You are on page 1of 28

Demystifying Machine

Learning for
Manufacturing:
Data Science for All
Jeff Kavanaugh
June 7, 2018
1 | Internet
1Infosys of Manufacturing Midwest 2018
Confidential
Today’s discussion

Research Machine Learning and IIoT and AI in Practice


Analytics Background
Industry 4.0 maturity index and Industrial IoT through facilities
framework energy management

Future proof: learning and Water Treatment Plant


communication through data
science and critical thinking Automotive OEM predictive
churn

2 | Internet
2Infosys of Manufacturing Midwest 2018
Confidential
www.infosys.design/plantio

3 | Internet
3Infosys of Manufacturing Midwest 2018
Confidential
Industry 4.0: Beyond the POC… the time to scale is upon us
Industry 4.0 integrates the physical and virtual worlds through technology enablers, which brings the fungibility and
speed of software to manufacturing operations. The potential value created by Industry 4.0 vastly exceeds the low-
single-digit cost savings that many manufacturers pursue today (acatech, Infosys, BCG, McKinsey, et al).

Disruptive technology enablers for Industry 4.0 are at a tipping point Industry 4.0 is changing manufacturing

25B
connected things forecasted to
250M
connected vehicles are forecasted
$493B
in digital revenue gains per
ship by 2020.* to have some form of wireless annum
network connection by 2020.*

$421B $907B 100x


in cost and efficiency gains in annual digital investments disruptive digital innovation is
per annum 100x faster than physical
disruption*

But are we ready?....


*McKinsey, acatech, Infosys, BCG research
4 | Internet
4Infosys of Manufacturing Midwest 2018
Confidential
Industry 4.0: Global study conducted on operations efficiency as a
driver for competitiveness
Dimensions Directional findings The opportunity

Engineering 18% • Industry 4.0 announced at Hannover


• Vast majority (82%) of Messe 2011, but systematic
Efficiency
companies are aware of 46% implementation still only 18%
Performance the high potential in
• Current speed of implementation
implementing Industry 4.0
OPERATIONS EFFICIENCY

Efficiency 35%
places the 2022 goal of 46% at risk
concepts
Maintenance • Reason: data hurdles and piecemeal
Efficiency
• 46% want to implement POC approach – unclear path
Industry 4.0 solutions 32%
Approach to overcome barriers:
Information systematically for enhanced 32%
Efficiency asset efficiency by 2022 1. Evaluate your digital maturity
2. Proof of concepts to demonstrate
• Only 30% have implemented 18%
Energy 15% business value, then scaled action
Efficiency
data-driven or intelligent
4%
services 3. Set clear targets
2017 2022
Service 4. Prioritize measures that will bring
Systematically implemented Potential recognized the most value to business
Efficiency
Partly implemented No awareness
5. Demonstrate courage, persistence

Source: Infosys and Institute for Industrial Management (FIR) at RWTH Aachen study conducted in 2015 and updated in 2017.
Sample size: 433 executives across industrial manufacturing sectors from China, France, Germany, UK and USA
5 | Internet
5Infosys of Manufacturing Midwest 2018
Confidential
Humans still matter! Industry 4.0 maturity is about more than the
technology, and poor reasoning skills are constraining progress

A survey by PayScale Inc., an online pay and


Their No. 1 complaint?
benefits researcher, showed 50% of employers
complain that college graduates they hire aren’t
ready for the workplace. Poor critical-reasoning skills
Source: UTD research study December 2017 and PayScale Inc., 2016 | Internet of Manufacturing Midwest 2018
6 6Infosys Confidential
Industry 4.0 maturity drives significant efficiency improvement,
and analytics is a fundamental requirement
Advanced Long-Term
analytics for Automated
simulation & decision
identification making and
Near-Term of most likely actions
scenarios
Big Data
analytics to
Up to date understand
digital model root causes
(Digital
Shadow ) to
show what’s
happening
now
Business
applications
connected to
each other
Computerization
E.g. CNC milling
machine but not
connected

