热连轧钢材力学性能在线预报的研究与应用

 传统的产品力学性能检测是一种建立在统计学随机抽样理论基础上的检验方法,即在实验室中对取样板卷的头尾部切割样品进行检测,检测结果代表整批产品的力学性能。由于钢材生产流程长,生产过程控制参数存在一定的波动,传统力学性能检测方法不能反应每一卷带钢的力学性能,所检测样品的代表性不够充分。随着工业互联网、大数据和人工智能技术的飞速发展,特别是工业大数据相关技术的发展和应用,为这一问题的解决提供了新的途径。以实现山东钢铁集团日照有限公司热连轧产品力学性能在线预报为试验对象,以热连轧产品生产全流程关键控制工艺参数为基础,采用神经元网络、随机森林等算法建立碳素结构钢、低合金高强度结构钢的力学性能预报模型,构建了一种基于工业大数据为基础的热轧产品力学性能预报系统,包括数据采集、数据清洗、模型训练、结果分析、再现性试验和在线应用。力学性能在线预报系统已成功运行2年多时间,系统的预测精度高、稳定可靠。预测结果精度在±6%以内的比例达到90%以上,MAPE(平均绝对百分误差)不大于4%,均低于再现性检测水平,预测结果完全可以取代检测试验;提高了生产效率,缩短了产品的检测周期,轧后即可掌握产品的力学性能,降低了生产成本,已成为生产运行过程不可缺少的环节。

Research and application of on-line steel mechanical property prediction in a hot strip mill

The traditional testing method of mechanical properties of steel products in a hot strip mill is an experimental method based on the statistical random sampling theory. The test samples are cut from the tail end of a hot coil and then tested in a laboratory. Test results represent the performance of the whole batch of steel products. Due to the long steel production process and certain fluctuations in the control parameters of the production process, the traditional mechanical property testing method cannot reflect the mechanical properties of each coil of strip steel, and the representativeness of the tested samples is not sufficient. With the rapid development of industrial Internet, big data and artificial intelligence technologies, especially the development and application of industrial big data-related technologies, new approaches are provided for solving this problem. Taking the realization of on-line steel mechanic properties prediction (On-line MPP)of hot tandem rolling products of Shandong Iron and Steel Group Rizhao Co., Ltd. as the test object, and based on the key control process parameter data of the whole process of the hot rolling products, the neural network, random forest and other algorithms are used to establish carbon structural and low-alloy high-strength construction steel mechanical properties prediction model, built a mechanical properties prediction system for hot-rolled products based on industrial big data, including data acquisition, data cleaning, model training, result analysis and online application. The on-line MPP has put into application for over two years and it has demonstrated well precision,high stability and reliability. The prediction accuracy of the model is within ±6%, the sample volume reaches more than 90%, and the MAPE (average absolute percentage error) ≤4%, which is lower than the reproducibility detection level., it can replace sampling inspection. As a result, shorten product inspection cycle, the mechanical properties of the product can be grasped after rolling. The system has become an important part in the production and operation process.