I’m building a radically new product - how do I use AI to save time?

Let's start with a typical engineering workflow

1. Define requirements
2. Iterate your design by running simulations
3. Build a prototype and test it
4. Figure out how to manufacture it
5. Launch and learn how your customers use your product

Machine Learning or AI is all about pattern recognition and fast model creation based on data. So we are looking for the processes that create the most data where help with pattern recognition or fast modelling can speed up the workflow.

Pre-design: You can use machine learning to learn what the market wants by analysing online or social media data using Natural Language Processing (NLP).

Iterative Early Design: You can change your workflow from - iterating between design changes and feedback through tests and simulation to - you build up a large database of designs (e.g. 10,000), you use that to teach an AI, and then the AI helps you to go through design iterations in hours not days.

Lots of engineers find this concept counter-intuitive because it feels inefficient to run 500 simulations instead of let's say 5. And they are right that computationally it is 100 times more expensive. However, companies like Google realise that computational cost is dwarfed by the value that accelerating your overall workflow by 90% creates. (check out the google press release)

Here is an example from Monolith: The team ran 600 CFD simulations up front and you can now find the design interactively in minutes playing with a dashboard.

Design Validation: Once you have a prototype, you need to understand how reliably it is going to work and when it might fail. For a smart meter, it means checking it out in different homes and seeing if it always works in different locations and temperatures. This step always requires either running lots of simulations or doing lots of tests to build a high-quality product and spend weeks looking at results to make sure it all works. This is where machine learning shines and really helps with identifying patterns and finding problems.

Here is an example from Monolith. Once you have finished the design of your aircraft, you want to understand how it will perform in different flight scenarios (altitude, weather, ...). The below dashboard helps you do that more quickly.

Production: Manufacturing processes are repetitive and create lots of data. Quality control using computer vision and predictive maintenance of the machines based on vibrational analysis is the name of the game here.

Connected engineering: Once your product is out there you want to learn how people are using it. Best example: nobody uses a fridge for its nominal design points. They are hardly ever optimised for usage. You want to learn what people do and upgrade your product next year.