Is Prompt Engineering the New Mantra?

AI has sparked both excitement about career growth and fear of job loss. It’s led to discussions on employment, upskilling, and development while prompting people from all walks of life to examine their learning paths. 

Generative AI and tools like Gemini and ChatGPT have popularized prompt engineering, and there are countless courses on the topic. Most experts have found that clear, contextual, iterative prompting can benefit general AI usage. Businesses also seek practical applications for measurable outcomes.

🔎 Example: Engineering firms will incorporate generative AI into their engineering workflows to achieve faster, higher-quality results with fewer personnel.

How To Prompt Like a Pro

In The Case for Common Sense in the Age of Physical AI, I used a simple example to illustrate the complexity of prompts. The article referenced Memorex Corporation’s 1970s wine glass commercial titled: Is it live, or is it Memorex? 

It explains resonance, where Ella Fitzgerald's singing frequency matched the natural frequency of a wine glass, causing it to shatter. This phenomenon occurred during the live performance and when played back on Memorex tape. We could use generative AI to analyze the factors leading to the wine glass shattering. 

💡Tip: Prompts must be refined with domain knowledge to achieve a comprehensive understanding appropriate for engineering applications. 

Framework for Domain-specific Generative AI

A solid grasp of first principles is crucial for domain-specific generative AI in complex physical systems. The following framework aids in the design and analysis within mechanical engineering. Other fields can also create frameworks based on undergraduate and graduate curricula.

These are undergraduate and graduate-level mechanical engineering courses that form the basis of domain-refined prompts:

• Engineering Graphics

• Kinetics

• Kinematics

• Statics

• Dynamics

• Fluid Mechanics

• Solid Mechanics

• Thermodynamics

• Dynamics of Structures

• Engineering Economics

• Electrical Circuits

Mechanical Engineering encompasses a comprehensive study of the various subjects detailed above. During the first and second years, foundational courses in mathematics, physics, and chemistry complement these. Overall, the curriculum lays the groundwork for incorporating domain-specific generative AI as an enhancement tool within mechanical engineering processes.

Revisiting the Wine Glass Lesson for Real-World Applications 

Consider resonance-induced mechanical vibrations in real-world applications, such as a hard disk drive. Hard drives still make up 50% of global data storage. A hard disk drive, 1 to 20 TB capacity in a server, has read/write heads that fly above the rotating disks while traversing them. 

As with the shattering wine glass from my previous newsletter, the read/write sub-system can be subject to faults from resonance-driven excessive vibration. Alignment with the frequency of an external stimulus, such as the rotation of a cooling fan in the server enclosure, can be a reason.

The resonance-driven arm vibration can lead to soft failures, such as a reduction in data throughput as read/write heads keep retrying. Or, in the worst case, a hard failure. Indeed, this phenomenon has resulted in numerous expensive issues in IT data centers—a story we’ll discuss in a subsequent newsletter. 

💡Tip: Domain-specific generative AI prompts can be part of the tool chest for designing hard drives and other components. 

Gen AI at Work

So, how could an engineer use generative AI? Let’s consider a day in the life of Chandrakant Patel, the mechanical engineer, in the 1980s and now.

I first started working in Silicon Valley 42 years ago as a hard disk design and manufacturing engineer. I created designs on a drafting table and performed detailed engineering analysis by hand with the help of a calculator such as an HP 15C. By the late 1980s, I started using computer-aided design (CAD) and computer-aided engineering (CAE). 

Today, I would use CAD-CAE and domain-specific Generative AI to augment my workflow. 

Case of Hard Disc Design with Gen AI augmentation

Hard drives have a knowledge base of more than four decades, much of it captured on the internet. Generative AI can help by summarizing hard drives' state of the art design. 

Basic questions such as “show me hard disk drive designs” and “show me head arm assembly designs” can provide a great background. These questions will return publicly available information in summary form.

Next, I will pose the following more specific question, shown below:

I have established that precision, vibration, and resonance-induced excessive vibration of structures are all key challenges. Fluid mechanics and heat dynamics are also important. All these are fundamental areas in the framework shown above.

Using Generative AI for Domain-specific Cases

When I get into the details of the head arm assembly design, generative AI becomes more domain-specific. As shown in the sketch, in a hard drive, an arm with a flexible (Flex) beam carries a read/write head. When the disc is spinning, the air gap between the disc and the aerodynamically shaped read/write head is 3 to 5 nm (i.e., three billionths of a meter). Oscillations can cause degradation in data throughput or, worse, a head crash. 

Having learnt the basics, such as the “wine-glass phenomenon” mentioned in my previous newsletter, I calculate the natural frequency of the arm and the flex (see figure) based on the:

• Material type

• Geometry

• Boundary condition

The boundary conditions define how the arm is held at its edges. Using my first-principles knowledge, I chose the boundary conditions as fixed at one edge and free at the other edges.

Having calculated the natural frequency of the arm, I return to generative AI with domain-specific questions.

🔎 Examples: “What are possible external vibrations that can cause a hard drive to fail?” or, “What are the external frequencies that can align with the head arm assembly in a hard drive?”

Parting Thoughts

In this newsletter, I showed the point where generative AI becomes more domain-specific. We’ll continue this example in the next edition, show the calculations, and tie it to a real-life case.

In the future, AI will help us calculate the natural frequency of the arm as a prompt by integrating into design software. However, given my experiences in engineering, I prefer to set up a problem and conduct a first-pass calculation by hand to get a “feel.”  

Generative AI also consumes a lot of energy. So, we should probably ask ourselves more often: 

About the Author

Former SVP, Chief Engineer, and Senior Fellow at HP, Chandrakant is a leader in AI, energy-efficient computing, and sustainability. He is an IEEE Fellow, ASME Fellow, member of the NAE, and the Silicon Valley Engineering Hall of Fame. Follow me on LinkedIn or email me at [email protected].

Want to stay ahead of the curve with insights into the newest advancements in Physical AI? Subscribe to Chandrakant’s newsletter at GenAI Works.

🚀 Boost your business with us—advertise where 10M+ AI leaders engage

🌟 Sign up for the first AI Hub in the world.

📲 Our Socials