Sharma Anmol, Saini Ravinder S, Kaushik Ashish, Okshah Abdulmajeed, Kuruniyan Mohamed Saheer, Gurumurthy Vishwanath, Vyas Rajesh, Binduhayyim Rayan Ibrahim H, Heboyan Artak
USICT, Guru Gobind Singh Indraprastha University, Sector 16C, Dwarka, Delhi, India.
Department of Allied Dental Health Sciences COAMS, King Khalid University, Abha, Saudi Arabia.
Sci Rep. 2025 Sep 1;15(1):32239. doi: 10.1038/s41598-025-17487-z.
Complex geometries achievable with resin-based 3D printing are susceptible to lower levels of surface roughness, particularly in areas where support structures are attached and removed. The slicing parameter serves as the cornerstone for developing a model for predicting the corresponding output. In the present research, a resin 3D printer is used to fabricate the specimens in accordance with the combination of important parameters that were recovered utilizing the design of the experiment (DoE). Layer thickness, infill population density, print angle, exposure time, and lift speed are the five factors used to build DoE, which consists of 32 ideal runs for assessing surface roughness (SR). SR is a critical factor that influences the durability and effectiveness of dental devices. In order to anticipate output, a model is developed. To choose the best modelling strategies, a comparison of three base model techniques, artificial neural networks (ANN), support-vector regression (SVR), and decision trees (DT), as well as two ensemble techniques, random forest (RF) and XGboost, is conducted. This work utilized hyperparameter tuning for model improvement and use RMSE and R as performance metrices for model efficiency. This application is still relatively underrepresented in the literature and often with isolated ML models rather than hybrid approaches. Among the three base models, SVR performs best with R and RMSE 0.96745 and 0.017974 at C = 5 and gamma = 1 resp. Ensemble techniques justifying clubbing perform better in all ways with XGboost showing R 0.99858 and RMSE 0.00346998 as the best among all techniques. This work helps dental professionals in utilizing ensemble ML to improve model efficiency and predictability.
基于树脂的3D打印所能实现的复杂几何形状容易出现较低水平的表面粗糙度，尤其是在支撑结构附着和移除的区域。切片参数是开发预测相应输出模型的基石。在本研究中，使用树脂3D打印机根据利用实验设计（DoE）恢复的重要参数组合来制造样本。层厚、填充体密度、打印角度、曝光时间和提升速度是用于构建DoE的五个因素，DoE由32次理想运行组成，用于评估表面粗糙度（SR）。SR是影响牙科器械耐用性和有效性的关键因素。为了预测输出，开发了一个模型。为了选择最佳建模策略，对三种基本模型技术（人工神经网络（ANN）、支持向量回归（SVR）和决策树（DT））以及两种集成技术（随机森林（RF）和XGboost）进行了比较。这项工作利用超参数调整来改进模型，并使用均方根误差（RMSE）和相关系数（R）作为模型效率的性能指标。该应用在文献中的代表性仍然相对不足，并且通常使用孤立的机器学习模型而非混合方法。在三种基本模型中，SVR表现最佳，在C = 5和gamma = 1时，R为0.96745，RMSE为0.017974。证明合并合理的集成技术在各方面表现更好，XGboost显示R为0.99858，RMSE为0.00346998，是所有技术中最好的。这项工作有助于牙科专业人员利用集成机器学习来提高模型效率和可预测性。
