www.pwc.

com/me

An introduction to
implementing AI in
manufacturing
Content page
Foreword 5

Deploying AI at scale 8

Four different stages of digital maturity 12

Implementing AI in the real world 14

Six building blocks for successfully 22

implementing AI

Summary 30

Authors and Contributors 32

Foreword

Artificial intelligence (AI) is gradually being implemented

in almost every aspect of our lives. In medicine, geology,
customer data analysis, autonomous vehicles and even art,
its applications are everywhere and its uses are constantly
evolving. However, AI has raised at least as many questions as
it has answered, including how the technology is defined and
used (assisted vs. augmented vs. autonomous intelligence, for
example), whether computers are capable of thinking in the
same way as humans (the so-called Turing test), the broader
impact of automation on society, and the unanticipated ethical
and moral dilemmas it may cause.

This white paper is the first in a global series on AI in

manufacturing published by PwC, and is meant for executives
in manufacturing and industrial companies who are looking
to implement AI in their organisation. It focuses on the use
of AI throughout the manufacturing value chain, including in
production, testing and engineering. The aim is to address
the business applications that can be supported by data,
technology and automated processes, and how they are
enabled by the right people, organisational structure and
culture. We want to share recent developments, raise questions,
pose challenges and bring cutting-edge research and ideas
closer to industry.

Dr. Anil Khurana Dr. Reinhard Geissbauer

Global lead, Industrials, Manufacturing Global lead, Digital Operations Impact
& Automotive Center

Anand Rao Steve Pillsbury

Global lead, Artificial Intelligence US lead, Digital Operations Impact Center

AI in Manufacturing 5
 Foreword
From digital novices to digital champions: How AI can transform manufacturing

As we embark on a new decade, it is evident that the 2020s will shape up to be incredibly exciting for
manufacturing as the digital world continues to transform – and, no doubt Artificial Intelligence (AI) will
be at the heart of this transition.

For manufacturers, AI promises to be a game-changer at every level of the value chain. Direct
automation, predictive maintenance, reduced downtime, 24/7 production, improved safety, lower
operational costs, greater efficiency, quality control, and faster decision making are just some of the
rewards on offer to organisations that embrace the transformation and master the implementation of AI
throughout their entire business.

For those who have not yet made much progress with the adoption of AI, the good news is that it’s not
too late to start, nor are you alone. While manufacturing is often considered to be at the forefront of the
application of new technologies, according to a PwC survey of manufacturing executives carried out
across 26 countries in 2018, only 9% have implemented AI in their processes to improve operational
decision-making. This is probably because the process of introducing AI is highly complex, time-
consuming, and capital intensive, and requires a comprehensive, systematic approach if it is to prove
successful.

Though it may be a daunting task, it is now widely accepted that every industry will be impacted
dramatically by Artificial Intelligence (AI) within the next five years. Organisations and industries that
embrace this technological revolution early on are going to reap huge benefits that will see them emerge
ahead of their competitors, while companies that don’t will risk losing their competitive edge.
The technology is now advancing so rapidly that organisations that don’t make their move into AI soon
will find themselves falling behind before they realise it.

This message on AI and 4IR technologies was one of the many that were stressed upon during the series
of over 40 high-level panel discussions that took place at the second edition of the Global Manufacturing
and Industrialisation Summit (GMIS) in July 2019 in Yekaterinburg, Russia. The Summit gathered over
3,000 visitors from all over the world to hear more than 80 leading experts drawn from government,
industry, technology, finance, and academia to discuss and debate how the manufacturing sector
can harness the power of AI and all other forms of Fourth Industrial Revolution technology to drive its
development in the coming years. Exploring the opportunities and challenges that arise as the Fourth
Industrial Revolution unfolds and plotting a path forward for the sector is at the heart of what the Global
Manufacturing and Industrialisation Summit aims to achieve.

In a similar way, this report by PwC, ‘AI in Manufacturing: Assessing the scale of opportunity’ offers
essential guidance for executives working in the manufacturing space who are looking to implement AI in
their organisation, with a focus on the use of AI throughout the manufacturing value chain. Inevitably, one
of the most influential factors in AI adoption is a company’s overall progress in digitising its operations.
By first assessing their current stage of progress, companies can take the next step in the transition from
being digital novices to digital champions, and in the process play a part in the transformation of the
manufacturing sector.

