擴展制造業(yè)AI應用報告

1、Executive SummaryAI in manufacturing is a game-changer. It has the potential to transform performance across the breadth and depth of manufacturing operations. However, the massivepotential of this new Industrial 4.0 era will only be realized if manufacturers really focus their efforts on where AI c

2、an 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 andanalysis. We analyzed 300 leading global manufacturers from four key segmen

3、ts automotive, industrial manufacturing, consumer products, and aerospace & defense - to understand the focus of their AI initiatives. We also spoke with 30senior industry executives, all of whom are involved in their organizations AI initiatives. Finally, we analyzed 22 AI use cases in manufacturin

4、g operations. These use cases were spread across seven broad functional areas, from inventory management through to production and quality control.The key findings that emerge from this analysis include:Europe is leading the way, with more than half of its top manufacturers implementing at least one

5、 AI use case in manufacturing operations (within Europe, Germany leads the pack, with 69% of its manufacturers implementing AI). Europe is then followed by Japan (30% implementing) and the US (28%).Three use cases stand out in terms of their suitability forkickstarting a manufacturers AI journey:Int

6、elligent maintenanceProduct quality controlDemand planningThese use cases have an optimal combination of several characteristics, that make them an ideal place to start:Clear business value/benefitsRelative ease of implementationAvailability of data e.g., performance data from machines and equipment

7、 for intelligent maintenance, pictures and videos capturing finished products for quality, etc.Availability of AI know-how and/or existing standardized solutionsThe opportunity to add features that aid visibility and explainability, allowing employees to understand how decisions are reached and easi

8、ng adoption by operational teams.In the final section of this report, we look at the critical success factors for scaling these use cases in operations:Deploy successful AI prototypes in live engineering environmentsThe first step in achieving scale involves bringing the AI prototype up to speed wit

9、h processing data in real time from the shop floor/production environment. To automate the collection of real-time, live data, the prototype needs to be integrated with legacy IT (such as MES and ERP) and industrial internet of things (IIoT) systems.Put down solid foundations of data governance and

10、AI/ data talentTo create a robust foundation for scale, and to encourage new implementations, manufacturers should design a data governance framework that defines critical processes related to the generation, management, and analysis ofdata. In addition, they need to deploy a data & AI platform a ce

11、ntral platform to store and analyze data using AI and to make it available to issue-specific AI applications. Alongside governance and platform, talent will also be a key building block, including manufacturing-specific expertise in AI, data science, and data engineering.Scale the AI solution across

12、 the manufacturing network Once the AI platform is ready, AI applications can be deployed and made available across multiple sites/factories. Performance needs to be continuously monitored for value generated, output quality and reliability.2Scaling AI in Manufacturing Operations: A Practitioners Pe

13、rspective4What is AI?Artificial intelligence (AI) is a collective term for the capabilities shown by learning systems that are perceived by humans as representing intelligence. Today, typical AI capabilities include speech, image and video recognition, autonomous objects, natural language processing

14、, conversational agents, prescriptive modeling, augmented creativity, smart automation, advanced simulation, as well as complex analytics and predictions.In the context of manufacturing operations, we found most AI use cases centered around the following technologies:Machine learning: The ability of

15、 algorithms and code to use data and automatically learn from its underlying patterns without being explicitly programmed.Deep learning: An advanced form of machine learning that uses artificial neural networks to analyze and interpret images and videos.Autonomous objects: Artificial agents such as

16、collaborative robots or autonomous guided vehicles that can handle a task given to them on their own3sg Operationng AI in Manufact4ScalierspectiveAI holds strong potential across the manufacturing value chainA range of leading organizations are using artificial intelligence in their manufacturing op

17、erations leveraging the benefits it offers over traditional methods: Bridgestone, the Japanese tire manufacturer, introduced a new tire assembly system “EXAMATION” to improve the quality of its tires. This system provides automatic control of quality assurance in the production process an approach t

18、hat was previously dependent on human skills and judgement. This system is equipped with anartificial intelligence tool that uses sensors to measure the characteristics of individual tires based on 480 quality items. EXAMATION uses this information to control production processes in real time, ensur

19、ing that all components are assembled under ideal conditions. This system helps promote ultra-high levels of precision in tire manufacturing, resulting in an improvement of more than 15% in uniformity when compared to a conventional manufacturing process.1 Danone uses machine learning to predict dem

20、and variability and planning. The new capability improved its forecasting process and led to more efficient planning between different functions, such as marketing and sales. It has led to a 20% reduction in forecast error and a 30% reduction in lost sales.2These examples provide compelling evidence

21、 as to why artificial intelligence is being adopted across manufacturing sectors. And, as Figure 1 shows, AI offers applications across the breadth and depth of manufacturing operations, from product development to quality control.To discover all potential applications of AI in manufacturing operati

