Rapid mechanical evaluation of quadrangular steel plates subjected to localised blast loadings

The design of modern military and naval platforms against weapon threats is often assisted by a combination of experimental, analytical and computational simulations. These tools provide relevant insights about material reliability, mechanical performance and platform design vulnerability to support the determination of safety critical aspects, such as response to blast and fragmentation loading. Analytical models are inherently simplified, limiting their ability to accurately model scenarios with complicated geometries and material properties, or highly non-linear loadings. Appropriate experimental and numerical modelling can overcome the limitations of analytical models but also require long lead times and high associated costs. These issues can be a point of concern for projects with strict development schedules, short time-to-solution, and limited resources. Machine learning techniques have proven viable in the development of fast-running models for highly nonlinear problems. The present work explores four models based on the Multilayer Perceptron (MLP), a type of Artificial Neural Network (ANN), for assessing the mechanical response of mild steel plates subjected to localised blast loading. Experiments combined with validated Finite Element Analysis (FEA) models provide a hybrid dataset for training ANNs. The resultant dataset is a combination of sparsely populated experimental data with a denser dataset of validated FEA simulations. The final results demonstrate the potential of ANNs to incorporate high strain-rate material response behaviour, such as that from blast loading, into optimised models that can yield timely predictions of structural response.