For EV Safety Tests, AI Is Here to Help

Supercharging the development of safe, high-performing EV batteries means nurturing trust in AI to drive competitive innovation.

April 1, 2025

Automotive engineering consultancy HORIBA Mira is integrating Monolith’s tools into its technical offer for enhanced battery and powertrain development. (Image: Monolith)

For the automotive sector, one of the most significant barriers to the development of safe, high-performing EV batteries is the complexity of developing and validating them. Testing campaigns, in particular, present unique challenges. Cycling protocols often require 5-7 days per test point at a minimum, and aging studies can extend to weeks or months. With the increasing pressures to get products to market faster, this is an area in which artificial intelligence (AI) can provide a lot of value.

Empowering engineers to use AI to tackle their most intractable physics challenges has been Monolith’s mission since day one. In pursuing this, our goal has been to supercharge time to market, slash testing costs, and facilitate higher-performing, safer products.

The intricate physics involved throughout the development process — including thermal, chemical, and electrical interactions — raises an essential question about our testing approach. While grid-based sampling has long served the industry in exploring multiple parameters such as temperature profiles, C-rates, and depth of discharge combinations, should we challenge our traditional testing strategies?

Rather than exhaustively testing every possible combination, is there an opportunity to develop more focused, efficient approaches that could transform how we validate and optimize these complex systems?

Design of Experiments

The complexity of battery models and their sensitivity to various factors means machine learning can be a useful tool for optimizing battery management systems. (Image: Monolith)

AI is transforming engineering testing, and one of its highest-impact applications is optimizing the Design of Experiments (DoE).

Many engineers are already familiar with optimal design principles — using statistical methods to determine the most efficient combination of tests that thoroughly cover the design space. However, there’s a practical challenge. While these methods can significantly reduce testing time and costs, planning optimal test sequences requires considerable time and statistical expertise. Test engineers and lab technicians, already pressed to complete existing test campaigns, rarely have the bandwidth to optimize their testing strategies. It’s a paradox. They need more efficient testing to save time, but don’t have the time to figure out how to test more efficiently.

That’s why we looked to develop a Next Test Recommender for this process — using machine learning to identify which tests will provide the most valuable insights for understanding the design space. However, we quickly learned that purely statistical approaches often clash with real-world laboratory constraints.

Pre-trained models can enable more accurate, valuable predictions for battery degradation and thermal propagation. (Image: Monolith)

This insight led us to develop an active learning approach that integrates practical testing limitations directly into our recommendation engine. For instance, when analyzing thermal degradation patterns, traditional statistical algorithms might suggest precise but impractical setpoints such as 23.7 °C (74.66 °F). With AI-driven solutions, we can instead map recommendations to actual equipment capabilities, such as using integer temperature steps, while maintaining statistical rigor.

Enhancing Testing Efficiency

This pragmatic approach can also be extended to voltage limitations, current constraints, and other real-world testing parameters to enhance testing efficiency. We have therefore extended this methodology to cathode composition optimization, where the parameter space becomes exponentially more complex. For example, a typical NMC cathode development program might explore variations in nickel:manganese:cobalt ratios, dopant concentrations, and coating compositions, creating thousands of potential combinations. Meanwhile, our active learning approach dynamically adapts its sampling strategy based on predicted performance gains and practical manufacturability constraints.

Optimizing your testing plan with machine learning can reduce the number of battery tests by up to 70 percent. (Image: Monolith)

When we work with battery manufacturers and reveal the benefits these technologies can bring, we often ask them, ‘If you could complete your validation testing 3-4 weeks faster, what would you do with that time? Would you accelerate your product’s journey to market or reinvest those weeks into additional safety testing and performance validation?’ This choice perfectly illustrates how optimized testing can drive competitive advantage or enhance product quality. It’s genuinely fascinating to see how different manufacturers approach this strategic decision.

Naturally, these changes come with challenges. The shift to AI-guided testing represents a significant change for battery engineers who have relied on decades of experience and well-established testing processes. We understand the hesitation; asking an engineer to move away from comprehensive testing approaches they’ve trusted for decades, to instead follow recommendations from a statistical model that suggests testing less, is a big leap.

But this transition doesn’t have to be abrupt. Starting by applying active learning to historical test campaigns clearly demonstrates how these new methods could have achieved the same insights with fewer tests. This retrospective validation has proven particularly effective in building confidence among skeptical teams. We’re now seeing these types of approaches gain significant traction; major cell manufacturers such as CATL fully integrate these machine learning methods into their R&D processes, achieving substantial efficiency gains in battery development.

Embracing AI’s Potential

It’s important to note that skepticism toward AI is reminiscent of past attitudes toward now-standard technologies. Thirty years ago, tools such as CFD were met with trepidation — an attitude defined by limited understanding. Today, engineers see them as indispensable. Similarly, adopting AI requires a shift in perspective and embracing its potential.

Self-learning algorithms have the power to completely change the way we develop EV batteries. They can optimize battery designs, identify the best materials, deliver advanced analytics, and simulate complex systems with unprecedented accuracy. These capabilities are vital for developing safe, high-performing, dependable products and solving intractable physics problems.

Our engineering partners have already begun to see the benefits of integrating AI and machine learning into their workflows. Whether saving time on product validation or uncovering deeper design insights, the results show much promise for where engineering testing is headed.

As trust builds and adoption grows, AI and machine learning will become essential tools in the engineer’s toolkit, driving innovation and efficiency in EV battery development to new heights.

This article was written by Richard Ahlfeld, Founder and CEO of Monolith (London, United Kingdom). For more information, visit here .