Badr Al-Olama
Global Manufacturing and
Industrialisation Summit (GMIS)

6 AI in Manufacturing AI in Manufacturing 7
 What is the potential of AI?

Simply put, AI is the simulation of human intelligence processes by computer systems2. Later in this paper
we will look in more detail at the four main types of AI: assisted intelligence, augmented intelligence,
autonomous intelligence and automation. Since the Industrial Revolution, factories have been rapidly and
inexpensively mass-producing products to satisfy demand. However, manufacturers today are confronted
with a wave of new challenges including:

Challenge 2
Digitised products

Deploying AI
at scale Challenge 1
The ‘personalisation’
of mass products
Challenge 3
Compliance with
environmental standards
and wastage

“
A glimpse into the future of manufacturing can be found at FANUC’s In addition, manufacturers are still contending with the problems they have always faced: higher
plant in Oshino, Japan. Here, at one of the largest manufacturers production costs, equipment failure, and bottlenecks in the supply chain.
of industrial robots in the world, the robots build, inspect and test
themselves. AI is helping to address some, if not all, of the above issues. PwC’s AI analysis in 2017, Sizing the prize
– What’s the real value of AI for your business and how can you capitalise, showed that global economic
FANUC’s complex of 22 sub-factories is significantly advanced. It Only 9% of output, as measured by gross domestic product (GDP), could be 14% higher in 2030 than baseline
is in fact the world’s first factory complex with robots that create manufacturing projections of $114 trillion, as a result of the expected growth of AI. This translates to an additional
digitised “offspring” capable of machine learning and operates on a 24 $15.7 trillion3 (Figure 1).
executives have
hour basis. The company demonstrates just how far the use of AI in
manufacturing processes has come since the 1960s, when ‘Unimate’, implemented AI The economic benefits of AI will primarily be the result of:
the first mass-produced industrial robot arm, joined the assembly line at • Productivity gains from businesses that automate processes and augment the work of their existing
General Motors, lifting hot pieces of metal and placing them in cooling labour force with different types of AI technologies.
liquid. • Increased consumer demand, resulting from the availability of personalised and/or higher-quality
digital and AI-enhanced products and services.
While FANUC exemplifies an AI-led future, many manufacturers today
still struggle to deploy the technology at scale. In fact, a 2018 PwC Figure 1: How AI could contribute to additional global GDP growth
Global Digital Operations study of 1,155 manufacturing executives
across 26 countries showed that only 9% have implemented AI in their
processes to improve operational decision-making1. So before we 15
examine the applications and benefits of AI in manufacturing, we must
examine what AI is, its potential and what drives its adoption.
10

0
2017 2018 2019 2020 2021 2022 2023 2024 2025 2026 2027 2028 2029 2030

Quality Labour productivity Personalisation Time saved

8 AI in Manufacturing Source: PwC analysis AI in Manufacturing 9
 Potential by sector Potential within manufacturing organisations

Since the manufacturing and construction industries are capital intensive by nature, the adoption of AI Figure 4 shows where AI capabilities are currently being used within manufacturing companies5. While
applications may increase the sectors’ contribution to GDP gains by more than 10% by 2030 (Figure 2). the technology has been implemented throughout the key parts of the business, companies have put
slightly more focus on adding AI solutions to their core production processes: product development,
engineering, and assembly and quality testing.
Figure 2: GDP gains in 2030 resulting from AI by industry sector (% of GDP)
Figure 4: AI implementation in manufacturing functions

24% 23% 20% 20% 20% 20% 19%

Assembly/ Product Procurement Order Maintenance Logistics End-to-end

quality development management supply
Source: PwC analysis testing & engineering chain
planning

Potential by country The core use cases for AI in manufacturing are becoming clearer as companies learn to better quantify
the value of the technology. However, PwC research has shown that uncertainty around return on
When we looked at the forecast GDP impact by country, we found that China is expected to investment (ROI) remains one of the major challenges to implementing it more widely6. As Figure 5
experience a GDP impact of $2.5 trillion, the largest absolute GDP gain as a result of AI. This is shows, many companies also struggle with collecting and supplying the data that an AI system needs
largely due to the size of the country’s manufacturing industry, as well as the significant emphasis to operate, and lack people in their workforce with the right skills to implement AI at scale.
China is placing on developing AI4.

Figure 5: Challenges for manufacturing companies with implementing AI at scale

Figure 3: GDP impact from AI in 2030 by industry, region and channel of impact (% of GDP)

Inevitably, one of the most influential factors in AI adoption is a company’s overall progress in
digitising its operations. The organisations that have made the most progress in digitising core
business processes also lead on AI adoption. In the Global Digital Operations study, PwC ranked
companies by digital maturity and grouped them into four categories: digital novices, digital followers,
digital innovators and digital champions7. The study showed that 69% of digital champions have
implemented, piloted or plan to use AI within their business, compared to 10% of digital novices.
Source: PwC analysis

10 AI in Manufacturing AI in Manufacturing 11
 For manufacturing companies considering introducing AI into their processes, it is important to define
their goal from the outset: are they seeking to automate repetitive tasks or are they fundamentally
changing the nature of work in their factories by having humans and machines collaborate with each
other to make decisions? In our view, the following four types of AI8 are in existence today:

