292151911 | Aleksei Gorodenkov | Dreamstime
665f161a81d78b59f784dc61 Dreamstime M 292151911

How one manufacturer made all its digitized data easily searchable. Hint: It was AI

June 5, 2024
Agrochemical company plugged natural language-based artificial intelligence into its PLM and, now using a “smart” search tool that was five years in development and prototyping, query time for plant employees has been shortened and problem-solving accelerated.

Process manufacturing plants generate thousands of data points every day, from automated sensor readings to notes and observations from plant operators. The answers to today’s problems are often buried beneath mounds of historical data; the trick is finding them.

But one artificial intelligence-enabled system, with the proprietary name Smart Search, helped one agrochemical company transform digitized data into a usable knowledge platform that empowers their people.

See also: Epicor eyes faster growth as AI tools find uses

Digital tools for shift handover, inspection rounds and plant process management (PPM) platforms are helping manufacturers collect and centralize data and notes for future retrieval and analysis. Digitizing this vast amount of information like historical trends and events into a centralized knowledge management system helps troubleshoot and allows faster decision-making.

See also: Survey: 90% of manufacturers are using AI, but many feel they lag behind competitors

These benefits depend on workers finding critical information when they need it. For the agrochemical company, that proved to be a challenge, even with all data digitized. A traditional keyword-based search system can produce thousands of results of varying degrees of relevance. If the user doesn’t know what keywords to use or the original entry is incomplete, finding the right information is like looking for a needle in a haystack.

Operators needed a system that would enable efficient access to all the historical knowledge captured in their PPM—even if entries were incomplete, improperly tagged, or used uncommon abbreviations or spellings. Ideally, they would be able to query the system using natural human language (e.g., “Why is this product brown instead of gray?”).

Podcast: Cybersecurity action steps and the dilemma of guarding private data

The system needed to be able to understand what was being asked and find the most relevant and helpful information in the historical dataset. That’s where AI helps their team.

Building a domain-specific search program

AI is a powerful tool for Smart Search. Using a form of AI called natural language processing (NLP), such a system can “understand” what users are asking in plain language. NLP also allows the system to find more relevant results by analyzing text for meaning instead of just looking for keywords.

But in a highly specialized domain such as agrochemical manufacturing, an off-the-shelf AI solution doesn’t cut it. Entry logs often contain technical terms and abbreviations specific to an industry or even a company. Entries may also be incomplete, contain misspellings, or use different languages or dialects across international locations. Commercially available NLP tools trained on well-written, grammatically correct text (usually in common English) were simply not up to the task.

Podcast: Generative AI on the plant floor

To solve the problem, eschbach worked with AI researchers at the University of Göttingen in Germany to adapt an AI search tool for the needs of the manufacturer. It started with a systematic customer research study with user groups, workshops, and onsite investigations to study the users’ workflows and requirements. Nearly five years of development, prototyping and beta testing resulted in a customized search solution that understands their language, workflows, and user needs.

Now, users can query their PPM in natural language to find historical data and get to the bottom of emerging problems. The search program has been well-accepted and widely used by workers at all levels in the company, from shop-floor operators to process engineers.

See also: AI abuse by cyberattackers is a people—not technology—problem

And as the company uses this search tool, it will learn more and more, increasing search efficiency at the plant. As others in the company globally access the search tool, it will be increasingly more valued as expertise from around the globe will be available to any operator, anywhere, at any time.

Prior to implementing the tool, it often took employees several minutes or longer to find the information they needed in the PPM—if they could find it at all. For many types of queries, only information from the last month was accessible.

See also: Your competitors are using AI. Why risk falling behind them?

Now, they can easily search through more years of digitized logs and data and find the answers they need in seconds. With NLP, results don’t depend on workers knowing the exact search term or entry logs being correct and complete.

The Smart Search tool has significantly enhanced the productivity and effectiveness of their workforce. The PPM now serves as an effective knowledge transfer platform, allowing employees to learn and benefit from historical knowledge and data.

By increasing access to key operational information and historical knowledge, the search program accelerates problem-solving and troubleshooting, reduces safety and compliance risks, and drives overall operational efficiencies.

About the Author

Andreas Eschbach

Andreas Eschbach is the founder and CEO of global software company eschbach and invented plant process management platform Shiftconnector, which helps production teams streamline shift-to-shift communications and enables better data sharing and workforce collaboration. He has led a variety of international software initiatives for major process manufacturing companies, especially in chemical and pharmaceutical industries.