Study On Prediction Of Compressive Strength Of Alkali Activated Slag Concrete After High Temperature

Conventional concrete uses silicate cement as the cementitious material,and its production process is energy intensive and polluting to the environment.The Asia Pacific Green Low Carbon Development Summit released the Emerging Economies CO₂Emissions Report 2021,which sets the goal of working towards"carbon neutrality"by2060.To further achieve this goal,the search for new sustainable cementitious materials that can replace silicate cement has become a hot topic of research.Alkali activated slag（AAS）cementitious materials are regarded as one of the most promising alternative materials due to their energy efficiency,waste efficiency,environmental friendliness and excellent mechanical properties.After decades of development in China,AAS concrete has has gradually overcome the problems of fast hardening and shrinkage.Concrete prepared from it has also found initial application in structural engineering.AAS concrete has now been successfully used in cast-in-place reinforced concrete structural projects,which proves the promise of AAS concrete engineering applications.In addition,the frequency of building fires poses a serious threat to the structural safety of concrete.Scholars at home and abroad have realized the importance of the high temperature resistance of concrete and launched research.The high temperature properties of silicate cement,especially the mechanical properties have been more comprehensive research and understanding.However,research related to AAS cementitious materials is still at a preliminary stage.In addition,the factors influencing the residual mechanical properties of AAS concrete after high temperature are complex and non-linear,making the tests of residual mechanical properties very difficult and even causing secondary damage to the structure.Therefore,it is of great scientific and practical significance to provide a scientific method for predicting the residual mechanical properties after high temperature by means of mathematical modelling to enrich the theoretical system of this material and to extend its engineering applications.The main studies in this paper are as follows.（1）In the AAS concrete system,a prediction model for the compressive strength of AAS concrete at different ages is developed based on BP neural networks,PSO-BP neural networks,alternating conditional expectation（ACE）and random forest（RF）,considering the influence of material factors on its initial compressive strength.The models are evaluated by means of the evaluation function and validated experimentally to determine the better model.In addition,the weights of the influencing factors are analysed based on the ACE and RF models to provide a reference for better predict of the initial strength of AAS concrete.（2）In the case of normal concrete,a prediction model for the loss of compressive strength of normal concrete after high temperature was developed based on four methods,considering the influence of material factors and high temperature operation mechanism on its residual compressive strength after high temperature.The better model was also determined through the evaluation function and experimental validation.The weight analysis was also carried out based on ACE and RF model to provide a reference for the prediction study of after high temperature compressive strength loss of concrete.（3）With reference to the factors influencing the compressive strength of AAS concrete and the loss of compressive strength of normal concrete after high temperature,the effects of material factors,high temperature operation mechanism and curing conditions on the loss of compressive strength of AAS concrete after high temperature were considered.The same four models established were evaluated using root mean square error（RMSE）and mean absolute error（MAE）,and the better prediction model for after temperature compressive strength loss of AAS concrete was finally determined through experimental validation.In addition,their influences were weighted by ACE images as well as importance analysis of RF factors of influence.The following conclusions can be drawn from the research in this paper.（1）Alkali concentration,water-glass modulus,water-binder ratio,specific surface area of slag and alkalinity coefficient were used as input parameters for the model to output the 3d,28d and 90d compressive strengths of AAS concrete.The data used for model training were obtained from published literature,with 171,309 and 146 groups respectively.It was shown that the PSO-BP model exhibited higher accuracy.And the prediction model at 90d age showed greater stability.In addition,the weight analysis of the influencing factors based on ACE as well as the RF model concluded that the water-binder ratio and modulus had the most obvious influence,followed by alkali concentration,and the alkalinity coefficient and specific surface area of slag were the smallest.（2）The water-binder ratio,heating temperature,heating rate,maintaining duration at target temperature,resting duration after cooling,and cooling mode（water cooling/natural cooling）were used as input parameters.As the data for building the model originated from different laboratories,it was influenced by the experimental environment and other unset experimental parameters,which eventually led to some discrete situations.In order to solve the problem of large dispersion,a more stable rate of compressive strength loss after high temperature was chosen as the output parameter.A total of 1803 sets of data were obtained through the collection and collation of literature data.The RF model showed a lower error and a higher stability of the relative error with an average error of only 5.7%after experimental validation.Similarly,based on the Gini index,the effect of heating temperature on the loss of compressive strength of normal concrete after high temperature was 67.3%;the water-binder ratio was 11.6%,the heating rate was 10.3%;the maintaining duration at target temperature was 5.6%,the resting duration after cooling was 4.2%;and the cooling mode was 1.0%.（3）A total of nine influencing factors were selected as input parameters:alkali concentration,modulus,water-binder ratio,curing temperature,heating temperature,maintaining duration at target temperature,heating rate,resting duration after cooling and cooling mode（water cooling/natural cooling）.The rate of loss of compressive strength after high temperature was also used as the output.The model training data was obtained from published literature,with a total of 259 sets.The results show that all four models can produce a better fit superiority,indicating a certain reference significance for engineering practice.The models were ranked in order of accuracy as RF>PSO-BP neural network>BP neural network>ACE.According to the response surface image of the ACE model,it can be seen that the most important influencing factor on the rate of compressive strength loss after high temperature of AAS concrete is the heating temperature,followed by the alkali concentration.This is consistent with the importance ranking of RF:heating temperature weight is 60.2%,alkali concentration weight is13.5%,curing temperature weight is 10.0%,water-binder ratio weight is 5.5%,heating rate weight is 3.0%,water-glass modulus weight is 3.2%,maintaining duration at target temperature weight is 1.7%,resting duration after cooling is 0.5%,and cooling mode weight is 2.4%.