Four different stages of digital maturity Figure 6: The four types of AI

Digital champions Digital followers Human in the loop No human in the loop

The enterprise has a clear strategic position Internal functions such as sales, 1 Assisted intelligence 3 Automation
in the marketplace with complex and tailored manufacturing, sourcing and engineering Hardwired/ AI systems that assist humans in Automation of manual and
specific making decisions or taking cognitive tasks, both routine
customer solutions, offered via multilevel are integrated and collaborate closely. But
systems actions. Hard-wired systems that and non-routine. This does not
customer interactions. These companies there is little activity beyond vertical digital do not learn from their involve new ways of doing things -
have implemented near-real-time end-to- integration within the company. The culture interactions. it automates existing tasks.
end integration and connectivity of their and workforce at these companies are not
value chain across internal and external yet digitally oriented.
networks. These companies know how to 2 Augmented intelligence 4 Autonmous intelligence
leverage technology to connect customers, Adaptive AI systems that augment human AI systems that can adapt to
partners, operations and people to create systems decision making and continuously different situations and can act
value through ecosystems in new ways. learn from their interactions with autonomously without human
Digital champions have built a digital culture humans and the environment. assistance.
by establishing new ways of working and
making substantial investments in training,
sourcing and developing new capabilities and
skills.
Of course, companies may want to use different levels of automation and machine learning for different
tasks. A step-by-step approach, i.e. starting with a ‘human-in-the-loop’ system and moving to a fully
autonomous process, can be important for building up trust in the data and algorithms, and ultimately
Digital innovators Digital novices drive higher adoption of AI solutions throughout the organisation.

The enterprise is digitally connected The company employs some isolated

to external partners and customers, using digital solutions and applications, but these
integrated platforms for information exchange exist at the functional or departmental level
and collaboration. But the horizontal within the organisation.
digitisation is limited to the immediate supply
chain, with no wider ecosystem for customer
solutions, technology or people. Digital
innovators prize digitisation and encourage
the workforce to help identify new digital
solutions, but their advances are limited in
scope.

12 AI in Manufacturing AI in Manufacturing 13
 Implementing AI
in the real world

In this section, we will look in detail at two companies

PwC has worked with to implement AI in their
manufacturing processes. Both examples started with
augmented intelligence solutions and have now moved
on to autonomous intelligence. In the future, we believe
that the level of automation and machine learning will
evolve further.

14 AI in Manufacturing AI in Manufacturing 15
 Case Study 1:
3B-Fibreglass
3B-Fibreglass (3B) is a developer and supplier of Problem solving with AI: Identifying a break Root-cause analysis: Initial investigations • However, adding exogenous information from the
glass-fiber products and technologies to reinforce before it happens indicated that the system knew that a break was sensors did not improve the forecast quality, nor
thermoplastic and thermoset polymers, which are going to happen. Using sensors (mainly monitoring did it entirely explain the observed patterns.
used by the automotive and wind-power industries, A major source of inefficiency in the production of thermocouples, electrical currents and gas flow),
and in high-performance composite materials. glass fiber is fibers breaking when the pulling force 3B collected process information from within the 3B’s experience proves that AI can add significant
It operates three state-of-the-art manufacturing is applied. 3B used AI to gain insight into what was furnace, channels and close to the bushings, which value in a manufacturing setting. It also proves that
facilities located in Battice (Belgium), Birkeland causing the breakages and to act on the results. could be used to help explain what triggered breaks. results do not always come easily: in this case,
(Norway) and Goa (India). Three years of data was analysed from hundreds of the computer-vision analysis was very successful,
When a single fiber breaks, it triggers a domino sensors, trying to make a link. Upon using a wide illustrating the power of deep-learning algorithms
3B’s manufacturing process consists of a hot phase effect in which all the fibers going through the range of techniques, it appeared that: capable of making predictions that a human cannot.
and a cold phase. In the hot phase, silica sand, bushing knit together and the entire flow stops. However, the root-cause analysis was more limited
limestone, kaolin clay and a range of additives The quicker an operator can repair this situation, the • There was a common cause behind the breaks in its achievement.
are melted and heated up to a temperature of easier it is to fix and efficiency losses are minimised. which was anything that influenced the
approximately 1,400°C to create glass. The glass structure of the glass. One reason for this is that AI teams can only draw
remains in the furnace for a highly variable amount A set-up was created in which a camera constantly • Time-series analysis indicated that break behavior conclusions when the right data is included in the
of time - the first molecules leave the furnace after monitored the flow of the fibers as they left a was strongly auto-correlated - that is to say, dataset. At 3B, very often the data was initially
a couple of hours and some molecules remain for bushing. A deep-learning ‘computer vision’ network a high break level in one time slot was likely to generated for overall process monitoring, not to
several days. then analysed the data. result in high break levels in the subsequent time have machine-learning algorithms applied to it.
slots. Autoregressive Integrated Moving Average
From the furnace, the stream of molten glass is In the first phase of the analysis the goal was for the Models (ARIMA) appeared to perform very well
guided through channels to bushings. These are AI to identify a break, which alerted an operator that on the timeline.
plates made of precious-metal alloy, with anywhere a problem had occurred on a particular bushing.
from 4,000 to 6,000 very fine holes. To make glass
fiber, the molten glass passes through the holes In the second phase, the system had to predict
(each of which is typically just 9-25 micrometers in that a break was about to happen, on average 75
diameter) and comes out as fine filaments. seconds in advance (ranging between 65 and 85
seconds). The system also indicated where the
Every furnace feeds a couple of dozen bushings. problem occurred, allowing further investigations
As the molten glass passes out through the holes, into the root cause.
a pulling force is applied to it, lengthening and
thinning the fibers. As soon as the fiber leaves the
bushing, the temperature drops instantly to room Better visibility into the
temperature over a distance of less than a meter.
manufacturing process

