Move Fast and Make Things

Rightly or wrongly, it’s commonly accepted that software or firmware is easier to change than hardware. “We’ll fix it in software,” is one of the most terrifying phrases a software engineer will ever hear. Physics can’t be swept under a rug made of binary. If I were to pick the most important contributor to our success at Woosh, it is our ability to experiment and prototype quickly in the physical world. Our culture promotes failing fast and rapidly iterating.

However, we’re a startup and don’t possess the resources of established tech companies. We’re resourceful out of necessity. For speed, we prefer to purchase components and equipment off the shelf, as long as the cost is reasonable (for us). When that’s not feasible (cost/time) or possible (doesn’t exist), we innovate and invent.

From left to right: our early pizza box and foldable filter prototypes. Final version on far right.

In 2021 we began exploring the concept of a reusable, foldable frame with replaceable filter media. Naturally, this required a lot of experimentation. Creating injection mold tooling is enormously costly and time consuming. We turned to 3D printing, as many others have done. However, that approach literally came with one big problem: an air filter frame is much too big to print on most 3D printers. We didn’t have tens of thousands of dollars to spend on a commercial large scale 3D printer. Instead, we modified cost effective open source Creality CR10 S5 printers.

We produced and shipped early samples using 3D printers until we moved to injection molding at scale.

By early 2022 we zeroed in on the frame design, begun prototyping the electronics, and started collecting data. Early analysis appeared to validate our core thesis: as the filter loaded with dust, differential air pressure between the front and back of the filter increased and could be measured electronically. Paradoxically, after a few months of gradual increase, the differential pressure would start to decrease, as if dust particles were leaving the filter. We soon discovered that our selected pressure sensor was clogging with dust as air passed through its tiny sampling ports, ruining its ability to measure pressure.

Obviously, without a solution to this problem, we wouldn't have our product. Furthermore, we couldn’t wait three months to validate each new approach or design change. It needed to be done in days, or we were going to run out of funding. To facilitate fast testing, we built a contraption affectionately named Big Blue.

Staying true to startup culture, Big Blue lived in my garage for about a year.

Big Blue simulates the conditions of an HVAC system. We built it using plywood, an off-the-shelf coaxial blower, Dwyer panel mounted pressure sensors, silicone caulk, weather stripping, tubing, wiring, and a Raspberry Pi Zero.

Fully automated remote control and data collection.

Over the following weeks we ran hundreds of experiments and collected tens of thousands of data points. From the resulting dataset we designed a novel and patent pending sampling method. This permits accurate differential pressure measurement while protecting against dust ingression. Additionally, we selected a new sensor that doesn’t require air (or dust) to flow through it. Most importantly, we simulated years of wear and dust loading with a nasty substance called ASHRAE Test Dust #1.

This should have be an episode of Dirty Jobs. Carbon black gets in/on everything and never fully washes off.

Big Blue worked so well that we built “mini blues” to simulate air pressure changes and run end-to-end testing of our digital stack. We re-purposed blowers originally designed for cooling rack-mounted servers to generate air pressure. We ran stress and longevity tests to verify battery life. These also let us simulate many different types of HVAC systems to verify our filter life algorithms. 

Also 3D printed with a custom Raspberry Pi HAT (more below).

By summer 2023 we were at last ready to launch! We found a contract manufacturer who was excited to help build our product. Great! Except, they had no idea how to run quality assurance on the finished assemblies, and we only had three weeks to provide them with a solution. However, by this point we knew a ton about 3D printing, PCB design, air pressure, Python scripting and all the things a Raspberry Pi can do. 

The Raspberry Pi is one of the most underrated pieces of technology to come out of the 2010s.

Our hardware team can design and fabricate a PCB in days. Leveraging this strength was paramount in making the production deadline. We built a custom Raspberry Pi HAT to program and send commands to the device-under-test, generate air pressure, control and measure electrical power, and even take pictures of the LEDs. We built and shipped three sets of manufacturing test jigs to the factory by the deadline.

Three Gateways being being programmed and tested.

Even though we’re well past our initial launch, we haven’t stopped innovating. We’re still 3D printing, cutting, gluing and drilling prototypes for new products we plan on launching later in 2025. Keep an eye on us!