基于大数据的力学性能预测与工艺参数筛选

发布时间：2018-07-21 12:43

【摘要】：神经网络数据样本的质量影响着模型的预测精度;选择有效的工艺参数进行建模可以提高模型的训练速度,节省训练时间。针对数据样本质量和输入工艺参数的选择问题进行了研究,采用了计算马氏距离的方法剔除异常点,改善数据样本的质量。基于采集到的热轧汽车大梁板的17个工艺参数的生产数据,采用贝叶斯神经网络建立力学性能预测模型。通过采用平均影响值筛选出对力学性能影响较大的工艺参数进行建模,以简化模型。结果显示:简化后的模型取得了较高的预测精度,对于抗拉强度和屈服强度,分别有96.64%和94.96%的数据预测值和实际值相对误差在±6%以内;对于伸长率,有96.64%的数据预测值和实际值的绝对误差在±4%以内。
[Abstract]:The quality of neural network data samples affects the prediction accuracy of the model, and the selection of effective process parameters for modeling can improve the training speed of the model and save the training time. The selection of data sample quality and input process parameters is studied. The method of calculating Markov distance is used to eliminate outliers and improve the quality of data samples. Based on the production data of 17 process parameters of hot rolled automobile beam and plate, a prediction model of mechanical properties is established by using Bayesian neural network (Bayesian neural network). In order to simplify the model, the process parameters which have great influence on mechanical properties are selected by using the average influence value. The results show that the simplified model has a high prediction accuracy. For tensile strength and yield strength, the relative error between the predicted and actual values is within 卤6% for tensile strength and yield strength, respectively, while for elongation, the relative error is within 卤6% for tensile strength and yield strength. The absolute error between the predicted value and the actual value of 96.64% data is within 卤4%.
【作者单位】：东北大学轧制及连轧自动化国家重点实验室;鞍钢股份有限公司第三炼钢连轧厂;
【基金】：钢铁联合基金重点资助项目(U1460204) 辽宁省自然科学基金资助项目(2015020180) 中央高校基本科研业务费专项资金资助项目(N140704002)
【分类号】：TG335.11

【相似文献】