Trace Id is missing
July 05, 2024

Recognizing the Essence of AI and Building the Future With Clients: MHI's DI to Create Proprietary Architecture Using Azure OpenAI Service

Since its establishment in 1884, Mitsubishi Heavy Industries, Ltd. (MHI) has been providing a wide range of products and services in various fields - including energy, engineering, defense and space, and mechanical systems, as a “comprehensive infrastructure company that supports peoples’ lives through engineering and manufacturing”. Among these is the field known as the “Energy Domain,” where the company's gas turbines, boilers, and other power generation equipment, as well as operation and maintenance solutions, support the large-scale power generation facilities that are indispensable to our daily lives.

MHI has declared its intention to become carbon neutral by 2040. To achieve this target, MHI has announced Mission Net Zero as its doctrine for action. Mission Net Zero calls on MHI to promote “energy transition” by implementing measures such as striving to eliminate carbon from existing infrastructure, establishing a hydrogen ecosystem and achieving a CO2 ecosystem. MHI is also pursuing “smart infrastructure,” introducing infrastructure that supports smart logistical systems; low-carbon, energy-efficient data centers; and automated mobility.

Mitsubishi Heavy Industries

“tomoni®” Accelerates Digital Innovation in Power Plants

MHI, which has diverse business domains ranging from the development of complex machinery to operation and maintenance, has established “ΣSynX(Sigma SynX)” in 2021 as a DI (Digital Innovation) brand to enhance the potential for value creation by combining the vast amount of data and know-how stored in these domains with external technologies and knowledge. Hiroyasu Ishigaki, Senior Engineer, Digital Strategy, Technology Strategy Department, Energy Domain, Mitsubishi Heavy Industries, Ltd, explains the ΣSynX brand.

“ΣSynX is based on the concept of ‘smart and connected.’ It promotes autonomous and intelligent solutions, consolidates digital assets that are unevenly distributed throughout the company and shares them with each business unit, and encourages collaboration between digital-related divisions such as ours and other units. Through this effort, we aim to provide our customers with new value that emerges from this.” told Ishigaki.

Along with the “Logistics Intelligence and Automation Project” and the “Remote Condition Monitoring Service,” ΣSynX is embodied in the intelligent solution “TOMONI®,” which accelerates the design, operation, maintenance and decarbonization of power plants.

More Than Just a Product - Solutions That Help Customers Form a Growth Spiral

The origin of the name TOMONI is “together with customers.” It also expresses the company's commitment to solving problems in collaboration with clients. According to Ishigaki, TOMONI has three functions: First, it enhances connectedness with customers through remote monitoring - by sharing customer data, MHI supports daily management and operation. The second function of TOMONI is to improve functionality. This feature supports advanced and visualized control-related components such as fuel efficiency and flexibility. The third function is smart maintenance. This feature is designed to enhance the sophistication and efficiency of management and operation work by properly organizing, visualizing, and sharing necessary information from past inspection data and documents.

“We believe that TOMONI is not only a product to be implemented, but that it is also an intermediary that contributes to the digitalization of our customers. The energy industry, where facilities and equipment are rarely replaced, tends to lag behind other industries in efficiency and digitalization. Our goal is to provide a solution that will help customers promote the digitalization of their operations and create a better growth spiral, starting with TOMONI,” stated Ishigaki.

Released in 2017, TOMONI has been continuously upgraded and has been installed in 90 domestic and overseas power plants. In the spring of 2023, a project was launched to further supplement its functions with the power of AI, which is currently undergoing internal trials and verification.

Promoting In-House Use of Generative AI Through “TOMONI TALK With ChatGPT”

“The catalyst for this project was the emergence of ChatGPT,” recalls Daisuke Goto, Group Manager of Modular Design Group, DPI Department, Digital Innovation Division, Mitsubishi Heavy Industries, Ltd. Since its release in 2022, generative AI has become a hot topic around the world, and of course MHI was also considering utilizing this technology.

Since MHI was an early adopter of Microsoft's Azure OpenAI Service - which ensures that customer data is not used for training of generative AI - there was no concern that the input information for the generative AI would be leaked outside the company. However, many of the group company's services handle confidential information, so they proceeded to carefully establish the guidelines first.

