ABSTRACT
Component reliability is a function of Component and System design, manufacture, assembly and installation. It is a fact however that regardless of how well these processes are implemented, a proportion of components will fail. It is also a fact that you do not know which component will fail or the time it will take. As a result we use “prediction models” to forecast reliability performance and expect accurate results. The most widely recognised such models are; MIL STD 217 (*1) Bellcore (*2), British Telecom’s HRD (*3) and CNET (*4). They all share in using a combination of the pragmatic and theoretical approach, and similar concepts such as the assumption of a “constant failure rate” during the “mature” period. However, although the base data used in many of these models has been subject to periodic change and the addition of new correction factors, the assumptions made in using this data have remained essentially static during a period when technology and its application have undergone considerable change. As a result these models are becoming increasingly unwieldy and in many cases totally inaccurate.
Within GPT-Public Switching we have experience with both Bellcore and HRD, although it is the latter which forms the basis for most Design, Engineering and Contract work. The significance of the degree to which the HRD model can predict to a reasonable degree of accuracy the ultimate reliability performance of a Telephone System has increased substantially over the last 5 years. This has resulted from greater emphasis by both manufacturer and customer on “cost of ownership” coupled with the need for reliability improvement that demands some realistic basis for comparison. As a consequence considerable effort has been directed during this period to producing a measure of both component and equipment performance in comparison to HRD (Issue 4) predictions.
This has resulted in considerable reliability improvements to our product, in addition a number of anomalies have been identified with both the data and the assumptions used by HRD.
285These covered; errors in predicting “low-technology” established components in contrast to for example complex IC’s where the model performed well; criticism of the use of the “Quality” and “Maturity” factors; and the need to overhaul but not necessarily eliminate the “constant failure rate” concept. This paper details this work and also seeks to look ahead to the requirements for the next generation of reliability model that will be required to overcome the more fundamental aspects of these problems.
