reliability engineering

A strategic job for preparing plans to reduce the failures and the cost of failures as a preventative measure to reduce the cost of unreliability. Acquires failure data and analyzes the data to quantify the financial impact and prepare long term solutions to prevent reoccurrences to improve reliability and uptime. Determines the cost advantages and proposes alternatives for solving the problem and recommends the alternative with the lowest long term cost of ownership. The purpose of these actions is to prevent failures.

Reliability growth models are important management concepts for making reliability visual with simple displays. The simple log-log plots of cumulative failures on the Y-axis against cumulative time on the X-axis often make straight lines where the slope of the trend line is highly significant for telling if failures are coming faster (b>1) which is undesirable, slower (b<1) which is desirable, or without improvement/deterioration (b=1), which usually drifts toward undesirable results. The reliability growth models are frequently called Crow-AMSSA plots in honor of Larry Crow's proof of why the charts work as described in MIL-HDBK-189 when he worked with AMSAA.

For expensive components and expensive tests, sudden death tests involve a few components that tie-up a test frame as they are heavily loaded under the same test loads/conditions with several items being run at the same time. When one of the items fails the entire test frame is shut down so that you have 1 failure (this is the sudden death!) and several suspensions because the unfailed units are survivors as the test is halted until the test frame is loaded with new samples for resumption of the life test. Opening the test frame (instead of tying up the frame until all samples have failed) is cost effective. If three units can be tested simultaneously and the test is halted on the first failure, then perhaps we will literally have only 4 failures and 8 suspensions for preparing the Weibull analysis. Will the 4 sample + 8 suspension data set be different than if all 12 samples had been run to failure?-the answer is yes, they will be different, but will they be significantly different-the answer is no to the significant difference. So, as with simultaneous testing the suspensions (censored data) become important details for use in the statistical analysis. Most sudden death tests are accelerated to generate the data in a short period of time although this carries the risk of introducing unexpected failure modes (but this can also be useful information for anticipating field failures).

Software does not wear out but it does fail and most failures are due to specification errors and code errors with only a few errors in copying or use. The only software repair is by reprogramming and adding safety factors is almost impossible. Software reliability improves by finding errors and fixing the errors but estimating the number of errors which canse failures is extremely difficult as many branches of software code may lie dormant and unused until special events occur to make the latent failures obvious. Software failures are not often time related but are more software code page dependent. Software reliability is improved by extensive testing to disclose the failures and then fixing them to repeat the test all over again to validate the fix did not generate more failures and to continue the search of other latent defects.