The rise of AI has led to dynamic and accelerating societal change. In the 2010s, practical research and development of AI, especially deep learning, made dramatic advances. Now, in the 2020s, we are entering the stage in which AI is extensively used in business. This article explains how AI is being used in the infrastructure and data services that Toshiba promotes.

 The world is struggling with countless problems, among them global warming, frequent natural disasters, and a global pandemic. The Toshiba Group is working to help solve these problems through a variety of initiatives.

 At the same time, society itself is also undergoing dynamic changes driven by technological innovation. With the coming of the 2020s, companies have begun accelerating the use of digital technologies in their organizations and business models. This is changing the face of society through so-called Digital transformation(DX). To meet the needs of this new era, the Toshiba Group is providing infrastructure and data services based on Cyber-Physical Systems(CPS) that leverage IoT, AI, and cloud technology.

 The Toshiba Group has developed extensive expertise and experience through its many years of infrastructure business. Toshiba uses various state-of-the art technologies to create these CPS. IoT technologies gather data from physical components and systems. This data is then analyzed in cyberspace, and AI technologies convert the results into added value. Through this process, Toshiba creates new services that help solve societal problems.

 AI technologies are indispensable for the advancement of CPS. Productivity improvements are achieved by performing advanced image inspections, defect cause analysis, process optimizations and so on. This is accomplished by collecting IoT data and using AI to analyze it. AI technologies also make it possible to gain various insights by identifying people and behavior, creating risk and demand forecasts, and detecting signs of potential failures. The new knowledge discovered through the use of AI is then fed back into the physical world, producing more advanced CPS.

 AI-based productivity improvements and insights help companies adapt to the new normal, accelerate their implementation of digital technologies, and contribute to the resilience of their business operations (Fig.1).

 A great deal of attention has been paid to AI in recent years, but the field's history is a long one. The term "artificial intelligence" was coined in 1956 at the Dartmouth Conference. This was followed by the First and Second AI Booms. The appearance of deep learning technologies in the 2010s caused a rapid acceleration in the evolution of AI. AIs were developed that were capable of beating human competitors at Go and in game shows. AIs become well-known and widely discussed. Natural language processing technologies also made dramatic advances, and are now used in many tools that people use every day, such as smartphone speech recognition and chatbots.

 It has been over a decade since the AI technology innovation brought about by deep learning. Now, with the coming of the 2020s, AI has advanced to the stage of extensive use in business (Fig. 2).

 The strengths of Toshiba's AI are being leveraged in social infrastructure and in a wide range of industrial fields, such as the energy and manufacturing fields. Our AIs are being used in smart factories, electrical plants, water treatment facilities, railroads, broadcast media, and more.

 However, deploying AI in industrial fields is no easy task. There are four reasons for this. First, the systems used in these fields are large and complex, and they handle massive amounts of operation data and intricately related parameters. Second, these fields have many mission critical systems which cannot be allowed to stop and which must rapidly recover from any outages. Third, they suffer from large amounts of missing data and have little data regarding abnormalities. Machine learning must therefore be applied to normal data alone, or to small amounts of data regarding abnomalites, in order to detect potential abnormalities. Fourth, a great deal of work in industrial fields is performed through collaboration between people and objects, so data must be collected not only for systems but also for the people that work with them.

We are solving these problems through the use of AI quality management and the development of systems in which AI constantly provides value by evolving in response to changes in external environments and business needs. These are made possible by the knowledge we have accrued regarding individual industrial fields through Toshiba's long years of experience, our track record of developing diverse AI technologies and solutions, and our expertise in building mission critical systems (Fig. 3).

 Let's look at Toshiba's AI quality management technologies and the initiatives it is using to create a platform that supports the development and operation of AI based on these quality management technologies. These technologies, and this platform, are essential for the ongoing use of AI.

 Toshiba has formulated unique AI quality assurance guidelines focused on five areas: (1) Creating highly robust AI models, (2) Using sufficiently high quality data and sufficient amounts of data, (3) Using flexible and responsive development processes, (4) Considering the quality of overall systems, and (5) Appropriately managing customer demands and expectations.

 Based on these guidelines, we have created our own MLOps (Machine Learning & Operations) Platform, which encompasses everything from the development of AI models to the operation of deployed AI and the ongoing provision of value while responding to changes in the external environment and business needs. Our data scientists, AI engineers, IT engineers, and service engineers can use this platform to coordinate closely and continuously provide customers with high quality AI services.

* Our AI quality management is introduced in detail in the fourth article of this article, and the MLOps Platform is introduced in detail in the second article of this article.

