Stem, Inc., is a leading provider of energy optimization services. The company’s PowerStore product predicts energy usage and deploys stored energy at precise times to reduce peak loads, enabling commercial and industrial customers to cut their energy costs.
Using Model-Based Design with MATLAB® and Simulink®, Stem completed the design, implementation, testing, and certification of the PowerStore energy storage system in just 13 months. David Erhart, vice president of engineering at Stem, explains how.
What led you to look for a new way of working?
As VP of engineering, my goal is to mitigate risk and control costs. Most traditional workflows are slow, expensive, and painful; they often require several costly board spins and lengthy hardware debugging sessions. Our company was in its early stages, so saving weeks of effort and tens of thousands of dollars in board-spin costs would be a big benefit.
Why did you select Model-Based Design?
When our chief power electronics engineer joined the company, we were using scopes to debug our inverter designs in the lab. Instead of tinkering with the hardware, he just took a few measurements and returned to his computer. After a couple of days, he showed us a plant model he had built that exactly mimicked the hardware behavior. In simulations, he identified the design problem and showed us how we could change the circuit and set inductor values to resolve it in the hardware. After the changes were made, the design worked immediately, and we decided to use modeling and simulation on a larger scale. With our PowerStore product, we used Model-Based Design right from the start.
What results have you seen so far?
We thought we were taking a risk by switching to Model-Based Design, but in hindsight the switch significantly mitigated our risk. We found we could speed up development by moving our controller design and hardware design forward in parallel. With an accurate plant model we could design our controller before the hardware was available—in fact, our power electronics engineer developed the controller himself in Simulink. We used simulations to find the optimal inductor and capacitor values. And we debugged our design via simulation instead of in the lab.
Once the hardware was ready, we generated C code from the model instead of writing code; it was like “printing” our design to the microcontroller. In the end, we needed just one additional board spin before moving to production.
We also achieved certification about 25% faster than usual because of Model-Based Design. PowerStore interacts with the grid, so it must be certified to IEEE 1547 for interconnecting distributed resources with electric power systems, among other standards. We simulated certification tests that our design initially failed, made changes to our controller model, re-generated code, and passed the tests the next day.