22、ons, we researched 300 global manufacturers.The 300 represented the top 75 global organizations in four manufacturing segments: automotive, industrialmanufacturing, consumer products, and aerospace & defense. We also conducted in-depth interviews with 30 senior executives from these segments to unde

23、rstand how theyare implementing and scaling AI (please see the research methodology at the end of the report for more details).30%reduction in lost sales achieved by Danone by using machine learning to predict demand General Motors “Dreamcatcher” system uses machine learning to transform prototyping

24、. The solution was recently tested with the prototyping of a seatbelt bracket part, which resulted in a single-piece design that is 40% lighter and 20% stronger than the original eight- component design.3Figure 1: AI has potential across the breadth and depth of manufacturing operationsDemand planni

25、ngAI enables organization to optimize product availability by decreasing out of stocks and spoilage. AI can also help with getting a better understanding of sales patterns.LOral uses AI algorithms to predict demand based on a wide variety of data gathered from social media, weather, and nancial mark

26、ets.4betterInventory ManagementAI can be used to get a understanding of inventory levels enabling organizations to plan ahead and avoid stock-outs ProductionTAKT can be reduced by using AI to streamline manufacturingprocesses, improving throughputMitsubishi Electric uses AI to automatically adjust r

27、ate, speed, acceleration, etc. of the industrial robots leading to the time reduction to 1/10th of conventional methodSafety AI is used to get a better understanding of risk factors within the shop oor and can helpsafer operationsSource: Capgemini Research Institute analysis.Product development/R&DA

28、I enables organizations to expediate product development and R&D by reducing the test times and driving more concrete insights from customer data and demandsIntel is using big data and AI platforms to create tests for hard to validate functionalities improving the targeted coverage by 230 x compared

29、 to standard regression testsProcess control AI can help organizations optimize processes to achieve production levels with enhanced consistency, economy and safety Unilever uses AI to inuence operations by predicting outcomes and improving eciency levels to optimise output.Quality controlProduct qu

30、ality inspections bring uniformity and eciency in quality control, using image-based and sensor-based processes.Bridgestone uses AI to promote high-level of precision in tire manufacturing, resulting in an improvement of more than 15% over traditional methods MaintenanceUsing AI, organizations can p

31、redictand prepare for asset failure, reducing (or even avoiding) downtime.General Motors uses computer vision to analyse images from robot mounted camerasto spot early signs of failing robotic partEnergy managementAI allows organizations to gain deeper insights in the energy use throughout the produ

32、ction process, resulting in reduced bills and more sustainable productionEurope leads AI deployment in manufacturing operationsOur research of 300 major manufacturers found that Europe leads all major manufacturing countries in implementing AI in manufacturing operations. As Figure 2 shows, more tha

33、n half of the European cohort are implementing AI solutions, with Japan and the US following in second and third. In Europe, Germany leads the way, with 69% of its manufacturers implementing at least one AI use case in manufacturing. Germany is followed by France (47%) and UK (33%). A growing realiz

34、ation of the scope of AI coupled with support from governments across countries is likely helping manufacturing companies in adopting AI in operations.Figure 2: Europe leads in implementing AI in manufacturing operationsTop global manufacturers implementing AI by country/regionEurope51%Japan30%Unite

35、d States28%Korea25%China11%Percentages represent the share of organizations in each country/region that are implementing at least one AI use case, out of total organizations from that country/region in the top 300 global manufacturers (top 75 from each of the four focus sectors of the study). Number

36、 of organizations in top 300 global manufacturers that are in: US = 89, Europe = 73 (including France = 20, UK = 15, Germany = 13), Japan = 44, China = 32, and Korea = 12. Together these countries represent manufacturers with $3.8 trillion of annual revenues among the top 300 global manufacturers.So

37、urce: Capgemini Research Institute analysis.Three use cases to kickstart AI in manufacturing operationsAs part of our research, we closely analyzed 22 unique use cases (please refer to appendix) against a range ofcharacteristics, from whether there is a clear benefits case to the availability of rel

38、evant data. The aim was to identify the best use cases for organizations to begin their journey, and the assessment characteristics included:Clear business value/benefits. Focusing on use cases where benefits are easily identified and quantified including in financial terms makes building the busine

39、ss case easier. For example, reducing downtime, improving OEE, reducing product defects and reducing inventory.Relative ease of implementation. Focusing on less complex use cases leads to shorter payback periods (usually in a matter of a few months) and a higher return on investment. This further st

40、rengthens the business case.Availability of data. For example, performance data from machines and equipment for intelligent maintenance, pictures and videos capturing finished products for quality, etc. Also, the data must contain enough occurrences of the particular issue that you want to predict i

