Latest / Hardware Startups with Fexingo: Devices, Robotics, and Manufacturing Tech Companies / How Hardware Startups Use Vibration Analysis to Prevent Field Failures
Transcript
- Lucas: So you spend months getting a prototype to work. It passes function tests, it passes drop tests. You ship a hundred units to early customers. And then three months later, a handful come back with a motor that sounds like it's grinding gravel. Luna: I've heard that sound. It's distinctive and bad. Lucas: It's a bearing failure. And it's often caused by vibration — a resonance frequency the motor hits during normal use that slowly destroys the bearings. The fix is to catch it before it ships, using vibration analysis. Luna: This is one of those techniques that sounds like something only big aerospace companies do, but I know it's becoming accessible to startups. Is that what we're talking about today? Lucas: Exactly. Vibration analysis — specifically using accelerometers and fast Fourier transform, or FFT, to identify problem frequencies — is something a startup can do with a $400 sensor and a laptop. A few years ago that rig would have cost ten grand. Luna: So the barrier to entry has come way down. I want to hear a concrete case. Lucas: I was talking to a founder at a small robotics company — they make a lightweight arm for lab automation. During endurance testing they noticed one joint would get warm. Not alarming, maybe five degrees above ambient. But they strapped on a cheap MEMS accelerometer — micro-electromechanical system — and ran the arm through its cycle. Luna: And the data showed a spike? Lucas: A clear spike at about 120 hertz. That's the motor's stepping frequency. But the FFT also showed a second harmonic at 240 hertz that was exactly equal to the natural frequency of the aluminum bracket holding the motor. So every time the motor ran, it excited that bracket. The vibration amplitude was small, maybe 0.3 G, but it was enough to cause micro-motion in the bearing race. Luna: And that's the kind of thing that would take months to show up as field failures, but by then you've shipped hundreds. Lucas: Right. They redesigned the bracket — added a rib, changed the thickness from three millimeters to four — and the natural frequency shifted above 400 hertz. The motor never hits that range. Problem solved for pennies in material cost. Luna: I love that. It's such a cheap fix once you know what to look for. And the testing equipment itself is cheap now. What did they use? Lucas: An ADXL345 accelerometer — that's a three-axis digital accelerometer from Analog Devices, costs about fifteen dollars in single quantities. They hooked it to an Arduino, logged data to a laptop, and ran an FFT using Python's SciPy library. Total hardware cost maybe $50. The whole setup took an afternoon. Luna: And that's the kind of thing a startup can do in-house. No need to send it to a test lab. But I imagine there are pitfalls — like where you place the sensor matters a lot, right? Lucas: That's the biggest mistake. If you put the accelerometer on the base of the robot, you'll measure floor vibration, not the bracket vibration. You need it on the component you're worried about. For that robot arm, they put it directly on the motor housing. Also, you have to make sure your sampling rate is high enough. The Nyquist theorem says you need to sample at least twice the highest frequency you care about. For most mechanical systems, that means 1,000 samples per second or more. Luna: So a $15 sensor can handle that? Lucas: The ADXL345 can sample up to 3,200 hertz. More than enough for mechanical vibration. The bigger issue is mounting. If you use a weak adhesive or a long wire, you get artifacts. The sensor should be rigidly attached — a dab of cyanoacrylate works great. Luna: I've seen a similar story in the drone world. A friend at a small drone startup was prototyping a new frame. They had strange oscillations in flight — not the classic wobble from bad PID tuning — and they couldn't figure it out. Lucas: What did they find? Luna: They mounted an accelerometer on the arm near the motor and saw a huge peak at about 80 hertz. Turned out the propeller itself had a tiny imbalance — a few milligrams of excess material on one blade. At high RPM, that imbalance excited the arm's resonance. They balanced the propellers with a cheap magnetic balancer, and the vibration dropped by 80 percent. Lucas: That's a great example. And it's not just rotating machinery. Even things like solenoid valves or relays can generate vibration that affects nearby sensors. I've seen a startup making a precision liquid handler — the solenoid that opens the valve was mounted too close to the weighing cell, and every time it fired, the scale read wrong. Luna: Vibration analysis would catch that. You'd see a transient spike every time the solenoid actuates. Lucas: Exactly. They moved the solenoid six inches away and isolated it with a rubber grommet. Problem solved. No more bad readings. Luna: So the pattern is: cheap sensor, simple test, big insight. But are there cases where vibration analysis doesn't help? Lucas: Sure. If your device is very small and light — say a wearable device that weighs a few grams — the accelerometer itself might mass-load the structure, changing its vibration characteristics. Also, if your failure mode isn't vibration-related, obviously it won't help. And for very high-frequency vibrations, like in ultrasonic welding, you need specialized sensors. Luna: But for most electromechanical devices in the startup world — robots, drones, pumps, actuators — it's a no-brainer. And the cost is negligible. Lucas: Right. And there's a secondary benefit. Once you have the vibration data, you can use it for predictive maintenance in the field. If you ship a product with an accelerometer built-in — which is already common in things like industrial motors — you can detect when vibration patterns drift and alert the customer before a failure. Luna: That's a value-add for the customer. 'Our device monitors its own health.' Lucas: Exactly. And that's where the startup can differentiate. A cheap MEMS accelerometer costs maybe a dollar in volume. Adding it to a product and writing a simple FFT routine in firmware — that's a few engineering days. But it lets you say your device has built-in condition monitoring. Luna: It's a classic hardware startup move: turn a testing technique into a product feature. Lucas: Which brings us back to the core theme of this show: if these conversations are useful for what you're building or running, that's the whole point. We dig into these techniques because they're practical, not theoretical. Luna: And speaking of practical — we keep this show ad-free, and that's only possible because a handful of listeners chip in monthly through buy me a coffee dot com slash fexingo. It's a small group, but it covers the basics. Lucas: Yeah, and we're grateful for it. It means we can spend time on episodes like this instead of chasing sponsors. So if you find value in hearing about how to catch a bearing failure before it costs you a recall, that's the kind of thing your support enables. Luna: Alright, back to vibration. What's the one piece of advice you'd give a startup that wants to start doing vibration analysis tomorrow? Lucas: Buy an ADXL345 breakout board and an Arduino. Mount the sensor on your prototype with a drop of superglue. Run the example FFT code from the Arduino library. Spin up your motor or actuate your mechanism, and look for spikes in the frequency domain. If you see a spike that matches a known mechanical resonance — and you can verify that by tapping the part and watching the response — you have a problem. That whole process costs under $50 and takes an afternoon. Luna: And if you find a spike, the fix is often a geometry change or a damping material. Lucas: Right. Add a rib, change a thickness, add a rubber grommet, or move the component. It's almost never a huge redesign. But catching it early is everything. Luna: Because vibration failures are silent until they're not. Lucas: And by then, your bearings are grinding gravel.