Hantsch.Vom Messwert zur Erkenntnis – Datenanalyse und künstlicheIntelligenz.In Treffpunkt Zukunft: Sensorik, Handwerkskammer Dresden, 2019.
CLOUD&HEAT Technologies GmbH. We build the most energy e�cient data centers. Worldwide. https://cloudandheat.com 13
References IV
[10] Andreas Hantsch and Sabine Döge.Assessment of micro-organism growth risk on filters with machinelearning.In Clima 2019 – 13th REHVA World Congress, Bucharest, Romania,26.–29. May 2019, 2019.
[11] Andreas Hantsch.Machine learning with little data - from digital twin to predictivemaintenance.In PyCon.DE & PyData Berlin, 07.-09.10.2019, 2019.
[12] Andreas C. Müller and Sarah Guido.Introduction to Machine Learning with Python: A Guide for DataScientists.O’Reilly Media Inc., Sebastopol, 2016.
CLOUD&HEAT Technologies GmbH. We build the most energy e�cient data centers. Worldwide. https://cloudandheat.com 14
References V
[13] Aurélien Géron.Hands-On Machine Learning with Scikit-Learn and TensorFlow:Concepts, Tools, and Techniques to build intelligent systems.O’Reilly Media Inc., Sebastopol, 2017.
CLOUD&HEAT Technologies GmbH. We build the most energy e�cient data centers. Worldwide. https://cloudandheat.com 15