You are on page 1of 30

PowerBI-PI-Azure ML

Analytics made simple

Jason McKittrick

Jo Dara

Bill Barna

Curt Hertler

jasonmck@microsoft.com

billb@nealanalytics.com
4/14/2015

jdara@osisoft.com

curt@osisoft.com

The world of data is changing

Analyzing data with Excel


Discover
Analyze
Visualize
Explore

Analyzing data with Excel


Discover
Analyze
Visualize
Explore

Analyzing data with Excel


Discover
Analyze
Visualize
Explore

Analyzing data with Excel


Discover
Analyze
Visualize
Explore

Share & collaborate with Power BI


BI Sites /
Dashboards
Shared Queries
Data Refresh
Data Search
Mobile Access
Natural Language
Questions

Share & collaborate with Power BI


BI Sites
Shared Queries
Data Refresh
Data Search
Mobile Access
Natural Language
Questions

Share & collaborate with Power BI


BI Sites
Shared Queries
Workbook

Data Refresh
Data Search
Mobile Access
Natural Language
Questions

Power BI
Cloud
On-Premise
On-Premise
DB

Data Management Gateway


Installed On-Premises

Share & collaborate with Power BI


BI Sites
Shared Queries
Data Refresh
Data Search
Mobile Access
Natural Language
Questions

Search for:
Public Data
Corporate Data
Shared Queries

Share & collaborate with Power BI


BI Sites
Shared Queries
Data Refresh
Data Search
Mobile Access
Natural Language
Questions

Share & collaborate with Power BI


BI Sites
Shared Queries
Data Refresh
Data Search
Mobile Access
Natural Language
Questions

Microsoft BI and the OSIsoft PI System


Evolution
Context
Visualize
vNext

Microsoft BI and the OSIsoft PI System


2015

Evolution
Context
Visualize
vNext

Azure ML

Azure Data Factory

Power BI Designer Preview


PI System 2015 -

PI Integrator for Azure

Operational Context to Real Time Data


Evolution
Context

Asset
Framework
Hierarchy

Visualize

Plant
Process
KPI

vNext
PowerView Report

KPI Total Template

KPI Laboratory Template

PI System Integration with Excel


Evolution

Excel
Worksheet

Power View
Excel 2013

Context
Visualize

Power Map
Excel 2013
Add-In

Power Query User Configurable

Data Model
(PowerPivot)

vNext
PI ODBC
2014

PI System

PI OLEDB
Enterprise

PI System Integration w/Excel & AzureML


Evolution
Context
Visualize

Excel
Worksheet

Power View
Excel 2013

Power
Query

Azure ML
Data Model
(PowerPivot)

vNext
PI Integrator
Azure SQL
Database

PI System

PI System 2015 Integration with AzureML


Evolution
Context

Azure
Data Factory

Azure SQL
Database

Visualize
vNext

Power BI Designer

Production

PI Integrator for Azure

Drilling

PI System 2015

IT Network

Azure ML
Studio

AzureML and PI System 2015


Self-Service
Data Science

Analyze datasets through


experimentation to find
best fit prediction models

Provides access to finalized


models as web service

Powerful new data imaging and predictive analytics solutions are making it possible for companies such as Shell BP PLC and
Chevron Corp. to map and exploit previously uncharted oil and gas fields. Wall Street Journal online, January 2, 2013

About Neal Analytics

Service Offerings

History

Founded in 2011
Offices in Seattle and Mumbai, India
OSIsoft & Microsoft Gold Partner

Capabilities

25 data scientists on staff


Predictive Analytics, data visualization and data engineering
Industry expertise in retail and energy

Technologies

Microsoft Azure, Azure ML, SQL Server and PowerBI


R, SAS, SPSS, and Weka
Sawtooth Technologies (Conjoint Analysis)

Approach

Demonstrate capabilities with Proof-of-Concept (POC)


Demonstrate value with Proof-of-Profit (POP) methodology

www.nealanalytics.com | Practice Director: Bill Barna (billb@nealanalytics.com)

Forecasting

Leverage time series data to improve resource, demand and price forecasting.

Visualization

Use data visualization technologies to improve comprehension by engineering


teams.

