When challenged by fragmented knowledge and time-intensive manual analysis, artificial intelligence helped to accelerate decision-making during operational events
Artificial intelligence (AI) is no longer a futuristic concept in the chemical process industries (CPI) — it is a present-day driver of operational efficiency and competitive advantage. In fact, more than 80% of CPI executives surveyed by IBM [1] acknowledge that AI will have a profound impact on their business within the next three years. Already, companies are deploying AI in areas such as research and development (R&D; 74%), manufacturing (61%), risk management (58%) and forecasting (47%) to reduce costs, improve product quality and accelerate innovation.
One company leading the charge in real-world AI implementation is TotalEnergies (Paris, France; www.totalenergies.com), which has introduced AI-powered knowledge assistants at its Antwerp refinery (Figure 1). In partnership with Sinequa by ChapsVision (www.sinequa.com), TotalEnergies developed and deployed AI assistants, internally code-named as MILAa (My Intelligent Learning Application) and JAFAR (Generative AI for Return of Experience) to overcome long-standing challenges in refinery knowledge management and root cause analysis. These tools use natural language processing (NLP), machine learning (ML) and intelligent search to unify operational data, reduce downtime and boost decision-making speed and accuracy.

FIGURE 1. TotalEnergies’s site at Antwerp, Belgium, with close to 1,700 employees, was the first of the company’s sites to leverage AI-powered search for feedback and analysis of incident causes (Photo credit: TotalEnergies)
Operational data challenges
Before deploying these AI solutions, TotalEnergies engineers faced significant hurdles in accessing critical operational data. Root cause analysis (RCA) reports, maintenance records and technical manuals were dispersed across siloed systems, stored in inconsistent formats and often available only in a single language.
This manual, document-heavy approach to problem-solving introduced three major risks:
- Extended downtime due to delays in identifying root causes of failures
- Inconsistent decision-making, because learnings were not easily shared across teams or sites
- Repeated operational incidents, costing millions in lost productivity, repairs and supply chain disruptions
In one example, multiple pump failures over a two-year period could have been avoided if relevant RCA data had been available in real time across teams.
Domain-specific AI assistants
To address these challenges, TotalEnergies partnered with Sinequa, a leader in enterprise intelligent search, to build AI assistants specifically tailored to the needs of industrial operators and engineers.
MILAa: A smarter approach to learning from past failures. MILAa was developed to consolidate more than 1,000 RCA documents from TotalEnergies’ Antwerp refinery into a centralized, AI-enhanced knowledge base. Built on Sinequa’s enterprise search platform, the tool leverages domain-specific ontologies to extract and organize structured information, such as equipment types, failure modes, downtime durations and the effectiveness of remedial actions. Engineers can interact with MILAa using natural language queries, significantly reducing the time required to locate relevant data. By eliminating the need to manually sift through dense technical reports, MILAa not only streamlines problem-solving but also facilitates cross-site learning — helping to standardize operational responses and preventive strategies across TotalEnergies’ global refinery operations.
JAFAR: From search to smart dialogue. To enhance usability, TotalEnergies deployed JAFAR, a generative AI assistant that converts static documents into dynamic, conversational insights. JAFAR extends MILAa’s capabilities by automatically translating technical RCA documents into French, Dutch, German and English, while preserving industry-specific terminology. This is possible thanks to a custom-built internal dictionary that supplements the base model, ensuring the AI understands TotalEnergies’ unique language and acronyms.
“Automatically translating these documents is no easy task, as they contain technical language and terminology specific to TotalEnergies’ core business,” said Pierre Jallais, lead architect for Smart Search Engines and LLMs at TotalEnergies. “Our business relies on many specific terms and acronyms tied to our activities, so with Sinequa, we were able to enhance the default model by integrating an internally developed dictionary. This provides JAFAR with more context, enabling it to better understand and process our unique documents.”
This domain-specific enhancement allows JAFAR to segment RCA content, extract failure patterns and summarize data in a format that is both technically accurate and easily navigable, dramatically improving how engineers interact with operational knowledge.
AI across the CPI
While TotalEnergies focuses on operational knowledge and equipment uptime, AI is revolutionizing other parts of the chemical sector as well. In R&D, machine learning helps scientists identify promising molecules, optimize chemical formulas, and forecast efficacy — accelerating product development cycles and reducing costs. In supply-chain planning, AI reduces forecasting error by as much as 50%, helping companies manage raw material procurement and reduce inventory waste. These tools also support predictive forecasting, allowing CPI firms to better anticipate pricing shifts and demand fluctuations, which in turn increases profitability and supply-chain agility.
TotalEnergies’ AI deployment is a natural extension of these broader innovations — applying similar techniques to equipment reliability, maintenance planning and operational safety.
Optimizing performance
Issues, disruptions and downtime can often cost refineries millions in operational losses. A global survey of over 3,000 plant maintenance decision-makers commissioned by ABB [2] reveals that over two-thirds of industrial businesses experience unplanned outages at least once a month, costing the typical business close to $125,000 per hour, or up to one million dollars per eight-hour shift. According to a study by Kimberlite [3], unplanned downtime in the oil-and-gas industry can cost offshore organizations an average of $38 million annually. Organizations with the worst performance saw financial impacts exceeding $88 million annually. Even a 1% downtime rate, equivalent to 3.65 days, can cost more than $5 million per year.
Since going live in February 2024, the MILAa and JAFAR systems have delivered measurable performance improvements at TotalEnergies’ Antwerp refinery. The implementation has reduced the time required to perform root cause analyses, significantly accelerating decision-making during operational events. In just the first six months, the insight gained from the tools informed decisions that helped prevent three major equipment failures, each of which could have resulted in millions of dollars in operational loss. ■
Edited by Dorothy Lozowski
References
- IBM, Research Insights: Optimizing the chemicals value chain with AI, www.ibm.com/thought-leadership/institute-business-value/en-us/report/chemicals-value-chain-ai, 2020.
2. ABB, ABB survey reveals unplanned downtime costs $125,000 per hour, press release, www.new.abb.com, Oct. 11, 2023.
3. As reported by Maxgrip, Understanding the Cost of Downtime in Business Operations, www.maxgrip.com/resource/article-the-cost-of-unplanned-downtime/, accessed April 25, 2025.
Author
Laurent Fanichet is advisor and vice president, corporate communications for Sinequa by ChapsVision (251 W 30thStreet, 8th Floor, New York, N.Y. 10001; Email: [email protected]; Website: sinequa.com), where he leads Sinequa’s global communications strategy and analyst engagement, and plays a critical role in supporting customer success initiatives. With over 20 years of software industry experience across Europe and the U.S., he previously served as vice president of marketing, where he built the U.S. marketing team from scratch, contributing to Sinequa’s growth in North America and helped establish Sinequa’s leadership with firms like Gartner and Forrester. Before joining Sinequa, he held senior roles at Quantum Corporation, Atempo and ILOG.