Improved overall efficiency & Overall

Equipment Effectiveness (OEE)

16 AI in Manufacturing AI in Manufacturing 17
 Case Study 2:
ZF Friedrichshafen
ZF Friedrichshafen AG (ZF) is a leading supplier to Example 1: Predictive maintenance for gear-part ZF employed preliminary data-preparation steps Next, the data was transformed using the fast-
the automotive industry worldwide, with more than production machines before applying AI algorithms to predict the fourrier method to analyse measurements across the
230 locations and almost 150,000 employees. ZF probability of the honing ring failing, in order to entire frequency spectrum, with additional statistical
recently enhanced its AI activities by establishing Honing is a process that finishes an automotive improve data quality and usability. As typically seen features examined based on the fourrier-transformed
its own technology center for AI9 and expanding its gear’s surface to specific dimensions. A honed gear in the field of data science, data preparation was data. On the basis of all these extracted features,
partnership with Microsoft to establish one of the produces minimal noise in a transmission, provides the most time-consuming part of the project, but supervised machine-learning algorithms could
most comprehensive digital cloud platforms in the a high threshold for wear and tear and operates paid off later when applying predictive algorithms. be applied for binary classification, i.e. to predict
automotive industry10. These strategic initiatives smoothly. In the ZF facility, honing gears is an The time-series data first needed to be prepared by probabilities for ‘fracture’ and ‘no fracture’. Various
are helping ZF to achieve ‘Vision Zero’, a future for essential process in the production of quality using missing value imputation, applying a rolling algorithms such as support vector machines,
mobility with zero accidents and emissions. gear parts. window, labeling each time window into ‘fracture’ random forest decision trees and artificial neural
and ‘no fracture’ and then extracting meaningful networks, were applied to train prediction models
ZF is using AI in manufacturing to automate In a honing machine, the honing ring experiences features from the prepared time series. In the next based on historical data and to test their accuracy
and optimise processes, in order to improve wear and tear and is eventually replaced just before feature engineering step, individual attributes were based on cross-validation methods. A random forest
its production output and quality. These are the end of its lifetime, to prevent it fracturing during analysed for anomalies. For attributes with high model showed the best predictive performance, with
two examples of how the company is using AI operations. If the honing ring fractures earlier variance in the measurements, trends over time the strongest single predictor being the kurtosis of
to improve gearbox production at its plant in than expected, it not only damages gear parts were computed using linear regression models and the fourrier-transformed data. The AI solution that
Saarbrücken, Germany. severely, but also results in unplanned production the number and level of top and bottom outliers was developed was capable of accurately detecting
downtime that impacts the level of overall were analysed using density-based models such 99% of tool fractures.
equipment effectiveness (OEE). OEE measures as DBSCAN. Finally, attributes with very low
how close factories are to perfect production – i.e. variance were discarded in order to reduce data
Predictive maintenance manufacturing only good-quality products, as fast dimensionality.
for gear-part production as possible, with no stoppages.
For the company, a negative impact on OEE means
machines
availability loss, performance loss and quality loss.
To tackle such negative impacts, ZF aimed to
• Improved Overall Equipment predict the fracture of the honing ring using data
Effectiveness (OEE) from vibration sensors. However, gathering, storing
and analysing data were bigger challenges than
anticipated because:
• Better understanding of
the production process and • Data had to be collected from multiple honing
equipment performance machines
• Sensors had a wide range of frequencies creating
a lot of values for each measurement. This kind
of dataset is hard to analyse using traditional
methods
• There were significant data-storage requirements
because sensors measure in milliseconds

