Industrial Data Science - Problems, Data sets & Techniques

This was a guest lecture given for the "Predictive Analytics: Applications of Machine Learning" course in UC Santa Cruz, Silicon Valley ext.

We look at the problems and solutions techniques that make up the Industrial Data Science landscape, for the entire Asset lifecycle. From its design (birth), building, operating, monitoring, maintaining and eventual decommissioning.

Ram Narasimhan

Transcript

1. Industrial Data Science Problems, Datasets, Algorithms Ram Narasimhan
4. Asset Lifecycle Decommission Modify/ Upgrade M aintain Plan/Design/ Build Operate/ Monitor
5. What sort of data would be available from these assets?
6. Design Data BOM OEM Specifications Operations Quality Alarms, Failures Maintenance Records Diagrams Images & Videos Tabular Time Series Data
7. Asset Perf Cost/ Revenue Analysis Performance Mgmt Lifing Studies Ops Mgmt Process Optimization Resource Optimization Anomaly Detection Event Detection Remaining Useful Life Failure Post-Mortems Root Cause Analysis Alarm Analysis Field Investigations False Pos Reduction Degradation Index Demand Forecasting Pricing Maintenance Optimization Preventative Maintenance Survival Analysis
8. Asset Perf Ops Mgmt
9. Operations Management • Managing all aspects of an assets operations • Maximize efficiency • Maximize utilization, minimize all fleet and asset-related expenses Staff Materials Equipement Technology
10. Asset Perf Cost/ Revenue Analysis Ops Mgmt
11. Cost & Revenue Analysis • Estimating the strengths and weaknesses of alternatives • Determine options that provide the best approach to achieve benefits while preserving savings. CBA has two main applications: 1. To determine if an investment or decision is sound – verifying whether its benefits outweigh the costs, and by how much 2. To provide a basis for comparing investments or decisions – comparing the total expected cost of each option against their total expected benefits
12. Asset Perf Cost/ Revenue Analysis Performance Mgmt Ops Mgmt Demand Forecasting Pricing
13. Performance Management The technique of getting the most out of each asset, over its lifetime • Manage the performance of a machine/asset/plant over its entire lifecycle • Involves monitoring the performance of an asset (onsite, or remotely)
14. Asset Perf Cost/ Revenue Analysis Performance Mgmt Ops Mgmt Resource Optimization Demand Forecasting Pricing
15. Resource Optimization • A set of processes and methods to match the available resources (human, machinery, financial) with the needs of the organization • Allowing the right amount of 'slack' (robustness vs. fragility) when optimizing • Resource/asset utilization leveling
16. Asset Perf Cost/ Revenue Analysis Performance Mgmt Ops Mgmt Process Optimization Resource Optimization Demand Forecasting Pricing
17. Process Optimization Challenge: How to produce more while spending less? Visibility & Coordination Streamlining Workflows Forecasting Changes Eliminating redundancies
19. Asset Perf Cost/ Revenue Analysis Performance Mgmt Ops Mgmt Process Optimization Resource Optimization Demand Forecasting Pricing Maintenance Optimization Preventative Maintenance
20. 5 Analytics Based M aintenance 4 Schedule Optim ization 3 Fleet-level M aintenance 1 Reduce M aintenance Costs 2 Repair vs Replace vs Renew 2 Maintenance Optimization
21. Preventative Optimization Policy - How Frequently? Work-scope - Bundling Tradeoff - Between downtime, Maint. Costs & Breakdown costs
22. Asset Perf Cost/ Revenue Analysis Performance Mgmt Ops Mgmt Process Optimization Resource Optimization Anomaly Detection Demand Forecasting Pricing Maintenance Optimization Preventative Maintenance
23. Anomaly Detection Anomaly detection is the identification of rare items, events or observations which raise suspicions by differing significantly from the majority of the data Techniques
25. Asset Perf Cost/ Revenue Analysis Performance Mgmt Lifing Studies Ops Mgmt Process Optimization Resource Optimization Anomaly Detection Event Detection Demand Forecasting Pricing Maintenance Optimization Preventative Maintenance
