Predictions, service failure

These examples illustrate the application of Weibull plots to service failure data in order to predict the probability of failure before the end of the service life, coupled with a power law to relate time to failure to the applied electric stress (voltage). 

This classification can be loosely linked to the purposes of testing. For quality control, fundamental properties are not needed, apparent properties will usually be acceptable, although functional properties would certainly be desirable. For predicting service performance, the most suitable properties would be functional ones. For design data, fundamental properties are really needed, although considerable help can often be got from functional properties. For investigating failures, the most useful test will depend on the individual circumstances but it is unlikely that fundamental methods would be necessary. 

This means that non-linear phenomena will happen in CII due to the coupling and complex interactions among its components. Not all outcomes to perturbations, even minor, can be predicted. Some of them will prove unmanageable, conducting the system to unwanted states and potential service failures. 

From the design point of view, it would be desirable to be able to predict the service life of a given bonded structure in the use environment this capability has yet to be developed. Some progress has been made, however, particularly in the application of reaction rate theory to predict the failure time of test specimens. 

There have been many other empirical approaches to predict service life from accelerated test results, some based on relations between change in polymer properties and exposure variables (78). A simple empirical method of estimating lifetimes by comparison with control materials has recently been proposed (79). The method consists of exposing the test materials in the laboratory accelerated test device simultaneously with several control materials with similar composition and construction to the test material and having a range of failure times outdoors. It requires, as do all service life methods, that the accelerated test produce the same failure modes as natural exposure. In addition, rank correlation between the two exposures should be very high. If these conditions prevail and the service life of the control materials is well defined, the service life of the test material can be bracketed by two control materials. 

All the dimensional problems associated with metals are present in plastics, but to a greater extent. In addition, there are some entirely new ones. All these problems are predictable and containable, but they can never be ignored. It is probably true to say that failure to allow for dimensional changes is responsible for more in-service failures with plastics than any other factor. The six prime causes follow ... 

The time to failure of specimens subjected to an accelerated lifetest is a random variable which is usually distributed according to a lognormal cumulative probability function (CPF). Once the shape and the values of the parameters characterising the CPF have been determined, it is possible to predict the failure rate of the sampled population as a function of the time in service." ... 

The reliability of a particular type of component will vary from production batch to batch, but it would be uneconomic to perform the variety of lifetests needed to predict service performance on every batch. However, by performing such a series of tests before a particular type of component is used in service, it is possible to identify and characterise the failure mechanisms, and to design simpler tests which can be regularly used to assess whether future production batches are likely to meet the required reliability standards. These simpler tests are mostly based on a specification of the maximum permitted failure fraction after a particular stress test. 

---