SUPERB AI.

Overview

The shipbuilding industry is rapidly embracing digital transformation, with smart technologies such as autonomous ships, digital twins, and smart production systems reshaping operations. Amid these advancements, AI-driven worker safety management has emerged as a critical focus.

A global leader in shipbuilding and offshore engineering, renowned for its world-class R&D, engineering expertise, and strong order backlog, reached out to Superb AI to maintain its continued growth momentum. The company, which actively sought to adopt AI to ensure workplace safety, decided to collaborate with Superb AI to enhance its worker safety monitoring system.

Challenge: The Dilemma of Securing Data on Workplace Accidents and Hazards

“Ensuring worker safety in shipyards is our highest priority. In particular, real-time detection of risks like worker falls is crucial, yet securing relevant data posed the biggest challenge. From a safety perspective, real accident footage should not exist, yet staging such scenarios for data collection was both costly and time-consuming.” – Team lead of the shipyard safety management team

The biggest challenge in developing an AI-powered safety monitoring system was securing high-quality data on hazardous situations. Real accidents should not occur, and staging safety incidents for data collection demanded extensive time and financial resources. Additionally, the AI model needed to be trained on diverse data reflecting a variety of work environments and conditions.

Solution: Unlock New Safety Possibilities with Generative AI by Creating Diverse Data on Hazardous Situations

"Superb AI’s generative AI technology provided the ideal solution for us. By leveraging real worksite images, we could seamlessly synthesize diverse hazardous situations that may occur at our sites, enabling us to efficiently generate high-quality training data to train our AI model."- Team lead of the AI development team

To address the customer’s challenges, Superb AI provided a tailored solution:

1. Realistic Synthetic Data Generation Based on Actual Work Environments

• Leveraged actual CCTV footage as backgrounds
• Applied SDXL inpainting models to naturally synthesize hazardous situations
• Created datasets reflecting a variety of workplace conditions and scenarios

Generative AI (SDXL inpainting model) synthesizing an image of a fallen worker in an industrial setting

Synthetic image of a fallen worker in an industrial setting

2. Efficient Data Generation System

• Intuitive user interface for easy dataset creation
• Automated metadata generation and validation
• Integrated data quality management system

3. Custom Demo Application

• Real-time data generation and validation
• Adjustable parameters for scenario customization
• Instant deployment and usability of generated data

Benefit: Faster AI Model Development and Enhanced Safety Management

"With generative AI, we rapidly secured training data for a wide range of hazardous situations. As a result, our AI model development became significantly faster, model performance improved, and real-time hazard detection capabilities in actual worksites were enhanced." - Project manager

AI object detection applied to a training image created by generative AI

AI object detection applied to a training image created by generative AI

Key achievements are as follows:

1. Accelerated Data Acquisition

• Eliminated the need for expensive staged-accident simulations
• Reduced AI model development time
• Streamlined data generation and integration

2. Enhanced AI Model Performance

• Improved hazard detection accuracy across multiple risk scenarios
• Increased real-time monitoring reliability
• Minimized false detections

3. Scalable AI System Development

• Enabled rapid collection of data on new safety threats
• Allowed continuous model performance optimization
• Facilitated expansion to other industrial settings

Conclusion: Pave the Way for the Smart Shipyard of the Future with Generative AI

This project demonstrated how generative AI can revolutionize safety management in the shipbuilding and offshore industries. Going forward, we will continue contributing to improved safety and efficiency in industrial worksites through innovative technology.

Appendix: SUPERB AI with AWS

Superb AI built efficient solutions using powerful AWS services.

The overall pipeline architecture for training and deploying an image generation AI model and running batch inference at scale using AWS infrastructure

First, we automate the training pipeline for the image generation AI model by combining AWS Step Functions, Lambda, and Batch. This enables efficient processing of large datasets and optimizes the training workflow.

The trained model is deployed using SageMaker Endpoint, Lambda, and API Gateway, providing customers with real-time access via API. This allows them to generate the desired images instantly.

Plus, for large-scale image generation, asynchronous inference is enabled using Step Functions, Lambda, and Batch. This allows customers to efficiently handle image generation requests for large datasets and receive results quickly.

By leveraging various AWS services, we build a comprehensive image generation AI solution that supports the entire process—from acquiring scarce datasets to model training, deployment, and large-scale inference.