Mechanical properties and multi-layer perceptron neural networks of polyacrylonitrile fiber reinforced concrete cured outdoors

Abstract

The application range of polyacrylonitrile fiber (PANF) in concrete engineering in alpine regions was clarified by performing six sets of mechanical tests of PANF-reinforced concrete (PANFRC) by outdoor curing. The results show that PANF can enhance the performance of concrete mainly in terms of the tensile performance. The splitting strength and flexural strength increased by 15.69% and 8.54% after 150 days, and the dosage used in the alpine region should be controlled between 1.2–1.5 kg/m3. Finally, based on the experimental results, the multi-linear regression prediction model (MLR) and multi-layer perceptron neural networks model (MLP) were used to predict the compressive strength (fcu), splitting strength (fsp), flexural strength (fts), tensile–compression ratio (fsp/fcu), ratio of flexural strength to compressive strength (fts/fcu), and ratio of flexural strength to splitting strength (fts/fsp) of each group of PANFRC specimens. The prediction model was constructed based on the fiber content (W), curing age (D), dynamic elastic modulus (E), and surface rebound hardness (R). The results of the model test show that MLR and MLP are reasonable for predicting the mechanical properties of PANFRC, and the prediction error of the latter is smaller. Moreover, R has a strong dependence on the mechanical evaluation index, the sensitivity coefficient Qik is 0.37, PANF has the least influence, and Qik is only 0.16. This study has reference significance for the application of PANFRC in alpine regions and provides useful ideas for the efficient and intelligent development of the construction industry in alpine regions.