Goto, who was watching the enormous task of establishing the guidelines from a neighboring department, thought, “Let's proceed with the implementation of the generative AI to the extent that it does not conflict with the guidelines. The summary and translation functions of the generative AI could be used immediately in our internal operations,” and so created a prototype chat application that utilized ChatGPT's functions with a low-code application. After a trial within the department, the in-house ChatGPT application was implemented on TOMONI in July 2023. It was named “TOMONI TALK with ChatGPT.”

Initially, it was a text-based chat application that could summarize and translate information, generate document drafts, and program code - not substantially different from ordinary ChatGPT in functionality. However, as soon as it was released, the number of internal users quickly increased.

“It's in a different league from the previous apps. The need for generative AI is definitely there." Convinced that this was the case, Goto participated in an in-house hackathon organized by the research department. There he met over 20 cross-functional teams, who came up with various ideas for utilization of generative AI, such as troubleshooting, organization of emails, and conference room reservation functions. Having experienced firsthand the enthusiasm within the company, Goto decided to further update the system.

“We had limited resources, so we decided to leave functions that could be solved by Copilot for Microsoft 365, such as email and meeting summaries, to them. Instead, we thought it would be more meaningful to focus on functions that would allow everyone in the company to utilize their own data and gain insights from it,” said Goto.

They began exploring the use of their own tailored generative AI, but they ran into a major obstacle in improving the accuracy of their data retrieval and responses.

“We evaluated Microsoft's API, ‘Azure OpenAI Service on your data,’ but it didn't seem to give us the results we wanted. We also referred to case studies of other companies that have implemented similar functions and found that there may be a way to do it, but we could not figure out the details. So we decided to consult with Microsoft to see if they could introduce us to a partner company that could support us,” said Goto.

In response to the consultation, Microsoft introduced Zen Architects, Inc. as a partner. They had the best reputation in providing accompanied support systems, and in developing a system utilizing Azure AI Service and Azure PaaS, which had already been used in TOMONI as a document search service.

“RAG” - a Method That Enabled TOMONI TALK with ChatGPT

“When Microsoft approached us, we thought that RAG (Retrieval Augmented Generation) might be effective,” said Kazuyuki Miyake, President of Zen Architects Co., Ltd.

Azure OpenAI Service would first create an answer to the question from the trained data. For generic questions, Azure OpenAI Service may be able to find the answer in that trained data. When it comes to more technical questions, however, it may not be able to find the answer in the trained data, and may deliver an incorrect answer. This is an instance of where the LLM (Large Language Model) is lying - also known as “hallucination.” A system that “lies” in a plausible manner is not reliable. On the other hand, fine-tuning LLMs to relearn large amounts of domain specific knowledge is not cost-effective, and there are limits on restricting “lying” as well.

“In order to solve this problem, as an extreme example we should interpret the meaning of each query without using any data from Azure OpenAI Service, search our in-house data according to that interpretation, and have Azure OpenAI Service create a response text based on the search results obtained. The method called RAG is utilized based on this concept,” explained Miyake. He also emphasized that the most important thing is the maintenance of in-house data, followed by the accuracy of the search engine, Azure AI Search.

“Azure OpenAI Service is just one component. If you focus too much on that particular functionality, we would miss the main point. Rather, we should focus on the data. The people at MHI were originally preparing their data with the idea of ‘connecting intelligently,’ so RAG was a good match for them,” according to Miyake.

Marveled By the Speed With Which Japan's Leading Experts Gathered and Proceeded With the Implementation

Zen Architects is certified as an Advanced Partner in Microsoft's Azure OpenAI Service reference architecture endorsement program and has already released RAG patterns that leverage the Azure OpenAI Service.

However, Atsushi Yokohama from Zen Architects, an Azure Expert and certified Microsoft MVP for AI since 2017, explained; “Considering the future vision of TOMONI, we thought it would be a better match to build a fully customized RAG instead of applying an existing pattern.”

As such, Zen Architects proposed the “Azure Light-up”; a support program led by Yokohama and other Azure and AI experts. This program consisted of a two day session to deepen understanding of RAG and ultimately to prototype a system that matched TOMONI's vision for the future.

Yoshiyuki Nagata, Deputy Manager of ICT Operations Group, Takasago Service Engineering Department, GTCC Business Division, Energy Systems, Mitsubishi Heavy Industries, Ltd. who participated in this program, was astonished at the speed at which the implementation was carried out with hands-on experience.

