A statistical approach to proof testing

Police body armour has undergone rapid evolution, and this is due in part to a relatively simple type approval process which matches relatively small numbers purchased by individual police forces. However, as body armour technology and usage has increased there has been a gradual change of emphasis from solving the immediate protection problems of highly specialised systems towards quality assurance towards of a standard item of equipment. In addition, as the amount and age of armour in service increases, concerns have been raised about methods for ensuring the continued performance over long periods of use or over long production runs. These factors have drawn attention to the statistical significance of existing proof tests, in which armour systems are subjected to small numbers of stabs or ballistic tests. A number of approaches have been suggested including the addition of a V50 ballistic limit test to provide a fully quantitative measure of performance [1]. However this approach also lacks statistical rigour and further enhancements such as regression analysis [2,3] have been suggested to remedy this. In the current work a different statistical approach is suggested in which conventional proof tests can be used to produce statistically robust data of known significance. Initial trials on current police body armour showed that ballistic penetration and knife penetration were similar as the data was highly random and it was difficult to statistically predict individual test results. Ballistic blunt trauma followed a more predictable pattern with simple and easily predicted test-to-test variation allowing good predictions to be made. For the knife and ballistic penetration tests two approaches have been investigated. One method is a point estimate approach that determines failure probability as a simple ratio of pass or fail. Therefore to achieve a failure probability of lower than 0.1 (10%) no more than 1 failure in 10 would be allowed. The second option would be determine how many successful tests were needed to be sure (for instance to 95% probability) that the failure rate was no more than 0.1. This second approach is more severe and it has been shown that at least 28 successful tests are required in order to be reasonably sure that the failure rate is less than 0.1. This paper will demonstrate the development of the statistical model which has been used within 2007 HOSDB body armour standards and shows how it is applied in both type approval and batch testing.