Big-Data Applications and Services Marcel Dix , Dr. Benjamin Klpper, - PowerPoint PPT Presentation

INDUSTRIAL DATA SCIENCE (IDS), 2017-09-05, DORTMUND, GERMANY Holistic Development of Industrial Big-Data Applications and Services Marcel Dix , Dr. Benjamin Klöpper, ABB Corporate Research

Data Analytics Exercise Spot the anomaly 350 300 250 200 150 100 50 0 0 5 10 15 20 25 30 35 40

Data Analytics Exercise Spot the anomaly now 350 300 250 200 150 100 50 0 0 5 10 15 20 25 30 35 40 Slide 3 September 4, 2017

Large Scale Monitoring Applications A problem we share with our customers ... thousands of signals ... hundreds of robots ... hundreds of pumps ... hundreds of inverter DC/DC stages per plant! per factory! per site! per field! Analytics and machine learning make large scale industrial monitoring affordable! Slide 4 September 4, 2017

Potential for Industrial Big Data Analytics Example Figures High Volume, e.g.: – > 300 GB measurements Data p.a. in a single refinery – 400 GB alarm data p.a. in a petro-chemical plant High Velocity, e.g.: – 66.000 sensors with sampling rates between 1s and 60s in a single refinery Big Data High Variety, e.g.: – Time-series, log-data, unstructured data, video-data Low Veracity , e.g.: – time-synchronizing, faulty readings, missing data Development Partners: Slide 5 September 4, 2017

INDUSTRIAL DATA SCIENCE (IDS), 2017-09-05, DORTMUND, GERMANY Introducing ABB The pioneering technology leader

Shaping the world through innovation Pioneering technology since 1883 1900 1920 1940 1960 1980 2000 2020 Founding fathers Steam turbine Turbochargers Electrical drive Gearless motor Variable speed Extended control system for drives drives systems locomotives GIS Ultrahigh voltage Gas turbine Electric propulsion HVDC systems Industrial robot Collaborative robot Slide 7 September 4, 2017

Corporate Research Providing technology for tomorrow‘s innovations ABB Corporate Research Key figures – ~ 130 mn US$ project volume Västerås – in 8 global research areas Ladenburg aligned to ABB core Krakow technologies Raleigh Dättwil – ~ 700 highly qualified scientists Beijing / Shanghai Bangalore and engineers – in 7 corporate research centers around the world Slide 9 September 4, 2017

Corporate Research Center Ladenburg Key figures 70 Inventions/ year Cooperations 30 with universities 8 Significant innovations /year ~100 120 Employees Scientific publications/ year 20 Mio. USD ~80 Students/year Project volume/year Slide 10 September 4, 2017

INDUSTRIAL DATA SCIENCE (IDS), 2017-09-05, DORTMUND, GERMANY Example Analytics Research Project The FEE Project https://www.fee-projekt.de/

Project Overview 12 Objective: Operator Support functions Early Warnings • Ad-hoc Analysis • Decision Support • BASF Heterogeneous Approach: Integrated Analysis of all plant data Mass data Measurements, engineering data, electronic • shift books,… Big Data Analysis Research Topics Algorithm development • Support System Indexing of and search in process data • Integration into real-time plant operation • Big data technologies and architecture • User Centered interaction concepts • Operator

Operator Interface for Anomaly Detection 13

INDUSTRIAL DATA SCIENCE (IDS), 2017-09-05, DORTMUND, GERMANY Analytics Research How is it done?

