Dr. Kai-Philipp Kairies working in a meetingDr. Kai-Philipp Kairies working in a meeting

This paper will explain:

On occasion, battery owners and operators will turn to lab testing to predict the operational behavior of their assets. This is where the problem arises. Relying on lab tests creates a false sense of security and often puts decision-makers on the wrong track. In some cases, building operational models based on lab tests even makes things worse.

Yes, you read that right. More information can reduce the quality of your forecast. This sounds counterintuitive, but bear with me.

Variability in Battery Production

Battery manufacturing involves intricate processes that result in substantial variations in cell characteristics, even within the same batch. Factors such as material inconsistencies, manufacturing tolerances, and environmental conditions during production contribute to this inherent variability.

As a result, no two batteries – even when produced on the same machine on the same day – are exactly alike. This leads to several engineering challenges in battery module and pack design.

How End-of-Line Battery Testing Helps

To form electrical components of practical voltage and capacity, individual battery cells are combined into modules and packs. Because the weakest cell defines the performance of such a serial connection, it is imperative only to combine similar battery cells. To ensure this, battery manufacturers conduct comprehensive end-of-line testing and sort their batteries according to the results.

Typically, batteries that show similar AC1000 results (i.e. have a comparable internal resistance when tested with a 1000Hz alternating current) are grouped together. The spread can be limited to acceptable ranges within each group by sorting and grading the cells. The groups containing the best-performing cells are then sold to the most important customers, the lowest-performing (but still OK’ish) cells go to less-important customers, and the rest goes to Alibaba.

The Batteries You Get for Testing

Now imagine you’re running a battery manufacturing site. Your production spread is typical (i.e. pretty large), but by sorting and grading, you ensure your customers get a product that works “good enough.” Now, imagine that one of your customers requires 50 cells to perform their own lab tests on. Which cells are you going to send them?

If you’re anything like the battery manufacturers I’ve spoken to over the last decade, you will take 100 cells from the top group and run a few additional tests on them, just to be sure. The top 50 cells of that exercise are then sent to the customer. After all, you don’t want them to find minor problems and start a discussion about your general production quality, right?

The Harsh Reality

Now what happens when you take those pre-tested cells and start building models on them? You’ll get a perfect model of battery cells that have nothing to do with the ones that might get you in the news one day. Because accelerated aging, performance loss, and safety issues don’t happen to the best cells. They happen on the outside of the normal distribution. In other words, the cells that will inevitably end up in your operational asset.

One particular example of this can be found in a recent analysis we did for an energy company during the commissioning of a BESS site in North America. Cloud diagnostics uncovered that the site consisted of batteries from multiple production batches with shocking deviations in their properties.

Batteries in Tests vs. Reality

The end of battery testing?

This article is not meant to challenge battery testing itself. Lab testing is absolutely needed to develop new batteries, thermal management systems or to create operation strategies. Lab testing just doesn’t add a lot of value when it comes to predicting the real-life behavior of a naturally diverse fleet of operational assets. If done wrong, it can even create blind spots that put investments and people at risk.

Instead of (mostly) relying on lab tests, the scalable and reliable approach to predicting individual battery behavior combines field data analysis, physics-based models, and artificial intelligence. The combination of these three capabilities can identify patterns, correlations, and anomalies that traditional laboratory tests systematically overlook – simply because the relevant anomalies were not present in the tested subsample.

In summary, differences between lab-tested batteries and real-world performance are substantial due to production variability and diverse usage conditions. Relying only on lab tests can give a false sense of certainty about battery performance in real settings. To tackle this, it's crucial to combine field data and advanced analytics to unlock the full potential of your batteries.

Dr. Kai-Philipp
About the author

Dr. Kai-Philipp


Kai-Philipp is the CEO and co-founder of ACCURE Battery Intelligence. He started his journey into the battery world as a test engineer for automotive Lithium-ion batteries in 2008. He has led R&D and consulting projects for mobility and energy companies around the world – from aging simulation to business model development. To recharge, he plays piano or goes to a boxing class.

Connect with me

About ACCURE Battery Intelligence

ACCURE helps companies reduce risk, improve performance, and maximize the business value of battery energy storage. Our predictive analytics solution simplifies the complexity of battery data to make batteries safer, more reliable, and more sustainable. By combining cutting-edge artificial intelligence with deep expert knowledge of batteries, we bring a new level of clarity to energy storage.  Today, we support customers worldwide, helping optimize the performance and safety of their battery systems. Visit us at accure.net.