10:40am (US Central) up 9:40am (US Mountain) 8:40am (US Pacific) - PowerPoint PPT Presentation

Ask a question! Use the chat tool or tweet using #iiotvirtualconf Up next 11:40am (US Eastern) 10:40am (US Central) up 9:40am (US Mountain) 8:40am (US Pacific) Transform your data into strategic business value with predictive analytics Moderator: Lucian Fogoros, IIoT World Speaker: Serg Posadas, Clockwork Solutions 1

2

You’ve never had more data on your strategic assets • historical data on operations, maintenance, and inspections • real-time and sensor data • digital and virtual asset models 3

But data ALONE provides NO strategic VALUE 4

Transforming data into timely insights and relevant actions gathering data + predictive modeling + actionable insights = Strategic Value 5

• How can I improve total effective equipment performance ? • How do I get the most value out of my assets? …So you can answer these • How do I avoid costs when operating and “million maintaining my assets? dollar” • How do I best manage spare parts to keep my assets questions running? • How do I best design my operation to employ my assets? • How do I maintain optimal uptime and asset- generated revenue ? 6

Historical business “intelligence” is looking back at historical data in attempt to react . Rear- facing BI is not accurate What Just Happened? 7

Strategic value Most solution providers What should we do? …but Next gen predictive • analytics ? how do What will happen? What just happened? you get traditional analytics traditional forecasting predictive analytics the right • rear facing • tied to historical data • forward looking • helpful after • doesn’t account for • provides warning and answers? the fact operation changes actionable insight • reactive • can’t anticipate dynamic • supports well-developed environment, aging, etc. strategies • quantifies risk 8

Traditional forecasting trades accuracy for ease of implementation ? Traditional forecasting analytics • Very likely to incur greater costs or • Limited to historical view more down time • Can’t include future events, • Lacks ability to support strategic policy changes, evolution of planning business environment • Does not measure risk 9

Predictive analytics provides a higher degree of accuracy Predictive analytics • Insights well beyond rear facing • Simulates operations hour by analytics hour, including failures, repairs, shipments, part buys, • Historical data only defines refurbishment, retirement, starting point obsolescence, etc. • Models future events for each • Provides a holistic view of asset and its components complex scenarios 10

Predictive analytics is determining industry leaders Within the next five years, advanced implementation of Industry 4.0 will become a ‘qualifier to compete’ and is also likely to be seen by investors as a ‘qualifier for funding’. Industry 4.0: Building the digital enterprise, PwC 2016 Global 11 Industrial Survey, April 2016

Challenges Data • Multiple sensors • High volume & velocity IIoT • Complex distribution of sources predictive Obstacles analytics • Simple data but requires advanced techniques • Combine asset health monitoring with maintenance & operations data • Need automation 12

Asset health monitoring for predictive maintenance analytics Benefits • Leverage advances in predictive health maintenance • Reduced unplanned downtime CBM • Control costs challenge Challenges • Data quality • Data structures • Volume & velocity of real-time and historical data sets • Prediction accuracy • False positives 13

Data aggregation • Automate data aggregation & transformation • Leverage latest techniques for data conditioning • Combine data silos sources for complete, accurate prognostics 14

Machine Learning & CBM 15

Machine Normal Normal Normal Normal Pre-Failure Failure Learning 1. Separate data into training and test sets. & CBM Training 2. Describe the data attributes in the training data set 16

Machine Failure Normal Pre-Failure Normal Pre-Failure Learning 1. Separate data into training and test sets. & CBM 2. Describe the data attributes in the training data set. 3. Apply a predictive technique 4. Evaluate predictor with test data and measure error. Test Predictions 5. If not satisfied, try another predictor. Repeat while minimizing error. 17

Pre-Failure Normal Pre-Failure Failure Normal Normal Machine Learning 1. Separate data into training and test sets. & CBM 2. Describe the data attributes in the training data set. 3. Apply a predictive technique 4. Evaluate predictor with test data and measure error. 5. If not satisfied, try another predictor. Repeat while minimizing error. 6. Select best prediction algorithm. New Predictions Predict based on new values. 18 7. May have to re-train as conditions evolve.

not so fast… sensor 1 sensor 2 CBM Sensor data is noisy & inconclusive reality sensor 3 sensor 4 sensor 5 19

train test Anomaly Detection Rate 65% CBM condition False Positive Rate 10% indicators 20

