Sensitivity testing and logit analysis: Two recent armaments engineering case-studies

Efficient statistical techniques for designing and analyzing experiments are often misunderstood or underutilized, despite possessing great potential when well-informed decision-making and significant cost-savings are desired. Destructive testing with binary response data is a `worst-case' scenario with regards to cost and efficiency; however, in armaments engineering (especially with explosives, energetic mixes, propellants, pyrotechnics, and other one-shot devices) binary response data is often all that is available. This paper illustrates modern methods and best-practices to employ when dealing with these types of analyses, and contrasts two very different types of testing strategies applied to the munitions used in training and battle by the Warfighter. First, we introduce the topic of test design and experimentation. To motivate the subject and illustrate some basic aspects of sensitivity testing, we present an example common to the general public: impact-resistant cases for smart-phones. Next, we elaborate on some of the finer details of Logit Analysis/Binary Logistic Regression and Generalized Linear Models (GLM), while highlighting some of the mathematical underpinnings inherent in these methods. An overview of testing strategies follows, which compares some of the different methods available for data generation. These methods can be roughly divided into two groups - those dealing with pre-manufactured test samples, and those where sequential testing is an option. Sequential tests are generally applicable when the stress level of interest is adjustable during the test's execution, which, with regard to sample size, allows for a more efficient test. Then we detail two recent successful examples in armament munitions testing: one where sequential testing was not possible (9mm ammunition propellant critical threshold development), and one where sequential testing was utilized (NMT - Networked Munition Technology). Finally, we discuss some best-practices, limitations, and rules-of-thumb to be mindful of when considering these methods for different applications.