So far we have introduced the idea of failure rate, and illustrated the facts that component or product life depends on the conditions, varies with time, and can be dramatically affected by the inclusion of defects. Now we need to address the question of whether a product is intrinsically sufficiently reliable, and in testing for this find out enough about the actual failure modes to be able to devise screens to prevent potential defects reaching the customer.
Unfortunately, time constraints dictate that we cannot duplicate intended life, but typically have to compress the test time to a reasonable value. Most reliability testing therefore applies more rigorous conditions to the device under test in order to accelerate failure, making it possible to get meaningful results within a short time. This gives both economic savings and quick turn-around during the development of new products or of mature products subjected to manufacturing and workmanship change.
The results from the tests are then extrapolated to give an estimate of the life for the product. The assumption made is that tests can be carried out under conditions of higher than usual stress (‘accelerated’ stress), and the effects of this stress can be represented by an acceleration factor A, where:
[life in use] = A × [life at accelerated condition]
Of course, this is only valid if:
The degree of acceleration may vary significantly, and some workers in this field define two types of test:
The danger is that high levels of stress may induce failure mechanisms that are not present in normal use.
In designing and interpreting accelerated tests, it is necessary to understand which stresses accelerate which failure mechanisms, and how varying the magnitude or rate of application of the stress influences product life. Table 1 suggests some of the links between mechanisms and stresses:
|wear-out failure mechanisms||acceleration stresses|
|fatigue crack initiation||mechanical stress or strain range, cyclic temperature range, frequency of cyclic stress|
|fatigue crack propagation||mechanical stress range, cyclic temperature range, frequency of cyclic stress|
|creep||mechanical stress, temperature|
|diffusion||temperature, concentration gradient|
|corrosion||temperature, relative humidity, contaminants|
|electromigration||current density, temperature, temperature gradient|
|dendritic growth||voltage differential, temperature|
|radiation ageing||intensity of radiation|
|stress corrosion||mechanical stress, temperature, relative humidity, contaminants|
|wear||contact force, relative sliding velocity|
Several types of stress may be used to accelerate failure, and accelerated tests should only be used when the correlation between test and field conditions is clearly understood:
[ back to top ]
The procedure for accelerated testing consists of:
The constants in a model are typically derived by testing sample populations at several stress levels and extrapolating these values to typical operational environmental conditions, as shown in Figure 1. The more sensitive the item under test is to the imposed stresses, the greater will be the slope of the projected line, and the more uncertain the projection for field life
after Quanterion Solutions
The validity of extrapolation depends on how accurately the model predicts acceleration. The mechanisms are different for temperature, humidity and mechanical stress, and not necessarily known with certainty. There is some discussion on these effects in this link (PDF file 341KB). This shows a number of different models that attempt to link failure under high stress to failure under normal working conditions. Typically, the lower the acceleration level, the closer the fit between prediction and actual results.
Even if you don’t have time to read the whole of the linked document, you need to be aware of the Arrhenius formula, as this is a common way of relating the rates of physical and chemical processes to temperature. Originally based on observations relating to biological models, in the past it has been a surprisingly good fit to electronic information. However, as we will see later, it must now be viewed with some scepticism.
[ back to top ]
Accelerated tests can be roughly broken into two types:
In the most general case, these accelerated tests can be carried out in three ways:
The last of these is common practice for electronic equipment, but there are variations in how the stress is applied:
In step-stress tests, progressively increasing stresses are applied to the same sample for constant time intervals until all samples have failed. Figure 2 is a schematic presentation of typical results. The embedded assumption is that the failure probability during each time interval is independent of previous history, but, because the effects of exposure at high temperatures are cumulative, the gap between temperatures has in practice to be kept equal and relatively large.
Step-stress tests generate data on product reliability more quickly than constant stress tests, but are more difficult to evaluate. Constant stress testing is therefore the most common procedure, with step-stress used mostly for quick comparison tests.
Failure mechanisms that are not apparent under normal conditions may become significant under accelerated stress, a phenomenon called failure mechanism shifting.
Because of inevitable process and material variations, there will be statistical distributions of time-to-fail results for each stress level. Combining these with the stress-life model, a statistical estimate can be made of time-to-failure under operating conditions.
You have developed a new consumer product that will be thrown away next Christmas, but needs to survive until then. What steps could you take to give confidence that your warranty returns will not be excessive?