(Sponsored Content)

By JOHN ERVIN, Chief Scientist, Peaxy, Inc.

By STEFAN GREENS, Director of Product Design, Peaxy, Inc.

Battery data visualizations can instantly bring to life the insights that data tables struggle to express, with uses that span across the entire lifecycle of the asset.

For R&D, product developers need to quickly see which battery designs are performant. Manufacturing engineers need to quickly see how yields could be improved.

If discrepancies occur when monitoring a grid-scale battery installation, you’ll want to investigate immediately. Any of these scenarios are only possible by viewing charts that provide live visualizations of your battery data at every level of the hierarchy.

While data visualization is what gets most people excited about working with their data, there is some preparation work required to get your data into the proper shape. In reality, within a single company or site installation, data can come from multiple sources, often from systems that don’t talk to each other, and can even be collected in spreadsheets on someone’s computer or in a handwritten document. Trying to visualize data from such a compromised data landscape will either result in an extremely limited view or prove practically impossible. This is where data parameterization comes into play.

Data parameterization means taking a large amount of data and generating a much smaller data set, thereby making it more easily digestible. A data set might be, for example, last year’s worth of OPC tags at 1 Hz from a power conversion system that is 200 tags wide. We often call these derived parameters “composite scores,” and they usually have values between 0-100. Their purpose is to characterize attributes such as overall performance, health, economic output vs. expected, or remaining useful life to name a few examples.

Several of these composite scores are typically generated from any single data parameterization exercise. In the PCS case, these could be scores which characterize the health of the inverter, the total lost in the step up transformer and so on.

An example of data parameterization that provides composite scores that allow you to tell at a glance whether battery designs are performant, or how yields can be improved (Source: Peaxy).

Data scientists might call this feature extraction, which is close to the concept, but not exactly what we mean here because it’s really at a lower level.

For instance, taking an SEM image of a Si/C colloid we are using to make a next gen lithium-ion anode, we might use “imageJ” to generate moments of the size distribution of that colloid. The average size, the variance of the size and so on would be the result of a feature extraction exercise. Of course we do that also, but when we speak of data parameterization, our approach is at a higher level than this.

You could argue that what we’ve described as data parameterization is just another way of saying KPIs. There is some truth to this, but the amount of data we use and the techniques we use to parameterize are generally a bit more sophisticated than your typical KPI. A KPI for a battery system might be total discharge energy over the battery’s life, or average round trip efficiency (RTE) for the last year. While this is also part of our data parameterization process, we go a step further and use these values, with other pieces of information, to generate composite scores.

This can’t be a one-off exercise. In order to drive data visualization, your data collection and parameterization has to be automated. What are the benefits? Of course this depends on the context. We have been fortunate enough to have worked with multiple organizations and their battery data challenges, and there are some common patterns.

In R&D, we’ve seen some tendencies to be initially resistant to too much automated data parameterization for fear of being prevented from doing all of the desired analyses. Just because you parameterize your data one way, however, doesn’t mean the data goes away and can’t be used for a parallel analysis. It’s a similar issue in manufacturing where there is often a question of whether data parameterization adds value above and beyond what the MES system provides.

In both of these scenarios, parameterization plays an important role in providing the foundation for visualized battery metrics with point and click drill-down that shows how the physical asset is performing in an easily digestible way. 

As mentioned in the beginning, battery data visualizations can instantly bring to life the insights that data tables struggle to express, but only if the charts fit the context. Before deploying a chart, it’s important to make sure it fits the needs of the use case, and that the needs in turn determine the charting solution you employ.

It’s often difficult to see potential red flags or anomalies when dealing with large datasets. There are, however, some strategies we can use to selectively drill down into a vast amount of data. 

The ability to drill down into the data is crucial, not only because trends may be difficult to identify and remedy at a macro level, but also due to the practical limitations of web browsers to present large amounts of data at once. It’s important to be able to drill down at a variety of levels, all of which map to real-life battery configurations.

Notice one block is discharging faster than others? You can easily zoom in on that block and see the relevant details instantly. Notice one of the block’s strings is under voltage? Zoom in on that string for a full view of what’s happening. At every level in the battery grid configuration, dynamic charts let you synchronize any number of attributes to the same live timeline, so that discrepancies between subcomponents or correlations between attributes literally jump out at you.

See emerging trends or anomalies in live data with dynamic charts (Source: Peaxy).

When analyzing historical data to satisfy reporting obligations or monitor warranty regimes, millions of granular data points per 24-hour cycle need to be conveyed accurately on a chart, and that’s a big challenge. Conventional decimation techniques can miss brief voltage spikes, so instead it’s recommended to periodically run data pipelines that incorporate every data point, no matter how anomalous or fleeting, into the plot. Since the primary goal is archival accuracy that can meet regulatory obligations, these static charts are pre-rendered daily, once the cycle is completed.

Create automated operational reports on full-screen resolution datasets with static plots (Source: Peaxy).

Finally, it’s important to allow the end user to define and design their own visualizations. This offers a number of advantages. For modern, cloud-based solutions, you must have the upfront configuration tools that allow you to perform deep analysis without needing to involve a programmer or write a single line of code.

For example, before you can train a regression analysis algorithm on historical data to generate performance predictions from live data, it’s important to get a feel for how the various features correlate, so that the dataset can be culled of any superfluous attributes. Some examples of point-and-click charting tools include:


John Ervin, Chief Scientist, Peaxy, Inc. – John has devoted his career to the commercialization of novel technologies and tools in the life science and data science sectors. His focus on bringing nascent technologies to market has made him a key part of growing customer bases from nothing to over 40 in multiple contexts. John has held management roles with Eksigent Technologies (acquired by AB/Sciex), PerSeptive Biosystems (acquired by AB), Silicon Kinetics where he is a founder and his current Chief Scientist role at Peaxy. He holds a BS from Cornell University, and a PhD in Physical Chemistry from the University of Illinois. View LinkedIn →

Stefan Greens, Director of Product Design, Peaxy, Inc. For over two decades, Stefan has focused on user-centric design and communication. He co-founded the Swedish design agency Äventyret, where he helped major institutions develop digital product strategies. Working for the Swedish foreign ministry, Stefan developed and implemented the national digital strategies for public diplomacy in China and the Middle East, and helped design Sweden’s current brand identity. Stefan’s user interface design experience covers data visualization projects, augmented reality and virtual reality. Stefan holds an MA in International Affairs and Economics from the Johns Hopkins School of Advanced International Studies. View LinkedIn ›

About Peaxy, Inc: Peaxy offers a battery intelligence platform that captures and analyzes customers’ data down to the serial number across the entire lifecycle, to help companies build better and run smarter. Deployments are typically done in 120 days with greater speed than an in-house development effort or a generalized analytics platform.

This post appeared first on Power Engineering.