Artificial Intelligence - Insights into the Mechanics of Biomaterials: Predicting the Compressive Load of Composite Sandwich Structures

Environmentally friendly sandwich structures made of composite materials are the key to sustainable innovation. They offer lightweight, durable solutions that reduce emissions in numerous industries. Machine learning (ML) streamlines the prediction of the mechanical behavior of materials, drastically reducing both the time and cost of material design and development. This study investigates the mechanical behavior of novel environmentally friendly composite sandwich structures under quasi-static out-of-plane compression loading using ML algorithms. The research focuses on assessing how the geometric dimensions affect the strength of these panels when used as load-bearing structures. Random forest (RF) and ridge regression methods were employed to assess the linearity and nonlinearity relationships within a database of 42,280 data points, focusing on the compression load as the target variable. The results show for the test data set that the RF model has a higher accuracy with an R-squared value of 0.9990, root mean square error of 0.1532 and a mean absolute error of 0.7030, which underlines the non-linearity of the relationship. Conversely, linear regression was found to be unable to effectively capture complexity, which is reflected in its lower performance. Furthermore, the uncertainty analysis shows that the RF model has a minimal bias and higher precision and clearly outperforms the ridge regression in terms of predictive accuracy and consistency. The analysis of the importance of the feature contribution reveals that the number of cupules has the greatest influence and the mechanical properties of the balsa wood have the least influence on the compressive strength of the composite panels. This study highlights the effectiveness of ML in predicting the mechanical behavior of environmentally friendly composite materials and structures, emphasizing the crucial role of geometric parameters and material properties in improving prediction accuracy and reliability.