Revolutionising Engineering Safety

Client Overview

Industry

Oil and Gas

Duration

1 year

Time

Aug'24 – Ongoing

Team

Team of 7

Using advanced AI and computer vision, it processes PFDs and PNIDs 50x faster than traditional methods, reducing costs by 93% and error rates by 90%, while enabling real-time safety decision-making across SATC's infrastructure.

The Challenge

Saudi Aramco Technologies Company (SATC) faced a critical challenge in their engineering operations. Their Quantitative Risk Assessment (QRA) process, essential for evaluating isolation procedures, EPS Transfer, and ensuring safety compliance, relied heavily on complex, time-consuming manual analysis of engineering diagrams. Engineers struggled with accuracy concerns and real-time decision-making bottlenecks that posed significant risks to operational efficiency and safety.

The stakes were high: these processes directly impact the safety of engineering operations across SATC's vast infrastructure, where any oversight could have serious consequences.

Solution: AI-Powered Intelligence

ProcessFlow AI emerged as SATC's game-changing solution, empowering the company to make faster, smarter decisions in safety and engineering while significantly improving operational efficiency

Data Foundation

Data Collection & Preparation

Open Source Training Data: 10+ public P&ID diagrams for initial model training
Proprietary SATC Data: 3,000 Saudi Aramco P&IDs with domain-specific symbols
Data Types: Engineering diagrams, Heat/Material Balance tables, equipment specifications
Annotation Volume: 500,000+ component annotations across diagrams

Our Data Engineering Role

Symbol Library Creation: Built comprehensive database of 2,000+ industry-standard symbols
Template Matching: Developed algorithms for component extraction from complex diagrams
OCR Pipeline: Custom text extraction for equipment tags and line numbers
Domain Adaptation: Fine-tuned on SATC-specific symbols and nomenclature

Architecture

AI Infrastructure

Model Selection: Custom vision transformers optimized for diagram analysis
GPU Server: A100 instance at 300 USD/month (cheapest market option for large models)
Processing Pipeline: Distributed processing across multiple models for efficiency
Caching Layer: Redis for frequently accessed diagram components

Technical Stack

Backend: FastAPI with async processing for high throughput
ML Framework: PyTorch with custom architectural components
OCR Engine: Combination of Easy-OCR and TR-OCR for text extraction
Database: PostgreSQL for component relationships and metadata
Deployment: Kubernetes for scalable container orchestration

Operational Costs

Traditional SME Approach: 45,000 USD/month

10 SMEs at 4,500 USD/month each
Processing: 2-3 P&IDs per day
Error rate: 30%

ProcessFlow AI Solution: 3000 USD/month

A100 GPU server: 3000 USD
Processing: 100+ P&IDs per day
Error rate: 2%

Explore More Breakthroughs

Spiral Health

AI-powered posture analysis delivers personalized physical therapy exercises in real-time through smartphones.

See Spiral in action