Prediction of blast loading in an internal environment using artificial neural networks

Abstract

Explosive loading in a confined internal environment is highly complex and is driven by nonlinear physical processes associated with reflection and coalescence of multiple shock fronts. Prediction of this loading is not currently feasible using simple tools, and instead specialist computational software or practical testing is required, which are impractical for situations with a wide range of input variables. There is a need to develop a tool which balances the accuracy of experiments or physics-based numerical schemes with the simplicity and low computational cost of an engineering-level predictive approach. Artificial neural networks (ANNs) are formed of a collection of neurons that process information via a series of connections. When fully trained, ANNs are capable of replicating and generalising multi-parameter, high-complexity problems and are able to generate new predictions for unseen problems (within the bounds of the training variables). This article presents the development and rigorous testing of an ANN to predict blast loading in a confined internal environment. The ANN was trained using validated numerical modelling data, and key parameters relating to formulation of the training data and network structure were critically analysed in order to maximise the predictive capability of the network. The developed network was generally able to predict specific impulses to within 10% of the numerical data: 90% of specific impulses in the unseen testing data, and between 81% and 87% of specific impulses for data from four additional unseen test models, were predicted to this accuracy. The network was highly capable of generalising in areas adjacent to reflecting surfaces and as those close to ambient outflow boundaries. It is shown that ANNs are highly suited to modelling blast loading in a confined internal environment, with significant improvements in accuracy achievable if a robust, well distributed training dataset is used with a network structure that is tailored to the problem being solved.

A Review of Blast Loading in the Urban Environment

“…Whilst this seems to be the case presented by Remennikov and Mendis [109], it should be noted that the geometries of the neural network models were the same as the database models. It may be more appropriate to say that neural networks can be used as an effective prediction tool, provided they are suitably trained using a thorough dataset to improve performance, similar to a conclusion determined by Dennis et al [110]. Likewise, Ruscade et al [111] use data-rich experimental work (30 pressure vs. time data sets per test) for the training of a numerical model, although no further detail is provided about the model.…”

Physics-informed regularisation procedure in neural networks: An application in blast protection engineering

“…A detailed discussion on different types of surrogate model is given in Jin (2005). One such surrogate model is an artificial neural network (ANN) and has been used extensively for this purpose in a variety of disciplines (Ahmadi, 2015; Kim et al, 2015; Papadopoulos et al, 2018; White et al, 2019) and in specific blast engineering applications (Dennis et al, 2020; Flood et al, 2009; Remennikov & Mendis, 2006; Remennikov & Rose, 2007).…”

“…Conversely, artificial neural networks (ANNs or NNs) can accommodate more variables when acting as a vector mapping model and provide the additional flexibility required to handle highly non-linear behaviour, as expected in extreme near-field blast events. They have been shown to accurately predict explosive loading in confined internal environments (Dennis et al, 2020), on a building behind a blast wall (Flood et al, 2009; Remennikov & Rose 2007) or along simple city streets (Remennikov & Mendis 2006).…”

A Critical Review of a Computational Fluid Dynamics (CFD)-Based Explosion Numerical Analysis of Offshore Facilities

“…Since all the hidden layers can contribute to the level of errors in the output layers to some degree, the output error signals are delivered backwards from the output layer to previous neurons in the hidden layer directly connected to the transient outputs. Once the error for each neuron has been calculated, the errors are then used by the neurons to update the weight of each connection until the network meets the condition that allows all the training patterns to be encoded [103].…”

“…Recently, a lot of researcher have tried to apply such technologies in gas dispersion or explosion analysis regarding typical ERA procedure as described in Table 6. Dennis et al [103] used ANN consisting of two hidden layers to predict the blast impulse based on validated numerical modelling data. Shi et al [104] presented Bayesian regularization ANN-based simplification for gas dispersion and explosion part in CFD-based explosion risk analysis.…”