热轧带材多缺陷和单缺陷表面质量综合预报

Comprehensive prediction of surface quality of hot-rolled strip with multiple and single defects

针对目前热轧带材表面质量缺陷类型多、产生机理不同,不能实现在线诊断的问题,提出融合各降维层深度置信网络模型的热轧带材多缺陷和单一缺陷表面质量问题综合预报技术。考虑传统深度置信网络在模型层数和节点数方面的不足,提出并建立了融合各降维层的深度置信网络模型,降低由于网络算法节点个数及层数不确定性造成的预报误差,并利用相关性将经典深度置信网络预报结果和各降维层的预报结果融合为唯一预报值。针对热轧带材生产过程表面质量多缺陷问题,考虑不同表面质量缺陷影响因素不同,建立了热轧带材表面质量多缺陷和单缺陷综合诊断策略模型。将所建模型应用到某热轧厂带材表面翘皮、铁皮印和毛刺等多缺陷预报,结果表明,该模型缺陷预报准确度达80%以上,这对热轧带材多缺陷表面质量具有良好的诊断效果。

In view of the problem that the surface quality defects of the hot-rolled strip are serious,have many types and different mechanisms,and cannot be diagnosed online.A method for the comprehensive diagnosis of surface quality problems of the hot-rolled strip with multiple defects and single defects was presented by combining the deep belief network model of each dimension reduction layer.Considering the traditional deep belief network model in the shortfalls,layer number and the number of nodes were put forward and set up the integration of each dimension reduction layer belief network model,nodes were reduced due to network algorithm and the layer number of forecasting error caused by uncertainty,and used the correlation between classic deep belief network forecasting results,and the forecasting results of dimension reduction layer integration was the only anticipation value.In view of the problem of surface quality multiple defects in the hot-rolled strip production,considering the different input parameters of different surface quality defects,a comprehensive diagnosis strategy for surface quality multiple defects and single defects of the hot-rolled strip was established.In a typical surface become warped in the process of hot rolling production line production leather,metal surface quality defects such as seal and burr problem,application integration of each dimension reduction deep belief network model and multiple flaws and single comprehensive diagnosis strategy,the results show that the surface quality defects of the model prediction accuracy were above 80%,for the hot-rolled strip surface defects with good diagnosis effect.