18 AI in Manufacturing AI in Manufacturing 19
 Example 2: Smart end-of-line testing in gearbox Afterwards, using an automated hyperparameter Today, most of ZF’s manufacturing data and
production tuning process, a variety of multiclass-classification systems reside on-premises. Over the coming years,
models with optimised parameters were trained to most of them are expected to move into cloud- Smart end-of-line testing
At ZF, end-of-line (EoL) testing is a process that find the best configuration for accurately predicting based systems. To prepare for this future scenario in gearbox production
ensures the highest quality gearboxes are produced. error classes and related rework actions. The and cover the current needs of AI use cases in
The EoL testbench measures the gearbox’s best-performing model was both accurate and fast manufacturing, ZF is implementing a hybrid platform
behavior using different sensors, including pressure, enough to allow real-time predictions for rework with some systems stored on-premises and others • Higher efficiencies in
temperature, vibration and sound, for which EoL recommendations on the testbenches. The model in the cloud. This is built on a similar technology rework
experts have defined thresholds using a complex was embedded into an end-to-end IT solution that and architecture for smooth transition between both • Reduced costs
manual process. The EoL testbench operator was enabled reliable daily operations with the AI solution worlds.
notified that a threshold had been exceeded in a on the shop floor. This augmented-intelligence • Better understanding of
failed test run and then addressed the problem solution enabled ZF to significantly enhance the EoL In addition, ZF is also enhancing its workforce product failures
based on his or her personal experience. However process, with potentially millions of euros saved per with the right analytics talent. It started by setting
AI-enabled assistance systems that support the year. up a central analytics lab within its IT innovation
decisions of testbench operators and rework department, to provide guidance around data-
personnel can significantly improve the efficiency Today, ZF still wants the EoL expert to be in the science topics and accelerate the development of
of machine hours, working time and material loop and manually validate whether clusters are analytics and AI use cases across ZF. The central
consumption. consistent and rework recommendations are correct. analytics lab followed a clear prioritisation process
In the future, this process can be fully automated, to focus on high-impact business problems.
ZF sought to improve its EoL process with a Big with the AI automatically and autonomously learning After conducting the first pilot projects in the
Data approach, i.e. using all available measurement new patterns and connections between error profiles central analytics lab, ZF realised that it had to add
data instead of only the first exceeded threshold. and rework recommendations. decentralised analytics teams to be closer to the
This was complemented by automating the process business domains and better address their specific
using AI to reduce human misjudgment. The foundations of scalable AI needs. The company set up multiple analytics teams
for specific domains, including one focused on
To identify patterns for rejected gearboxes using AI These examples are just two of many AI use cases manufacturing. This team now works hand-in-hand
solutions the following steps were taken: that exist across ZF’s multiple plants and divisions. with production and quality experts and receives
• Aggregating millions of measurement points However, in line with the main challenges for support from the central analytics team where
from the EoL testbench to relevant features implementing AI at scale (Figure 5), ZF has realised needed - for example, providing guidance on how
• Reducing the number of features using it needs to put in place the foundations for scalable to use specific components of the new platform. In
dimensionality reduction methods like PCA technology and a skilled workforce for AI. addition to the central and decentralised analytics
and tSNE teams, executive support and external partners
• Analysing the data using machine-learning The company is currently building a central ‘data providing their outside-in perspective were critical
clustering algorithms such as DBSCAN or lake’ and an enterprise platform for advanced success factors for ZF.
k-Means analytics and AI, to solve the increasing challenges
of Big Data and enable various AI use cases. To — Based on information from ZF Friedrichshafen AG, which does
not necessarily reflect PwC’s opinion.
The results from the algorithms were presented allow fast and scalable data processing, as well as
to EoL experts, combined with historical data on the integration of the latest innovative AI algorithms,
rework actions taken and their results. This allowed AI solutions are being developed based on open-
the EoL experts to assess the consistency of each source Big Data management technology and
cluster, based on error profiles and rework actions. machine-learning libraries.
For every consistent cluster, a label was applied so
that supervised machine-learning algorithms can be
trained to predict the specific label class.

20 AI in Manufacturing AI in Manufacturing 21
 Business Applications Talent and organisation

What are you using analytics for? Do you have the right skills and
structure?
• Identification and prioritisation of the
core business use cases • Data scientists
• Multi-departmental lenses • IT specialists
• Make use of available data • Business analytics translators
• UX designers

Six building blocks

• Center of competence / shared
services
• Chief Data Officer
• Advisors & start-ups

for successfully Data Process

implementing AI
Do you have access to the right data? Are your business processes optimal?

