【作者】 邢邦圣；

【导师】 杜长龙；

【作者基本信息】 中国矿业大学 ， 机械设计及理论， 2013， 博士

【摘要】 冷轧带肋钢筋因其优良的综合机械性能和粘结锚固性能，在建筑领域得到了日益广泛的应用，带来了明显的社会效益和经济效益。随着冷轧带肋钢筋产品应用层次的不断发展和应用领域的进一步拓宽，对产品机械性能和工艺水平提出了更高的要求。针对目前冷轧带肋钢筋生产中存在的产品机械性能不稳定、合格率低等问题，研究了冷轧带肋钢筋产品机械性能智能预测方法和产品冷轧工艺优化模型，为冷轧工艺规划和产品质量控制提供一快速、精确、经济的新途径。针对冷轧带肋钢筋产品机械性能和冷轧工艺参数间的物理关系极为复杂，难以直接建立两者之间的显式方程，研究了样本空间划分对于产品机械性能预测的意义，提出了基于原材料初始强度和工艺参数向量间距离划分样本空间的方法，为实现在较少数量训练样本前提下产品机械性能的智能预测奠定了基础。建立了冷轧工艺参数和产品性能参数间的线性直接映射预测模型、线性和非线性回归分析预测模型，并在各样本子空间和全空间范围内对产品机械性能进行了预测。结果表明，对于线性直接映射模型和线性回归分析预测模型，工艺参数向量的降维处理对提高其预测性能具有积极意义；而对于非线性回归分析预测模型，足够数量的实测样本数据对保证其预测精度更具积极意义。构建了基于BP神经网络的冷轧带肋钢筋机械性能预测模型，研究了隐含层节点数和网络误差等参数对产品性能预测精度的影响，并在各样本子空间和全样本空间范围内，对冷轧带肋钢筋产品机械性能进行了预测。结果表明，BP网络的合理结构、隐含层节点数和网络误差合理取值等因素，对保证冷轧带肋钢筋性能预测模型的训练效率、预测精度等具有重要意义。建立了基于径向基函数网络的冷轧带肋钢筋机械性能预测模型，研究了RBF神经元层宽度系数和网络误差对RBF网络逼近性能的影响，并在各样本子空间和全样本空间内对产品机械性能进行了预测。结果表明，在神经元层宽度系数和网络误差等取值合适的情况下，基于RBF网络的冷轧带肋钢筋性能预测模型具有较高的预测精度；按照工艺参数间距离划分样本空间，对于基于RBF网络的预测模型更具有积极意义。针对工艺实验法规划冷轧工艺成本高、周期长等缺点，研究了冷轧带肋钢筋加工工艺的多目标优化模型，提出了基于遗传算法和径向基函数网络的冷轧带肋钢筋加工工艺优化方法。利用遗传算法的全局搜索能力和径向基函数网络的高精度逼近性能，快速、准确地确定满足产品机械性能要求的工艺参数优化组合，为制定和优化冷轧带肋钢筋生产工艺提供了一条经济、有效的途径。更多还原

【Abstract】 Cold rolled ribbed steel bars have been increasingly applied in the constructionfield, because of their excellent comprehensive mechanical performance andanchoring bond performance and bring about obvious social and economic benefits.With the further development of the application level and application domain of theproducts, higher requirements on the mechanical performance and the technologicallevel are put forward, which needs precision prediction of the product performanceand optimization of the processing technology.Aiming at the instability of mechanical performance and low pass rate in coldrolled ribbed steel bars production, the intelligent prediction method for the product’smechanical performance and the optimization model of cold rolling process areresearched, so as to provide a fast, accurate and economical new approach for coldrolling process planning and product quality control.Aiming at the complexity of the physical relations between cold rolled ribbedsteel bars’ mechanical performance and cold rolling technological parameters, and thedifficulty of establishing the explicit equations between the two, the meaning ofdividing the variable space for economical prediction of the product performance isresearched in this thesis. The methods of dividing the space according to the initialstrength of the raw materials and the distance between the vectors of the processparameters are proposed, laying the foundation for predicting the product performanceintelligently and accurately with small number of training samples.The prediction models based on linear direct mapping, linear and nonlinearregression analysis are established, and the product performance are predicted withthe models in the subspaces and the total space of the samples. Test results show thatdecrease on the dimension of the technical variable vectors is of positive significanceto the prediction precision for the linear direct mapping model and the linearprogramming analysis one, and that a sufficient number of precision sample data is sofor the nonlinear programming analysis model.The prediction models based on BP neural network are established, and theinfluence of the hidden neurons number and the network error on the productperformance prediction is researched. And then, the product performances arepredicted in the subspaces and the total space of the samples. Test results show thatproper structure of the network, appropriate number of the hidden neurons and right value of the network error are of significance to the training efficiency and theprediction precision for BP network model.The prediction models based on radial-basis function network are established,and the influence of the spread coefficient and the network error on the predictionprecision is researched. And then, the product performances are predicted in thesubspaces and the total space of the samples. Test results show that RBF network withproper spread coefficient and network error is superior to the other prediction modelsresearched in the thesis; and that dividing the sample space according to the distancebetween the vectors of the process parameters is of positive significance for RBFnetwork.Aiming at the high cost and low efficiency of the experimentation method forcold rolling process planning, the multi-objective optimization model of theprocessing technics of cold rolled ribbed steel bars is established, and the method ofoptimizing the processing technics with genetic algorithm and RBF network isproposed. It has been shown that the method can quickly determine the propertechnological parameters meeting the property requirement of the product, whichprovides an economical and effective way to make and optimize the processingtechnology.更多还原