基于代理模型的零件表面粗糙度加工参数优化

发布时间：2018-03-11 09:34

  本文选题：加工参数优化　切入点：响应面法　出处：《西南交通大学》2015年博士论文　论文类型：学位论文

【摘要】：在金属切削过程中,加工参数不仅影响了切削过程的生产效率,而且还影响了切削力和切削零件的表面质量。实际上,切削力和零件的表面质量是密切相关的,切削力也是影响表面质量的因素。所以加工参数的优化研究对改善切削过程中的切削力和加工表面质量有非常重要的意义。本文以切削力和表面粗糙度为直接对象,应用基于响应面法的代理模型和人工神经网络模型技术,对加工参数进行优化。本论文主要进行了以下几个方面的工作：(1)在建立切削力模型的基础上,分析研究了影响切削力的主要因素。在本论文的研究工作中,除了把切削速度、切削深度、进给量作为关键的加工参数外,还把铣刀的刀齿数引入作为一个重要加工参数。在分析基于响应面的加工粗糙度代理模型的基础上,通过实验设计确定了四个关键加工参数的范围和水平。(2)进行了基于响应面的粗糙度代理模型的实验研究和灵敏度分析。首先根据四个关键加工参数的水平,通过切削实验采集了有关加工参数和表面粗糙度的样本数据集,在此基础上建立了基于二次响应面法的加工参数和表面粗糙度的代理模型。其次,进一步通过代理模型的灵敏度分析,简化了模型中对粗糙度不明显的项。最后利用代理模型对加工参数进行优化,切削实验表明基于响应面法的表面粗糙度代理模型的误差在可接受的范围之内。(3)基于响应面法进行了影响切削力和表面粗糙度的加工参数优化研究。根据四个关键加工参数的水平,通过实验采集有关粗糙度和切削力数据的基础上,建立了表面粗糙度和切削力的线性和二阶响应面代理模型,并通过切削实验对代理模型进行了验证,确保了代理模型的可靠性。应用建立的代理模型,对加工参数进行了单目标和多目标加工参数优化。同样,切削实验表明基于代理模型,针对表面粗糙度和切削力的单、多目标优化是可靠的。(4)进行了基于人工神经网络模型的加工参数单目标和多目标优化的研究。在研究人工神经网络模型结构和原理的基础上,应用人工神经网络建立了基于切削力和表面粗糙度的加工参数单、多目标优化模型,切削实验表明基于神经网络模型的切削力和粗糙度模型能够很好地逼近真实的切削模型。论文工作在试验设计的基础上,通过多种代理模型研究加工参数对切削力和表面粗糙度的影响规律,以及应用代理模型对加工参数单、多目标优化的方法,切削试验证结果表明所建立的各种代理模型和优化方法是有效和可行的。这项研究有三种技术。这些技术的响应面法(RSM),田口和人工神经网络(ANN)。在响应面方法中,敏感性分析被用于以评估每个切削参数的重要程度。单响应表明,在这项研究中与所使用的其他加工参数之间比较更为重要。在多响应(双响应)表面粗糙度为独立响应,而切削力被称为伴随响应。田口技术应用研究中具有相同的加工参数和相同的水平,取样结果与全因子设计相比结果非常好。使用田口方法的取样的数量等于三分之一全因子法中使用的数据。采用神经网络方法对加工参数进行优化,得到满意的结果。
[Abstract]:In the process of metal cutting, machining parameters of cutting process not only affects the production efficiency, but also affect the surface quality of the cutting force and cutting parts. In fact, the cutting force and surface quality of parts is closely related to the factors of cutting force also affects the surface quality. The optimization of processing parameters to improve the cutting force in the process of cutting and machining surface quality is very important. In this paper, the cutting force and surface roughness as direct object, application agent model of response surface method and artificial neural network model based on the technology of processing parameters are optimized. This paper mainly discussed the following aspects: (1) based on the cutting force model, mainly analyzes the factors that influence the cutting force. In this study, in addition to the cutting speed, cutting depth, feed rate as the key processing The parameters, the cutter tooth number of milling cutter is introduced as an important processing parameters. Based on the analysis of response surface roughness agent model based on four key processing parameters and the level determined by the experimental design. (2) were studied and the sensitivity analysis of response surface roughness model. Based on four key process parameters, by cutting experiment collect the machining parameters and the surface roughness of the sample data set is established based on the agent model of machining parameters of two order response surface method and surface roughness based on. Secondly, through further sensitivity agent model analysis, simplified the roughness is not obvious. In the final model using proxy model to optimize machining parameters, cutting experiments show that the surface response surface method of rough degree of error in model based on agent Within the acceptable range. (3) response surface method is the effect of cutting force and surface roughness of machining parameter optimization based on degree. According to the four key processing parameters collected by the experiments based on the level of roughness and the cutting force data, a linear surface roughness and the cutting force and the two order the response surface model, and through the cutting experiment of agent model is validated to ensure the reliability of the model. The agent model based on the single objective and multi-objective optimization of machining parameters on machining parameters. Also, the cutting experiments show that based on the agent model, the surface roughness and the cutting force of single and multiple targets the optimization is reliable. (4) the processing parameters of artificial neural network model of single objective and multi-objective optimization based on artificial neural network. Based on the model structure and the principle of the application. The artificial neural network to establish the cutting force and surface roughness on machining parameters based on single and multi objective optimization model, cutting experiments show that the cutting force based on neural network model and roughness model can well approximate the real cutting model. This paper based on experimental design, through a variety of proxy processing model parameters the law of cutting force and surface roughness, and the application of processing parameters of single agent model, the multi-objective optimization method, cutting test results of various agent model and optimization method proposed is effective and feasible. This study has three kinds of technology. The response surface method of these technologies (RSM), Tian and artificial neural network (ANN). The response surface method, sensitivity analysis is used to assess the importance of each cutting parameter. The single response shows that in this study with other use and Work is more important. The comparison between parameters in multi response (dual response surface roughness) as an independent response, and the cutting force is called with the response. With processing parameters and the same level of application of Taguchi, sampling results compared with full factorial design results very well. The number of sampling using Taguchi method the use of 1/3 factor method is equal to the data. By using neural network method to optimize machining parameters and obtain satisfactory results.

【学位授予单位】：西南交通大学
【学位级别】：博士
【学位授予年份】：2015
【分类号】：TH16
，

本文编号：1597593