• Single source of truth • Data democratisation

• Data quality handling • Data governance (security, privacy
• External data sources and services etc.)
• Unstructured data extraction • Knowledge exchange
• Real-time data processing • Collaboration
Manufacturing companies such as ZF show that to take a • Data dictionatries • Agile development
lead in using AI solutions at scale, specific technological
investments and organisational decisions are required. In this
section, we look at the necessary business, technical and
organisational changes that will form the basis of successful
AI implementation. These fall into six categories:
Technology Culture

How can you build a data-driven

Do you have the systems and tools?
culture?
• Central data platforms
• Experimentation mind-set
• Cloud services
• Data-driven decision-making
• Big Data architecture
• Adoption of analytics tools
• Open source tools
• Trust in data and algorithms
• loT connectivity
• Machine Learning (ML) / AI
• User Interface (UI)

Business
Technical
Organisational

22 AI in Manufacturing AI in Manufacturing 23
 1. Business applications 2. Data

Business applications must be the starting point for an AI strategy. Manufacturers need to ask themselves Data is the foundation of any AI endeavor. As a result, reliable and accurate data acquisition, management
where they want to use AI in the near, mid and long term, and gather all the use cases across their and governance are key to effectively applying AI algorithms to the company’s processes. Sensor data from
operations in a structured way, grouping them together by function and prioritising projects based on connected factory equipment is a key data source in the manufacturing sector, so the production line and
expected business outcomes and effort of implementation. This helps to identify potential pilot and factories play a critical role in the data-acquisition process. Many enterprises are building ‘data lakes’ to
‘lighthouse’ use cases to lead the way in the company’s AI plans. Such early and high-profile use cases collect raw data from sensors, Manufacturing Execution Systems (MES), maintenance processes and quality
are a key success factor, because they help to drive adoption of AI across the business. Once the overall checks in one central place. They are also enriching it with external data to get a 360° view of both their
vision and specific use cases have been established, the technology and organisational requirements will production process and the product.
follow on.
These large data-management systems require proper data cataloging and data lineage to keep track of all
available information and flows, and to make the data easily accessible to a large number of users.
Manufacturers should start by mapping their main data objects, such as production facilities, machinery and
Insights from the field products, and the associated data sources to understand the data volumes, velocities and varieties they will
be dealing with. Furthermore, they need to define data quality metrics and systematically monitor these to
create awareness of their importance, which is often a major challenge for implementing AI.

To create value with AI, 3B is not only aiming for Insights from the field
the low-hanging fruit, but also the problems with
larger value potential. Addressing these helps to
convince leaders about the value of AI.

Dimitri Laurent, Global Operations Director and Glass Science

and Technology leader at 3B

Think big but start small. You have to think about

what it means to scale and adopt something AI isn’t a kind of magic that
across the entire enterprise. Building showcases
helps to create awareness and prove the value,
works by throwing data into
to get more people on board. algorithms. You need to get
Georg Gabelmann, Data Science Manager, IT Innovation at ZF your data right; the amount
and quality of data is key,
so companies must keep
track of all the details when
collecting and aggregating it.

Dimitri Laurent, 3B

24 AI in Manufacturing AI in Manufacturing 25
 3. Technology 4. Talent and organisation

Technology for Big Data, Analytics and AI is still rapidly evolving, often causing uncertainty for companies More data and new technologies also require people with specific analytical skills in the manufacturing
around what their future IT architecture should look like, and which tools and vendors they should field. Companies have started to hire data scientists over the past few years, but are still struggling to
choose. Manufacturers should start using a ‘functional reference architecture’, which maps which tools find the right organisational set-up to make effective use of these new skills alongside those of traditional
the company needs to gather, store, manage and process data, as well as the necessary analytics engineers.
and visualisation tools. Based on this functional architecture, they can define their requirements for an
evaluation of the most suitable technologies available in the market and also set out the technical and We typically see manufacturers on a transformational journey, starting with fragmented and uncoordinated
infrastructure set-up they want to use. use of data and AI experts, moving on to a more centralised organisational model (for example, an AI
center of excellence or lab, as described at ZF), which drives the AI maturity level of the organisation.
Some open-source technologies have become the de-facto standards for Big Data management (these In the final stage, AI capabilities may become decentralised again, as a fully integrated part of the
include HDFS11 and Spark) and AI (for example, TensorFlow in Python), and have been integrated into organisation.
most commercial Big Data and AI platform services (for example on Microsoft’s Azure cloud-computing
service or Amazon Web Services). These technologies allow for scalable distributed data processing and A central AI team should be comprised not only of data scientists, but should also include data engineers,
the training of sophisticated machine-learning models. These machine-learning models enable machines data stewards, solution architects and analytics translators. These central resources then collaborate with
to detect complex patterns and anomalies. teams from the various manufacturing functions to jointly develop and put into practice AI solutions for
specific use cases. Often, the central team also drives the development of a central AI platform, based on
Manufacturers should give special focus to ‘time series’ data processing and analytics capabilities to best the technologies described above, and this means hiring platform architects and DevOps staff.
handle data streams from sensors in a production environment.

