AI Engineering Blog | Read the AI Articles | Monolith

How to Learn Faster and Find Better Designs at the Same Time?

Written by Dr Joël Henry | Jul 29, 2025 12:44:27 PM

AuthorDr Joël Henry, Lead Principal Engineer, Monolith

Read Time: 5 mins

In engineering, testing and design exploration can be slow, expensive, and often driven by intuition and experience. Often, test plans are fixed from the beginning, losing the opportunity to refine the plan as more knowledge is acquired.

At Monolith, we’ve already been helping engineers learn the whole picture faster with our Next Test Recommender (NTR). Now, we’re introducing a new feature that helps teams zoom in and find the best design or test sooner: NTR-Optimisation. This article explains how both approaches work, what makes them different from traditional optimisation, and when to use each.

 

How the Next Test Recommender (NTR) helps you learn faster

 

Imagine you’re handed a painting you’ve never seen before, and you can only look at a few small snapshots of it to figure out what’s on the canvas. If your goal is to describe the entire painting, you’d want those snapshots spread out: capturing different regions and features so you can piece together a complete picture.

This is what the Next Test Recommender (NTR) does. In the context of battery development, suppose you’re trying to understand the effect of different charging profiles on battery life and performance. Instead of running every possible test, NTR suggests the next most informative test, focusing on where your current knowledge is the weakest.

The aim isn’t to find the single best charging profile immediately. It’s to build a reliable understanding of battery behaviour across the whole operating space, so you can later make accurate predictions, create calibration tables, or validate performance over a wide range of scenarios.

 

NTR samples the design space as if it were discovering and understanding an entire painting: focusing on regions likely to provide more information (e.g. not sky).
 

How NTR-Opt helps you find your target faster

 

But what if you don’t need the full painting? What if your goal is only to find the flying bird in the picture - the best part, or in engineering terms, the optimal design or operating condition?

This is where NTR-Optimisation comes in. Using the same battery example, say you want to identify the fastest possible charging profile that still preserves battery health.

Rather than exploring the entire space, NTR-Optimisation focuses on finding the optimum quickly: suggesting each next test based on what you’ve already seen, zeroing in on promising regions without needing to cover every detail. Although you don’t explore everything, you learn enough about the less interesting regions to know they’re unlikely to contain your optimum, so you can confidently stop testing there.

It’s okay if you don’t map the whole landscape, as long as you have high confidence you’ve found the best answer.

 

NTR-Optimisation samples the design space as if it were trying to find a flying bird in the painting: focusing on regions of interest (e.g. sky).

 

How it’s different from traditional optimisation

 

Traditional optimisation typically relies on fast simulations or on models that have already been trained with a large amount of test data. This means you first need to run enough tests or build an accurate model before you can start optimising, which can be slow and costly when working with real physical tests. In contrast, NTR‑Optimisation is adaptive and data‑driven: each recommended test depends on what you’ve learned so far.

Returning to the painting analogy, a traditional optimisation method would first collect enough snapshots to build a fairly good (though still blurry) picture of the whole canvas. That rough view then guides you to where the bird is most likely to be, so you can focus your next tests there.

The trade‑off is that you still need enough initial points to sketch out that rough picture in the first place; without them, the optimiser wouldn’t know where to look - and could be detrimental, sending you towards unpromising or misleading areas instead.

 Traditional optimisation first gathers enough points to model the entire design space (see blurry background), then exploits that model to identify the optimal region.

 

When to use which

In the end, there isn’t a single “right” or “wrong” method; the key is to choose the one that best fits your goal:

  • NTR: when you want to learn quickly and build a predictive model for future use. Ideal for tasks like mapping battery performance, creating calibration tables, or covering a wide design space.
  • NTR‑Optimisation: when you want to find quickly, to identify the best design or test configuration without having to explore everything.
  • Traditional optimisation: when you already have detailed models or fast simulations and can afford a broader, more exhaustive search.

Together, these tools help engineers get more value from every test, whether the aim is to see the whole picture or to find the best answer fast. If you want to learn more about Monolith’s new feature NTR-Optimisation, watch the webinar here.

 

About the author

Dr Joël Henry is our Lead Principal Engineer and has worked with the majority of our clients across automotive and battery testing to create lasting value for them. He has won the Imperial College Award for the best PhD thesis and has previously worked on improving test and simulation methods in the aerospace industry.