Big Data And Predictive Health Maintenance

PREDICTIVE ANALYTICS AND DESIGN OPTIMIZATIONFOR EVERY EXPERTISEBig data and predictivehealth maintenanceSergey Morozov, CEONovember, 2016

About DATADVANCE DATADVANCE is an independent software vendor specialized indevelopment of process integration, data analysis and designoptimization software. DATADVANCE has been incorporated in 2010 as a result of acollaborative research program by: Institute for Information Transmission Problems – one of the leading mathematicalcenters with three Fields prize winners on the staff, and Airbus – a global leader in aerospace and defense industry. Using our software Airbus reduced design lead time by up to 10%* Used across all Airbus engineering departments Used in Airbus customer service department 100 active users and 200 engineers trained Disruptive approach to engineering data analysis and optimization led to drasticimprovements, and now Datadvance’s platform enables this for all* Airbus press release2DATADVANCE confidential - do not distribute under any circumstances

Our products and services Two product lines based on the same platform and math core: pSeven DSE – a software platform to build, explore and operate predictivemodels at the product design stage powered by pSeven Core, a softwarelibrary of advanced data analysis and optimization mathematical methods pSeven PHM – a software platform for operational predictive maintenance We integrate elements of artificial intelligence into customers’IT systems Our domains of excellence: Computed Aided Design and Engineering Automation of engineering and manufacturing processes Operational Predictive Maintenance3DATADVANCE confidential - do not distribute under any circumstances

The world is changing: Industry 4.01stMechanization, waterpower, steam power42ndMass production,assembly line, electricity3rdComputer, industrialrobots, automationDATADVANCE Conference 20164thCyber-physical systemsPredictive analytics

Industry 4.0: Cyber-physical systems & Predictive analyticsDataReports/analyticsWhat has happened?Why did it happen?MonitoringWhat is happeningright now?PredictiveanalyticsWhat is going to happenin the future? Products dominated by mechanical components are replaced by smart and connected systems integratingmechanical, electrical, controls. Smart sensors collect data during product manufacturing and service, generating vast amount of data. Internet connection is capable to transfer big amount of data to people, machines or services companies. Machine learning algorithms process data to optimize product behavior, operations and maintenance.5DATADVANCE Conference 2016

Operational Predictive MaintenanceOperational predictive maintenance Online and real-time monitoring of assets/equipment using various embedded sensors Real-time prediction of the condition of assets/equipment using machine learning techniques Determines the operational status of equipment Evaluates present condition of equipment Detects abnormal conditions in a timely manner Maintenance at appropriate or practical time, i.e. if any particular asset requires maintenance Initiates actions to prevent possible forced outagesBenefits: Significant reduction in unplanned machine downtime Minimization of production losses Increase of customer quality perception and satisfaction6DATADVANCE confidential - do not distribute under any circumstances

Predictive maintenance explainedScheduled/PlannedMaintenanceScheduled maintenanceno informationEquipment s and costsornormalMaintenancenormalDATADVANCE Conference 2016Time

Why railways need Operational Predictive Maintenance and Predictive Analytics? Large historical datasets from diagnostic and monitoring systems open a wayto create high-quality predictive models – the main enabler for service-byforecast approach. OPM allows to reduce OPEX of infrastructure and to increase reliability andsecurity at the same time. For example: Reduced locomotive availability due to unexpected breakdowns Rolling stock failures like broken wheels or valve failures in tankers Unfulfilled customer orders and SLA warranties High network congestion and mission failures, like derailments Poorly functioning signals and wayside equipment Unnecessary train stops due to malfunctioning wayside equipment Failures in locomotives, railcars and commuter trains8КОНФИДЕНЦИАЛЬНОOver 400 mln. is lostannually in the U.S.due to asset failureswithin Class I Railroads

General scheme of Operational Predictive Maintenance solutionsOnlineOfflineMaintenance Strategy Implementation(Decision Making / Asset Management)Data analysis and VisualizationData analysis, Construction of models andInterpretationData StorageData collection and TransmissionFlows of data and models9DATADVANCE confidential - do not distribute under any circumstances

pSeven PHM: Intellectual data analysis for operational predictive maintenanceOnlineOfflineMaintenance Strategy Implementation(Decision Making / Asset Management)Data analysis and VisualizationData analysis, Construction of predictivemodels and InterpretationUsersData Storage Data scientists AnalystsData collection and TransmissionFlows of data and models10DATADVANCE confidential - do not distribute under any circumstances