Predictive Maintenance

Optimize maintenance schedules by extending SCADA systems with predictive


analytics.

Optimization

Improve asset utilization with constraint-based modelling.

Decision Makers
are drowning in
data

Managers are challenged to see through all of the rapidly growing


volumes of data in order to understand whats really happening
Opportunities are missed or never even realized
Too many vendors/tools/platforms to choose from
Economic considerations grow profits or cut costs

Machine Learning
and Predictive
Analytics

Demand forecasting to optimize supply


Resource forecasting to effectively meet demand
Price forecasting to optimize investments
Proactively manage risks based on probabilities
Schedule maintenance to minimize operational impact

Competitive Advantage

Machine
Learning

Optimization

Whats the best that can happen?

Predictive Modelling

What will happen next?

Forecasting

What if these trends continue?

Statistical Analysis

Why is this happening?

Alerts

What action is needed?

Query & Drill-Down

Where exactly is the problem?

Ad Hoc Reports

How many? How often? Where?

Standard Reports

What happened?

Degree of Intelligence

Analytics

Access &
Reporting

Multiple Algorithm Evaluation

Tank Forecast Development

Tank Level

55
45

Linear Regression

35

Random Forest

25

Neural Network

15

Level

5
1

11

16

21

26

31

36

41

Day

46

Parallel model development in Azure ML allows for rapid


solution testing and development
Advanced machine learning algorithms (Neural Network
Regression, Poisson Regression, Decision Forest Regression)
allow leverage of state of the art in forecasting technology
Model cross validation will ensure model generalizes well to
new tanks

Parallel Feature Testing


Forecast Horizons
Forecast

Accuracy

Impact

1 hour

High

Route

1 day

High

Schedule

1 week

Medium

Schedule

Event Classification Model Training


Sensor Variables
Revolutions per minute

Equipment Failure
Tool Failure Risk
Drill bit

Weight on bit
Slow pump pressure
Drill bit selection

Drill string

Etc.

Motors/
pumps

Event Classification predictor based on historical drilling


information
Anomaly detection algorithm identifies unusual behaviour
in components for further investigation
Early warning system for component failure allows for
better management of replacement parts and predictive
maintenance

Anomaly detection model


Sensor Signals

Automatic detection
Spikes
Dips,
Level changes
Trends
Extreme values

Component Failure Prediction

Anomaly

Classification

Anomaly Detection

Strength

Known events

Unknown Events

Normal

Approach

Supervised learning

Unsupervised

Anomaly

output

Foam Events

Foam Precursors

Time

Support Vector Machine

Drilling Optimization

Drilling Rate

Advanced machine learning algorithms enable


tailored models to optimize drilling conditions for
various soil conditions

Drilling Rate

1.5
1

Soil 1
Soil 2

0.5

Clustering analysis can identify best performing


drilling parameters based on soil and rock conditions

Soil 3

0
C
C
C
C

Reduce drilling time and costs by decreasing time


to drill

Torque

*Size proportional to incremental Rate of Penetration (ROP)

Drilling Speed Predictive Strength

Drilling Optimization
Stratified Rock Formations

Rotation Speed

Control Variables
Drill bit

Weight on bit
Torque

Torque

Drill Bit
Soil Type
0%

20%

40%

60%

80%

100%

Weight on
bit

Rotation
speed

Field Resources

Analyst

Optimization Engine

Azure ML

Python

Trucks Maintenance

PI AF and notification

PI Server and MDB

Sensors
Site Operation
27272727

PI ACE

Data Center

Accepted by Azure ML

Competitive Model
Existing forecasting model
Neal Analytics model
The actual result

Methodology

Choose subset of the data for the experiment


Use external variables as explanatory variables
Train models with 2009 to 2013 data
Evaluate 2014 data with legacy system
Document 2014 outcome
Determine lift from new methodology

Production Forecast Error


Legacy
System

New
System

Lift

POC performance is extrapolated to


indicate solution performance.
Once the improvement in
forecasting is estimated, an ROI
business case for the final
solution is created.
Time: ~ 6 to 8 weeks

jasonmck@microsoft.com

jdara@osisoft.com
curt @osisoft.com

BillB@nealanalytics.com