Download Using Big Data for Machine Learning Analytics in Manufacturing White Paper Manufacturing

Manufacturing
White Paper
Using Big Data for Machine Learning
Analytics in Manufacturing
About the Authors
Jiby Joseph
Business Consultant, Manufacturing Innovation and Transformation Group, Tata
Consulting Services (TCS)
Jiby has over 10 years of industry experience in the manufacturing and retail domains
across Supply Chain Management, Strategic Cost Reduction, Business Process Reengineering, Lean Six Sigma, and New Product Development. Jiby holds an engineering
degree, along with an Executive Diploma in Management from the Indian Institute of
Management (IIM) Calcutta. He is a trained Lean Six Sigma Black Belt professional.
Omar Sharif
Business Consultant, Manufacturing Innovation and Transformation Group, TCS
Omar has over eight years of experience in manufacturing, in the areas of Strategic
Supply Chain Consulting, Logistics and Transportation, Warranty and Aftersales, Six
Sigma and Quality, and Training. He has worked on projects with leading global auto
OEMs, aero-engine OEMs, and large chemical companies. His current focus areas include
Analytics and Big Data. Omar holds an engineering degree along with a Post Graduate
Diploma in Management.
Ajit Kumar
Business Consultant, Manufacturing Innovation and Transformation Group, TCS
Ajit has over 10 years of experience in manufacturing operations and consulting. His
functional expertise comprises Industrial Operations, Supply Chain, and Quality and
Manufacturing Analytics. He has successfully executed various cost saving, process
improvement and business intelligence (BI) projects for leading manufacturing
companies.
About the Authors
Saurabh Gadkari
Business Analyst, Manufacturing Innovation and Transformation Group, TCS
Saurabh has over three years of industry experience in the manufacturing and telecom
domains. He has worked in the areas of research and analytics across Supply Chain, Sales
and Marketing, Warranty, and New Product Introduction. His current focus is on BI and
Big Data Analytics. Saurabh is an engineering graduate with a Diploma in Management.
Aditya Mohan
Business Analyst, Manufacturing Innovation and Transformation Group, TCS
Aditya has over seven years of work experience in the IT industry, with over three years
in the manufacturing domain as a functional consultant. He has worked in ERP
development and implementation across small and medium businesses in India. He
holds a Diploma in Management from IIM Lucknow.
Machine Learning as a concept has been in existence for many decades now. However, most
manufacturing operations — such as repairing an aircraft engine, planning the product mix
in cement production, or ensuring energy control in a large facility — are still largely
dependent on experience-based human decisions. The advent of Big Data technology,
coupled with efficient data storage mechanisms and parallel processing frameworks, has
found new use for the petabytes of data generated by manufacturing operations. Applying
Machine Learning techniques to the shop floor has enabled increased accuracy in decisionmaking and improvement in performance.
This paper explores how Machine Learning algorithms, in conjunction with Big Data
technologies, can help manufacturers bring about operational and business transformation.
Contents
1. The Evolution of Machine Learning
6
2. Approach to Machine Learning
6
3. Applications of Machine Learning in Different Industries
8
4. Case in Point: Application of Machine Learning for Predictive
Maintenance in Automotive Industry
10
5. Conclusion
13
The Evolution of Machine Learning
Artificial Intelligence (AI), a concept that came into existence in the 1990s, is fast gaining popularity across
industries. Deep Blue (the chess-playing computer developed by IBM)¹, Watson (the artificially intelligent computer
system specifically developed to answer questions on the quiz show Jeopardy!)², and Google Chauffeur (the
software powering Google's driverless car)³ are some landmarks in the field of AI, where computer programs have
surpassed the capabilities of the human mind.
Machine Learning is a part of AI that continuously observes a series of actions performed over a period of time, and
puts this knowledge to use by devising ways to perform similar processes better, in a new environment. In 1959,
Arthur Samuel defined Machine Learning as the field of study that gave computers the ability to learn without
being explicitly programmed. From initial efforts to explore whether computers could play games and mimic the
human brain, this study has now grown into a broad discipline with the ability to produce statistical and
computational theories of learning processes.
