Abstract: Sintering process is a key process affecting the quality of piezoelectric ceramics,which involves many factors and has the characteristics of nonlinear and hysteresis,leading to the difficulty in ensuring the quality of the finished products.To solve this problem,two indirect quality indexes,average grain size and sintering density were proposed by analyzing the change of ceramic microstructure during sintering process,and the relationship between the two indexes and the piezoelectric properties was analyzed.The quality prediction model was established to realize the quality prediction and control of sintering process.By ensemble learning CatBoost algorithm and Byesian Optimization Hyperband (BOHB) hyperparameter optimization algorithm,the BOHB-CatBoost quality prediction model was established.Finally,the performance of the model was evaluated by combining RMSE and R2,and compared with other prediction models.It was verified that the model had higher prediction accuracy and robustness,which could guide for the sintering process of piezoelectric ceramics significantly.

Key words: piezoelectric ceramics, quality prediction, Byesian optimization hyperband hyperparameter optimization algorithm, CatBoost algorithm

摘要： 烧结工艺是影响压电陶瓷成品质量的关键工艺,涉及影响因素众多,具有非线性、滞后性的特点,导致烧成品的质量难以保证。针对这一难题,通过分析烧结过程中陶瓷微观结构的变化,提出平均晶粒尺寸和烧成密度两个间接质量指标,并与压电性能指标间的关系进行分析,建立质量预测模型,实现对烧结工艺的质量预测及控制。通过采取集成学习CatBoost算法,并结合贝叶斯超频带(BOHB)超参数优化算法,以五折交叉验证的方式建立了BOHB-CatBoost质量预测模型。最后,结合RMSE和R2两个指标评估模型的性能,并与其他预测模型进行对比,验证了该模型具有更高的预测精度以及稳健性,对压电陶瓷的烧结生产过程具有较好的指导意义。

关键词: 压电陶瓷, 质量预测, 贝叶斯超频带超参数优化算法, CatBoost算法