Source: Industrie 4.0 Maturity Index, acatech study supported by Infosys, 2017 7 | Internet
7Infosys of Manufacturing Midwest 2018
Confidential
Machine Learning is an important component in Industry 4.0 analytics
Other AI Offerings

Deep Neural Networks Unstructured Text


Analytics

Computer Vision

Video Analysis
Time & AI Infrastructure

Applied Machine Learning Names Entity


Extraction

Image Insights /
Classification
Labeled Y Comparison Language
Data ? Chat Bots Translation
Y N Clustering /
Predict
Anomaly Detection
Categories?
N
Predicting Y Regression
Text Extraction
values ? N Dimension Knowledge High-Fidelity Speech
Reduction Management Synthesis

Data Analytics Cognitive

8 | Internet
8Infosys of Manufacturing Midwest 2018
Confidential
Machine Learning involves solving business problems using 25+
algorithms segmented into 4 groups
Solving a Machine Learning (ML) Different algorithms are better suited We have found the Python scikit-learn
problem depends on finding the for different types of data and flowchart useful for selecting ML
right algorithms for the business different problems algorithms specific to business
problem problem and available data

Yes/No;
Old Faithful
quality
(Describes
pass/fail
relationships)

Groups into
similar Addresses
characteristics too much
sensor data!
Infosys Confidential
http://scikit-learn.org/stable/tutorial/machine_learning_map/index.html 9 | Internet
9 Infosys of Manufacturing Midwest 2018
Confidential
Machine Learning platforms (tools) have a large library of algorithms,
designed to address different types of business problems
Classification Regression
Identify category to which an object belongs Predicts a continuous-valued attribute associated with an object
Applications: spam detection, image recognition, quality P/F Applications: forecasting, pricing determination

Support Vector k-Neighbors Random Forest / Stochastic Gradient Stochastic Gradient Lasso Ridge Elastic Net
Machine (SVM) Classification Decision Trees Descent (SGD) Descent (SGD) Regression
Classifier Regressor

Clustering and Anomaly Detection Dimensionality Reduction


Automatic grouping of similar objects into sets Reduce the number of random variables to consider
Applications: customer feed (Twitter) segmentation, grouping data Applications: visualization, sensor feeds, efficiency improvement

K-Means Gaussian Mean Shift Spectral Randomized Principal Locally Linear Kernel Spectral
Clustering Mixture Models Clustering Component Analysis Embedding Approximation Embedding
(PCA)
Infosys Confidential 10 | Internet
10Infosys of Manufacturing Midwest 2018
Confidential
Machine learning techniques are organized by ability to learn

Traditional Data Analysis  “If” and “Else” decisions designed by humans,


coupled with functions ( e.g. Excel functions), to
process data or adjust to user input
Raw Data Analyze and Use Model/  Changing the task might require a rewrite of the model
Write Rules Improve

Supervised  The training data one feeds to the algorithm


includes the desired solutions, called labels
Machine Learning  Based on learning, algorithm provides outputs for
new real time inputs
Raw Data Features / Train and Trained Deploy /  Examples: Classification, Regression
Labels Evaluate Model Improve

Unsupervised  Only the input data is known, and no known output


data is given to the algorithm. They are usually
Machine Learning harder to understand and evaluate
 Examples: Clustering (Identifying topics in a set of blog
Raw Data Algorithms Cluster / Use in supervised posts, segmenting customers into groups with similar
Anomalies learning preferences, detecting abnormal access patterns to a
website)

Reinforcement  The learning system, called an Agent, can observe


the Environment, select and perform Actions, and
Machine Learning Reward / get Rewards in return, or Penalties in the form of
Penalty negative rewards
 Examples : DeepMind’s AlphaGo, Walking Robot,
Goal Initialize Agent Environment Automated Trader
Infosys Confidential 11 | Internet
11Infosys of Manufacturing Midwest 2018
Confidential
Action
Common machine learning use cases in a manufacturing context

