By Andy Tomaswick
March 29, 2025
Deciding how to power a CubeSat is one of the greatest challenges when designing a modular spacecraft. Tradeoffs in solar panel size, battery size, and power consumption levels are all key considerations when selecting parts and mission architecture. To help with those design choices, a paper from researchers in Ethiopia and Korea describes a new machine-learning algorithm that helps CubeSat designers optimize their power consumption, ensuring these little satellites have a better chance of fulfilling their purpose.
The power situation for CubeSats is complicated, to say the least. They are typically powered via solar panels, which have to deploy from the “U” structure that all CubeSats are designed around. However, even if they are deployed successfully, they are subjected to wide ranges in solar radiation and temperature, causing dramatic swings in their overall power output.
According to the authors, power system faults cause around a quarter of all CubeSat mission failures. Several design choices, such as multi-input multi-output (MIMO) converters, can already help alleviate that. However, managing this type of power distribution system also comes at a cost, as it is designed to perform a function called Maximum Power Point Tracking (MPPT).
Fraser discusses space power with Dr. Stephen Sweeney
Simply put, MPPT is a control algorithm that attempts to get the highest possible power out of the system given any environment it finds itself in. For example, suppose the incidence of solar radiation dips slightly because the CubeSat is at a non-optimal orientation to the Sun. In that case, the MPPT algorithm tells it to reorient itself so that the highest solar irradiance level falls on its solar panels.
Several control algorithms are designed to optimize the MPPT of a CubeSat. These include creative algorithms named Perturb and Observe (P&O), Incremental Conductance (InC), and Particle Swarm Optimization (PSO). While all of these are relatively effective at optimizing MPPT, ranging from 88-94% efficiency, they all share the same weakness—they’re not adaptive, and their parameters must be set before the CubeSat launches.
Enter one of the most popular artificial intelligence algorithms – deep learning. The authors describe the development of a Deep Feedforward Neural Network (DFFNN) linked with a standard proportional-integral controller that outperforms all other MPPT algorithms. Its efficiency, which they calculated at 97% based on trials with simulated data of a year-long mission, also increases the overall power efficiency of the system as a whole, as well as lowers the “power ripple” – changes in the power output supplied by the system that can introduce “transients,” or temporary changes in voltage and current that can potentially damage components.
Powering a CubeSat is a key design consideration, as discussed in this video series on building one.
Credit – Building A CubeSat YouTube Channel
The algorithm has some downsides, as it is very computationally intensive, like all machine learning. To deal with that problem, the novel algorithm uses a technique called linear tangents and Neville Interpretation. This mathematical technique breaks down polynomial problems into much easier-to-solve equations, simplifying the calculation of the CubeSat’s desired trajectory.
Every little bit counts when it comes to improving CubeSat performance, and this paper contributes to that effort. A 3% improvement may not seem like much, but when thousands of hours of engineering and testing are at stake, even small improvements can be life-changing.
Learn More:
Abagero et al. – Deep Learning-Based MPPT Approach to Enhance CubeSat Power Generation
UT – A 3U CubeSat Could Collect Data During an Asteroid Flyby
UT – A CubeSat Mission Will Detect X-rays from GRBs and Black-Hole Mergers
UT – The First Cubesat With a Hall-Effect Thruster has Gone to Space