Dealing with limited data in ballistic impact scenarios: an empirical comparison of different neural network approaches

In the domain of high-speed impact between solids, the simulation of one trial entails the use of large resources and an elevated computational cost. The objective of this research is to find the best neural network associated with a new problem of ballistic impact, maximizing the quantity of trials available and simplifying their architecture. To achieve this goal, this paper proposes a tuning performance process based on four stages. These stages include existing statistical techniques, a combination of proposals to improve the performance and analyze the influence of each variable. To measure the quality of the different networks, two criteria based on information theory have been incorporated to reflect the fit of the data with respect to their complexity. The results obtained show that the application of an integrated tuning process in this domain permits improvement in the performance and efficiency of a neural network in comparison with different machine learning alternatives

neural network; limited data; bootstrapping; cross-validation; ballistic impact; numerical simulation; machine learning classifiers

Dealing with limited data in ballistic impact scenarios: an empirical comparison of different neural network approaches Articles