The Analytics process at ABB From Value Proposition to Continuous Value Delivery Work with Customer to Analysts Investigate Analysts Implement Data Deploy results for Identify Value Proposition Available Data Analytics Techniques continuous application 1 2 3 4 What are Customer’s pains Explore available data Develop analytics models Validate results on actual and gains ? fleet Plan data collection Design based on analytics Can analyzing data help ? question and available data Develop best visualization Collect sample data with end user (service staff, Leverage domain knowledge Variety of approaches customer) Explore data and formulate available From Value Proposition to hypothesis Deploy approach on the ABB specific analytics questions No cookbook for selecting analytics architecture Clean & prepare the data the best approach Develop the right thing Use the right data Understand the methods Make it repeatable Slide 21 September 4, 2017

Value Proposition in Analytics Projects 1 Common Value Proposition 1 Values: Prevent or minimize unexpected downtimes and minimize maintenance costs Maintenance Examples: Root-Cause Analysis, predictive/preventive maintenance, spare-part optimization Asset Values : Improve efficiency of assets or small subsystems (e.g. drive-train, PV installation) 2 Optimization Examples: Pump efficiency, PV output optimization Operation Values: Improve Operator efficiency , effective operation of systems 3 Support Examples: Recommender systems, anomaly detection, event prediction, process troubleshooting Operation Values : Improve efficiency of production processes 4 Optimization Examples: KPI systems/dashboards Slide 22 September 4, 2017

Project Example FEE – Big Data for Operator Support 1 Description of Application scenarios Current State: Desired State: Who: FEE Support: Operator in the control room (and process engineers) ▪ Identify suspicious signals and What: providing relevant data for diagnosis Monitoring of the plant in ‚calm‘ situations ▪ How: Desire: Browsing operator screens and trend display for ▪ (1) Fast visual impression on suspicious signals abnormalities in the process Is only done in ‚calm‘ situation without stress ▪ (2) Put into context to historical ‚normal‘ and ‚abnormal‘ signal paths Problems: (3) Providing extended context (1) Risk to simply overlook a suspicious signal (relevant alarms, operator notes, (2) Monitoring without broad coverage in stressful documents) situations Paper Prototypes (3) Difficult for unexperienced operators to judge the ‚ suspiciousness‘ of signals Development Partners: Slide 23 September 4, 2017

Project Example FEE – Big Data for Operator Support 1 Scenario Canvas for Application Partner Workshops Does a Can it be How early Access to prediction predicted? is it domain help? needed? knowledge Can it be Does detected? How to find detection Is there in historic help? sufficient data? data Is a project worthwhile? Development Partners: Slide 24 September 4, 2017

Project Example FEE – Big Data for Operator Support 2 Domain Specific Data Exploration and Preperation Tools Graphical Search on Plant Topology Full-Text Search across all data Tool for typical data quality tasks Guided process to cleansed Link assets found in notes and One access point to alarms, data set alarms to process signals operator notes, manuals… Identifies low quality and redundant signals Quickly find relevant events Explore process signals directly in the tool Smoothing and handling Understand context of events missing values Development Partners: Slide 25 September 4, 2017

Project Example FEE – Big Data for Operator Support 3 Live time- Database time-series DB series Q Time Series Transformation The distance between a live Sliding data time- Window W series and the most similar Representation subsequence from historical database is Distance/ used to Similarity Evaluation Dist( Q , W ) calculate the anomaly score. Anomaly Score Development Partners: Slide 26 September 4, 2017

Project Example FEE – Big Data for Operator Support 4 Big Data Architecture for Anomaly detection in chemical plants new data Analysierter Zeitraum: 14 Monate (e.g. via OPC) Analysierte Signale: 982 Virtueller Cluster: 4 Physikalische Rechner Apache Kafka 12 virtuelle Knoten, 8 cores per node, 8GB RAM per node Apache Spark Apache Hadoop Streaming HDFS Batch Calcuation Time Nodes 12 min 39 sec 12 14 min 43 sec 9 Apache Spark Core 31 min 35 sec 3 >16 h Ohne Parallelisierung Streaming Time Nodes Apache HBase 15 Sekunden 3 Development Partners: Slide 27 September 4, 2017

INDUSTRIAL DATA SCIENCE (IDS), 2017-09-05, DORTMUND, GERMANY Holistic Development of Industrial Big-Data Applications and Services Summary

Conclusion 4 1 2 3 Develop the right thing Use the right data Understand the methods Make it repeatable Slide 29 September 4, 2017