A branching algorithm to reduce computational time of batch models: Application for blast analyses. Issue 2 (June 2023)

• Record Type: Journal Article
• Authors: Dennis, Adam A; Smyl, Danny J; Stirling, Chris G; Rigby, Samuel E
• Is Part Of: International journal of protective structures. Volume 14:Issue 2(2023)
• Page Start: 135
• Page End: 167
• Publication Date: 2023-06
• Subjects: Risk -- probability -- computation time -- data -- batch; Buildings -- Protection -- Periodicals; Buildings -- Blast effects -- Periodicals; Building, Bombproof -- Periodicals; Structural analysis (Engineering) -- Periodicals; Shock (Mechanics) -- Periodicals; 624.176
• DOI: 10.1177/20414196221085720
• Languages: English
• ISSNs: 2041-4196

Abstract

Numerical analysis is increasingly used for batch modelling runs, with each individual model possessing a unique combination of input parameters sampled from a range of potential values. Whilst such an approach can help to develop a comprehensive understanding of the inherent unpredictability and variability of explosive events, or populate training/validation data sets for machine learning approaches, the associated computational expense is relatively high. Furthermore, any given model may share a number of common solution steps with other models in the batch, and simulating all models from birth to termination may result in large amounts of repetition. This paper presents a new branching algorithm that ensures calculation steps are only computed once by identifying when the parameter fields of each model in the batch becomes unique. This enables informed data mapping to take place, leading to a reduction in the required computation time. The branching algorithm is explained using a conceptual walk-through for a batch of 9 models, featuring a blast load acting on a structural panel in 2D. By eliminating repeat steps, approximately 50% of the run time can be saved. This is followed by the development and use of the algorithm in 3D for a practical application involving 20 complex containment structure models. In this instance, a ∼20% reduction in computational costs is achieved.