Today, although the field of Machine Learning is still nascent, it has found its way into daily user experience through
applications like Google Maps⁴ that present accurate geographical data from satellite view to street view, and
Netflix,⁵ which simulates the user experience through patterns of movie viewing habits. Other examples include
applications used for speech and gesture recognition (Kinect), natural language processing (Siri), facial recognition
(iPhoto), web search, spam filters, ad placement, credit scoring, fraud detection, stock trading, and drug design.
Big Data technology, with the capacity to process large volumes of data, is accelerating the growth of Machine
Learning applications. Apache Mahout, a machine learning library for Hadoop, has a collection of scalable Machine
Learning algorithms, executed in quick cycles with the innovative 'MapReduce' technology.⁶ These algorithms are
remarkable in their ability to bring out hidden relationships among data sets and make predictions.
Approach to Machine Learning
While Machine Learning techniques have found an increasing level of applicability and relevance to real world
scenarios, they pose a few implementation challenges.
First among them is the lack of expertise in applying Machine Learning techniques to business problems. Not many
data scientists have experience in the manufacturing industry, along with a strong knowledge of statistics and the
ability to derive analytical insights from manufacturing data.
[1] ‘Deep Blue, Icons of Progress’ IBM, accessed on 5 May 2014,
http://www-03.ibm.com/ibm/history/ibm100/us/en/icons/deepblue/
[2] ‘IBM's Watson supercomputer crowned Jeopardy king’, BBC News, 17 February 2011, accessed on 6 May 2014,
http://www.bbc.com/news/technology-12491688
[3] ‘Inside Google's Quest To Popularize Self-Driving Cars’, Popular Science, 18 September 2013, accessed on 6 May 2014,
http://www.popsci.com/cars/article/2013-09/google-self-driving-car
[4] Andrews Ng, ‘Lecture 1, Machine Learning’, Stanford,, 22 July 2008, accessed on 6 May 2014,
http://www.youtube.com/watch?v=UzxYlbK2c7E at 54 minutes
[5] ‘Collaborative Filtering’, Centre for Computational Statistics and Machine Learning, accessed on 6 May 2014,
http://www.csml.ucl.ac.uk/courses/msc_ml/?q=node/40
[6] The Apache Mahout™ Machine Learning Library, accessed on 8 May 2014, http://mahout.apache.org/
6
The second challenge is the lack of a culture that can apply the Machine Learning process to day-to-day operations.
This does not mean the mere usage of mobile devices with system failure prediction capabilities, but rather a high
level of involvement from the operations team in the process of Machine Learning, for better prediction accuracies
and process improvement results.
The third challenge is the availability of the right data from various operations and processes. Machine systems such
as Programmable Logic Controllers (PLC) and Supervisory Control and Data Acquisition (SCADA) may capture a lot of
machine data, but this data may not be relevant. PLC and SCADA do not store the entire data set required for
creating a predictive analytics solution based on Machine Learning.
The fourth challenge is the lack of technological competence in using Big Data for Machine Learning algorithms.
While software vendors have a growing list of Machine Learning algorithms, they are mostly unsupervised learning
algorithms that are used to derive inferences without setting the expected responses. Applying them to a specific
problem requires a lot of effort in fine tuning the parameters and validating the results.
As these challenges would be more or less relevant for different organizations, the implementation approach must
be tailored for specific requirements. The reference model depicted in Figure 1 enumerates the various elements to
be considered for the successful implementation of Machine Learning.
Ensure management commitment and investment
for people, process, data, and technology
Prioritize
business
challenges
Gain
competence
on Big Data
technology
and forge
alliances with
technology
partners
Deploy and
monitor
solution
Test model
for continuous
improvement
Build
data
infrastructure
Machine
Learning
Framework
Set up the
Big Data
platform
Prepare and
understand
the data
Collaborate
with
universities
and get
training on
Machine
Learning
models
Develop right
Machine
Learning
models
Figure 1: Approach to implementation of Machine Learning techniques
7
An organization needs to establish a clear link between its business imperatives and Machine Learning program
strategy. The initiative requires widespread executive sponsorship and business commitment to yield the expected
benefits. The following checklist can help in the implementation process:

