ページの本文へ

Hitachi

2018年8月2日

Insights Laboratoryメンバであり、北米社会イノベーション協創センタのチーフデータサイエンティストであるChetan Guptaと、日立ヴァンタラ社のJames Destro、日立コンサル社のRajesh Devnaniが、産業分野に向けたAI技術、ならびに、日立がグローバルのお客さまやパートナーとどのように協創し、社会や組織が直面している課題を解決しようとしているか議論を交わした。

What is industrial AI about?

Chetan
Let me set the context. For the last several years, we have been doing a lot of work in Industrial AI. To me it means using AI and machine learning to solve real world problems that our industrial customers are facing. On one axis, there are the horizontals such as maintenance, operations, quality and safety; and on the other axis there are descriptive, predictive, and prescriptive analytics. Over the last several years we have solved some interesting and difficult problems in each of these areas, and we are now looking at end-to-end multi-scale and multi-objective optimization.

Rajesh
Many people may not realize that AI technology isn’t new – it’s actually about 70 years old. Over the years, applications for AI technology have largely been focused in the consumer sector. Industrial AI is a fairly new phenomenon and stands to have a far more significant impact, driving operational efficiencies and intelligence that was never possible before now. We’ll now be able to generate completely new business models and monetize more value from the application of industrial AI.

James
It really started with the launch of the internet of things (IoT). IoT-connected machines and devices have started generating vast quantities of data that require greater agility and responsiveness than human analysis can deliver in a timely manner. This has driven the need for machine learning or artificial intelligence with analytics so that we can derive more value as the machines, control processes and control systems are learning and improving.

Rajesh
Data is the key. We need the ability to understand the environments where people or entities access data. Although we can make data much more accessible across organizations, the information is often proprietary. An increased willingness to share data has pushed research forward, but for us, it remains a challenge because of its inherently proprietary nature.

James
There's a willingness to share data around methodologies and training in the areas of safety and the environment because they aren’t competitive differentiators. When methods start to create differentiation, however, things change.

Rajesh
Global efficiencies around data-sharing are increasing, but there's both collaboration and competition. The collaboration is evident where there’s willingness to solve some common industry and societal problems. Organizations are far more likely to share data if it is not of a sensitive or proprietary nature.

James
One emerging business model we're seeing evolve from data-sharing is the ability of companies to monetize their data. You can think about monetization in a couple of ways: The first is data can be used internally to increase productivity and efficiency. The second is bringing value to the supply chain – upstream and downstream. For example, publishing the usage statistics for pieces of machinery, which can help a vendor create a better piece of equipment. Data is a differentiator. In the future, I expect we’ll see this shift toward outcomes that are generated from the data as the differentiators.

Rajesh
Yes, data will continue to be a strategic asset and a differentiator, but there are places where you're better off collaborating to benefit industry as a whole. People will be looking at which categories of data they should be collaborating on to solve industry-wide problems and which categories they should hold close as proprietary.

Which verticals are ready?

Rajesh
I think of this more from a use case perspective than a vertical perspective. The first is maintenance. A lot of people view maintenance as a siloed situation, which it is not. It impacts the whole value chain. And it's not just about uptime. When maintenance improves, quality improves, as does customer satisfaction. There are many positive upsides to maintenance.

A second use case is advanced process control (APC) across both process and discrete manufacturing. APC is really about how you apply AI to improve the performance of a process? It involves many technologies and essentially helps define a target operating envelope for effective & optimized operations. You can have an impact in real time, gain a lot more efficiencies and realize significant improvement in product quality.

Predictive quality is a third use case which is gaining prominence based on AI Computer Vision & Image Analytics capabilities. Then there is also demand forecasting, supply chain optimization and the list goes on. There are at least 100 different use cases that are important across industries and technologies that are poised to disrupt industry in a big way.

James
The industries that are seeing challenges are the ones that are directly facing disruption. They're also trying to achieve productivity. To get to that next generation of productivity increase, companies need to make changes.

Chetan
Companies that are feeling disruptive pressures are much more likely to innovate. The manufacturing industry is keenly interested in digital innovation, but some manufacturers are mature, and some are just starting out on their IoT journeys. This presents challenges for anyone trying to build solutions.

James
The challenge is finding repeatable value across this entire chain. There are platforms that can help you get to data and process it faster in the initial phase, but we need to operationalize what we learn from that data and the analytics and then integrate it into the work process.

Chetan
You're coming from the client’s perspective, and that is a major challenge when building the capability of your workforce for maintaining your solution. We need to think about maintaining or building workflows that can build these AI- and IoT-based solutions. From a research perspective, we need to simplify the use and deployment of analytics-based solutions.

Rajesh
There is a trade-off between creating a platform that is one-size-fits-all and building focused applications that serve business needs. We have to look at it from a holistic level rather than at the technology level. Looking at a customer’s complete value-chain, we can identify opportunities for where and how to apply the technologies that will benefit them most. The business case holds the value. Repeatability is essential for us to take it to a volume of customers and realize the right business value for our customers and ourselves.

