Software Architecture
for Responsible Artificial
Intelligence Systems:
Practice in the Digitization
of Industrial Drawings

1. Introduction

Purpose of responsible AI in industrial drawing digitalization

Challenges in digitizing historical drawings

Complexity in historical drawing formats, domain-specific knowledge, and high error risk in manual digitization.

Architectural Patterns: MVC, Layered Architecture

Model View Controller (MVC) for modular design

Layered Architecture: Separates concerns, enhancing usability and style adaptability

Federated Learning for privacy

Federated Learning: Secures data by training models locally

Collaborative multimodel fusion

Integrates specialized AI models (e.g., text and object detection)

End-user satisfaction (usability, interactivity, etc.)

Interactivity, usability, and interpretability scored highly

Performance Gains: P&ID tasks completed in days instead of months

Federated model accuracy improvement

Adapting to diverse drawing standards

Limited samples for specific element training

Responsible AI software architecture

Responsible AI Principles: Aligns with EU's seven AI ethics principles (e.g., transparency, accountability).

Role of AI technologies (OCR, computer vision, etc.)

Computer Vision for element detection

OCR for text extraction

Knowledge Graphs for relational data

Quality attributes (Usability, High performance, Accuracy)

Usability: Non-AI professionals should navigate easily

Performance: Efficient processing for large, multi-page drawings

Security & Privacy: Protection through federated learning

Application in digitizing historical piping & instrumentation diagrams

Performance analysis: accuracy, time, availability

Benefits of AI in drawing digitization

Reduces time and errors in industrial drawing digitization

Future improvements (interaction methods, expanded AI stack)

More interactive options (e.g., voice control)

Expanding AI models to support various industrial applications