George Hieber is an acknowledged expert on dynamics and fixture design, who for many years taught TTi's popular course Fixture Design for Vibration and Shock Testing.
Advocates of Highly Accelerated Life Test (HALT) and Highly Accelerated Stress Screen (HASS) procedures point to some apparent benefits to product design and manufacture. (1)
As can be inferred from the acronyms, these procedures are designed to save time in the evaluation of the sturdiness of equipment. The difficulty is that any attempt to “speed up” failure modes requires that the intensity of the specific environmental stimulus be increased, with the attendant problem that failure modes can be generated which will not happen in the actual use of the product in the field.
Environmental stress screening (ESS) is a process which attempts to weed out weak parts or poor workmanship by applying environmental stresses which are not necessarily representative of field environment. Because of this, vigilance must be exercised to insure that unrepresentative failure modes are not causing good parts to appear as giving unsatisfactory performance.
As this screening process is accelerated, there is an accompanying increase in the need for very careful evaluation of failures.
When it comes to HALT, however, a more serious situation arises. It can be argued that this procedure is not based on sound engineering principles. This procedure applies environmental stimuli, primarily vibration and temperature, in an effort to determine the “fundamental limit of the technology, or destruct limit” of the device under test. These environmental stresses are not meant to simulate field environments, but are stepped up to well beyond the expected field environments.
This procedure will certainly result in a rugged design, but will it be a good, much less, an elegant design? It's unlikely. After all, the goal of good design is to provide a device which will perform reliably during its lifetime, with minimum complexity, cost and weight. In order to do this, it is necessary to design for the expected end use, and that means devising tests which realistically assess the product’s reaction to real field stimuli. This has been emphasized in the military starting with MIL STD-810D, where test tailoring was recommended. The goal of this recommendation is to encourage suppliers to tailor their environmental tests upon the results of their own field measurements, rather than test to values in existing specifications which may be too general in scope. In contrast, HALT concentrates on ruggedizing design without respect to end use stimuli.
To illustrate the misuse of HALT, consider applying HALT on a wristwatch design (admittedly, some hyperbole is used here), the testing would not apparently be satisfied until the watch design is altered to survive several hammer blows and extended temperature exposure to 300°F. This could result in a very rugged, expensive and heavy timepiece which may not be any more accurate than the original watch design, but could withstand external forces and temperatures which it would not reasonably experience.
Aerospace structures are designed to withstand expected maximum loads; after testing to these loads, the loading is increased to ultimate, which is the expected maximum load times a safety factor. The structure must withstand this. The loading is then further increased where it is hoped that the structure will gracefully fail (as opposed to catastrophic failure) just above ultimate, which would prove it to be a very efficient design. Note that there is no effort made to increase the strength needlessly above this point.
With much smaller structures, such as electronic chassis and printed circuit boards, there is no justification to go through the sophisticated stress analysis alluded to above. In these cases, it is more probable that a small structural part will fail during testing because of the difficulty of predicting load paths in complex equipment. In this event, some repeated testing may be necessary.
It may be deemed proper in some cases to increase the stress levels in steps to prevent a catastrophic failure due to some unexpected weakness. But the expected field environment must be used to tailor the test. There is an exception, however: Even if a device is going to be used in a benign environment (TV set, laboratory instrument), where the field use is at room temperature and vibration is limited to footsteps passing nearby, it is proper to run an exposure such as the “Minimum Integrity Test” as called out in MIL-STD-810 to make sure that the design will survive assembly operations and reliably perform after shipping and handling.
The philosophy of the Minimum Integrity Test is similar to that of a very subdued HALT, in that an environmental stimulus is applied which is not representative of expected field environment and is used to evaluate quality of design.
As stated above, a significant problem with both HALT and HASS is associated with the terms “Highly Accelerated”. The major concept here is that damage due to environmental exposure in the field can be equaled in a shorter time in the laboratory by increasing the intensity of that exposure. Although this is true, various subtle and hidden factors can cause misleading and erroneous results.
Take structural fatigue for example. It is accepted that structural fatigue due to stress cycling can be caused by either:
and is plotted as so-called S-N curves for various materials. (N is the number of load reversals which will cause structural failure at peak stress S). It can be shown that on a log-log plot, these curves become straight lines, defined by an intercept and a slope.
N × Sb = C (eq. 1)
where b is the slope, and C is a constant.
Note: This equation is valid between the yield stress and the endurance stress.
If a critical part is subjected to a certain reversed stress in the field for a known number of stress reversals, then it can be shown that the same damage could be caused in a much shorter time by using the relationship shown in the equation. But there can be problems:
For sine excitation:
 (eq. 2) |
For random excitation:
 (eq. 3) |
where:
| T0 | is the field data reference time in hours, | |
| T1 | is the reduced test time in hours, | |
| G0 | is the reference sinusoidal vibration level (peak acceleration G’s) | |
| G1 | is the increased sinusoidal vibration level in order to reduce the test time to T1 | |
| W0 | is the reference random vibration level (Acceleration [Power] Spectral Density) | |
| W1 | is the increased random vibration level in order to reduce the test time to T1 | |
| n | is an exponent which relates stress to damping energy, | |
| b | is the slope of the S-N curve. | |
| m | the exponents b/(n-1) and b/n are also referred to as material constant m |
These equations can be used to determine the excitation change required to cause a certain desired change in test duration.
A number of metallic structural materials exhibit an S-N slope close to 9, so frequently books and papers on the subject tend to use b = 9. When it comes to damping, the subject is more complex and the value of n depends on the relationship between vibration input and stress response which is non-linear. It can be shown (3) that with stresses below 80% of the endurance stress limit, n = 2.4 while at higher stresses, n = 8.
