Let’s say that you’re responding to a new RFQ from an automotive manufacturer. You need to organize the requirements of the manufacturer, comply with all the standards and regulations, and follow the specifications.
But wait – this sounds like an RFQ your company responded to a year ago.
Could you use AI to look through old responses to look for similarities and differences? And while you’re at it, how about having AI take a first pass at generating the response as well?
This may sound like the future, but in this case – the future is now. Let’s break this down.
Step 1: Compare goals and constraints
You could do this manually. You could compare the specific design goals for the RFQ and similar proposals your company has submitted in the past. These might include improving performance, reducing weight, lowering manufacturing costs, or meeting new regulatory standards. You’ll also want to compare proposal constraints like materials, dimensions, safety factors, manufacturability, and compliance requirements. Or, if you’re using AI, you could just ask “Please find all similar proposals with similar goals and constraints.”
Step 2: Collect and Prepare Data
For this step, you’d want to examine all historical design data including CAD models, schematics, simulations, and test results from previous designs. You may want to annotate prior designs with relevant metadata, such as functionality, performance metrics, material properties, or manufacturing parameters. Finally, if you’re using AI, you’ll want to convert all data into standardized formats that AI tools can process (e.g., STL, STEP, or OBJ files for CAD). If you’re using AI, you could ask “based on the similar proposals you just pulled, please list historical design data that might be relevant to the new RFQ.”
Step 3: Leverage Generative Design Tools
Now you’re ready to generate preliminary responses and designs for the current RFQ. To do this, you can use AI-driven algorithms to generate optimal designs based on defined inputs (e.g., loads, forces, and constraints). You can also use AI for material distribution recommendations with the goal of creating lighter and more efficient designs while maintaining strength and performance. You may want to use AI to adjust key parameters and evaluate how they influence performance or cost.
Step 4: Train custom AI models or choose a partner who has done so
It’s easy to say, but much harder to do: You’ll need an AI model that has been trained on historical designs and all of their associated performance outcomes to predict and propose new designs. Preferably, your model will not just be trained on your designs but other designs from outside the organization as well. You can use (or find a partner who uses) Generative Adversarial Networks (GANs) as well. GANs can be used to create variations of existing designs by learning patterns from historical data. It’s also possible to use reinforcement learning to iteratively improve designs by testing different configurations in simulated environments.
Step 5: Put human feedback in the training loop
The importance of human feedback cannot be overstated. It will be important to collaborate with engineers to review AI-generated designs, ensuring feasibility and alignment with project goals. It will also be important to update the AI with feedback from design successes and failures to improve future outputs.
Step 6: Implement, monitor, and iterate
To implement this process, you will probably need to generate manufacturing instructions. Luckily, with AI, this is somewhat simple. You can use AI to create or adapt CAM files, tooling designs, or 3D printing instructions from the finalized designs. You’ll also want to build or test the prototypes for real-world testing and refinement, monitor how the AI-generated designs are performing in the real world, and use performance data to train future AI models, creating a cycle of continuous improvement.
I hope this helps and if you’d like to work with us at Humaxa, we’d be delighted to work side-by-side with you to shape the future of using AI to generate engineering designs!
Carolyn Peer
CEO/Co-founder, Humaxa