Order to Cash Core Manufacturing Procure to Pay Record to Report

Demand estimation - order Contracts analysis for


Predictive maintenance FP&A Forecasting
quantities named entity

Anomaly detection: credit Tech support / knowledge Real-time monitoring of


Commodity price forecasting
risk base foreign exchange

Order entry automation In-line quality inspection Consistent supplier terms

Automated inventory Defect root cause and


Long tail spend analytics
stocking for service truck corrective action

Simplification and
Production planning and Demand forecasting using
automation of manual
scheduling sales pipeline
services billing

Infosys Confidential 12 | Internet


12Infosys of Manufacturing Midwest 2018
Confidential
Industry 4.0:
Illustrative Case
Industrial
Studies
Examples
13 13 | Internet
Infosys of Manufacturing Midwest 2018
Confidential
13
Energy matters! Industrial IoT aids energy optimization in Infosys
campuses
Business Need Sustainability initiative at Infosys and implementation using IIoT solution

• Chillers
• HVAC Central Command Center Demand Management
• Generators
• Elevators
• Sewage
Treatment
plants
• Solar power
plants Digital Twin and Optimum Predictive
Operating Conditions Maintenance
Large Campuses
80 million+ square feet Assets Managed Solution Approach

Business Benefit 46% reduction in per-capita energy consumption over 8 years


$100 million savings over 3 years
14 14 | Internet
Infosys of Manufacturing Midwest 2018
Confidential
Visibility

Business Value
Maturity Level/
Centralized command center for real-time visibility
Visibility Transparency Predictability Adaptability

Path of development

Real-time data acquisition

Visibility to key operating parameters


15 15 | Internet
Infosys of Manufacturing Midwest 2018
Confidential
Business Value
Maturity Level/
Transparency and the Digital Twin
Analyzing performance – As Designed vs As Installed vs As Operated Visibility Transparency Predictability Adaptability

Path of development

As Designed As Installed As Operated


Plot of Critical Performance Parameters
• Condenser Water Delta (leaving temp – entering temp) Digital Model complements
• Chiller Water Delta (leaving temp – entering temp)
• Evaporator Small Temp Diff (Ref. Sat temp– Chiller
physical assets
Water leaving temp)
• Condenser Small Temp Diff (Ref. Sat temp– Study operating conditions,
Condenser Water leaving temp)
• Chiller Working Hours
trends and performance

16 16 | Internet
Infosys of Manufacturing Midwest 2018
Confidential
Predictability

Business Value
Maturity Level/
Implemented advanced analytics on chiller data for event detection and prognostics
Visibility Transparency Predictability Adaptability

Path of development

Correlation Exploratory Data


Data Collection Data Cleansing Event Detection Prognostics
Analysis Analysis

• Identification of key performance indicators


• Exploratory analysis and visualization of data
• Event detection – Hotelling’s T-squared and quartile-
based method
• Prognostics – ARIMA model with xreg variable
• Knowledge model development

17 17 | Internet
Infosys of Manufacturing Midwest 2018
Confidential
Example: greenfield waste water treatment plant’s pumping station

FOCUS
The strict environmental permit must be • Plant: State-of-the-art waste water treatment plant in Europe
fulfilled which means that the effluent
from the pumping station to
environment is not acceptable

TARGET
To find optimal design solution to fulfil the
required availability and safety with minimum
lifecycle cost
• Three operating scenarios:
1) Average flow
EVALUATE
To create a RAMS simulation model of the 2) Average + Industry peak flow
different design alternatives with different
operation and maintenance scenarios 3) Peak flow (heavy rain, flood)

• Three different design solutions for the pumping station to be analyzed


RESULTS Case 1: 4 big pumps and 3 small pumps (original design requirement
To find out design solution to fulfil the from the Client)
required availability and safety with Case 2: 3 big pumps and 3 small pumps
minimum lifecycle cost
Case 3: 2 big pumps and 3 small pumps

18 18 | Internet
Infosys of Manufacturing Midwest 2018
Confidential
Moving from a traditional RAMS* to AI-based RAMS design enabled
faster decision-making with more accuracy

System Design and Realization Processes Integrated Operation


Defined design solutios:
Supplier-specific Selected WP- Technical Performance & Availability
- Design basis, specification, Supply management
Work Packages suppliers Operability & Maintainability
objectives & requirements Safety & O&M Cost