Goal : Develop accurate prognostic Requirement : Study Condition Indicator (CI) across lifetime of a component. CI Value Sufficient data set Sufficient data set Time series behavior of each CI Complete predictive Operating Hours view CI Value Ideal data set 10 Minor maintenance Operational profile 8 actions Add maintenance and operational events change Location of operations 6 • Maintenance Actions 4 • Operational Profile Changes 2 • Operation Locations 0 0 100 200 300 400 Operating Hours 21

Individual CI’s Receiver Operating Characteristic Receiver Operating Characteristic Receiver Operating Characteristic 1.00 1.00 1.00 0.90 0.90 0.90 0.80 0.80 0.80 0.70 0.70 0.70 0.60 0.60 Sensitivity Sensitivity Sensitivity 0.60 0.50 0.50 0.50 0.40 0.40 0.40 0.30 0.30 0.30 OBE IDA3 MDR 0.20 0.20 0.20 0.10 0.10 0.10 0.00 0.00 0.00 0.00 0.20 0.40 0.60 0.80 1.00 0.00 0.20 0.40 0.60 0.80 1.00 0.00 0.20 0.40 0.60 0.80 1.00 1-Specificity 1-Specificity 1-Specificity Super Receiver Operating Characteristic Receiver Operating Characteristic Receiver Operating Characteristic 1.00 1.00 1.00 0.90 0.90 0.90 COMBINED 0.80 0.80 0.80 0.70 0.70 0.70 SUPER CI 0.60 0.60 Sensitivity 0.60 condition Sensitivity Sensitivity 0.50 0.50 0.50 0.40 0.40 0.40 0.30 0.30 0.30 ODA1 0.20 0.20 OFM0 0.20 IFM0 0.10 0.10 0.10 indicator 0.00 0.00 0.00 0.00 0.20 0.40 0.60 0.80 1.00 0.00 0.20 0.40 0.60 0.80 1.00 0.00 0.20 0.40 0.60 0.80 1.00 1-Specificity 1-Specificity 1-Specificity • Individual CIs often may not have sufficient prognostic power • Leverage Super CI to increase predictive resolution 22

Determine the best fit 23

1.4 CI Value 1.2 Degradation 1 0.8 0.6 0.4 COMBINED SUPER CI 0.2 0 500 400 300 200 100 0 Hours to Removal Determine 200 175 150 125 100 75 50 Hours to Removal LDA 35% 37% 57% 59% 71% 93% 100% the best fit Gaussian Naïve Bayes 39% 42% 46% 53% 68% 89% 100% K Neighbors 65% 57% 57% 58% 68% 82% 100% QDA 37% 39% 43% 51% 65% 84% 100% Linear Support Vectors 38% 38% 42% 47% 60% 77% 80% Non-linear Support Vectors 29% 30% 34% 43% 55% 77% 88% Logistic Regression 37% 37% 39% 43% 55% 70% 56% Stochastic Gradient Descent 21% 15% 24% 34% 35% 59% 4% Ridge Classifier 35% 34% 37% 43% 53% 73% 68% • Evaluate many prognostic algorithms to determine best fit 24

Healthy Anomalous Component Component with >100 hours to causal removal Anomalous COMBINED Anomaly Component SUPER CI detection • Identify distinct probability distributions for Super CI indications • Produce failure lead time 25

• Account for spikes and dips in the data • Manage data quality with cleansing and transforation • Abating Determine the optimal time for maintenance False positives • Analysis select wait time required to distinguish between false positive and true anomaly • Reduces negative impact on maintenance and supply 26

• Provide repair lead time – Reduce wait times for maintenance & parts – Optimize labor • Avoid catastrophic failures • Reduce logistics response time CBM • Control impact of failures on operations results • Extend asset life • Minimize unplanned downtime 27

• Parts with no sensors • Long term strategies • Impact on costs • Inventory Optimization But What • Future Performance metrics About … 28

• Parts with no sensors • Long term strategies • Impact on costs • Inventory Optimization But What • Future Performance metrics About … Asset Life Cycle Management 29

Asset Sustainment Operations Logistics Maintenance Life Cycle Supply Benefits Management • Strategic approach to long-term asset planning (LCM) • Accurately managing future costs and expenses • Maximizing uptime and revenue Uses • Managing components and assets with and without sensors • Accounting for changes in operations, upgrades, … • Anticipating dynamic conditions & evolving environment • Evaluating alternate future scenarios 30