Kernel extreme learning machine and finite element method fusion fire damage prediction of concrete structures

To achieve reasonable fire damage evaluation of concrete structures, a model-driven and data-driven fusion prediction framework is proposed in this investigation. In the framework, finite element method (FEM) coupled with a thermo-mechanical damage model is used to provide forward response calculation of concrete structures under the combined action of high temperature and external forces. Kernel extreme learning machine (KELM) is utilized to invert the thermal and mechanical performance parameters in finite element computation with aid of the measured response data. Additionally, sand cat swarm optimization (SCSO) algorithm is utilized to improve inversion performance. Fire damage of a concrete column and a concrete frame structure is studied and compared with the corresponding experiments. Through comparison, it can be found that the fire damage simulation of the two examples can match well with the corresponding experimental results. The results support that the proposed model-driven and data-driven fusion prediction framework with aid of KELM coupled with a SCSO and FEM coupled with a thermo-mechanical damage model can be utilized to support a useful tool for fire damage prediction of concrete structures.