Additive manufacturing of earthen materials reinforced with natural fibers is gaining attention as a sustainable approach for fabricating soil-based structures with enhanced performance. In this study, an explainable machine learning model is developed to predict the printing height in direct ink writing of wood particles reinforced clay composite. The studied factors included the ink composition (wood particles (0, 5, and 10 wt.%), water content (50, 55, 60, 65, 70 wt.%)) and the printing parameters (extrusion rate (16.161 mm3/s having 6mm nozzle diameter and 3.865 mm3/s having 3 mm nozzle diameter), and printing speed (30, 60, and 90 mm/s)). The role of factors governing buildability is investigated, showing that the printing height is a factor of both the ink composition and printing parameters. It shows the interaction between ink and printing parameters and explains how such an interaction impacts the collapse mechanism (i.e., slumping, buckling, tilting) of freshly-printed composite. Analysis of variance (ANOVA) and machine learning modeling (decision tree and TreeNet gradient boosting machine) are used to predict the printing height and relative importance of the governing factors. The results showed that the addition of 10wt.% of wood particles to the clay matrix increased the buildability by 28%. Increasing the extrusion rate also enhanced the printing height by 78.2%. Also, increasing the water content from 50wt% to 70wt% reduces the printing height by ~ 76%. Overall, an increase in printing height was observed at lower water content and printing speed, but larger extrusion rate (i.e., larger nozzle diameter and layer height). The impact of wood particles on the printing height is nonlinear and significantly interacts with the water content and extrusion rate. Wood particle reinforcement improves the interlaminar shear strength of printed layers and positively addresses billowing formation. This offers an opportunity to support the circular economy using wood residues and sawdust in sustainable manufacturing using earthen materials. Machine learning models could successfully predict the printing height with TreeNet model (R2test of 94.69%) outperformed the decision tree (R2test of 84.89%). Using an explainable machine learning model such as decision tree offers insight into the optimal design of ink composition and printing parameters for desired buildability.

Predicting printing height in direct ink writing of wood-clay composites using explainable machine learning

期刊论文
发表日期：

2025年8月15日

查看全文

Biva Gyawali,
Kai Bentley,
Abbas Hosseini,
Ramtin Haghnazar,
Devin Roach,
Pavan Akula,
Kamran Alba,
Vahid Nasir

摘要

Additive manufacturing of earthen materials reinforced with natural fibers is gaining attention as a sustainable approach for fabricating soil-based structures with enhanced performance. In this study, an explainable machine learning model is developed to predict the printing height in direct ink writing of wood particles reinforced clay composite. The studied factors included the ink composition (wood particles (0, 5, and 10 wt.%), water content (50, 55, 60, 65, 70 wt.%)) and the printing parameters (extrusion rate (16.161 mm³/s having 6mm nozzle diameter and 3.865 mm³/s having 3 mm nozzle diameter), and printing speed (30, 60, and 90 mm/s)). The role of factors governing buildability is investigated, showing that the printing height is a factor of both the ink composition and printing parameters. It shows the interaction between ink and printing parameters and explains how such an interaction impacts the collapse mechanism (i.e., slumping, buckling, tilting) of freshly-printed composite. Analysis of variance (ANOVA) and machine learning modeling (decision tree and TreeNet gradient boosting machine) are used to predict the printing height and relative importance of the governing factors. The results showed that the addition of 10wt.% of wood particles to the clay matrix increased the buildability by 28%. Increasing the extrusion rate also enhanced the printing height by 78.2%. Also, increasing the water content from 50wt% to 70wt% reduces the printing height by ~ 76%. Overall, an increase in printing height was observed at lower water content and printing speed, but larger extrusion rate (i.e., larger nozzle diameter and layer height). The impact of wood particles on the printing height is nonlinear and significantly interacts with the water content and extrusion rate. Wood particle reinforcement improves the interlaminar shear strength of printed layers and positively addresses billowing formation. This offers an opportunity to support the circular economy using wood residues and sawdust in sustainable manufacturing using earthen materials. Machine learning models could successfully predict the printing height with TreeNet model (R²_test of 94.69%) outperformed the decision tree (R²_test of 84.89%). Using an explainable machine learning model such as decision tree offers insight into the optimal design of ink composition and printing parameters for desired buildability.