Define the business case with a focus on prioritized opportunities, possible solutions, and return on investment.

Set in place a core team with expertise in business processes, Machine Learning models, data architecture, and Big
Data technology. Build a team of data scientists with such expertise, either by hiring external resources or through
internal grooming.

Get guidance from renowned universities on established Machine Learning models.

Deploy dedicated operational resources, build the infrastructure to source data, and set up the Big Data
technology platform.

Institute change management programs for improved processes and intelligent ways of working.
Applications of Machine Learning in Different Industries
Machine Learning can be applied to high volumes of data in order to gain deeper insights and to improve decision
making. Figure 2 depicts some emerging applications of Machine Learning.






Predictive maintenance or
condition monitoring
Warranty reserve estimation
Propensity to buy
Demand forecasting
Process optimization
Telematics
Manufacturing








Predictive inventory planning
Recommendation engines
Upsell and cross-channel
marketing
Market segmentation and
targeting
Customer ROI and lifetime
value
Retail
Aircraft scheduling
Dynamic pricing
Social media – consumer
feedback and interaction
analysis
Customer complaint
resolution
Traffic patterns and
congestion management
Travel and
Hospitality







Risk analytics and regulation
Customer Segmentation
Cross-selling and up-selling
Sales and marketing
campaign management
Credit worthiness evaluation
Financial Services





Alerts and diagnostics from
real-time patient data
Disease identification and risk
stratification
Patient triage optimization
Proactive health
management
Healthcare provider
sentiment analysis
Healthcare and
Life Sciences