RAMS Definition of Maintenance Development and Optimization of the Plant specific maintenance
Identification of
Design Categories for RCM- Planning of CBM, service program to achieve required availability
critical RCM positions
Process positions TBM and CM task and safety with minimum costs

Drishti 4.0 Operational


Excellence AI Platform
Application of
Prognostics and
Healty Baseline
Healty Management Instrumented process
Identify parameter
Yes Automated identification
Alarm Anomaly? In-Situ Monitoring
and data capture
Continue
monitoring No

Parameter Isolation Physics of


RAMS database
Failures Model

Data-Driven Models Failure Definition

Remaining Useful
Life Estimation

* RAMS = Reliability, Availability, Maintainability, Safety 19 19 | Internet


Infosys of Manufacturing Midwest 2018
Confidential
Visualization was an important tool to take decisions and
interventions based on analytics recommendations

Plant performance
KPIs:
RAMS and Risk
Index

Downtime insights:
Troubled asset, reason
for failures, cost savings

20 20 | Internet
Infosys of Manufacturing Midwest 2018
Confidential
Condition-based monitoring used AI to recommend pump
maintenance and proactive interventions

Results of RAMS analyses of three design solution cases with current maintenance service program (without
CBM= Condition Based Maintenance) and with Drishti 4.0* operational excellence AI platform (=with CBM)

Pumping station Case 1: Case 2: Case 3:


operational time 30 a 4 big and 3 small pumps 3 big and 3 small pumps 2 big and 3 small pumps

Maintenance service program Without CBM With CBM Without CBM With CBM Without CBM With CBM

Reliability and Availability


requirements are fulfilled YES YES YES YES NO NO

Total Life Cycle costs (€)


2,307,358 2,107,910 1,743,175 1,568,232 1,486,811 1,308,584

Case 1 without CBM: Case 2 with CBM: recommended Case 3:


The current design design solution Not acceptable
solution 1)No environmental risks because of
2)LCC cost are 740k€ less than the violation of
original design solution environmental
RAMS design savings 565k€
permit
* Dhristri 4.0: dhristri.com
With CBM LCC savings 175k€
21 21 | Internet
Infosys of Manufacturing Midwest 2018
Confidential
Machine Learning
in action:
Churn Prediction
Major Auto Manufacturer
22 22 | Internet
Infosys of Manufacturing Midwest 2018
Confidential
The customer digital services cycle can be defined in the shape of a
funnel, and at each stage, there is churn. How do we reduce churn?

• Purchased vehicle

• Enrolled in trial (1 year)


Find my car
• Converted to paid subscription Remote
Climate Start
• Renewed paid subscription
Typical
cloud
• At each stage of the funnel, we lose customers connected
– What can we do to reduce churn at each stage? features
• By using customer and vehicle data across each stage,
we can use machine learning to predict a customer’s
likelihood to churn (i.e. customer does not progress to
the next stage of the funnel) Remote
Lock/Unlock

Are there usage trends or customer behaviors that can


predict a customer’s likelihood of churning?
23 23 | Internet
Infosys of Manufacturing Midwest 2018
Confidential
Initial subscriptions present the biggest opportunity for improvement

Luxury Brand Mass Market


Brand
Select models

Current annual sales: ~300,000 Annual sales (projected 2019): ~1,500,000

13% 64% 23% 9% 46% 45%


33,600 vehicles 243,200 vehicles 64,400 vehicles 136,800 vehicles 699,200 vehicles 684,000 vehicles

Enrolled in trial Enrolled in trial but did Did not enroll in Enrolled in trial and Enrolled in trial but did Did not enroll in trial
and converted not convert to paid subscription trial converted to paid not convert to paid subscription
to paid

Increasing initial paid subscriptions is largest improvement opportunity


24 24 | Internet
Infosys of Manufacturing Midwest 2018
Confidential
We gathered all relative usage and subscription data
• We gather data for new vehicles that were sold and enrolled in a trial of one month (in 2016)
– These vehicles were up for renewal in the following year, total vehicles in sample: ~24,000
Vehicle Service
Sale 2016 2017 Renewal
Trial Period

Sep Oct Nov Dec Jan Feb Mar Apt May Jun Jul Aug Sep

• Next, we collected all usage and sales data for these Usage
Analysis
vehicles for the month before the renewal (~35,000 total commands)
– What specific commands were used by each vehicle? e.g. Remote Lock, Remote Start, Vehicle Finder, etc.
– What year / model was the vehicle?
– What was usage on the weekday vs the weekend for each vehicle?
– What metropolitan area was the selling dealer in?
– Did the customer subscribe to paid services?

