1、Scaling AI in Manufacturing Operations: A Practitioners Perspective Executive Summary AI in manufacturing is a game-changer. It has the potential to transform performance across the breadth and depth of manufacturing operations. However, the massive potential of this new Industrial 4.0 era will only
2、 be realized if manufacturers really focus their efforts on where AI can add most value and then drive the solutions to scale. To understand whether organizations are focusing on the most promising use cases, and then achieving scale with the solution, we have undertaken significant research and ana
3、lysis. We analyzed 300 leading global manufacturers from four key segments automotive, industrial manufacturing, consumer products, and aerospace that it can be prototyped, i.e. its development and implementation process are made standardized and repeatable; and the prototype is now ready to be depl
4、oyed at scale. Figure 10: Recommendations to scale AI in manufacturing operations Source: Capgemini Research Institute analysis. I. Deploy successful AI prototypes in live engineering environments a. Implement the AI application to process real-time data from the shop floor So far, the implementatio
5、n of the use case as a pilot/POC has happened in a sandbox or controlled environment. As a result, the system has been trained and tested on a limited set of data. Before the AI application can begin to handle many possible scenarios, it needs to be trained to a level where its accuracy is sufficien
6、tly high for a production environment. Siddharth Verma global head and VP, IoT Services, Siemens pointed out that organizations need to be ready for a few false starts (a lesson they learned during the initial days of an intelligent maintenance implementation). “We used AI to predict failures in fan
7、s that handle exhausts from an autoclave,” he explained. “In the early days, when the accuracy of the system was low, it predicted a few failures which turned out to be false alarms. At these points, it is important to remind everyone that it is a prediction which has a probability 20Scaling AI in M
8、anufacturing Operations: A Practitioners Perspective of being right or wrong. As accuracy improved, the system was able to predict many failures in advance and saved a lot of cost and downtime, proving its worth.” Testing on real time data not only builds accuracy, it also ensures the solution is up
9、 to the standards demanded by a factory floor environment. b. Create robust integrations with legacy IT systems and industrial internet of things (IIoT) systems Our research on scaling AI in the automotive industry found that integration issues with existing systems and tools is the biggest technolo
10、gical challenge standing in the way of at-scale AI.19 We believe that organizations can overcome this hurdle by proactively building in legacy IT integration as a key ingredient of the scaling process. Legacy manufacturing systems such as enterprise apps for product lifecycle management, manufacturi
11、ng execution systems (MES), and enterprise resource planning (ERP) systems have multiple data sources. These can be valuable inputs to the AI applications. Neeraj Tiwari, director manufacturing JV Organization at Fiat Chrysler China, explains how standardization of equipment and systems across the o
12、rganization makes integration easier. “We have a centralized process for purchase of equipment, their subsystems, and associated software. This brings a level of standardization and makes integrating AI applications much easier and results in far fewer issues. Not only this, we are able to easily re
13、plicate the applications to new plants with a fraction of efforts, say 15 to 20% of the effort required for the first implementation. So, lessons learnt in continuous improvement drive are horizontally deployed.” In addition to these data sources, AI systems will sometimes need more granular data. T
14、his would come directly from machines and equipment, such as IoT systems. Industrial IoT systems are increasingly emerging as a key source of data for AI applications, with two types of integrations needed: 1. Running AI computations at the edge, also known as “edge intelligence” (for instance, in t
15、he plant, in the assembly line, close to the asset) for making immediate tactical decisions 2. Collecting and processing IoT data in a central storage repository (for instance, strategic learning for optimization purposes). II. Invest in laying down strong foundations: governance, platform and talen
16、t a. Design a data governance framework and build a data Scaled implementation here refers to ongoing implementation across all sites/enterprise wide with full scope and scale. 19. Capgemini Research Institute, “Accelerating automotives AI transformation,” March 2019. 20. BMW, press release, “Fast,
17、efficient, reliable: Artificial intelligence in BMW Group Production,” July 2019. 21. DriveSpark, “Nissans Idea: Let An Artificial Intelligence Design Our Cars”, September 2016. 22. Food Drink & Franchise, “How Carlsberg is using AI to help develop new beers,” January 2018. 23. PR Newswire, “Canon S