Power usage analytics
Seismic data processing
Carbon emissions and trading
Customer-specific pricing
Smart grid management
Energy demand and supply
optimization
Energy, Feedstock,
and Utilities
Figure 2: Machine Learning applications across industries
8
Manufacturing organizations generate a lot of data in the course of operations, but currently, are not collecting,
storing, managing, and using this data judiciously in order to improve process performance. Machine Learning
systems can estimate the predicted outcome accurately based on training set data or past experiences. By
gathering valuable insights for better and more accurate decision-making, Machine Learning systems can help
manufacturers improve their operations and competitiveness.
The following are some potential game changing use cases from across industries.
Case 1 – Condition Monitoring
Airline companies no longer pay upfront for engines; instead they pay per hour of ‘time on wing’, a measure of
operational reliability of the engine or aircraft system. This forces engine manufacturers, currently engaged in
diagnostics of engine defects for service requirements, to try to improve the engine’s reliability. Big Data
applications for Machine Learning techniques carry out pattern matching for fault isolation and repair support
using multiple operational and external parameters received in real-time from sensor data. This helps
manufacturers accurately predict failure in engine operations well ahead of time, thus increasing the service
revenue and reducing the cost of service.
Similarly, the heavy engineering industry is moving away from the typical long term service contract to an
‘Analytics-as-a-Service’ model. This enables manufacturers to predict the health of the equipment in real time,
allowing customers to release the equipment for maintenance only when necessary. Machine Learning techniques
such as neural networks, support vector machines, and decision trees are capable of identifying complex
interdependencies within operational parameters and detecting anomalies that can lead to equipment failures.
Manufacturers of power generation turbines, electrical substation equipment, and building equipment have
implemented these models and techniques. The vehicle telematics industry is also using this model, enabling a
shift in the service business for automotive dealers from regular service visits to service based on analytical
findings.
Case 2 – Quality Diagnostics
Machine Learning techniques can potentially eliminate the process of testing, by predicting quality early on in the
manufacturing process. This can change the paradigm for precision manufacturers who are currently unable to
detect micro shrinkage or porosity of castings, and for engine manufacturers spending hundreds of hours in test
rigs before shipment. Big Data applications collect data from manufacturing operations and from the various
processes across the supply chain, to decode the behavior of material transformation and engine operation in order
to detect potential defects. ⁷
[7] ‘ Igor Santos, Javier Nieves, Yoseba K Penya, and Pablo G Bringas, ‘Optimizing Machine-Learning-Based Fault Prediction in Foundry Production’,
Deusto Technology Foundation, accessed 10 May 2014,
http://paginaspersonales.deusto.es/ypenya/publi/penya_DCAI09_Optimising%20Machine-learningbased%20Fault%20Prediction%20in%20Foundry%20Production.pdf
9
Case 3 – Energy Optimization
With growing privatization and volatile markets, the power generation and transmission industry is looking at
energy service deals as a new avenue for growth. Machine Learning techniques can help these power generation
and transmission organizations predict the demand fluctuations from energy consumption patterns and arrive at
the optimum demand response in real time, thereby optimizing the demand and supply from various sources at
minimum cost. Building management companies can also benefit from such an application, since it helps them
optimize their energy consumption and keep energy costs in check.
Case 4 – Demand Prediction
Machine Learning techniques based on natural language processing and speech recognition can process large
amounts of social media data. This can be helpful for the automotive industry and other B2C markets. Insights from
these pockets of data help in near-accurate demand prediction in response to an organization’s and its competitors’
market campaigns and product launches. Several Big Data applications are available in the text analytics space to
analyze data from various sources like call centers, blogs, survey results, and visit notes.
Case 5 – Propensity to Buy
With competition getting fiercer by the day, automotive Original Equipment Manufacturers (OEMs) are keen to
understand factors influencing the propensity to buy and the appropriate mix of customer incentives to fuel their
product demand. Though estimation of the propensity to buy is not a new concept, it has so far been limited to
major events, such as new product launches, and a narrow set of data. Machine Learning techniques can use vast
amounts of data to unearth vital insights like consumer attitudes and perceptions towards the brand. This can help
marketers improve conversion rates and formulate successful and economical up-selling, cross-selling, and
retention strategies for targeted customer segments.
Case in Point: Application of Machine Learning for
Predictive Maintenance in the Automotive Industry
Business Scenario
Consider the scenario of a stamping plant at an automotive OEM that manufactures vehicle panels. The operation
involves an intensive workload, requiring high availability of hydraulic press lines. The Overall Equipment
Effectiveness (OEE) of the press line was as low as 65 percent, with the breakdown time ranging from 17-20 percent.
Though the press manufacturer offered closed loop control systems, they were limited to validation against a static
range of values and did not read all the variables affecting the failure event. Hence they were not effective in
improving the OEE. In addition, the maintenance process was largely based on preventive scheduling, leading to
high unplanned downtime and maintenance cost as well as lost capacity during the maintenance tasks.
10
The OEM wanted to improve the equipment availability through accurate prediction of potential events such as
part failure and functional degradation. Through this proof of concept, TCS recommended the use of Machine
Learning analytics to achieve the same.
Business Process Change
The process of maintaining the press line was studied in detail. It was found that Maintenance Engineers spent a lot
of effort in attending to breakdowns and consequently, less time was spent visiting the floor and allocating
resources for planned maintenance. A Machine Learning solution could enable the condition monitoring process,
reducing the need to attend to breakdowns (see Figure 3).
Traditional
Shop floor visit
for equipment
observation
Study the
equipment log
Review
absenteeism
and critical
spares availability
Identify any
abnormalities and
decide on
maintenance task
Allocate
resources
for planned
maintenance
Review and
approve the
spare orders
30% time in analyzing the problem
Receive
notification
on equipment
breakdown
Redeploy the
people to
attend breakdown
maintenance
Diagnose the equipment
fault by analyzing the
performance indicators with
the known techniques
Identify
potential
options for fixing
the problem
If unsuccessful, try
another option
until problem is fixed
Try one of
the options
70% time in fixing the problem
Big Data
Real time observation
of failure probability
over mobile
Follow the directions
for maintenance event
based on probability
Review the
absenteeism and
critical spares
availability
Review and
approve the
spare orders
Resource allocation
for planned
maintenance
70% time in analyzing the problem
Receive
notification
on equipment
breakdown
Redeploy the
people to attend
breakdown
maintenance
Study the trend of operation
parameters to accurately
identify the right root cause
Identify the right
option for fixing
the problem
Execute the
option to fix
the problem
30% time in fixing the problem
Standard
Process change
Figure 3: Business Process Changes for a Maintenance Engineer at the OEM Plant
11
Analytical Solution
Stochastic Gradient Descent (SGD), a Machine Learning technique, was used to predict failure events using the
sensor data of all three machine systems in the press line – hydraulic press, blank holder, and lifter. The technique
maximizes the likelihood of classification into the defined categories using Logistic Regression. The algorithm uses
weight factors associated with the sensor data to classify the equipment events into the two categories: 'fail' and
'not fail'. It uses an iterative process to calculate new weight factors through the observation of equipment data.
The Predictive Maintenance Solution
Sensor data on 15 operating parameters (such as oil pressure, oil temperature, oil viscosity, oil leakage, and air
pressure) was collected from the equipment every 15 seconds for a period of 12 months. The components of the
solution output are depicted in Figure 4.
Drill down to derive equipment
level OEE and identify the
bottlenecks
Calculate the OEE of the entire plant
(OEE = Availability* Quality *
Performance)
Plant OEE
Time to
Fail
Estimate the time to fail in
correlation with machine
event failure probability
OEE by
Equipment
Probability of
Failure
Establish failure probability of
the machine based on the real
time operational parameters
Figure 4: The Predictive Maintenance Solution for the OEM Plant
12
Benefits of the Solution
The potential benefits that the OEM can realize by implementing this solution are:

