Stochastic material model based nonlinear finite element analysis of reinforced concrete beams

Abstract

implementing a machine learning algorithm can resolve the issue. This thesis presents an optimized Python-based nonlinear finite element analysis programme with a stochastic material model defined by the Gaussian process regressor. The overall process is decomposed into modules with vectorized Tensor library for seamless and intuitive machine learning integration. The analysis using the machine learning material model showed a corresponding structural response with a constitutive material model. Furthermore, the Python code with optimized structure using both CPU and GPUs for accelerated computation showed the drastic reduction of time of nonlinear layered finite element beam analysis. This suggested method is expected to shorten the processing time and reduce the hardware performance requirement of the programme while accurately simulating the real-world structural response of the large-scale structure with complex material behaviour.

Finite element analysis has been extensively implemented numerical analysis methodology for a long time. However, in the case of analysing a large-scale structure with a nonlinear finite element method, computing cost drastically increases, which means maximizing the efficiency of the programme and vectorizing the overall process for parallel processing with hardware acceleration becomes essential. Furthermore, defining the constitutive model for new material with a complex stress-strain relationship based on experiments is challenging