Chetan
From an R&D perspective, feasibility is key. We need repeatable building blocks that solve specific business problems under specific conditions, and can be applied across verticals. We call these building blocks “solution cores.” All of the solution cores are repeatable analytics and are developed to meet the needs of the customer. In research, we work with first of a kind challenges, identify and develop the underlying analytics required, and then extract and abstract our repeatable components. We are doing this systematically to solve the problems of operations, maintenance, quality and so on.

Optimization and decision making are also difficult problems. If you predict that a machine is going to fail at plant A, how can you prevent or mitigate the effect across the supply chain?

James
Absolutely. The fact that the research is founded on actual customers and customer outcomes gives credibility to usability and our ability to scale. It is critical to the process.

Keeping humans in the loop

Chetan
There is an ongoing concern about AI versus humans. AI is not a replacement for humans, but an aid to humans. So, instead of moving to completely autonomous systems, I suggest we identify targets where we must improve human performance – to augment human capacity.

Many industries are facing an aging workforce and are challenged to retain skills and know-how to remain competitive. This is particularly true in developed countries, so the bigger picture is one of sustainability and progress. We’ve talked with companies that have many senior employees. They want to capture an operation model so that they can transfer knowledge to the next generation. As an AI researcher, my perspective is often about getting to the point where I can use AI to make recommendations to fulfill specific needs. Before autonomous systems in industrial applications, we need to focus on building "industrial recommendation systems."

James
We’re also seeing the need for a learning process so that customers can learn to trust AI, machine algorithms and machine-generated data. This is needed for both the business decision maker and the action taker.

Rajesh
We have some fully automated manufacturing lines today, but these robots work in a very narrow context. There are parts of operations that might be automated, but not where you need to view problems from different perspectives. We have to find the right applications and work with our customers to help them operationalize AI in their business operations.

James
The challenge arises today when the task being performed outgrow an individual or group’s ability to process all the information in a timely manner. That's when the augmentation starts to happen. For example, think about the disruption to passengers, air traffic control and other systems when a plane is delayed. It’s more information than we can process to reschedule all the passenger connections in a timely way. In this case, we need humans to make decisions on top of the recommendations that come from a system. By combining AI with human decision-making, the rescheduling of a delayed flight can be shortened from 3-5 days down to one day.

The flip side occurs when a task is so simple and repetitive that automation can take the place of humans. We've seen that in manufacturing when the decision-making process is so limited that we can program it in and trust that the programming captures all of the outliers. Those are the two extremes of the use cases.

The use cases in the middle are the most challenging—cases where you use AI and machine processing, but the result isn't that much better.

Integrating AI for a better future

Rajesh
In terms of adoption, there are clearly a few challenges. First and foremost, leadership needs to take a leap of faith and start using technology for guidance in making strategy decisions. That's going to be one imperative for AI adoption.

The second issue is operationalizing AI. To make AI-based insights work, you need to integrate it into your business workflow so that it is seamlessly and fully synchronized with your business applications.

The third challenge is around change management. It is really about positioning AI in the best light and showcasing how it helps different stakeholder groups to enable ease of adoption.

Chetan
We also have to think from a design perspective and the user interface (UI). I think of voice as a UI. What UI changes do we need to make for the successful operationalization of these AI technologies? How do you adapt to gain human trust in the AI? The trust is very important, and when you hear a human voice, somehow you find it more trustworthy.

James
The UI is critical. The voice interface is interesting in an industrial setting where people may be wearing gloves or constrictive pieces of equipment, or they have their hands full. That voice interface is going to make AI more adoptable in some environments. And it is still critical that the information as well as the process can be captured and shown in a way that people can trust.

Rajesh
Over time we’ll move toward a more explainable AI concept, where we can explain our suggestions, and I expect confidence levels will be built on this. Once we get to a stage where we are able to explain our algorithms and say, "This is why you're getting this set of recommendations," people will be more trusting of the process.

James
In the consumer space, AI makes recommendations to me based on what it's learning from my buying behavior. A "wrong choice" is not going to impact my relationship with a retailer. But the result of a bad decision in the industrial world could mean more down time, increase hourly costs or jeopardize personal safety. We have to give a logical explanation as to the processes, the steps, and the potential outcomes of decisions.

Chetan
I agree, the cost of “a mistake” in the industrial sector is totally disproportionate compared to the consumer space. This is something we in the AI community need to be especially careful about. We can tell people with some degree of certainty how accurate things are, but we can’t yet say algorithms are 100% accurate. We need to wrap the machine learning results under an optimization problem, so that optimal decisions can be made. This is why humans are an important part of the discussion. We also need to work on models that are explainable and can increase the confidence in the "judgment" of AI systems.

Another challenge is how to combine "individual AIs." For example, if I have an AI based method for operational optimization and an AI for quality enhancement, how do we bring them together to manage tradeoffs? How do I apply AI globally and bring it together? There are many interesting challenges in industrial AI – and a lot of fun times ahead. I'm quite excited about the future.

Rajesh
It’s obviously a difficult challenge—getting the right data in the right security format to make those kinds of leaps. But we are getting there. In the next decade or so, we’ll solve most of these issues. Not to the level of general artificial intelligence, but at least with regard to industrial AI. Industrial AI we will gain significantly in maturity.