Using the values of b = 9 and n = 2.4, the exponent
or material constant m is calculated as:
MIL-STD-810E (4) used b and n values which were averaged from several structural materials, giving exponents which were then rounded off. For sine testing where G0 represents the field data, the exponent m is rounded off to 6.
For random testing where W0 represents the field power spectral density, the exponent m is rounded off to 4. MIL-STD-810E also states, vaguely, that "other values may be appropriate."
Since the values of b and n are chosen to provide average results for various excitation, they are in fact arbitrary and their use can lead to erroneous results.
It is strongly recommended that the value of b and n be derived from information on the specific materials used in a critical structure, rather than taking for granted that exponents 6 and 4 are applicable.
The following examples illustrating the potential problems that can occur if care is not taken. Consider a device with a design life of 1000 hours operating at a certain stress level. It is desired to apply a random vibration test to cause the same damage in an accelerated time of one hour. The random excitation equation can be used to determine the excitation level required, provided that the value of the S-N curve slope (b) and the value of exponent (n) which relates stress to damping energy are known.
Using the values of the material constant m as given in MIL-STD-810E for random vibration m = 4, based on the S-N slope of 9, with n = 2.4, with T0 = 1000 hours =103, T1 = 1 hour
The exponent b/n = m = 4 and dividing the exponents on each side of the equation by m and re-arranging,
W1 = W0 × 10-3/4 = W0 × 10-0.75 = W0 × 5.6
W1 = 5.6 × W0
This means that in order to complete the test in one hour the random vibration stress level must be increased by a factor of 5.6.
What if the critical part of the device used in the above example is made of a material which has a S-N slope b significantly different from the average value of 9 used in the MIL-STD-910E? The material is 17-4PH stainless steel whose S-N slope b is 17.6
Using a damping energy exponent n = 2.4,
To obtain the same stress levels determined in example A,
namely 5.6 : 1, the time of the test T1 is calculated as follows:
Since then 
Using a damping energy exponent of n = 5 and with b = 17.6,
the time of the test , then
The above examples illustrate that to test a device to a certain stress level with different exponent values, the test time can vary from 12 seconds to 60 minutes to 2.3 hours.
Thus the test could be either overtest or undertest. Specific S-N data should be used, rather than average, and more experimental material damping data should be generated.
So, it can be seen that significant detailed information must be available regarding the material of a critical part, if a fatigue phenomenon is to be evaluated in an accelerated test by increasing the stimulus level. If the stimulus is random excitation, an interesting approach has been developed in the automotive field (5), wherein accelerated tests can be performed without having to increase the stimulus level. In this method, the field records are examined and only those regions are selected where the vibration is high enough to cause significant fatigue damage accumulation. These regions are pieced together to form the test program. As random signals spend most of the time at low levels, it is apparent that time savings can be effected if these lower level stimuli are left out, with the result that a practical test acceleration can be accomplished without having to increase stimulus levels. This method has been successfully applied with significant test time reductions (1/3 to 1/4 test time required compared to field exposure times).
A reason for these particular results being so dramatic is that road vibration tests have shown that the random excitation is not strictly Gaussian (6). Significantly higher spikes occur (due to potholes and RR crossings, for instance) than would be dictated by the Gaussian probability density curve for a given G rms value. But the occurrence of these damaging levels is quite low, leading to the substantial test time reduction.
A note of caution should be added here. Whether the test acceleration is accomplished by increasing the level of the stimulus or by stacking together the higher inputs as just described, the result is that the structure is going to be worked harder, so that the material is going to heat up more so than in the field environment. This would probably be no problem with conventional structural metals, but if the critical items happened to be elastomeric isolators, for instance, they are likely to be destroyed (as has happened). The reason is that elastomers generate a great deal of internal heat while providing damping, and their poor heat conductivity leads to overheating. So if there is insufficient time between input spikes for heat dissipation, overheating will occur, resulting in a failure not necessarily representative of a real problem. This is another indication of the importance of knowing what the critical item is, and of knowing the physical characteristics of the material involved.
An environment which can frequently be more troublesome than vibration and shock is temperature. And again, the practice of accelerated testing by increasing temperatures significantly higher than field conditions can cause failures which do not represent likely field failures. An investigation (7) has shown that at least eight different failure mechanisms having to do with electronic circuits can be exacerbated by testing at higher temperatures than that existing in the field. This is not just to imply that the failure will occur sooner at higher temperature than at lower temperature — this, after all, is the very hoped for result of accelerated testing. Rather, the investigations showed that failures could be induced in the artificial laboratory environment which would not occur in the field environment. The reason is that there are certain physical and/or chemical reactions which are triggered at specific energy levels, so an effort should be made to understand a particular failure mechanism before deciding that a laboratory failure is indicative of a design fault.
Mechanisms such as creep, diffusion, corrosion, charge spreading and slow trapping (either of the latter two phenomena cause maloperation of semiconductor junctions) are all temperature dependent, and the important point is this: If a particular temperature sensitive failure mechanism is suspected in the field, and an accelerated temperature test is run in the laboratory to determine if the device will survive its life requirement, it is possible that a failure may occur in the laboratory at this higher temperature which is due to a different failure mechanism; one which would not happen in the field. Unless a careful examination of the failure is performed, it would be assumed that the design is unsatisfactory, when it may be perfectly acceptable. In this arena, again, it can be seen that accelerated testing has its own set of hobgoblins.
The topic covered in this technical paper is covered in various TTi courses (see our schedule for the next presentation):
toll-free tel. 866-TTi-4Edu (866-884-4338)
E-mail: Training@ttiedu.com
© Technology Training, Inc. 2006
050919 BJA