To perform analytics across lines, processes, and plants, companies need a standardized way to identify similar data. That’s where Sight Machine’s new namespace manager comes in.
SAN FRANCISCO—One of the first artificial intelligence (AI) small language models (SLM) for manufacturing tackles a core data governance challenge: mapping the multitude of factory data naming schemas into enterprise-wide unified namespaces or corporate data dictionaries.
The small language model, aptly dubbed Factory Namespace Manager, was recently introduced by Sight Machine, the provider of a platform for data-driven manufacturing and industrial AI. It is one of the first partner-enabled adapted AI models for manufacturing offered within the Azure AI model catalog, announced by Microsoft in November, Sight Machine said in a release.
Factory Namespace Manager uses AI to fill a crucial gap in the technology needed to create a unified namespace: mapping between the original data field names and the corporate standard. This enables manufacturers to integrate factory data with enterprise data systems for end-to-end optimization, according to the release.
Sight Machine’s SML is a customized, fine-tuned version of Microsoft’s Phi-3.5 small language model. Unlike large language models (LLMs), which are general purpose software trained on vast amounts of data, SLMs are used to focus on specific types of work and require less omputing resources, offering strong performance at low cost and low latency, the company said.
“Our solution addresses a widespread challenge in the manufacturing industry, converting decentralized naming systems into a single corporate standard,” said Sight Machine Chief AI Officer and Co-Founder Kurt DeMaagd, in the release. “This has become an acute problem as more clients push factory plant floor data to the cloud, removing data from its original context, and making the management of that data increasingly difficult.”
Individual plants often have thousands of data sources from multiple generations of machinery that are frequently 10 or 20 years old. Typically, the data streams aren’t labeled in a standardized format that makes clear where the data comes from and what it represents. To perform analytics across lines, processes, and plants (and even between otherwise-identical machines with different data labeling), companies need a standardized way to identify similar data.
Today, creating this translation layer requires a heavy investment of time by subject matter experts with extensive knowledge on the nuances of both the legacy and the target naming schemas, and is thus typically done manually for a small subset of data, according to Sight Machine.
“I’ve spoken to dozens of industrial companies about their current and potential use of AI in factory operations and the overwhelming feedback I hear and see in IDC survey data is that most companies are struggling to leverage AI effectively at scale due to the condition of their data,” said Jonathan Lang, research director of Worldwide IT/OT Convergence Strategies at IDC, in the release. “They have this dilemma that contextually similar data is formatted in multiple ways and is difficult to source and normalize amidst a complete lack of historical governance and data architecture. What I’ve heard loud and clear is that technology that helps to solve this challenge and reduce the labor requirement to decipher data will be readily adopted.”
The bottling company Swire Coca-Cola USA is reportedly planning to use Factory Namespace Manager to efficiently map its extensive PLC and plant floor data into its corporate data namespace.
“We are working with Factory Namespace Manager to recognize patterns in the data we’ve manually translated, and then applying the patterns to the rest of our factory data,” said Swire Coca-Cola USA Vice President of Data and Insights Bharathi Rajan, in the release. “This will make it much easier to get relevant data to frontline workers, to inform decision making, and to integrate production insights into other parts of the company, such as supply chain. This is one of the most useful applications of AI we’ve seen in manufacturing, and we’re excited to put it to work.”
Artificial intelligence is interwoven into Sight Machine’s Manufacturing Data Platform, which uses machine learning and other AI techniques to identify and optimize how machine settings, raw materials, and production practices interact to determine throughput, quality, sustainability, and other key manufacturing metrics.
“The collaboration between Microsoft and Sight Machine will give manufacturing organizations the ability to build AI solutions through Azure AI Studio and Microsoft Copilot Studio that deliver real value and advance business transformation,” said Satish Thomas, corporate vice president of business and industry solutions at Microsoft, in the release. “Factory Namespace Manager applies SLM AI technology to a high-impact use case with strong potential ROI for companies pursuing data-driven manufacturing.”
Sight Machine’s AI offerings include Factory CoPilot, which uses generative AI technology to offer an intuitive, “ask the expert” experience for all manufacturing stakeholders. Built using Microsoft Azure OpenAI Service, Factory CoPilot can automatically summarize all relevant data and information about production in real-time (such as for daily meetings). It can then generate user-friendly reports, emails, charts, and other content (in any language) about the performance of any machine, line, or plant across the manufacturing enterprise, based on contextualized data in the Sight Machine platform, the release stated.
Sight Machine said it also offers Blueprint, AI-driven tag-to-asset mapping software for clients that have large volumes of poorly identified data sources. It uses AI to map each data source (tag) to a specific asset (machine). Sight Machine built Blueprint in partnership with Microsoft and NVIDIA, the company said.