基于L-M算法网络的卫生瓷干坯强度的预测

Dry Strength Prediction of Sanitary Ceramics by an L-M Algorithm Artificial Neural Network

采用Levenberg-Marquardt算法对BP神经网络进行改进,对获得的实验数据进行训练,首次建立了卫生陶瓷凝胶注模成型工艺中单体、引发剂、交联剂、催化剂含量和坯体干燥强度的映射网络模型,从而可利用该模型来预测在一定的有机成型添加剂含量下卫生陶瓷的干燥强度。结果表明,其预测平均误差小于0.6%,说明神经网络用于卫生陶瓷凝胶注模成型后性能的预测是完全可行的。

A prediction model for dry strength of ceramic sanitaryware shaped by gelcasting based on improved BP artificial neural net was developed by using the Levenberg-Marquardt training algorithm. The nonlinear relationship in monomer content, initiator content, catalyzer content, crosslinker content and dry strength was established. Dry strength performance of ceramic sanitaryware could be predicted by means of the trained neural net from the giving datum. The rcsults showed that the average prediction error of dry strength was low than 0.6 % ; and the as-established model was suitable for the dry strength prediction of ceramic sanitarywarc.