18、howcases Solutions to Help Enterprises Embrace Lean Document Management Processes at the 2018 Association for Manufacturing Excellence,” October 2018. 24. Customer Think, “How AI is reshaping the food processing business, January 2019. 25. Micron Insights, “Case Study: Micron Uses Data and Artificia
19、l Intelligence to See, Hear and Feel,” November 2018. 26. Thales Aerospace Magazine, “Assuring air safety and reliability The Shift to Predictive Maintenance,” May 2019. 27. Enterprise IOT Insights, “Nokia claims first “real-world” 5G smart factory trial with Telia and Intel,” April 2018. 28. Capgem
20、ini Research Institute focus interview. 29. Boeing, Innovation Quarterly, “A collaborative effort in machine perception research aids airplane inspection,” November 2018. 30. Stottler Henke, “Aurora,” accessed November 2019 References 25 Research Methodology We interviewed over 30 senior executives
21、from the manufacturing sector, drawn from the following segments: 1. Industrial manufacturing 2. Automotive 3. Consumer products 4. Aerospace & defense These executives fulfilled one of four specific roles and all were involved with their organizations AI initiatives: 1. Department/function head in
22、one or more manufacturing plant(s) e.g., maintenance, production, quality 2. Plant leadership (plant manager/director) 3. Director/VP operations (corporate/multi-country responsibility) 4. AI heads/ heads of innovation/ chief digital officers We also conducted extensive secondary research, examining
23、 the AI initiatives being tested and implemented by the top 75 organizations in each of the four segments (by annual global revenue). We analyzed company websites, annual reports, press releases and articles, investor and media presentations, earnings calls transcripts, leadership interviews, as wel
24、l as official social media information. Our aim was to enlist as many AI use case implementations as possible. We found 102 AI implementations among these 300 organizations, with some having several implementations underway. We found that many of the 102 had significant similarities. We were therefo
25、re able to identify 22 unique use cases among them. These use cases belong to different functions from demand planning to maintenance (see appendix for details). Chief Digital Offi cer/AI Heads Head of Operations Departmental Managers Plant Managers 29% 29% 29% 13% Executives interviewed by designat
26、ion 26Scaling AI in Manufacturing Operations: A Practitioners Perspective FunctionUse cases Product development/R&D New product development Product validation in R&D Product enhancement Demand Planning Demand planning/forecasting Inventory Management Order optimization Standardized communication wit
27、h suppliers using NLP Inventory planning Process Control Real time-optimization of process parameters Optimize equipment changeover Production Optimizing overall productivity in the product line Reduction in TAKT time Computer vision for product identification Layout planning Collaborative robots (c
28、obots) Quality Control Product quality inspection Predicting final product quality Maintenance Intelligent maintenance Energy management Spotting anomalies in communications network Worker safety Scrap/wastage reduction Increasing equipment efficiency Appendix All prominent AI use cases in manufactu
29、ring operations Source: Capgemini Research Institute analysis. N = 300 largest organizations in industrial manufacturing, automotive, aerospace & defense, and consumer products. 27 Automotive Product quality inspection The BMW Group uses AI to evaluate component images from its production line, allo
30、wing it to spot, in real time, deviations from the standard.20 Intelligent maintenance General Motors uses a computer vision system to analyze images from cameras mounted on assembly robots, to spot early indications of a failing robotic part. In the pilot phase, the system was mounted on 7,000 robo
31、ts and was able to detect 72 instances of component failure. This helped them prevent massive potential downtime costs, which can reach at least $20,000 per minute for an organization of General Motors size. Product validation Nissan is in test phase of using AI to design its cars. To comply with ne
32、w regulations (e.g., new safety specifications), AI is used to modify an existing car, while keeping in mind the knock- off effect of the modification.21 Consumer products Product enhancement Carlsberg has implemented a “Beer Fingerprinting” project, which is developing sensors that can differentiat
33、e between various flavours of beers. Today, no technology can effectively and quickly discriminate between flavors. Carlsberg processes the resulting data via AI and uses the information to develop new beers and enhance the quality of existing beers.22 Product quality inspectionCanon is using AI in