The ability to predict a failure event before it occurs, with an accuracy of more than 92 percent and a cut-off
probability of 40 percent

Increase in OEE from the industry average of 65 percent to a benchmark figure of 85 percent

Greater efficiency in the scheduling and planning process, ensuring minimal loss of production

Improvements in asset reliability and product quality
Conclusion
Across industries, Big Data technology has tremendous potential to leverage Machine Learning capabilities in
enabling accurate decision-making for superior performance. There are many applications of Machine Learning
techniques in the manufacturing industry, but successful implementation requires commitment from top
management to enable changes in processes, active involvement of operational resources, availability of data, and
collaboration with academia and technology partners with expertise in Machine Learning models and Big Data
technology. The solution for predictive maintenance analytics using Stochastic Gradient Descent, as presented in
this paper, demonstrates how Machine Learning can enable accurate prediction of failure events in the press line.
Recent developments in advanced computing, analytics, and low cost sensing have the potential to bring about a
transformation in the manufacturing industry. The implementation of Machine Learning and Big Data may drive the
next wave of innovation and may soon prove to be an unavoidable tactical move in achieving higher levels of
optimization.
13
About the Manufacturing Solutions Unit
Global manufacturers are trying to reduce operational expenditure, invest in process improvement,
utilize existing capacity optimally and increase efficiencies, while maintaining product quality and
meeting safety and regulatory norms. TCS' Manufacturing Solutions provide you the bandwidth to
innovate on business models, leveraging contemporary technology solutions.
We believe in leveraging learning from across the segments in developing business solutions. Be it
in applying the concepts of lean new product introduction from discrete industries to a chemical
manufacturer, or leveraging the aerospace industry experience in service management for the
automotive sector, our dedicated Manufacturing Centers of Excellence (CoEs) under these focus
vertical industries are continuously looking at breakthrough solutions. Clients can benefit from our
rich experience in both the discrete (automotive, industrial machinery and equipment, aerospace)
and process industries (chemicals, cement, glass and paper).
Contact
For more information about TCS’ Manufacturing Business Unit, visit:
http://www.tcs.com/industries/manufacturing/Pages/default.aspx
Email: [email protected]
Subscribe to TCS White Papers
TCS.com RSS: http://www.tcs.com/rss_feeds/Pages/feed.aspx?f=w
Feedburner: http://feeds2.feedburner.com/tcswhitepapers
About Tata Consultancy Services (TCS)
Tata Consultancy Services is an IT services, consulting and business solutions organization that
delivers real results to global business, ensuring a level of certainty no other firm can match.
TCS offers a consulting-led, integrated portfolio of IT and IT-enabled infrastructure, engineering and
assurance services. This is delivered through its unique Global Network Delivery ModelTM,
recognized as the benchmark of excellence in software development. A part of the Tata Group,
India’s largest industrial conglomerate, TCS has a global footprint and is listed on the National Stock
Exchange and Bombay Stock Exchange in India.
IT Services
Business Solutions
Consulting
All content / information present here is the exclusive property of Tata Consultancy Services Limited (TCS). The content / information contained here is correct at
the time of publishing. No material from here may be copied, modified, reproduced, republished, uploaded, transmitted, posted or distributed in any form
without prior written permission from TCS. Unauthorized use of the content / information appearing here may violate copyright, trademark and other applicable
laws, and could result in criminal or civil penalties. Copyright © 2014 Tata Consultancy Services Limited
TCS Design Services I M I 07 I 14
For more information, visit us at www.tcs.com