Headwaters announced that it had built a construction planning risk assessment support system used by approximately 3,000 J-POWER Group employees. Combining OCR, RAG and Azure OpenAI, the system supports hazard prediction and risk mitigation at worksites.
Headwaters announced that, as part of its AI platform business, it would develop and provide a risk assessment support system for J-POWER Group construction plans. The system uses OCR to read construction plans, searches and references relevant past documents through RAG, and uses Azure OpenAI to extract risk items and propose countermeasures. It is intended to prevent serious accidents at worksites and standardize operations.
In the demonstration, common tests were conducted for 3 types of work, including transmission tower foundations and tree clearing. Recall improved from 67.6% to 76.3%, while precision rose from 35.0% to 46.1%. The number of AI-generated information items was reduced by approximately 13.4%, an effect expected to reduce the workload involved in verification.
A feature of the system is that AI does not replace human judgment and is limited to providing support. It adopts a human-in-the-loop approach, with final decisions made by people. Going forward, the company plans to expand the types of work and documents covered and extend the system's use beyond the power and energy sectors to other areas of social infrastructure.
The source is a PR release from Headwaters, reported by VOIX. Further updates and expansion of the system's scope of application will be watched.