Introduction
Nuclear power ranks among the many world’s most regulated industries. AI and particularly generative AI have created sufficient impression that thought leaders rank it amongst different transformative “basic goal applied sciences” reminiscent of electrical energy and the steam engine. Harnessing AI to reimagine nuclear operations throughout the business means extra carbon-free nuclear power for electrical grids and information facilities, which the Worldwide Power Company estimates demand to double by 2026. In September 2024, Westinghouse unveiled its HiVE™ AI system, powered by its fine-tuned bertha™ generative AI mannequin, reworking how prospects collaborate with Westinghouse.
Constructing a Higher Information Administration Resolution
Westinghouse’s digital transformation began greater than 5 years in the past with a deep bench of information and nuclear specialists and over 70 years’ value of cleaned and contextualized industrial information distinctive to the nuclear world. Nevertheless, the crew wanted to enhance the corporate’s information infrastructure if it needed to appreciate its AI ambitions. The prevailing on-premises analytics database lacked some vital scalability options and choices. With no scalable cloud resolution, the info crew struggled with an absence of computing sources, an incapacity to quickly experiment with large quantities of information, and restrictions on safely sharing information throughout purposes.
To construct a world-class, nuclear-specific AI functionality, Westinghouse wanted a greater resolution. Westinghouse determined to construct on the Databricks Information Intelligence Platform, a transfer that will show essential in its mission to drive innovation. The nuclear business has all the time been deeply dedicated to security and lowering danger, with each element inspected and controlled. Managing and securing essential nuclear information is just not negotiable. With this in thoughts, Westinghouse got down to design an information spine that would host AI purposes for a number of the most trusted utilities on this planet. Databricks was the best accomplice to assist Westinghouse obtain this purpose.
As Westinghouse got down to design an information spine so safe and sturdy that it might host AI purposes for a number of the most trusted utilities on this planet, it turned to Databricks. The Databricks crew rapidly turned a “guiding gentle” for Westinghouse, offering essential help because the Westinghouse infrastructure crew took the lead in configuring our techniques to satisfy the nuclear business’s strict regulatory necessities. Westinghouse was capable of leverage Databricks’ state-of-the-art governance with Unity Catalog. It was constructed in response to greatest practices outlined within the Databricks AI Safety Framework (DASF), complementing Microsoft’s sturdy safety requirements. These foundations bolster the credibility of Westinghouse’s information administration practices and provides its prospects peace of thoughts, which is important in an business the place belief and reliability are paramount.
When it got here time to modernize how the info was organized, the Databricks skilled providers crew delivered. Collectively, Westinghouse and Databricks created a scalable and multi-tiered analytics atmosphere, full with an ML Ops course of that streamlines the complete machine studying lifecycle. This basis additionally featured a strong prototyping atmosphere, together with devoted workspaces, for testing and deploying AI fashions, all backed by a safe and dependable information lakehouse structure.
The brand new infrastructure instantly saved tons of of hours yearly for the Digital Optimization Providers enterprise unit and allowed the Westinghouse crew to reinvest of their product traces to incorporate AI for customer-facing purposes and providers.
To make this imaginative and prescient a actuality, Westinghouse had to make sure that its information was correctly ready, managed, and ruled. That’s the place Databricks’ highly effective applied sciences, together with Auto Loader, Photon engine, and Lakeflow Jobs, actually shined. Then, when Westinghouse wanted real-time insights into its information high quality and pipeline efficiency, they tapped into options like Lakehouse Monitoring and Expectations. Now, with Unity Catalog (UC) governing its information, Westinghouse has full visibility into its information’s journey, from supply to vacation spot. Within the nuclear business, every thing revolves round security and belief. As Westinghouse continues to develop pioneering new AI options, Databricks providers reinforce the belief Westinghouse earns for managing information securely and reliably.
Accelerating AI in a Advanced Business
On September 4, 2024, Westinghouse launched its HiVE™ nuclear particular AI system and its bertha™ generative AI mannequin to the world. Not solely has the Westinghouse crew quickly superior its AI capabilities utilizing the Databricks Information Intelligence Platform, however it will possibly now create future AI merchandise and options restricted solely by creativeness.
To help in growing bertha™, Westinghouse leveraged the Databricks Mosaic AI Agent Framework, to quickly consider numerous foundational fashions and GenAI techniques. Utilizing Databricks Experiments and MLFlow, Westinghouse performed speedy experimentation to find out the very best fashions, whereas logging statistics to judge efficiency. This method enabled Westinghouse to speed up the event of its customized Generative AI resolution.
Westinghouse can now leverage its superior information infrastructure to create options throughout the nuclear business. For instance, giant industrial services talk and retailer monumental portions of information. With an structure constructed on Databricks, Westinghouse maintains an AI resolution to extract, cleanse, and retailer machine information from over 200 nuclear services worldwide. One other instance consists of an AI utility designed to course of video information in real-time inside Strain Water and Boiling Water Reactors with the potential to detect particles not less than 90% higher than guide inspections and save as much as 25% on inspection prices.
Lastly, one other nice instance consists of leveraging the bertha™ generative AI mannequin to generate licensing information and documentation dramatically sooner. Historically, it will possibly take months to manually compile new nuclear web site licenses or environmental assessments. This can be a essential step in streamlining nuclear growth.
The Databricks infrastructure has freed information and nuclear specialists to deal with nuclear innovation. Because of this, the Westinghouse information scientists delivered 4 proofs of idea in December 2024, two production-grade techniques within the first quarter of 2025, and helped generate 45 distinct innovation concepts within the first two months of 2025.
Conclusion
The Westinghouse-configured Databricks Information Intelligence Platform removes big limitations to attaining Westinghouse’s AI ambitions. Now, Westinghouse can scale compute, quickly and safely experiment with mass quantities of manufacturing information, and share data securely throughout purposes. Westinghouse HiVE™ nuclear-specific AI system prospects respect the facility of auditability, enter and output transparency, real-time information processing, and operational analytics. The Westinghouse groups worth the unimaginable and adaptable partnership with Databricks to create a novel platform that positions it for continued pioneering AI innovation.
“With Databricks all the time offering the most recent options that hit the market, Westinghouse is ready to frequently incorporate new AI capabilities for our prospects.”
— Catherine Stanley, Information, Digital, and AI Supervisor at Westinghouse