“I had a vague idea of what we could do with RAG, but it was taking shape right in front of my eyes. Seeing the blueprints drawn on the whiteboard on the first day being implemented by the second day was very inspiring for our engineers," said Nagata.

“MHI had already tried ‘Azure OpenAI Services on your data’ from the beginning. In addition, they already had the idea that Azure OpenAI Service is just one component and that database maintenance, not LLM fine tuning, is what is important, which I believe contributed to the positive effect,” evaluated Miyake.

“Just Like Magic" Ripple Effect Created by Specialists’ Skill

The release of the improved version of TOMONI TALK with ChatGPT was met with a remarkable response within the company. Currently, there are 200-300 active users per day, which is a consistently higher usage rate than other internal business applications.

“Another significant benefit beyond the practical use of TOMONI TALK with ChatGPT is that we have received feedback from users in various departments on the usage of it and on how to improve their own department's operations by utilizing generative AI. I am convinced that we will be able to further expand usage, such as with translation and proofreading of documents for overseas markets,” Goto stated with a smile.

He also added that, “As awareness for AI grows within the company and more ideas are gathered, the scope of our efforts will also expand. We are excited for this kind of multiplier effect.”

“The development team is feeling a positive response to the project,” says Nagata. MHI's development team has found a veritable way to utilize the cloud through this project.

“We have been using the cloud for a long time, but until now it was only a system that had been moved to the cloud. However, the experience of building a system using a combination of PaaS (Azure OpenAI Service) and other SaaS has given us the confidence to respond quickly and flexibly to other projects while using multiple tools,” told Nagata.

Miyake of Zen Architects strongly agreed, saying, “That's exactly what we wanted to convey.”

“We have been advocating for more than 10 years that if you are serious about cloud computing, you should use managed PaaS and SaaS. We are very happy that developers have come to understand this, and it gives us a lot of confidence,” added Miyake.

“I, too, was one of those who were impressed.” Ishigaki followed. “Watching Zen Architects building architecture in their sessions was like witnessing magic. I realized how beneficial it is to collaborate with a partner who understands the essence of what they are doing. I am spreading the word to encourage my colleagues to study cloud computing and get certified in Azure,” said Ishigaki.

The Next Goal is "TOMONI Copilot" - Powerfully Driving Customers' DX

Further development of TOMONI TALK with ChatGPT will focus on expanding the project internally and throughout group companies, as well as on making the service available to TOMONI user companies.

“We hope that TOMONI TALK with ChatGPT serves the user companies as a trigger to realize the importance of integrated data, just as we have realized this time. We would also like to pursue the development of ‘TOMONI Copilot’ that utilizes various data accumulated in TOMONI," said Ishigaki.

Mitsubishi Heavy Industries has succeeded in developing an application that will lead to the future by recognizing the essence of generative AI and teaming with a skilled partner. Goto, who played a central role in the project, said, “Our awareness has changed dramatically through this collaboration. We would like to continue to work on cloud native technology, so we would like to ask for your continued support,” expressing his expectations for Microsoft.

Nagata then requested, “We need to continue to provide this TOMONI for many decades, since power plants have a long lifespan. We would like to have a long-lasting relationship with Microsoft in order to sustainably deliver an excellent product to our customers.”

Throughout this interview, we were very impressed by the attitude of everyone, being committed to working together to achieve DX not only for their own companies, but also for their clients and partners. We at Microsoft very much share this attitude. As partners walking together toward the future, we will continue to strive for providing better services.

“We hope that TOMONI TALK with ChatGPT serves the user companies as a trigger to realize the importance of integrated data, just as we have realized this time. We would also like to pursue the development of ‘TOMONI Copilot’ that utilizes various data accumulated in TOMONI”

Hiroyasu Ishigaki, Senior Engineer, Digital Strategy, Technology Strategy Department, Energy Domain, Mitsubishi Heavy Industries, Ltd.

Take the next step

Fuel innovation with Microsoft

Talk to an expert about custom solutions

Let us help you create customized solutions and achieve your unique business goals.

Drive results with proven solutions

Achieve more with the products and solutions that helped our customers reach their goals.

Follow Microsoft