热轧带钢力学性能在线预测模型的开发与应用

热轧带钢力学性能在线预测技术能够优化生产工艺、改善成品质量。为此,在传统数学模型的基础上,采用Bayes神经网络建立了热轧带钢力学性能在线预测新模型。介绍了基于Bayes理论方法的神经网络、数据预处理方法、数据平台的搭建,力学性能在线预测模型输入参数的选择,以及基于某1 780 mm热轧带钢生产线,对以SPA-H、510L、610L为代表的典型钢种带钢力学性能的预测精度。结果表明,屈服强度预测相对误差在±6%范围内,抗拉强度预测相对误差在±6%范围内,伸长率预测相对误差在±4%范围内。在线预测系统的开发与应用不仅能优化生产工艺,而且还能减少成品力学性能的检测样本量,降低了生产成本。

Development and application of on-line prediction model for mechanical properties of hot rolled strips

The production process can be ptimized,and the products can be improved by the on-line prediction technology of mechanical properties of hot rolled strip.Based on the traditional mathematical model, a new on-line prediction model for the mechanical properties of hot rolled strip was established by using Bayes neural network.The neural network based on Bayes theory method, the data preprocessing method, the data platform construction, the selection of the input parameters of the on-line prediction model of mechanical properties were introduced as well as the prediction accuracy of the mechanical properties of typical steel strips, such as SPA-H, 510L and 610L, based on a 1 780 mm hot rolled strip production line. The results show that the relative error of yield strength prediction is in the range of ±6%, the relative error of tensile strength prediction is in the range of ±6%, and the relative error of elongation prediction is in the range of ±4%. The development and application of the on-line prediction system can not only optimize the production process, but also reduce the sample amounts for measuring mechanical properties, and reduce the production cost.