What is accelerated life testing?
Many products are designed to last years, but nobody can wait years for test results. Accelerated life testing raises the stress that drives the dominant failure mechanism, measures the life at several stress levels and extrapolates to normal use with a physical model. Done well, it gives a life prediction in weeks and shows how much margin a design has.
Life–stress models
| Model | Stress | Life relationship | Parameter |
|---|---|---|---|
| Arrhenius | Temperature | L ∝ exp(Ea / kT) | Activation energy Ea (eV) |
| Inverse power law | Voltage, load, pressure, vibration | L ∝ 1 / S^n | Exponent n |
| Peck, Eyring | Temperature and humidity | Arrhenius × humidity term | Ea and humidity exponent |
| Coffin–Manson | Thermal cycling | N ∝ 1 / ΔT^m | Exponent m |
Accelerated life testing example
Power supply modules are tested at 85, 105 and 125 °C, 12 units at each temperature. Weibull analysis of each group gives characteristic lives of 6,200, 2,350 and 980 hours. The Arrhenius fit is excellent and gives an activation energy of 0.57 eV, typical of electronic failure mechanisms. At the use temperature of 40 °C the predicted characteristic life is 86,800 h (9.9 years), with a 95% confidence interval of 84,200 to 89,400 h. One hour at 125 °C equals about 89 hours of use. To demonstrate 90% reliability over 5 years (43,800 h) with 90% confidence, assuming β = 2, the team needs to run 20 units for 517 hours each at 125 °C without a failure.
Tips
- Use at least three stress levels, spread as widely as the design allows, so the model can be checked.
- Confirm the failure mode at each level by failure analysis; a new mode at the highest stress invalidates the fit.
- Get the life at each level from a Weibull analysis, and use the same life measure (η or B10) at every level.
- Be careful with long extrapolations: the further the use condition is from the tests, the more the prediction depends on the model.
How to use this Accelerated Life Testing tool
- Choose the model. Arrhenius for temperature, inverse power law for voltage, load, pressure or vibration.
- Enter the use condition. Enter the stress the product sees in normal use, for example 40 °C.
- Enter the test results. For each test stress, enter the life measured there, such as the Weibull η or the mean life.
- Read the prediction. The tool fits the model, gives Ea or n, the life at use with a confidence interval and the acceleration factor of each level.
- Plan a test. Enter the life and reliability to demonstrate, confidence, β and units to get the zero-failure test time at the highest stress.
- Export. Export the life–stress plot and results to PDF or Excel.
Frequently asked questions
What is accelerated life testing?
Accelerated life testing (ALT) runs products at higher stress than in normal use, such as higher temperature, voltage or load, so that they fail in weeks instead of years. A life–stress model fitted to the results at several stress levels then predicts the life at normal use conditions.
What is the Arrhenius model?
The Arrhenius model describes how temperature speeds up chemical and physical degradation: life is proportional to exp(Ea / kT), where T is the absolute temperature, k Boltzmann's constant and Ea the activation energy of the failure mechanism. It is the standard model for temperature-accelerated tests of electronics, adhesives, batteries and many materials.
What is the acceleration factor?
The ratio of life at use to life at test stress. For Arrhenius, AF = exp[(Ea / k)(1/T_use − 1/T_test)] with temperatures in kelvin. An acceleration factor of 89 means one hour at the test condition equals about 89 hours of use.
What is the inverse power law?
A model for non-thermal stresses such as voltage, mechanical load, pressure or vibration: life is proportional to 1/S^n. The exponent n is found from the slope on a log–log plot.
How long should a zero-failure test run?
With a Weibull shape β, testing n units for time t each without failure demonstrates reliability R at a life L with confidence C when t = (L / AF) × [ln(1 − C) / (n ln R)]^(1/β) (Lipson equality). More units or a higher acceleration factor shorten the test.
What are the risks of accelerated testing?
Too much stress can create failure mechanisms that never occur in use, which makes the prediction meaningless. Keep the stresses within the product's design limits, check that the failure modes are the same at every level, and use at least three stress levels.