Headwaters has built an AI-powered construction plan risk assessment support system for J-POWER, and is preparing and improving the production environment to help prevent field hazards from being overlooked and standardize risk assessments.
Headwaters announced that it had built a construction plan risk assessment support system used by approximately 3,000 J-POWER group employees. The system is designed to enable risk assessments that are less dependent on individual experience when extracting hazards at worksites. AI output is positioned as a reference for cross-checking, not as a substitute for the final decision.
The system combines OCR processing of construction plans, RAG-based searches of related documents and risk extraction support using Azure OpenAI. It provides functions for checking hazards, scoring their severity, displaying mitigation measures and collecting feedback on responses. The intended operating model is for personnel to make the final decision after cross-checking the results against site conditions.
The initial verification evaluated coverage and accuracy for transmission tower construction. Following migration to the production environment, the company is making phased improvements, including screen enhancements and the addition of input parameters. It plans to expand the applicable work types and add document types going forward.
In verification results for July to September 2026, recall across 3 work types, transmission tower foundations, tower assembly and tree clearing, improved by 8.7 percentage points from 67.6% to 76.3%, while precision improved by 11.1 percentage points from 35.0% to 46.1%. The number of generated items per case was reduced by approximately 13.4%, from 145.8 to 126.3.