After getting the right data, we can build a model to answer the overarching question:
Which customers will subscribe?
25 25 | Internet
Infosys of Manufacturing Midwest 2018
Confidential
After a number of tuning iterations, the model enables churn
prediction on an individual basis
• The classification model was tuned over multiple iterations using Microsoft Azure, in order to find the
ideal level of accuracy and resiliency measured with the Receiver Operating Characteristic (ROC)
– Certain data was removed from the model to improve model quality
This curve would
indicate we could
repeat customers extraneous data fields (e.g. VIN) predict every single
customer perfectly
(impossible!)
VIN: SN0001
Model: SUV MODEL A
Weekday uses: 48 Renewal This straight line
Weekend uses: 18
Remote: 0 probability: would indicate we
Status: 0
73.9% are guessing
Finder: 31 randomly
Lock: 35

This is the
Over the course of 100+ iterations, the machine learning accuracy of our
algorithm uses the training set to build a decision tree model using limited
based on the input data. Sample decision branches: data – this can only
• Is weekend usage greater than weekday usage? improve as we add
• Was the car purchased in an area with extreme weather? more data points,
e.g. demographics
The tuned model enables us to determine the churn probability for each customer
Infosys Churn Model POC, 2018 26 26 | Internet
Infosys of Manufacturing Midwest 2018
Confidential
We can now choose customers to address, to maximize profitability

VIN Scored Probabilities


• We can choose which ~22,000 customers (VINs) SN00001 0.121978149
customers to reach out to 0% likely 100% likely SN00002 0.48944521
to renew to renew SN00003 0.054602593
• This enables better SN00004 0.196847379
efficiency of resources Less than 10% Less than 25% SN00005 0.128807783
likely to renew likely to renew Actual model output
(VIN data is anonymized)

Sample implementation using How effective customer


machine learning output* Net Profit (in 000's) incentives are (“change
Per Year, 1 year old vehicles Only
Campaign Conversion Rate their minds”)
## 2% 3% 4% 5%
• Incentives are 4% effective

Probability of renewal
5$ 16 $ 38 $ 59 $ 81
• Customers < 20% likely to renew 10 $ 12 $ 49 $ 87 $ 125
15 $ (9) $ 44 $ 96 $ 149
• Profit in first year: $96k 20 $ (36) $ 30 $ 96 $ 161
Additional net profit per
• This is maximum cumulative profit for the scenario 25 $ (59) $ 13 $ 85 $ 158
year (000s)
30 $ (80) $ (3) $ 74 $ 150
35 $ (95) $ (15) $ 65 $ 145
We choose the retention (renewal) probability of 40 $ (109) $ (26) $ 57 $ 141
customers to address
* Using Azure Machine Learning Studio

Based on incentive effectiveness, we can maximize value by choosing the targets


This is relevant for many manufacturing scenarios involving diminishing returns
27 27 | Internet
Infosys of Manufacturing Midwest 2018
Confidential
Continuing the conversation….

Jeff Kavanaugh
Partner, Manufacturing
Infosys Consulting
jeff_kavanaugh@infosys.com

Adjunct Professor
University of Texas at Dallas
jeff_kavanaugh@utdallas.edu

@jeffkav http://bit.ly/2qzanfr https://www.infosys.com Foundational skills for learning


/age-of-ai/ in the age of AI (Amazon, B&N)
www.infosysconsultinginsights.com www.infosys.design/plantio

28 28 | Internet
Infosys of Manufacturing Midwest 2018
Confidential

You might also like