This initiative is an AI-powered tool developed for the Saudi Water Authority that automatically reads and interprets complex engineering diagrams used in water desalination plants. It identifies equipment, classifies plant processes, and allows engineers to ask questions about the diagrams through a chat interface. It was developed to reduce the time and manual effort required to analyse these technically complex documents.
Initiative overview
Water desalination plants rely on highly detailed engineering diagrams known as Piping and Instrumentation Diagrams (P&IDs) to document the equipment, piping, and processes that make up their operations. Manually reading and interpreting these diagrams is time-consuming, technically demanding, and prone to human error. The Saudi Water Authority identified an opportunity to apply generative AI to automate this process, reducing the burden on engineers and technicians while improving the speed and accuracy of information retrieval.
The initiative delivers an AI-powered system that automatically processes P&ID documents, extracts key information about equipment and components, and classifies the type of desalination process depicted. The system then generates structured summaries of the workflow for each process, giving engineers a clear and concise overview without needing to manually trace through complex diagrams. A user-friendly dashboard presents this extracted data in a visual and accessible format. Additionally, the system includes a conversational chat interface, allowing users to ask questions directly about the content of specific P&ID documents and receive instant, contextual answers.
The core objectives of the initiative are to automate data extraction from P&IDs, improve the accuracy of equipment and process identification, streamline asset management through comprehensive equipment listings, and enhance operational efficiency by reducing manual intervention and minimising downtime. The system is designed to process data with minimal latency, comply with relevant data privacy and security standards, and requires minimal training for end users.
Looking ahead, the system is designed with scalability in mind, with the capacity to expand across multiple desalination plants and accommodate increasing volumes of P&ID data. As the tool matures through user feedback from engineers and technicians, it is expected to be institutionalised as a standard part of the Authority's plant management workflow, reducing reliance on manual document interpretation and supporting more informed decision-making across operations.
Other relevant details
Results, outcomes and impacts: The initiative is currently being deployed. A proof of concept was completed with promising results. The system achieved greater than 90% accuracy in identifying desalination process types and equipment from P&ID diagrams, meeting the initiative's core performance target. Notably, the AI-powered approach processed and interpreted diagrams approximately ten times faster than manual methods, representing a substantial gain in operational efficiency for engineers and technicians. As the initiative moves toward full deployment, expected impacts include further reductions in manual effort, improved equipment cataloguing accuracy, and faster decision-making supported by the conversational chat interface. Success will continue to be measured.
Challenges and lessons learned: A key technical challenge is the variability and complexity of P&ID documents, which often differ in format, notation, and level of detail across plant sites. Ensuring the AI achieves consistent accuracy across non-standard diagrams requires careful model design and validation. Early lessons highlight the importance of involving engineers and technicians from the outset, as their domain expertise is essential for validating extracted data and workflow summaries. For initiatives of this kind, success depends on strong subject-matter expert engagement, high-quality and representative training data, realistic accuracy targets, and institutional commitment to embedding the tool into existing operational workflows rather than treating it as a standalone solution.
Additional insight: The Generative AI for Engineering Design initiative is more than an automation tool; it is a strategic asset that aligns the Saudi Water Authority with the goals of Saudi Vision 2030, demonstrating how national utilities can apply cutting-edge AI to manage complex, safety-critical engineering data at scale. A distinctive feature of the system is its multimodal accessibility. Recognising that engineers and technicians are frequently on-site at desalination plants rather than at a desktop, the conversational chat interface is designed to be mobile-responsive, ensuring that critical P&ID intelligence is accessible anywhere, at any time. The initiative prioritises sovereign data management, ensuring all P&ID data which carries significant operational and national infrastructure sensitivity is processed and stored in compliance with local data residency requirements and national security protocols.