How does it work? Typical data analysis pipeline11DATADVANCE Conference 2016

: Airbus Real Time Health MonitoringВидео-ролик в отдельном файле(3 минуты)Video in a separate file (3 min)Video online (3 min)We developed machine learning and predictive modeling capabilities of AiRTHM(Airbus Real Time Health Monitoring)12DATADVANCE confidential - do not distribute under any circumstances

: Prediction of failures of aircraft auxiliary power unitGoal Predict failures of APU to improve maintenance procedureData for model training 30 aircrafts and about 200 parameters per aircraft Learning data set: 3 years ( 400 flights during an year) Model testing: 0.5 year in operationBenefits of predictive maintenance Early warning about some types of failures Detection of failures with an accuracy of about 90% (9 correctlypredicted failures account for 1 false alarm) Cost reduction associated with downtime of plane due tounexpected failures by 30%13DATADVANCE Conference 2016

Benefits of predictive maintenance for airlinesPredictive maintenance (aka condition-based maintenance) technologies: perform maintenance at an appropriate time, and before the equipment loses optimum performance or fail reduce disruptions to facility operations and increase equipment availability monitor the condition of in-service equipment.Benefits: “Predictive maintenance can increase aircraft availability by up to 35%”, – Luiz Hamilton Lima, vice president ofservices and support at Embraer Adopting predictive maintenance through the use of data analysis can reduce maintenance budgets by 30-40%,reports claim.Sizing the benefits: 10 000/hour – cost of keeping a commercial passenger jet grounded*. 95 000 hours – Delta Airline delay from July, 2015 to July, 2016. 30 mln./year – potential savings if just 10% of the delay hours is because of maintenance* According to the aerospace arm of SAP, the software group14DATADVANCE Conference 2016

: Analysis of accident of power plant gas turbine Information about accident Location: Some power plant in Russia Accident data: May 2016 Estimated gas turbine maintenance cost: 10 mln. Euro Turbine has 100 sensors Data from sensors was stored but it was not analyzed online Would it be possible to predict accident and stop turbine beforeaccident in case of online monitoring and diagnostics?15DATADVANCE Conference 2016

: Analysis of accident of power plant gas turbineAccidentThe turbine can be stopped 2 weeksprior to the accident!Potential cost savings: 10 mln. Euro!Systematic deviation from thenormal stateAccident early precursor16No domain specific knowledgewas used.DATADVANCE Conference2016 Just pure data analysis!

: Railway Monitoring System: Incident RankingProblem Railway monitoring system automatically logs a large amount of complexalerts (incidents). Incidents should be handled manually by operators. Moscow Railway: 5000 incidents a day, 24x7, 4 hours to fix Vast majority of incidents ( 97%) are not related to real failures, and occurdue to unplanned maintenance and flaws of diagnostic procedures. As a result, operators spend much of their time on non-critical incidents anddo not have time to handle all incidents, missing the real system failures. Moscow Railway: Wrong ranking is the reason of 54% of “missed” incidentsProject scope: Automatic incident ranking by importance with machine learning on realhistorical data (5.5 bln alarms, 4.5 years) Predict probability of failure Root cause analysis for major accidents recommendations on the of diagnostic tools coverageResult - Significant increase in situation center efficiency: 2x times increase in reaction time 5x times load dropProject run by Telum, our partner company in railway industry.17КОНФИДЕНЦИАЛЬНО

Conclusions Industry 4.0 is already here, with its challenges and opportunities! Operational predictive maintenance enabled by big data and advances in machine learning allows to Significantly reduce unplanned asset/machine/infrastructure downtime Reduce OPEX of infrastructure Increase reliability and security at the same time Increase of customer quality perception and satisfaction DATADVANCE and Telum, our partner company in railway industry, are your reliable partners toimplement efficient predictive analytics solution for your railway company.18DATADVANCE Conference 2016

pSeven PHM: Key features of the platform Build and explore predictive models Data import, cleaning and pre-processing Feature selection and extraction Advanced data analysis mathematical methods– Efficient in-house methods for anomaly detection and failure prediction for multidimensional data– Clustering on graphs for automatic extraction of system components, including in-house approaches– Modern methods of robust classification, including imbalanced classification SmartSelection technology to select the mode efficient data analysis method Post-processing and visualization Deploy predictive models as services Package and export predictive models Publish and deploy to pSeven model server (Model as a Service) Integrate predictive models into existing IIoT ecosystem Powerful cloud software platform with reach data pre- and postprocessing capabilities20DATADVANCE confidential - do not distribute under any circumstances

“Predictive maintenance can increase aircraft availability by up to 35%”, –Luiz Hamilton Lima, vice president of services and support at Embraer Adopting predictive maintenance through the use of data analysis can reduce maintenance budgets by 30-40%, reports claim. Sizing the