34、product quality. Its solution automatically identifies defects by analyzing images of the inspected parts.23 New product development Kelloggs has launched an AI system that helps customers decide which recipe should be chosen to make a product of their choice on their website Bear Naked. This techno
35、logy helps the consumer giant come up with final products that the consumers actually want.24 Most implemented use cases by sector 28Scaling AI in Manufacturing Operations: A Practitioners Perspective Industrial Manufacturing Product quality inspection The process that Micron uses to produce memory
36、technologies on its silicon wafers is highly complex and precise, with potential defects largely invisible to the human eye. As a result, there is a high potential for errors in the process. Micron spots any defects using computer vision, which has improved manufacturing efficiency and effectiveness
37、.25 Intelligent maintenance Thales SA, a leading supplier of electronic systems to aerospace and defense companies, collects historic and current data on parts failures. Drawing on the data, it has developed an AI algorithm to predict potential problems. This can be used to identify when parts might
38、 fail, allowing it to make proactive maintenance decisions for its customers. 26 Real time optimization of process parameters Nokia launched a video application that uses machine learning to monitor an assembly line process in one of its factories in Oulu, Finland. It alerts the operator of inconsis
39、tencies in the process so that issues can be corrected in real time.27 Aerospace and Defense Intelligent maintenance Airbus is using AI to anticipate when its trimming machines are going to fail. Airbus uses this information to determine the root cause issue and plan maintenance, thereby avoiding ex
40、pensive downtime.28 Product quality inspection Boeing is using computer vision for aircraft inspection. Ground crews scan various parts of an airplane with an augmented reality headset and other necessary hardware, capturing images. These images are then transmitted to a back-end processing platform
41、 where computer vision techniques are deployed to identify if certain abnormalities exist.29 Real time optimization of process parameters Bombardier has partnered with Aurora to strengthen its resource planning and scheduling using its AI enabled tools. As a result, Bombardier can schedule its airpl
42、ane assembly operations more quickly and is able to handle production rate changes more effectively.30 29 Abhishek Jain Consultant, Capgemini Research Institute Abhishek is a consultant at the Capgemini Research Institute. He loves to explore how disruptive technology is changing consumer behavior a
43、nd businesses. Jerome Buvat Global Head of Research and Head of Capgemini Research Institute Jerome is head of the Capgemini Research Institute. He works closely with industry leaders and academics to help organizations understand the nature and impact of digital disruption. Amol Khadikar senior man
44、ager, Capgemini Research Institute Amol is a Senior Manager at the Capgemini Research Institute. He leads research projects on key digital frontiers such as artificial intelligence, data privacy, and the future of work to help clients devise and implement data-driven strategies. Yashwardhan Khemka M
45、anager, Capgemini Research Institute Yash is a manager at the Capgemini Research Institute. He likes to follow disruption fueled by technology across sectors. The authors would like to specially thank Subrahmanyam Kanakadandi and Jeff Theisler for their contributions to the report. The authors would
46、 also like to thank Jean-Pierre Petit, Fabian Schladitz, Sbastien Guibert, Manuel Chareyre, Francois Calvignac, Jacques Bacry, Philippe Ravix, Philippe Sottocasa, Jacques Mezhrahid, Alexandre Embry, Moise Tignon, Balaji Thiruvenkatachari, Yves R Vergnolle, Lukas Birn, Fabian Schladitz, Mihir Punjabi
47、, Ron Tolido, Ingo Finck, Sanjeev Gupta, Patrick Sitek, Vivek Kotru, Bobby Ngai, Karen Brooke, Melanie Daubrosse, Adam Bujak, Valerie Perhirin, Mark Kirby, Sandhya Sule, Eric Reich and Chloe Duteil. The Capgemini Research Institute is Capgeminis in-house think tank on all things digital. The Institu
48、te publishes research on the impact of digital technologies on large traditional businesses. The team draws on the worldwide network of Capgemini experts and works closely with academic and technology partners. The Institute has dedicated research centers in India, the United Kingdom, and the United
49、 States. It was recently ranked Top 1 in the world for the quality of its research by independent analysts. Visit us at About the Capgemini Research Institute Anne-Laure Thieullent Managing Director, Artificial Intelligence & Analytics Group Offer Leader Anne-Laure leads the Artificial Intelligence & Analytics Capgemini Group Offer (Perform AI), one of Capgeminis 7 Group Portfolio Priorities. She advises Capgemini clients on how they should put AI technologies to work for their organization, with trusted AI at scale services for busin