Machine Learning-Based Predictive Model for Tensile and Flexural Strength of 3D-Printed Concrete

Abstract

The additive manufacturing of concrete, also known as 3D-printed concrete, is produced layer by layer using a 3D printer. The three-dimensional printing of concrete offers several benefits compared to conventional concrete construction, such as reduced labor costs and wastage of materials. It can also be used to build complex structures with high precision and accuracy. However, optimizing the mix design of 3D-printed concrete is challenging, involving numerous factors and extensive hit-and-trail experimentation. This study addresses this issue by developing predictive models, such as the Gaussian Process Regression model, Decision Tree Regression model, Support Vector Machine model, and XGBoost Regression models. The input parameters were water (Kg/m3), cement (Kg/m3), silica fume (Kg/m3), fly ash (Kg/m3), coarse aggregate (Kg/m3 & mm for diameter), fine aggregate (Kg/m3 & mm for diameter), viscosity modifying agent (Kg/m3), fibers (Kg/m3), fiber properties (mm for diameter and MPa for strength), print speed (mm/sec), and nozzle area (mm2), while target properties were the flexural and tensile strength of concrete (MPa data from 25 literature studies were collected. The water/binder ratio used in the dataset ranged from 0.27 to 0.67. Different types of sands and fibers have been used, with fibers having a maximum length of 23 mm. Based upon the Coefficient of Determination (R2), Root Mean Square Error (RMSE), Mean Square Error (MSE), and Mean Absolute Error (MAE) for casted and printed concrete, the SVM model performed better than other models. All models’ cast and printed flexural strength values were also correlated. The model’s performance has also been checked on six different mix proportions from the dataset to show its accuracy. It is worth noting that the lack of ML-based predictive models for the flexural and tensile properties of 3D-printed concrete in the literature makes this study a novel innovation in the field. This model could reduce the computational and experimental effort required to formulate the mixed design of printed concrete.

Development of Prediction Method for Dimensional Stability of 3D-Printed Objects

“…Despite the utilization of a limited number of datapoints (81 datapoints), the regression models exhibited R 2 values exceeding 90%, indicating high accuracy. Compared to the recently research [29], R 2 values for GPR and SVM were higher: 12.99-18.58% or 3.23-10.00% higher than their study, respectively. Furthermore, it can be inferred that increasing the number of datapoints would lead to higher accuracies in the regression models.…”

Material Extrusion Filament Width and Height Prediction via Design of Experiment and Machine Learning

“…We then made some comparisons to previously reported studies. Ali et al [32] predicted the tensile strength and flexural strength of 3D-printed concrete materials by four machine learning methods, with R 2 values ranging from 0.7253 to 0.8785. Wang et al [30] predicted the drug loading efficiency of 3D-printed drugs in relation to material and process parameters, with R 2 values ranging from 0.678 to 0.93.…”

“…Machine learning constitutes yet another avenue for exploring the intricate nexus between the dimensions of 3D printing structures and their process parameters. It is a branch of artificial intelligence that focuses on developing statistical models and algorithms, enabling computers to learn adaptively from existing data and evolve without hard coding [32]. Machine learning has been successfully applied in various fields such as healthcare, energy, materials, and manufacturing [33][34][35][36].…”

“…Given the divergence in model performance, scholars often undertake comparisons across various models while addressing the same problem and employing identical datasets. Ali et al [32] selected Gaussian process regression models, DT regression models, SVM models, and XGBoost regression models to predict the tensile, compressive, and flexural strength of 3D-printed concrete materials, and their SVM model demonstrated the most favorable performance. Similarly, Wang et al [30] studied the drug loading efficiency of 3D-printed drugs using five methods: DT, RF, KNN, XGBoost, and a light gradient boosting machine (LightGBM), with the DT model emerging as the top performer.…”

Advancing Construction 3D Printing with Predictive Interlayer Bonding Strength: A Stacking Model Paradigm

“…This entails evaluating the printing outcomes in advance, thereby assisting in avoiding unnecessary physical experiments. However, current research on predicting the performance of 3D printing concrete using ML algorithms is relatively limited, with only a few scholars attempting to predict parameters such as compressive strength [ 17 ], flexural strength [ 18 ], tensile strength [ 18 ], rheological properties [ 19 ], printability [ 20 ], etc., while no one has yet used ML methods to predict IBS. IBS is a critical parameter determining the quality of 3D printing concrete, highlighting the necessity of its advance prediction.…”