AI Agent in High-Speed Propeller Design: ShipTechAI Case Study

As a marine engineer with 37 years of experience, I've witnessed the transformation from manual calculations to CAD design. Now, AI agents are bringing revolutionary changes to engineering design. Through the ShipTechAI project, I've experienced firsthand how AI agents can dramatically improve design efficiency while maintaining engineering rigor. This article shares real cases and insights from this transformation.

1. The Limitations of Traditional Propeller Design

1.1 Complexity of Traditional Workflow

As a ship designer, I've spent decades designing high-speed propellers. Traditional design processes typically include:

• Manual database queries: Finding reference cases from design manuals and literature
• Complex calculations: Thrust calculation, cavitation check, strength verification
• Repeated modifications: Manual parameter adjustment based on experience
• Long cycles: Initial design takes 4-6 hours, multiple iterations even longer

1.2 Limitations of Traditional Methods

These limitations led me to consider: Can we use AI technology to improve design efficiency? This question inspired the ShipTechAI project.

2. ShipTechAI Project Background

2.1 Project Vision

ShipTechAI aims to create an intelligent tool for surface piercing propeller (SPP) design. The core objectives are:

• Data digitization: Digitize authoritative Rolla SPP database
• Intelligent design: AI-assisted rapid preliminary design
• Visual analysis: 3D visualization of design results
• Practical application: Provide professional, usable engineering tools

2.2 Technical Challenges

As someone who learned Python just 10 days before this project, I faced several major challenges:

1. Database processing: How to digitize and interpolate 101 Rolla SPP design points
2. Algorithm implementation: Complex 3D interpolation and performance prediction
3. Interface design: Create intuitive, professional engineering interface
4. Engineering verification: Ensure calculation accuracy and reliability

With only 10 days of Python experience, completing such a complex project seemed impossible. But AI agents changed this reality.

3. AI-Assisted Development: Deepseek and Claude Collaboration

3.1 The Power of Multi-Agent Collaboration

In the ShipTechAI project, I adopted a multi-AI-agent collaborative approach:

3.2 Real Development Cases

Case 1: 3D Interpolation Algorithm Implementation

The Rolla SPP database contains 101 design points requiring 3-dimensional interpolation (diameter, pitch ratio, blade area ratio). This is core but complex:

• Challenge: How to accurately interpolate in 3D space
• Deepseek solution: Provided complete scipy interpolation code
• Claude optimization: Improved boundary condition handling and error checking
• Result: Implemented high-precision 3D interpolation in 2 hours

Case 2: Intelligent Diagnostic System

Users may input unreasonable parameters during design. How to intelligently identify and give suggestions?

• Challenge: Engineering knowledge combined with smart judgment
• My role: Provide engineering criteria and domain knowledge
• Claude design: Implement multi-level diagnostic logic
• Deepseek implementation: Fast code generation
• Result: Comprehensive diagnostic system covering 6+ common scenarios

3.3 Division of Labor Insights

After two months of practice, I discovered the optimal collaboration pattern:

Task Type	Primary AI	Reason
Core algorithm	Deepseek	Fast, accurate calculations
Architecture design	Claude	Strong systematic thinking
Error debugging	Claude	Excellent at logical analysis
Code generation	Deepseek	High generation speed
Code review	Claude	Thorough review
Optimization suggestions	Claude	Comprehensive consideration

4. Key Technical Implementation

4.1 Database Digitization

The Rolla SPP database is an authoritative reference in the industry but exists only in paper form. We:

1. Manual digitization: Transcribed 101 design points into structured data
2. Data validation: Cross-checked multiple literature sources
3. Database design: Created efficient query structure
4. Interpolation algorithm: Implemented 3D interpolation for intermediate states

4.2 Intelligent Design Process

User only needs to input basic parameters:

• Ship speed and displacement
• Number of propellers
• Installation position (fully submerged / surface piercing)

System automatically completes:

1. Resistance estimation: Calculate ship resistance based on speed and displacement
2. Power distribution: Reasonably allocate total power
3. Propeller design: Match optimal diameter, pitch, blade area ratio
4. Performance prediction: Calculate efficiency, thrust, torque
5. Intelligent diagnosis: Identify potential problems and give suggestions

4.3 Visualization

Used Plotly to implement:

• 3D propeller model: Intuitive display of design results
• Performance curves: Efficiency vs. advance coefficient relationship
• Comparison analysis: Compare multiple design schemes

5. Real Application Results

5.1 Efficiency Improvements

Comparing traditional methods vs. ShipTechAI:

Design Phase	Traditional Method	ShipTechAI	Efficiency Increase
Preliminary design	4-6 hours	10 minutes	96%
Parameter optimization	2-3 hours	30 minutes	83%
Scheme comparison	Half day	1 hour	75%
Report generation	2 hours	5 minutes	96%

5.2 Real User Feedback

5.3 Unexpected Benefits

Beyond efficiency improvements, ShipTechAI brought unexpected value:

• Knowledge inheritance: 37 years of design experience solidified into tool
• Lower entry barrier: Newcomers can do professional design
• Quality improvement: Reduce human errors, improve consistency
• Innovation stimulation: More time to explore innovative designs

6. Experience Summary and Future Outlook

6.1 Key Insights

1. "Domain Knowledge + AI Agent" is the optimal combination

• AI provides technical implementation capability
• Professionals provide domain knowledge and verification
• The two are indispensable

2. Multi-agent collaboration better than single agent

• Different AIs have different strengths
• Reasonable division of labor improves efficiency
• Cross-validation improves reliability

3. Engineering thinking remains core

• AI is a tool, not a substitute for thinking
• Problem decomposition and systematic design still rely on humans
• Engineering verification and quality control require professional judgment

6.2 Advice for Other Engineers

6.3 Future Plans

ShipTechAI is just the beginning. Future development directions include:

• Database expansion: Add more propeller series data
• Function enhancement: Add cavitation analysis, strength calculation
• Multi-language support: Serve global users
• Cloud deployment: Provide online services
• Community building: Attract more engineers to participate in development

7. Conclusion

As a 59-year-old engineer, developing ShipTechAI taught me: Age is not a barrier to embracing new technologies, but a mindset issue. With the help of AI agents, we can achieve things that were previously unimaginable.

The value of AI agents is not replacing engineers, but empowering them. It allows us to:

• Escape tedious repetitive work
• Focus on creative and high-value work
• Transform decades of experience into shareable tools
• Lower industry entry barriers and promote knowledge dissemination

If I can develop professional engineering tools with just 10 days of Python experience, so can you. The key is: dare to try, embrace AI, and maintain engineering rigor.

The AI era is already here. As engineers, we can choose to resist, or we can choose to embrace and lead. I chose the latter and saw unlimited possibilities.