bringing data science to manufacturing - internet of … · bringing data science to manufacturing...

Post on 12-May-2018

214 Views

Category:

Documents

1 Downloads

Preview:

Click to see full reader

TRANSCRIPT

#IoTMan

Bringing Data Science to ManufacturingPower Tool Maker Increases Visibility & Decreases Complexity with Machine Learning, Data Science & IoT technologies

Coop, Ph.D.Sr. Data ScientistBig Data & Analytics, Digital Accelerator

Stanley Black & Deckerrobert.coop@sbdinc.com | www.stanleyblackanddecker.com

Disclaimer: The views and opinions expressed in this presentation are those of the author and do not necessarily reflect the policy or position of Stanley Black & Decker. Examples of analysis performed within this article are only examples. Assumptions made within the analysis are not reflective of the position of Stanley Black & Decker.

#IoTMan

Background and overview

Who am I, who is Stanley Black & Decker, and what will I be talking about?

March 8th, 2017Coop, Ph.D. 2

#IoTMan

Background

• How do we solve problems we only partially understand?

– Data mining

– Image recognition

– Predicting machine failure

• Given enough information, find an approximate solution– Large amount of data

– Set of “problem/solution” examples

Robert (Bobby) Coop, Ph.D. Computer Engineering with a focus in Machine Learning

March 8th, 2017Coop, Ph.D. 3

Coop, Ph.D.

Sr. Data ScientistBig Data & Analytics, Digital AcceleratorStanley Black & Deckerrobert.coop@sbdinc.comwww.stanleyblackanddecker.com

#IoTMan

Who is Stanley Black & Decker?Consumer and professional tools, storage solutions, and other products.

March 8th, 2017Coop, Ph.D. 4

#IoTMan

Who is Stanley Black & Decker?

• Stanley Security Solutions

• Hospital & Healthcare

• Fastening Solutions

• Infrastructure Products

• Pipeline Services

But wait, there’s more!

March 8th, 2017Coop, Ph.D. 5

• Growing by acquisition presents many challenges

• Different information systems

• Different levels of technological advancement

• Culture shock

#IoTMan

Stanley Engineered Fastening – Global Automotive

Twenty plants over six countries. Approximately 75% of the cars on the road contain SEF parts.

March 8th, 2017Coop, Ph.D. 6

#IoTMan

Stanley Engineered Fastening

• 6 locations

• Billions of fasteners shipped annually

–Over 1,500 products

• Demanding customers, harsh penalties

–Defect rates

–Delivery on time in full

North America Automotive

7March 8th, 2017Coop, Ph.D.

#IoTMan

Overview

•The goal

•The process

•Case study

•Thinking big

•Thinking small

•Bonus material!

It’s all great stuff!

March 8th, 2017Coop, Ph.D. 8

#IoTMan

The goal

What are we trying to accomplish, and how do we start?

#IoTMan

Smart FactoryOptimizing operations

Where do we start?

March 8th, 2017Coop, Ph.D. 10

#IoTMan

What makes a good project?

• The scenario to avoid

– Talking with the business

– Identifying some data and analytics

– Successfully completing a project

– Nothing changes and everything goes back to the way it was

• What went wrong?

– Lack of a standard project intake process leads to wasted effort

• Potential projects must be able to answer standard questions

– What data will be used for this project, who is the subject-matter expert for the data, and how many hours a week

can the SME contribute to this effort?

– How will success be defined, how will failure be defined, and how will we determine when the project is done?

– Most important question:

Distinguishing between “neat” and “valuable”

March 8th, 2017Coop, Ph.D. 11

“If we finish this project and are wildly successful, how will the day-to-day operation of the

business change?”

#IoTMan

The process

Approaching the problem, planning the solution, executing the plan

#IoTMan

Cross Industry Standard Process for

Data Mining (circa 2000)

• Understand business problem

• Discover, explore, understand existing data

• Iterate! ‒ What questions about the data can the business answer?

• Extract, transform, load, clean data

• Produce a model

• Iterate!‒ Are there additional data transformations that can help?

• Evaluate performance of model

• Iterate!‒ Does the model perform the task required by the business?

• Deploy into production

• Iterate!‒ Find the next problem and do it again!

The importance of structure: If there’s no battle plan then how will you win?