26. Lifing Studies The process of testing or modelling of durability of components Lifing: the context of lifespan of a component and turning it into a verb • Impact of fatigue on a component’s lifespan • the act of improving fatigue life or the life of the material in general
28. Asset Perf Cost/ Revenue Analysis Performance Mgmt Lifing Studies Ops Mgmt Process Optimization Resource Optimization Anomaly Detection Event Detection Remaining Useful Life Demand Forecasting Pricing Maintenance Optimization Preventative Maintenance
29. RUL Remaining Useful Life Estimate of the remaining time (days|years) that an item, component, or system is estimated to be able to function before warranting replacement.
30. What Information/Data could we use to build a Remaining Useful Life Model?
31. We could look at: Asset age, fleet lifetime Asset usage Service & Maintenance Records Inspection Data from the equipment
32. Asset Perf Cost/ Revenue Analysis Performance Mgmt Lifing Studies Ops Mgmt Process Optimization Resource Optimization Anomaly Detection Event Detection Remaining Useful Life Degradation Index Demand Forecasting Pricing Maintenance Optimization Preventative Maintenance Survival Analysis
33. What is Survival Analysis?
34. What is Survival Analysis?
35. A non-parametric statistic is not based on the assumption of an underlying probability distribution Parametric statistics assumes that sample data comes from a population that can be adequately modelled by a probability distribution that has a fixed set of parameters. Most well-known statistical methods are parametric. Parametric Non-Parametric
36. This statistic gives the probability that an item will survive past a particular time t. At t = 0, the Kaplan-Meier estimator is 1 As t goes to infinity, the estimator goes to 0 Parametric methods assume that the underlying distribution of the survival times follows certain known probability distributions. Popular ones include the exponential, Weibull, and lognormal distributions Parametric Non-Parametric
37. Insurance 05 • Time until claim Manufacturing 04 • Time to Failure Retail 03 • Detergent runs out Online Retail 02 • Next Phone Geological 01 • Next earthquake
38. What is a Degradation Index?
39. What is a Degradation Index? (2)
40. Asset Perf Cost/ Revenue Analysis Performance Mgmt Lifing Studies Ops Mgmt Process Optimization Resource Optimization Anomaly Detection Event Detection Remaining Useful Life Failure Post-Mortems Degradation Index Demand Forecasting Pricing Maintenance Optimization Preventative Maintenance Survival Analysis
41. Failure Post-Mortems Failure has already occurred • Why • One-off? • Recall • New Maintenance program? • Policy changes
42. Asset Perf Cost/ Revenue Analysis Performance Mgmt Lifing Studies Ops Mgmt Process Optimization Resource Optimization Anomaly Detection Event Detection Remaining Useful Life Failure Post-Mortems Root Cause Analysis Alarm Analysis Field Investigations False Pos Reduction Degradation Index Demand Forecasting Pricing Maintenance Optimization Preventative Maintenance Survival Analysis
43. One lesson to take away...
44. Data Science + ML through the entire Asset Life cycle Plan/ Build Operate Upgrade/ Monitor Preventative Mainteance MTBF Maintenance Optimization Root-cause Maintain Demand Forecasting Cost Analysis Pricing Resource Optimization Performance Management Utilization Efficiency Alerts/Alarms CBM Anomaly Detection RUL
45. Thank you! @ramnarasimhan
46. Industrial Data Science Problems, Datasets, Algorithms Ram Narasimhan
48. Asset Perf Cost/ Revenue Analysis Performance Mgmt Lifing Studies Ops Mgmt Process Optimization Resource Optimization Anomaly Detection Event Detection Remaining Useful Life Failure Post-Mortems Root Cause Analysis Alarm Analysis Field Investigations False Pos Reduction Degradation Index Demand Forecasting Pricing Maintenance Optimization Preventative Maintenance Survival Analysis