In the manufacturing industry, fragmented data presents a significant challenge. This data, generated from a myriad of sensors, machines, and systems, often lacks standardization, making it difficult to manage, integrate, and analyze. As manufacturers strive to optimize production, reduce downtime, and enhance decision-making, the need for a unified approach to handling this data becomes increasingly critical.
Microsoft, in collaboration with its partners like Sight Machine, is at the forefront of addressing this challenge. Advanced AI technologies and solutions integrated into platforms like Microsoft Fabric are transforming the way manufacturers handle plant floor data. These initiatives aim not only to assist manufacturers in managing data more efficiently but also to fully use industrial data to enhance productivity, improve efficiency, and achieve cost savings.
In AI, small language models and large language models serve distinct purposes, each offering unique advantages. Small language models are specialized and efficient, focusing on specific tasks or domains. This specialization allows small language models to provide highly accurate and relevant insights tailored to industries such as manufacturing. Due to the smaller size, small language models require fewer computational resources, making them more cost-effective and faster to deploy. This efficiency is crucial in manufacturing environments where real-time data processing and decision-making are essential.
Large language models, on the other hand, are general-purpose models trained on vast amounts of data, making them versatile but also resource intensive. While large language models excel in scenarios requiring broad language understanding, they can be less precise for specialized tasks.
Fine-tuning small language models can enhance their performance for specific tasks by customizing pre-trained models with additional training on targeted datasets. This approach allows small language models to achieve higher accuracy and relevance in their designated areas, making them more effective for specialized applications like manufacturing. Fine-tuning is also more cost-effective and efficient compared to training large language models from scratch, as it requires fewer computational resources and reduces operational costs. One of the key advantages of fine-tuning is the ability to control the model’s responses. Fine-tuned models are optimized for specific tasks, ensuring consistent and predictable behavior. This is crucial for applications where precise and reliable outputs are necessary.
For example, in manufacturing, fine-tuned models can be tailored to understand and respond accurately to industry-specific terminology and requirements. Fine-tuning also allows for better implementation of responsible AI practices, preventing unintended behaviors and ensuring models adhere to ethical guidelines. Using Microsoft Azure OpenAI Service, manufacturers can fine-tune small language models to address unique challenges.
Microsoft introduces new adapted AI models for industry
Microsoft partner, Sight Machine, has developed Factory Namespace Manager, a small language model specifically for manufacturing, using a fine-tuned version of Phi-3.5 SLM. Factory Namespace Manager is among the first partner-enabled adapted AI models for manufacturing available in the Microsoft Azure AI Foundry model catalog. It addresses a critical data governance challenge in the manufacturing industry: the standardization of factory data naming conventions. In many manufacturing environments, data is generated from a wide variety of sensors, machines, and systems, each with its own naming schema. This lack of standardization can lead to significant difficulties in managing and integrating data across different sources.
Factory Namespace Manager solves this problem by using AI to map the multitude of factory data naming schemas into unified corporate-standard namespaces or data dictionaries. This process enables manufacturers to integrate factory data with enterprise data systems, facilitating end-to-end optimization and improving overall operational efficiency. By creating a unified namespace, the tool helps ensure that data from different sources can be easily understood, analyzed, and utilized for decision-making.
“Our solution addresses a widespread challenge in the manufacturing industry, converting decentralized naming systems into a single corporate standard. 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.”
Kurt DeMaagd, Sight Machine Chief AI Officer and Co-Founder
This solution is particularly valuable for companies with extensive and diverse data sources from multiple generations of machinery, which often lack standardized labeling. Factory Namespace Manager makes it easier to manage and leverage this data, ultimately enhancing productivity and reducing the complexity of data management. By using AI, this tool bridges a significant gap in technology, enabling the mapping between original data field names and corporate standards. This capability allows manufacturers to seamlessly integrate factory data with enterprise data systems, facilitating end-to-end optimization.
The additional fine-tuning of AI models within Factory Namespace Manager helps affirm the tool can adapt to specific manufacturing environments and data sets, enhancing its accuracy and effectiveness. By adhering to principles of responsible AI, such as fairness, transparency, and accountability, the tool not only improves operational efficiency but also ensures ethical and trustworthy AI deployment in manufacturing.
“We really came to appreciate the importance of responsibly trained models. Even when dealing with seemingly mundane manufacturing data, it was essential to apply responsible AI principles correctly to prevent models from misbehaving. Support and guidance from Microsoft helped us improve the efficiency of fine-tuning and helped ensure the models developed were robust and reliable.”
Kurt DeMaagd, Sight Machine Chief AI Officer and Co-Founder
Sight Machine and Microsoft have developed additional innovative tools aimed at enhancing manufacturing productivity and efficiency.
AI and data revolutionize industries at Microsoft Ignite 2024
The post The future of manufacturing: AI for data standardization appeared first on Microsoft Industry Blogs.
Source: Microsoft Industry Blog