 The design, procurement, manufacturing, and maintenance processes of manufacturing industry production sites encompass a wide range of needs. These include the ability to accurately and rapidly assess responses to quotation requests for items to be procured, improvements to the work efficiency of operators at production sites, improvements to accuracy of image inspections, yield improvements, greater equipment maintenance efficiency, manufacturing parameter optimization, and optimization of inventories of products for which support is being offered.

 We provide a wide range of solutions that meet the needs of manufacturing industry customers using SATLYS, Toshiba's analytics AI that brings together the AI utilization knowledge of the entire Toshiba Group.

 These solutions have long histories of actual use in Toshiba Group manufacturing sites, and their effectiveness has been thoroughly verified. That's why we are so confident in our one-stop AI solutions, which represent the culmination of our extensive knowledge of production site needs. Let's look at some of these solutions, together with examples of their real-world deployment.

 First, let's look at one of our solutions that apply AI to image inspection. This case example comes from a Toshiba Group worksite that manufactures cast metal components.

 During the cast metal component manufacturing process, an inspector visually inspects the surfaces of processed components to check their quality. However, with this inspection method, pass/fail decisions vary from inspector to inspector. This leads to a lack of uniformity in inspection results. Furthermore, inspectors become fatigued when conducting visual inspections for long periods of time, which can result in a decline in the quality of their inspections. Training inspectors and developing their skills also takes a tremendous amount of time. The succession of skills and knowledge is no easy task.

 We solved these problems through image analysis using SATLYS. We collected inspection results from experts and provided them to SATLYS as training data. We then used cameras to take pictures of the processed surfaces of components and had SATLYS determine their pass/fail status. This reduced the amount of time required to make quality evaluation decisions and produced more uniform inspection quality. It enabled anyone to perform inspections at the same level as experts and assisted with the succession of experts' skills and knowledge.

 Another example is the use of SATLYS image processing technologies and deep learning in rating the grain of metallographic structures, which requires high-precision image analysis.

 We applied deep learning to diverse training data that reflected the knowledge of expert personnel. This made it possible to achieve a level of recognition accuracy almost equal to that of inspectors with extensive image analysis experience, something which would not be feasible when using a rule-based approach. The rating of the grade of metallographic structures was previously determined through visual inspection by an inspector. We used AI automate this inspection. This produces more uniform levels of inspection accuracy and significantly reduces the time and labor involved in inspections. We provide this AI through our METALSPECTOR/AI service.

 Now, let's look at some specific examples of using AI in applications such as customer procurement operations and Toshiba Group maintenance operations.

 The first is our use of AI in procurement operations. In the past, responses from suppliers to quotation requests needed to be confirmed by employees. However, this process was time-intensive and required the knowledge of veteran purchasing personnel who could rapidly discover incorrect values and areas requiring improvement. We used AI to analyze data such as past quotations and related information and automatically calculate estimate values. This was accomplished as one of the functions of Meister SRM, a strategic procurement solution supplied by Toshiba Digital Solutions. Calculating estimate values makes it possible to identify issues with suppliers' quotations, accelerating and improving the efficiency of the quotation process.

 The second is our use of AI in maintenance operations. We used AI to optimize maintenance component inventories. AI analyzed past maintenance history data and predicted device failures. These predictions were then used to optimize inventory quantities. Through this, we were successfully able to reduce maintenance component inventories by 30%. We provide this AI in the form of SATLYS KATA Inventory Optimization of Maintenance Parts.

 Furthermore, we have used AI in our warehouse picking work to reduce operation times by almost 20% by analyzing operator activity histories and optimizing the working environment accordingly. This was achieved by using wearable devices and business intelligence tools to visualize work contents (using "SATLYS KATA Work Activity Estimation").

* Maintenance component inventory optimization is introduced in detail in DiGiTAL T-SOUL Vol.24 #04.  Work behavior inference is introduced in detail in DiGiTAL T-SOUL Vol.31 #03.

 We are improving work efficiency in various areas by using AI to perform analysis and visualize current conditions.

 Toshiba Digital Solutions is developing and supplying solutions that leverage diverse AIs to meet actual worksite needs. The Toshiba Group's field knowledge, track record, and expertise are what have enabled us to create the AIs that are being used in industrial fields. The Toshiba Group is actively working to apply AI to business activities and is accruing experience that will be vital to supporting its many customers. The AIs that Toshiba develops, backed by this experience and track record, will become part of the Second Stage of AI, used to their fullest in various customer business applications.

Solving the problems faced by society is a pressing issue for us. We will continue to support our many customers through the use of Toshiba's AI solutions.

 Please contact us to find out what Toshiba can do for you.

• TOSHIBA Cyber Physical Systems