Tailorability is an important advantage of composites. Incorporating new bio-reinforcements into composites can contribute to using agricultural wastes and creating tougher and more reliable materials. Nevertheless, the huge number of possible natural material combinations works against finding optimal composite designs. Here, machine learning was employed to effectively predict fracture toughness properties of multiscale bio-nano-composites. Charpy impact tests were conducted on composites with various combinations of two new bio fillers, pistachio shell powders, and fractal date seed particles, as well as nano-clays and short latania fibers, all which reinforce a poly(propylene)/ethylene–propylene–diene-monomer matrix. The measured energy absorptions obtained were used to calculate strain energy release rates as a fracture toughness parameter using linear elastic fracture mechanics and finite element analysis approaches. Despite the limited number of training data obtained from these impact tests and finite element analysis, the machine learning results were accurate for prediction and optimal design. This study applied the decision tree regressor and adaptive boosting regressor machine learning methods in contrast to the K-nearest neighbor regressor machine learning approach used in our previous study for heat deflection temperature predictions. Scanning electron microscopy, optical microscopy, and transmission electron microscopy were used to study the nano-clay dispersion and impact fracture morphology.

中文翻译：

---

多尺度生物纳米复合材料断裂韧性的机器学习预测

可定制性是复合材料的一个重要优势。在复合材料中加入新的生物增强材料有助于利用农业废弃物并制造更坚固、更可靠的材料。然而，大量可能的天然材料组合不利于寻找最佳复合设计。在这里，机器学习被用来有效预测多尺度生物纳米复合材料的断裂韧性特性。对具有两种新生物填料、开心果壳粉和分形枣籽颗粒以及纳米粘土和短拉塔尼亚纤维的各种组合的复合材料进行了夏比冲击试验，所有这些都增强了聚（丙烯）/乙烯-丙烯-二烯单体矩阵。使用线弹性断裂力学和有限元分析方法，获得的测量能量吸收用于计算应变能释放速率作为断裂韧性参数。尽管从这些冲击测试和有限元分析中获得的训练数据数量有限，但机器学习结果对于预测和优化设计是准确的。与我们之前研究中用于热变形温度预测的 K 最近邻回归器机器学习方法相比，本研究应用了决策树回归器和自适应提升回归器机器学习方法。使用扫描电子显微镜、光学显微镜和透射电子显微镜研究纳米粘土分散和冲击断裂形貌。