March 8th, 2017Coop, Ph.D. 13

#IoTMan

Case study

Optimizing operations within Stanley Engineered Fastening

March 8th, 2017Coop, Ph.D. 14

#IoTMan

• Increase Overall Equipment Effectiveness

• Suspicion – maintenance downtime causing

inefficiencies

Teaching a data scientist how to manufacture

Important Note: This is not real data. All data and performance information is presented for illustrative purposes only and was synthetically created.

March 8th, 2017Coop, Ph.D. 15

#IoTMan

• Examining the data– Exploring database

– Database relationships

Teaching a data scientist how to use a manufacturing execution system

March 8th, 2017Coop, Ph.D. 16

#IoTMan

• Data exploration and preparation tends to take the majority

of the time for any project

Some data is hard to understand. Other data is harder.

March 8th, 2017Coop, Ph.D. 17

#IoTMan

March 8th, 2017Coop, Ph.D. 18

Important Note: This is not real data. All data and performance information is presented for illustrative purposes only and was synthetically created.

Visualizations can help ensure that the business and the

data scientists share an understanding of the data

#IoTMan

March 8th, 2017Coop, Ph.D. 19

Important Note: This is not real data. All data and performance information is presented for illustrative purposes only and was synthetically created.

Visualizations can help ensure that the business and the

data scientists share an understanding of the data

#IoTMan

March 8th, 2017Coop, Ph.D. 20

#IoTMan

•Statistical, advanced analysis – Compare sensor machine logs, down-time

events, part defect rates

– Classical and advanced techniques used

•Problem – why wasn’t there

enough data?

Predictive maintenance

March 8th, 2017Coop, Ph.D. 21

Important Note: This is not real data. All data and performance information is presented for illustrative purposes only and was synthetically created.

#IoTMan

• Data was being overwritten– 1000 log entries per machine (approximately 48 hours)

• Data architects create database which maintains historical state

Recovering lost data

March 8th, 2017Coop, Ph.D. 22

#IoTMan

March 8th, 2017Coop, Ph.D. 23

#IoTMan

Evaluate and ITERATE

• Even with accurate modeling, lift was not ideal

• Scheduling is the better goal

Should we be modeling downtime?

March 8th, 2017Coop, Ph.D. 24

Important Note: This is not real data. All data and performance information is presented for illustrative purposes only and was synthetically created.

#IoTMan

Scheduling, changeover time, next stepsDowntime due to changeovers and scheduling optimization

March 8th, 2017Coop, Ph.D. 25

Important Note: This is not real data. All data and performance information is presented for illustrative purposes only and was synthetically created.

#IoTMan

Scheduling, changeover time, next stepsDowntime due to changeovers and scheduling optimization

March 8th, 2017Coop, Ph.D. 26

Important Note: This is not real data. All data and performance information is presented for illustrative purposes only and was synthetically created.

#IoTMan

Thinking big

Bringing change to organizations

March 8th, 2017Coop, Ph.D. 27

#IoTMan

Leadership and engagement

• Top-level support– Recurring strategic meetings

– Specific assigned roles for research and testing

• Select a flagship location– Test-bed for technology, vendors

– Regular tours from VIP groups- drive progress and reinforce support

• Driving progress within a business unit– Engage

– Educate

– Re-engage

• Driving progress across the enterprise– Data science center of excellence

– Embedding resources in business units

Change can’t happen overnight

March 8th, 2017Coop, Ph.D. 28

#IoTMan

Thinking small

Bringing change to factories

March 8th, 2017Coop, Ph.D. 29

#IoTMan

Change within the factory

• Data collection, operational rigor– How is information collected?

– How is analytics presented?

– How is progress measured?

• Practical audit procedures– Is the system being used as intended?

• Big brother, incentives, and messaging– System usage as basis for reward

– Advanced analytics as a force multiplier

Operational culture

March 8th, 2017Coop, Ph.D. 30

Important Note: This is not real data. All data and performance information is presented for illustrative purposes only and was synthetically created.

#IoTMan

Thanks for your attention!

Questions, comments?

Bonus material!In-depth tutorials for constructing predictive maintenance systems.

Predictive maintenance: https://goo.gl/0XE7w2

Estimating remaining engine life: https://goo.gl/4x5KHG

top related