Insights from the field

Insights from the field

Process engineers are

AI is often misunderstood. smart, but they speak a
Applied machine learning is different language than
AI, and data science is the data scientists. They
key. ZF’s first step towards often express requests
AI is machine learning. rather than specific needs,
Georg Gabelmann, ZF
which makes it hard for
data scientists to really
understand the problem
and come up with the best
solution.

Dimitri Laurent, 3B

26 AI in Manufacturing AI in Manufacturing 27
 5. Processes 6. Culture

To operate efficiently, enterprises need to define a minimum level of AI governance and processes. Finally, manufacturers need to enable a data and AI-driven culture. To meet this objective, they must build
An example of such a process is AI use-case pipeline management, to identify and evaluate new use trust in data and algorithms by not only educating the workforce about AI and its capabilities and value,
cases in the business on an ongoing basis, plan their agile development and ensure a smooth transition but also its risks and limitations. At the same time, they need to manage the concerns of their workforce
into operations. Another example is a defined development process for AI solutions. These typically and create a compelling vision of effective human-machine collaboration to counter fears of AI replacing
build on the cross-industry standard process for data mining (CRISP-DM) and define the phases of data thousands of jobs in manufacturing.
analysis and AI solution development.
Senior executives need to drive the change by showing commitment and ‘connecting’ with AI methods
Data and analytical model governance is a key capability for effectively operating AI solutions in and technologies. They can also foster innovation by supporting a fail-fast experimentation approach for
manufacturing. Enterprises need properly to define data ownership, access and security along with AI use cases. And while failure is condoned, success should be duly celebrated and communicated to
AI model performance criteria. Furthermore, they need to assure AI model fairness, explainability and drive adoption of AI across the organisation.
robustness, as well as consider ethics and regulation, as outlined in PwC’s Responsible AI Toolkit12.

Insights from the field

Insights from the field

At ZF, we are managing

use cases in a structured
process, from first
ideas through demand In the process engineering
clarification, use-case department, 3B is already
evaluation and prioritisation, seeing a change in culture
to the planning of concrete towards strongly leveraging
implementation projects. data and AI. Three years
This structured approach ago it was still hard to get
assures the best use of people excited about these
available resources for topics, but now they are
business value. coming up with ideas for
Georg Gabelmann, ZF
using AI by themselves.

Dimitri Laurent, 3B

28 AI in Manufacturing AI in Manufacturing 29
 Summary

AI is both a powerful source Companies such as ZF have

of disruption and a tool to shown that to take a lead in using
gain competitive advantage. AI solutions at scale, specific
Manufacturing firms that fail to technological investments and
recognise the importance of AI are organisational changes are required.
likely to lose their competitive edge.

The six building blocks for

While manufacturers have started to successful AI implementation
implement AI across the value chain, span business, technical and
companies are most commonly organisational factors:
using it in core functions such as
assembly and quality testing, product 1. Business application
development and engineering. 2. Data
3. Technology
4. Talent and organisation
5. Process
6. Culture
Manufacturers are implementing AI
primarily to:

1. Automate manual and cognitive While the current adoption rate in

tasks manufacturing is low, the prevalence
2. Assist in making decisions or of AI is expected to increase
taking action significantly by 2030.
3. Augment decision-making
through continuous machine
learning

The organisations that have made

the most progress in digitising core
business processes are also leading
in AI adoption.

30 AI in Manufacturing AI in Manufacturing 31
 Authors and
Contributors
Dr. Bashar El-Jawhari, Partner, Middle Koen Cobbaert, Director, Belgium
East Koen Cobbaert is a director in the
Dr. El-Jawhari leads the Industry Management Consulting practice of PwC
4.0, procurement and supply chain Belgium. His focus has always been on
competency with a focus on the optimising the supply chain, with a strong
Resources and Industrials (R&I) industry focus on the tactical and strategic level.
sector in the ME. He has deep expertise Over the years, Koen has built expertise in
in manufacturing, cost reduction, the application of analytics and machine
productivity improvement and operating learning in supply chain and operations
models. In addition, Dr. El-Jawhari is a
certified Master Black Belt in Lean Six
Sigma and was North America Director
management. Koen holds a Master degree
in Industrial Management from KULeuven. References:
of Supplier Development and Quality Andreas Odenkirchen, Senior Manager,
Assurance and Global Manager of Germany
Order Fulfillment at two Fortune 500 Andreas is a Senior Manager in PwC’s
heavy equipment manufacturers. He has technology consulting practice, focused
advised multiple academic departments on designing and implementing data-
and research centers in the US. Dr. El- driven and AI-enabled solutions for
Jawhari received his PhD and Masters industrial enterprises. Prior to joining PwC,
from the University of Wisconsin- Andreas worked as a Data Scientist and
Madison. Strategy Consultant at IBM’s Global Center
of Competence for advanced analytics.
Sachin Halbe, Director, Middle East PwC’s Global Digital Operations Study, 2018. Digital Champions.
1

