Reliatools

Reliability Test Methods: HALT, ALT, Run-to-Failure, Degradation, RDT and Burn-In

Every reliability test answers one question. Using the wrong method answers a different question than the one the program actually asked. This guide covers what each method is for, what it proves and what it cannot prove, and where it fits in product development.

Reliability testing is the main way engineering teams find and confirm failure modes before customers do. It is also where a lot of validation budget is wasted. Typical waste includes a HALT result quoted as an MTBF, an accelerated test with an activation energy nobody verified, or a success-run test sized without a mission profile. The methods below complement each other. The skill is knowing which question each one answers.

MethodQuestion it answersOutputWhen
HALT / step stressWhere does the design break, and how much margin is there?Operating and destruct limits, weak pointsEarly design, prototypes
Accelerated life testing (ALT)How long will it last under use conditions?Life estimate through an acceleration factorDesign verification
Run-to-failure (RTF)What is the actual life distribution, and is the acceleration model right?Weibull parameters, stress-life curveDesign verification, model building
Degradation testingHow fast is performance drifting toward the limit?Degradation rate, time to thresholdWear-out mechanisms, maintenance intervals
Reliability demonstration (RDT)Does the product meet its target, at stated confidence?Pass/fail against R at CValidation, before release
Burn-inCan early-life defects be removed before shipment?Screened production unitsProduction

HALT: find the limits, not the life

Highly Accelerated Life Testing pushes prototypes well beyond specification, using cold and hot steps, rapid thermal transitions, random vibration, and then combined stresses, until something fails. Each failure is analysed, fixed if it is relevant, and the test continues. The result is a map of the product’s operating limits and destruct limits, and of the margin between those and the specification.

HALT is qualitative. It is a fast way to find weak points and compare design generations, but its results cannot be converted into an MTBF, a FIT rate or a field life. The full HALT guide covers classical versus rapid HALT and how to judge whether a HALT failure matters in the field.

Step stress

HALT is built on the step-stress method: stress is raised in fixed increments, with a dwell at each step, until failure. A typical HALT uses about 10 °C per thermal step, with a dwell long enough for the part to stabilise (which depends on its thermal mass), and 3–5 Grms per vibration step. Step stress finds the stress at which a part or assembly breaks down. It does this quickly and with very few units, which makes it a natural companion to the slower life-oriented methods below.

The step sizes, dwell times and starting points all have to be planned before the chamber is booked. The HALT/HASS Test Plan Builder lays out a step-stress profile from the product’s specification limits.

Accelerated life testing: compress time with a physics model

Accelerated life testing (ALT) speeds up aging by raising a stress that drives a known failure mechanism, such as temperature, humidity, temperature swing, load or cycle rate. Years of use are compressed into weeks. The conversion back to field life is the acceleration factor (AF), and the AF comes from a physics-of-failure model. For temperature-driven chemistry, that model is Arrhenius:

AF=exp⁡[Eak(1Tuse−1Ttest)]AF = \exp\left[\frac{E_a}{k}\left(\frac{1}{T_{use}} - \frac{1}{T_{test}}\right)\right]

EaE_a is the activation energy (eV), kk is Boltzmann’s constant (8.617×10−5 eV/K), and temperatures are in kelvin.

The hardest part of ALT is the model coefficient, here the activation energy, not the chamber. Literature values are a starting point, but they belong to a specific failure mechanism. A real component or assembly usually has several competing mechanisms, each with its own activation energy, and a literature value borrowed for the wrong mechanism can be off by an order of magnitude or more. Identify the physics of failure first, then choose the model. Cyclic thermal fatigue uses Coffin-Manson, not Arrhenius.

The Arrhenius calculator computes the AF and the equivalent test time. Whether that AF can be trusted is a question only the next method answers.

Run-to-failure: the test that checks all the others

Run-to-failure (RTF) operates units at a defined stress, with no maintenance or intervention, until they fail, and records each time to failure. RTF is among the most valuable reliability tests a program can run, because it is the one that confirms the acceleration factor every other test depends on.

Run to failure at two or three different stress levels and the results form a stress-life curve, also called an S-N or Wöhler curve. On log-log axes the curve is close to a straight line. Extrapolate that line to the use stress and the team has a field-life estimate backed by its own data instead of a borrowed coefficient.

