A comparative study of neural network model and LOLIMOT for predicting superplasticiser dosage for self-compacting concrete containing supplementary cementitious materials

The use of superplasticiser (SP) has been vital to highly flowable concrete such as Self-Compacting Concrete (SCC). Advanced computing methods (Fuzzy Logic models and Neural Networks etc.) have been utilized to predict mechanical and durability parameters of SCC in the last decade. This research investigates the different approaches with Local Linear Model Tree (LOLIMOT) and Artificial Neural Network Model (ANN) to the prediction of the superplasticiser dosage needed to achieve satisfactory rheological characteristics of SCC. The variables and data considered for the prediction were obtained from 48 different SCC mixtures containing several types of Supplementary Cementitious Materials (SCMs): Metakaolin (MK), Ground Granulated Blast-furnace Slag (GGBS), Fly Ash (FA) and Waste paper Sludge Ash (WSA). A correlation analysis was used to identify the significance of each input variable then model input variables were selected for Local Linear Model Tree (LOLIMOT) and Artificial Neural Network Model (ANN). Several models were constructed to examine the effects of the various combinations of variable selections on the accuracy of output of models. The performance of models was evaluated by the required neurons for a specific error and the errors for the specific amount of neurons. The results show that the proposed LOLIMOT and ANN models can be used in the prediction of superplasticiser dosage of self-compacting concrete containing SCMs. Finally, the partitioning of input spaces in LOLIMOT models was extracted and a sensitivity analysis was performed with a trained LOLIMOT inference system.