Vasić Milica V, Awoyera Paul O, Onyelowe Keneddy C, Mydin Md Azree Othuman, Barišić Ivana, Grubeša Ivanka Netinger
Centre for Materials, Institute for Testing of Materials, Beograd, Serbia.
Department of Civil Engineering, Prince Mohammad Bin Fahd University, Al Khobar, Saudi Arabia.
Sci Prog. 2025 Apr-Jun;108(2):368504251348050. doi: 10.1177/00368504251348050. Epub 2025 May 30.
The shaping and drying of ceramics are a critical yet complex process that directly influences ceramic materials' final properties and performance. Predicting key parameters such as the coefficient of plasticity, mass loss during drying in the air at the critical point, and shaping moisture is essential for optimizing these processes. This study analyzes the dataset of the clays of various chemical compositions to predict and reveal the most important influences on the shaping and drying parameters in producing ceramic tiles. The data are then employed to develop and compare four advanced machine learning models. The models were evaluated using the most important performance metrics such as the coefficient of determination (²), mean absolute percentage error (MAPE), mean absolute error (MAE), and root mean squared error (RMSE). Extreme Gradient Boosting (Gradient Boosting) emerged as the most reliable model, with 0.9871 R², 0.2672 RMSE, 0.2086 MAE, and 1.61% MAPE. Support vector regression and artificial neural networks also delivered strong performances, while random forest, though competitive, was slightly less accurate. Furthermore, model interpretation methods in machine learning analysis provided valuable validation of the predictive capabilities of the models and the influence of key input features. The advanced machine learning techniques in optimizing ceramic shaping processes offer a robust predictive toolkit for enhancing efficiency, reliability, and sustainability in ceramic materials engineering. It is seen that the AlO levels up to 23% had little effect on plasticity and drying susceptibility, with significant changes occurring above 28%. The critical FeO content is found between 1.5% and 1.7%, and SiO of up to about 62%. The findings of this study offer valuable decision-support tools for ceramic manufacturers, raw material suppliers, and process engineers, enabling more informed material selection, reduced waste, and improved product consistency across the industry.
陶瓷的成型和干燥是一个关键但复杂的过程，直接影响陶瓷材料的最终性能和表现。预测诸如可塑性系数、临界点在空气中干燥时的质量损失以及成型水分等关键参数，对于优化这些过程至关重要。本研究分析了各种化学成分的黏土数据集，以预测并揭示对生产瓷砖的成型和干燥参数最重要的影响因素。然后使用这些数据来开发和比较四种先进的机器学习模型。使用诸如决定系数（R²）、平均绝对百分比误差（MAPE）、平均绝对误差（MAE）和均方根误差（RMSE）等最重要的性能指标对模型进行评估。极端梯度提升（梯度提升）成为最可靠的模型，R²为0.9871，RMSE为0.2672，MAE为0.2086，MAPE为1.61%。支持向量回归和人工神经网络也表现出色，而随机森林虽然具有竞争力，但准确性略低。此外，机器学习分析中的模型解释方法为模型的预测能力和关键输入特征的影响提供了有价值的验证。优化陶瓷成型过程中的先进机器学习技术为提高陶瓷材料工程的效率、可靠性和可持续性提供了一个强大的预测工具包。可以看出，AlO含量高达23%时对可塑性和干燥敏感性影响不大，而在28%以上会发生显著变化。临界FeO含量在1.5%至1.7%之间，SiO含量高达约62%。本研究的结果为陶瓷制造商、原材料供应商和工艺工程师提供了有价值的决策支持工具，有助于在整个行业中做出更明智的材料选择、减少浪费并提高产品一致性。