Stress-life curve built from run-to-failure testsFailures at three elevated stress levels fall along a straight line on log-log axes. The fitted line is extended to the lower use stress to estimate field life.log(life to failure)log(stress)Test level 3Test level 2Test level 1Use stressPredicted field life
Run-to-failure at two or three elevated stress levels gives a stress-life line. Extrapolating it to the use stress is valid only while every level fails by the same mechanism.

RTF data also produces the standard reliability metrics: Weibull shape and scale, reliability at a stated confidence, the failure rate at a given time, and B10 life. All of these can be checked against the mission profile or duty cycle. Once the stress-life relationship is established, ALT can be used with confidence to verify later design changes, which can shorten the development cycle substantially.

The condition that makes it work: every stress level must fail by the same mechanism. If the highest level brings in a new failure mode, those points do not belong on the line, and extrapolating through them produces a confident, wrong answer. Do failure analysis on every failure, not on a sample of them.

Fitting the failure times is where the Weibull analysis calculator comes in. It handles suspended units too, so a unit removed from test without failing still counts.

Degradation testing: measure the drift, skip the wait

Degradation testing applies continuous stress or use and tracks a performance parameter as it declines over time, for example resistance, leakage, light output, seal force or capacity. It resembles run-to-failure, but it is usually much shorter, because the units do not have to fail. Once the degradation path is modelled, the time at which each unit would cross its failure threshold can be projected.

This makes degradation testing a practical substitute for run-to-failure on long-lived products, and it is the natural method for wear-out mechanisms. Run at constant stress until the degradation trend is well enough established to extrapolate. Watch for a change in slope, which often signals that a second mechanism has started. Because the method gives a rate, it is also the right basis for maintenance schedules and replacement intervals.

Reliability demonstration testing: prove the target

Reliability demonstration testing (RDT), often run as a success-run or test-to-bogey, happens late in development to show that the product meets a stated reliability target at a stated confidence. Its output is pass or fail. The classic zero-failure sample size is:

n=ln⁡(1−C)ln⁡(R)n = \frac{\ln(1 - C)}{\ln(R)}

For example, demonstrating R=0.90R = 0.90 at C=0.90C = 0.90 takes 22 units tested to one bogey with no failures.

That formula is the easy part. An RDT is only meaningful when three inputs are right: the sample size, the acceleration factor that converts test time to field time, and the mission profile that defines one bogey. If any of the three is wrong, the demonstration proves something other than the target.

RDT usually takes one of two forms:

Sample size depends on the claim being made. The reliability sample size calculator covers success-run, failures-allowed, extended-bogey and Weibayes plans. The mission profile builder defines what one bogey represents.

Burn-in: screen out infant mortality

Burn-in runs production units at elevated stress, usually temperature and voltage, to make weak units fail before they ship. For example, a processor manufacturer might run every device at elevated temperature and voltage for 24–48 hours, so that units with latent manufacturing defects fail in the factory instead of at the customer.

Burn-in only pays off when the population actually has an early-life failure tail, meaning a decreasing failure rate (Weibull β<1\beta < 1). Applied to a population that is already in its useful-life or wear-out region, it removes no defects. It only uses up life every unit was supposed to deliver. The bathtub curve and Weibull β guide explains how to tell which region a population is in.

Choosing the temperature, voltage and duration is a balance: enough stress to bring out defective units, not so much that good units are damaged. That balance cannot be set with confidence without the methods above: HALT for the limits, ALT and run-to-failure for the acceleration factor. The Burn-In Wizard turns those inputs into a screen duration.

Putting the methods in order

  1. Concept and early design: HALT and step stress to find weak points and establish margins.
  2. Design verification: ALT, run-to-failure and degradation testing to estimate life and confirm the acceleration model.
  3. Validation: RDT to demonstrate the reliability target at confidence against the mission profile.
  4. Production: burn-in, or HASS, where the population has an early-life defect tail worth removing.

Each step depends on what the earlier ones established. An RDT built on an acceleration factor that no run-to-failure ever confirmed is only as good as the coefficient it borrowed. To assemble all four stages into one plan, with test durations driven by the mission profile, start with the Test Plan Generator.

Reliability testing is not a box to tick before launch. It is the evidence that a product will do what it was designed to do, for as long as it was designed to do it. Used in the right order, these methods find problems while they are still cheap to fix and give release decisions numbers they can defend.

Want help applying this to your validation plan? Contact Reliatools.