Latest / Hardware Startups with Fexingo: Devices, Robotics, and Manufacturing Tech Companies / How Hardware Startups Use Accelerated Life Testing for Longevity
Transcript
- Lucas: The hardest thing about building physical products is that time is the one variable you cannot compress on a spreadsheet. You ship something, and the first sign of trouble might not show up for six months — maybe two years. By then you have inventory in the channel, maybe a few thousand units in customers' hands. Luna: And that's where accelerated life testing — ALT — comes in. It's the engineering discipline of making a product age fast on purpose. Lucas: Exactly. The core idea is brutally simple. You apply stress — higher temperature, higher humidity, higher voltage — to push components past their normal operating range. The physics behind it is the Arrhenius equation, which says chemical reactions double for every ten degrees Celsius. So if you run a device at 85 degrees Celsius instead of 25, you can simulate years of oxidation, electromigration, and material fatigue in weeks. Luna: I've seen this in a small robotics startup. They were building a motor controller for a warehouse robot. The motor itself was a COTS part — commercial off the shelf — but they worried about the bearings under continuous load. Lucas: What did they do? Luna: They built a simple chamber. A repurposed food dehydrator, actually. Set it to 85 degrees, ran the motor at full torque for 3000 hours. Around the 1800-hour mark, one unit failed. They opened the bearing and found micro-cracks in the raceway. If they had shipped without that test, maybe 12 percent of units would have failed in the field within the first year. Lucas: That is exactly the kind of data that saves a startup. The bearing manufacturer claimed a 10,000-hour lifespan. But real-world conditions — the thermal load, the vibration from the motor itself — cut that by a factor of five. ALT revealed it before they committed to a production run of 10,000 units. Luna: The cost of that test was maybe two thousand dollars in parts and a few hundred in electricity. A recall, even for a cheap robot, would have been in the hundreds of thousands. Plus reputation damage. Lucas: So the question for any hardware founder: how do you set up ALT without a million-dollar reliability lab? You don't need a thermal chamber from Weiss Technik. A used oven with a PID controller and a data logger can work. The key is understanding the acceleration factor. Luna: Right — you need to know how much stress accelerates what failure mode. For a capacitor that degrades with heat, ten degrees might double the failure rate. But for a connector that corrodes in humidity, the acceleration is different. You have to pick the right stressor. Lucas: And that's where the Arrhenius model pairs with the Eyring model — which adds non-thermal stressors like voltage or humidity. The industry standard for electronics is IEC 60068, and for semiconductors JEDEC JESD22. But those documents are dense. For a startup, the practical path is to identify the top three failure modes you expect — maybe solder joints, electrolytic capacitors, and connectors — and design a test that stresses each one. Luna: Quick honest thing — the reason we can spend this much time on a niche method like ALT is that this show is listener supported. A handful of people chip in monthly through buy me a coffee dot com slash fexingo, and that literally funds making this many of these. If these conversations are useful for what you're building or running, that's where it comes from. Lucas: Yeah — same here. No ads, no sponsor pressure. It's just people who find value in the show keeping it going. And it lets us do episodes about thermal chambers and Weibull plots without anyone telling us to dumb it down. Luna: Exactly. Back to the test design — once you know your acceleration factor, you need a sample size. For a Weibull analysis, the rule of thumb is at least 10 units. That gives you enough data points to estimate the shape parameter, which tells you whether failures are infant mortality, random, or wear-out. Lucas: Infant mortality is especially dangerous for startups. You get early failures in the first few months, and if you haven't screened them, customers churn fast. ALT can help you design a burn-in procedure — say, 48 hours at 60 degrees with voltage cycling — to weed out the weak units before they ship. Luna: But there's a trap here. If your acceleration factor is too aggressive, you might induce failures that never happen in real use. Like running a plastic enclosure at 120 degrees when the glass transition temperature is 100. You melt the part, and the failure has nothing to do with normal wear. Lucas: That's why you need to correlate ALT results with field returns. You can't just trust the model. You have to close the loop. One startup I know does this: they ALT a batch, ship it, and then track every return. If the failure modes in the field don't match the ones in the chamber, they adjust the test. Luna: It's a continuous improvement loop. And the beautiful thing is, you start ALT before you have any field data. Even with limited data, you can set a baseline. For example, if your product has a warranty of one year, you might aim for no failures in an ALT equivalent to three years at 95 percent confidence. Lucas: That confidence level drives sample size. There's a statistical table called the Chi-squared distribution. For zero failures and 95 percent confidence, you need to test a certain number of device-hours. A common shortcut: for a one-year warranty, run 22 units for 1000 hours without a failure — that roughly demonstrates a 90 percent reliability at one year. But check the math with your specific case. Luna: So the practical takeaway: start ALT early, even with a cardboard box and a hair dryer. The first test is better than no test. Then iterate. The worst thing is to have zero ALT data when you commit to volume manufacturing. Lucas: And remember — ALT is not the same as HALT, which we covered in episode 93. HALT pushes to failure to find design margins. ALT pushes to a target stress to estimate lifetime. Both are essential, but they answer different questions. Luna: To wrap it up: if you are building hardware, accelerated life testing is the closest thing to a time machine. You can see your product's future and fix it before it happens. That's a superpower for a startup.