CS – Communication & Systèmes / 1CONCEPTEUR, OPÉRATEUR & INTÉGRATEUR DE SYSTÈMES CRITIQUES www.c-s.fr
MACHINE LEARNING& IKATS PLATFORM
& PREDICTIVE MAINTENANCE
December 2018, Craïova
CS – Communication & Systèmes / 2/ 2
MACHINE LEARNING : ASSESSMENT
Algorithms no major “scientific” obstacle
› Predictive “data driven” modelling
› Strong theoretical foundation
Supervised/ unsupervised training, …
› Many “use cases”
Evaluation toward reproducibility and validation
› Articles with “words” more than “figures”
› Evaluation criteria not enough specified nor standardized;
criteria definition is challenging
CS – Communication & Systèmes / 3/ 3
INDUSTRIAL MACHINE LEARNING
Dealing with time series data is a key requirement of industrial
machine learning that distinguishes it from consumer applications
Specific criteria are required to make industrial machine learning
applicable
(Industrial) Machine learning has to be deployed :
› In integrated software systems /solutions (as compared to hardware
equipment)
› With on-going or recurrent upgrades
From an industrial and operational point of view, it’s a disruption
CS – Communication & Systèmes / 4/ 4
IKATS PLATFORM
IKATS Platform
Machine learning
AlgorithmsTime Series
Exploration
Visualization
Advanced Analysis
Industries Academia
Interactive Toolkit for Analysis of Time Series
CS – Communication & Systèmes / 5/ 5
IKATS : TECHNICAL OUTLOOK
Infrastructure: storage, network
Data Organisation and distribution
Computing and analytics
Re
ss
ou
rce
man
ag
em
en
t
User Workbench& Data Viz
Ku
be
rne
tes
CS – Communication & Systèmes / 6/ 6
DEALING WITH TIME SERIES…
08
Dedicated Time Series
Scalable for Big Data
Cloud compatible
User Friendly
Product “3 in 1”
Evolutivity
Data Viz & Analysis
Open Source
CS – Communication & Systèmes / 7/ 7
SCIENTIFIC INTERESTS
Evaluation / comparison of algorithms
“Reproducibility” / Reuse
› Archive / record of experiments
› Meta-parameters tuning
Leverage “multidisciplinary approach”, between
› Applied mathematics and machine learning
› Information Technologies and big data
› Each respective discipline:
Engineering,
Physics
Medicine, e.g. physiology,
Biotech,
Sociology
…
CS – Communication & Systèmes / 8/ 8
STATUS
IKATS
› Is developed and leaded by CS, within an overall project associating LIG
(Laboratoire Informatique de Grenoble)
› https://ikats.org
› https://github.com/IKATS/IKATS
CS – Communication & Systèmes / 9/ 9
AN INDUSTRIAL OBJECTIVE : PREDICTIVE MAINTENANCE
CS – Communication & Systèmes / 10/ 10
PREDICTIVE MAINTENANCE CHAIN
Acquisition Pre-Processing
DataLakeArchiving
Post ProcessingAdanced AnalysisMachine learning
Expertises Data Scientist
& Specialists
Other Data & Information Systems
Continuous AnalysisAlarm
ReportsDecision support tools
Sensors
ModèlesPrédictifs
Inventory Management,
…
Maintenance Realization
MaintenancePreparation
Time Series
IKATS
CS – Communication & Systèmes / 11/ 11
LESSONS LEARNED
From an industrial and operational perspective,
› a first tough challenge is in data collection and analysis. It’s a prerequisite
› “Transfer learning” might be an approach to be favoured,
CS – Communication & Systèmes / 12/ 12
THANK YOU
Top Related