James
I think there will be adoption by specific industries first. There will be an acceleration once there's a key win, not by a particular company, but a key win by an industry participant. To have a massive change to the industry, it's going to take a major win. It's coming. We have customers testing the water. But we need to get to operationalizing and productionizing those systems, and that’s when we’ll see it happen.

Chetan
We are trying to solve problems quickly and address difficult challenges. There's an array of problems and an array of industries that we can touch as a community, and it’s fascinating.

Rajesh
AI will become much more mainstream, visible and embedded in the core business fabric. The way we adopt technologies, including AI, is going to become more seamless. There’s going to be more attention to making AI more user friendly, simpler, and an integral part of the business.

James
For me, it's fascinating to see where the best talent and the best minds are focused. With Industry 4.0, artificial intelligence could cause a massive shift in how we think about productivity and where to focus top talent. The next generation is open to this and that’s where things start to change. Mindsets will change from making money from clicks to changing the world through Social Innovation. That's exciting.

The need for innovation ecosystems

Chetan
What do you think of having a common marketplace where there’s an understanding of policies regarding the sharing of data and where everyone can benefit equally? How important is it to build consortiums? Should Hitachi take more initiative?

James
I think it depends on the issue. There are some consortiums and standards bodies, but these are at the transaction level where they're only sharing data so they can continue to do business. The approach needs to be focused on a specific industry segment where we can provide differentiation in the data processing and/or the analytic power. Changes will occur when you can crowdsource analytic insight and AI types of algorithms. And that will be on a specific subset of information.

We start small and it’s something that grows. To realize the larger value of sharing information is going to take an ecosystem and not just a single player.

Rajesh
It's all about who takes the leadership in establishing such marketplaces. Ultimately, they have to go narrow – industry by industry – and address the unique set of challenges on how to collaborate, not only from a data perspective, but also from the business organization perspective. That requires focusing on the issues in each industry. As Hitachi we offer a wide & deep footprint of both Operational (100+ years) and Information (55+ years) technology expertise based on our experience across multiple industry verticals. We are seeing the onset of comprehensive industrial ecosystems that strive to optimize across entire value chains (including networks of suppliers and customers). A marketplace for leveraging Industrial IoT applications at scale is likely to develop in the near future and provided standardized plug n play modules that can serve individual use cases with minimal customization. At Hitachi, we are well poised to take a leadership position in establishing these Industrial IoT marketplaces and supporting ecosystems to help deliver a significant business impact.

The next conversation in this series will focus on industrial IoT. Learn how Hitachi is working on a global scale to innovate for a better future for all. With customer-driven solutions and co-creation in every phase of development, Hitachi is collaborating with customers, partners, and educational institutions worldwide to accelerate the resolution of issues facing society and organizations today.

プロフィール

※ 所属、役職は公開当時のものです。

Chetan Gupta, Ph.D.

Chief Data Scientist & Architect
Hitachi America, Ltd.
Global Center for Social Innovation–North America

As head of the Industrial AI Laboratory, Chetan is responsible for leading a large team of data scientists, architects and developers engaged in developing cutting edge solutions and opening new frontiers in the area of industrial analytics. The Lab develops fundamental horizontal technologies to build solutions such as predictive maintenance, quality, and operations monitoring and control, for industry specific verticals such as mobility, mining, building energy management systems.

Chetan received his Ph.D. in Mathematics and M.S. in Mathematical Computer Science and Chemical Engineering from University of Illinois, Chicago. He has close to 50 patents either granted or under review and more than 40 publications in the area of data mining/machine learning, data stream systems, complex event processing, workload management.

James Destro

Vice President, Product Management
Hitachi Vantara

James currently heads the Product Management for IoT Applications at Hitachi Vantara. His focus is on creating horizontal IoT applications that leverage advanced analytics including artificial intelligence and vision analytics. These applications will focus on driving outcomes specifically in the areas of maintenance and manufacturing. His background includes more than 20 years of industrial enterprise software leadership driving operations intelligence, advanced analytics, and optimization for industrial giants including ExxonMobil, Siemens, Dow Chemical, BASF, Saudi Aramco, Schneider Electric and General Electric.

James holds a Bachelor’s degree in Chemical Engineering from Rensselaer Polytechnic Institute.

Rajesh Devnani

Vice President, Global Solutions Management
Hitachi Consulting Corporation

Rajesh heads the Global Solutions Management function and is the Global Predictive Maintenance Solution Lead at Hitachi Consulting. He focuses on creating differentiated world-class solutions leveraging the best capabilities of the Hitachi Group across digital technologies. Under his leadership, Hitachi Consulting has developed unique, market-facing solutions addressing key digital transformation opportunities.

Rajesh’s background includes two decades of manufacturing industry experience across ERP programs, business consulting and IoT/analytics-led digital transformation engagements. He holds a Bachelor of Electrical Engineering degree and MBA in Marketing and Finance. He also holds multiple professional certifications including CSCP and SAP certifications. Rajesh speaks at multiple industry/ professional forums and writes blogs on key technology topics.