41、n the future, such as a fault or substandard quality. When there are not enough occurrences, simulation data can be used. In many cases, the necessary data will come from machines equipped with IoT sensors.Availability of the AI know-how, existing standardized solutions, and IT infrastructure requir

42、ed to deploy AI in production, at scale. Based on the popularity of these use cases, many vendors have been quick to translate those into AI offerings, which they bundle into packaged solutions, and which can be customized to specific use cases.The opportunity to add features which aid visibility an

43、d explainability, allowing employees to understand how decisions are reached and easing adoption by operational teams.We believe that three of these 22 use cases serve as a good starting point for manufacturers to focus their efforts, as they possess an optimal combination of the factors listed abov

44、e.These three use cases are:Intelligent maintenanceProduct quality inspectionDemand planning.Intelligent maintenanceIntelligent maintenance of plant machinery and equipment is the “l(fā)ow hanging fruit” of AI adoption across industries. When applied to bottleneck resources, its ROI can be significant,

45、as we saw in the industry examples earlier. Beyond minimizing downtime, AI-enabled intelligent maintenance also reduces maintenance costs and increases productivity. It is relatively easy to implement, given availability of good quality dataand the expertise to analyze it in business context. Severa

46、l integrated solutions are available, both from specialized startups and large players. Intelligent maintenance adds value in a few variants:Predicting when machines/equipment are likely to fail and recommending optimal times to conduct maintenance (condition-based maintenance).Analyzing root causes

47、 and identifying drivers of machine downtime to prevent future breakdowns. For instance, General Motors analyzes images from cameras mounted on assembly robots, to spot signs and indications of failing robotic components with the help of its supplier. In a pilot test of the system, it detected 72 in

48、stances of component failure across 7,000 robots, identifying the problem before it could result in unplanned outages. According to the Robotic Industries Association, the cost of just one minute of production-line downtime for a company like General Motors can be as high as $20,000.Correlating the

49、impact of events and issues on machine efficiency and breakdowns. For instance, Volvo uses large- scale datasets in its Early Warning System. Every week, the system analyzes over one million events that occur during machine operations, such as temperature increases or abnormal pressure readings. Thi

50、s allows the organization to assess their impact on breakdown and failure rates.Minimize production losses and maximize OEE (overall equipment effectiveness)Ensuring you have the “right alerts at the right time”. This is to avoid too many false positives that would render the solution unusable (a pi

51、tfall with many intelligent maintenance solutions). It also allows you to factor in the“time to action”. In other words, when the alert should be raised to ensure the necessary steps can be taken to avoid a predicted failure.The head of digital innovation of a large European engineering firm, outlin

52、es how intelligent maintenance creates benefitsin multiple ways. “We have implemented predictive maintenance, using AI for our tuning machines, which run 24/7,” he explains. “Previously, every unplanned stoppage resulted in lost production time the time it took to complete maintenance and get the ma

53、chine back up. Now, we have dataon these past failures and their possible causes and can predict when the next issue is likely to occur. This allows us to save not only production hours lost, but also unplanned maintenance cost and man-hours, leading to big savings.”Figure 3: Using AI for intelligen

54、t maintenance in manufacturing1The AI system istrained using data from pastmachine failures6Actual data from failures is fed back into the AI system toimprove its accuracy in future5Alerting service personnelwhen fault probability rises over a thresholdIdentifying key drivers of equipment breakdown

55、out of a large number of possible causesOptimal times to conduct maintenance to minimize production losses2Sensors from plant equipment3continuously collect data onvarious operational parameters that aect machine performanceThis data is collected/uploaded in data storage4The AI-based system analyzes

56、 this data and makes a varietyof recommendations while improving correctness of its own predictionsExpected benetsHigh uptime and availability, leading to high overall equipment eectiveness (OEE)Low maintenance costAvoiding loss of production Low spare part inventorySource: Capgemini Research Instit

57、ute analysis.The power of this approach can be seen in the example of a leading automotive manufacturer that was struggling to reduce machine stoppages and minimize productionlosses.9 It wanted to identify machines and production lines in advance where faults are likely to occur, jeopardizing sales

58、and final deliveries to customers. With an AI-enabled predictive maintenance solution, it was able to accurately identify machines and lines that were most likely to fail andtake proactive remedial action. For instance, in a month where significant failures were anticipated, intelligent maintenance

59、allowed 300 additional cars to be produced. This was inaddition to the output that might otherwise have been lost because of downtime and maintenance.To arrive at these predictions, the organization deployed a supervised machine learning model. The model analysedhistorical reasons for failure, focus

60、ing on information captured during stoppages. For example, it collected data on the duration of outages and the reasons for failure. Figure 4 shows how the AI model works in action predicting a possible loss of approximately seven cars at Line “PUR”, Station “CT229”, in the first shift.Figure 4: Pre