Sachin is a member of the Industry 4.0 2

Definition from WhatIs.com, Rouse, M. (Ed.). (n.d.). What is AI (artificial intelligence)?
practice in PwC ME. He has extensive (https://searchenterpriseai.techtarget.com/definition/AI-Artificial-Intelligence)
experience in digital manufacturing and
supply chain management. He has led
3
PwC’s AI analysis, 2017. Sizing the prize – What’s the real value of AI for your business and how can you capitalize.
multiple operational improvement and 4
PwC’s technical report on AI, 2018. The macroeconomic impact of artificial intelligence.
business transformation projects in GCC,
Asia Pacific and USA across sectors such 5
PwC’s Global Digital Operations Study, 2018. Digital Champions.
as energy, healthcare, transportation and 6
PwC’s technical report on AI, 2018. The macroeconomic impact of artificial intelligence.
industrial products. Sachin holds an MBA
from Kellogg School of Management, 7
PwC’s Global Digital Operations Study, 2018. Digital Champions.
USA.
8
PwC’s AI analysis, 2017. Sizing the prize – What’s the real value of AI for your business and how can you capitalize.
Martin Whyte, Director, Germany 9
ZF’s press release, 2019. ZF Establishes Its Own Technology Center for Artificial Intelligence and Cybersecurity.
Martin Whyte is a Director within PwC’s (https://press.zf.com/press/en/releases/release_3841.html)
technology consulting practice leading
the PwC Europe digital operations
10
ZF’s press release, 2019. ZF expands Partnership with Microsoft to Develop Digital Services.
(https://press.zf.com/press/en/releases/release_3743.html)
analytics team. He focuses on
envisioning, designing and implementing 11
HDFS is the primary data storage system used by Hadoop applications. It employs a NameNode and DataNode
data-driven solutions in manufacturing, architecture to implement a distributed file system that provides high-performance access to data across highly scalable
supply chain and procurement. Prior to Hadoop clusters.
this Martin was in different positions with 12
PwC’s Responsible AI toolkit, 2019. A practical guide to Responsible Artificial Intelligence (AI).
other technology consulting firms always
focusing on data analytics solutions.

32 AI in Manufacturing AI in Manufacturing 33
About the Global Manufacturing and Industrialisation Summit (GMIS)

The Global Manufacturing and Industrialisation Summit (GMIS) was established in 2015 to build bridges between
manufacturers, governments and NGOs, technologists, and investors so that they can harness the transformative
power of the Fourth Industrial Revolution. A joint initiative by the United Arab Emirates and the United Nations Industrial
Development Organization (UNIDO), GMIS is a global platform that presents stakeholders with an opportunity to shape
the future of the manufacturing sector and contribute towards global good by advancing some of the Sustainable
Development Goals.

The first two editions of the Global Manufacturing and Industrialisation Summit were held in Abu Dhabi, United Arab
Emirates, in March 2017, and Yekaterinburg, Russia, in July 2019, respectively, with each edition welcoming more than
3,000 high-level delegates from more than 40 countries.

GMIS 2020, the third edition of the Global Manufacturing and Industrialisation Summit, will be held alongside Hannover
Messe, the world’s largest industrial trade fair, between 20 and 21 April 2020 in Hannover, Germany, and will focus on
glocalisation in pursuit of inclusive and sustainable global value chains.

You can find more about GMIS at:

www.gmisummit.com

Follow GMIS on:

Facebook: @GMISummit Instagram: @gmisummit

Twitter: @GMISummit LinkedIn: GMIS - Global

Manufacturing & Industrialisation
Summit

34 AI in Manufacturing
www.pwc.com/me
©2020 PwC. All rights reserved

At PwC, our purpose is to build trust in society and solve important problems. We’re a network of firms in 157 countries with over 276,000 people
who are committed to delivering quality in assurance, advisory and tax services. Find out more and tell us what matters to you by visiting us at
www.pwc.com.
Established in the Middle East for 40 years, PwC has 22 offices across 12 countries in the region with around 5,600 people. (www.pwc.com/me).
PwC refers to the PwC network and/or one or more of its member firms, each of which is a separate legal entity. Please see www.pwc.com/
structure for further details.
36 AI in Manufacturing