Prediction of the frost resistance of high-performance concrete based on RF-REF: A hybrid prediction approach. (23rd May 2022)

• Record Type: Journal Article
• Authors: Wu, Xianguo; Zheng, Shiyi; Feng, Zongbao; Chen, Bin; Qin, Yawei; Xu, Wen; Liu, Yang
• Is Part Of: Construction & building materials. Volume 333(2022)
• Publication Date: 2022-05-23
• Subjects: High-performance concrete -- Frost resistance prediction -- Relative dynamic modulus of elasticity -- Random forest -- Index screening; RF Random forest -- REF Recursive feature elimination -- RMSE Root mean square error -- R2 Goodness of fit -- SVM Support vector machine -- ANN Artificial neural network -- ML Machine learning -- GGBFS Ground granulated blast furnace slag -- OBB Out of bag; Building materials -- Periodicals; 624.18
• DOI: 10.1016/j.conbuildmat.2022.127132
• Languages: English
• ISSNs: 0950-0618

Abstract

Highlights: RF combined with REF was proposed to predict frost resistance of concrete. REF helps to identify the key factors of mix proportion influencing frost resistance. The parameters of the proposed RF model were determined to improve predication accuracy. The applicability of the proposed RF-REF model was verified in the case of diversified mix. Abstract: Infrastructure projects in extremely cold areas have high requirements regarding the frost resistance of concrete. To more efficiently design concrete mix proportions in engineering applications to meet the requirements of frost resistance (in terms of the relative dynamic modulus of elasticity after freezing and thawing cycles), an intelligent framework based on the random forest (RF) algorithm is developed to predict the frost resistance of concrete. First, orthogonal tests and engineering data are used to obtain a dataset of raw material mix proportions for concrete and the corresponding frost resistance index values. Second, the RF and recursive feature elimination (RFE) are combined to eliminate coupling factors and noise and determine the optimal factor index system for the concrete mix proportion considering the corresponding influence on frost resistance, and the RF is used to establish a regression prediction function between the concrete mix proportion and frost resistance. Finally, the State Key Project of the Songtong Expressway in Jilin Province is used as a case study to verify the feasibility of the proposed method in a cold region, and the results show that (1) RF-RFE can effectively screen important indicators. After screening, the characteristic parameters are the water binder ratio, cement dosage, coarse aggregate dosage, fine aggregate dosage, compound superplasticizer amount and fly ash dosage. (2) The frost resistance of concrete can be accurately predicted by the RF model. The root mean square error ( RMSE ) of the test set is 0.077, and the goodness of fit ( R 2 ) is 0.9578. The proposed framework can be used to effectively predict the frost resistance of concrete to improve its durability, thus providing a good foundation for the optimization of the concrete mix design in practical engineering and excellent economic benefits and application prospects.