基于多信息和多模型融合的刀具磨损预测性评估的方法研究

发布时间：2017-12-30 23:13

本文关键词：基于多信息和多模型融合的刀具磨损预测性评估的方法研究　出处：《中国科学院大学(中国科学院沈阳计算技术研究所)》2017年硕士论文　论文类型：学位论文

【摘要】：本文在工业大数据和PHM的应用背景下,利用数据驱动式的分析流程对数控铣刀磨损的预测性评估方法进行了深入的研究和分析。在研究过程中,本文首先对铣削过程中采集的信号进行了充分地探索和分析,主要包括信号无效值的截断处理、异常值的过滤处理,信号周期性、平稳性以及功率和能量特性等方面的分析。接着,对预处理后的信号进行了特征提取,分别用统计方法从时域中提取统计特征、用FFT变换从频域中提取频谱和能量特征、用WT变换从时频联合域中提取小波系数和能量分布比特征。此外,按照特征所属的信号类型和所属的轴向对提取出来的特征进行了划分和融合,用于多信息特征融合的实验研究。本文中通过基于F-test检验的评估值和基于互信息(MI)的评估值对特征进行选择,以提高模型拟合的速度和准确性。在本文中,分别用决策回归树(DTR)和支持向量回归(SVR)模型对多信息融合的特点和影响刀具磨损的主要因素进行了实验验证和分析。结果表明,一般情况下多信息融合的效果要优于单信息,并且得出铣削力信号特征和X轴上特征是影响刀具磨损的主要因素。最后,本文引入了机器学习领域的集成方法作为多模型融合的策略,并用回归树作为集成方法的基础学习器,对刀具磨损进行了评估和预测。在本文基于多模型融合的刀具磨损评估过程中,从模型的准确度、稳定性和适用性三个方面上,通过实验验证和对比分析方式分析了多模型融合方法和单模型方法在刀具磨损预测上的性能。结果表明,基于集成方法的多模型融合策略在上述三个指标中取得的效果明显优于单模型,从而说明了基于集成方法的多模型融合策略能够有效地用于刀具磨损的评估和预测。
[Abstract]:The industrial application background in big data and PHM, using data driven analysis process of assessment method of NC cutter wear is studied and analyzed. In the process of research, this paper gives a full exploration and analysis of the signal acquisition in milling process, including the truncation signal. The processing value, filtering the abnormal value, periodic signal, analysis of stability and power and energy characteristic. Then, the preprocessed signals were extracted respectively, using statistical methods from time domain to extract the statistical feature extraction, spectrum and energy characteristics from the frequency domain using FFT transform to extract and the energy distribution of wavelet coefficients than the features from the joint time-frequency domain using WT transform. In addition, in accordance with the characteristics of axial signal type characteristics belong and belongs to extract was divided and fusion for multi Experimental study on the characteristics of information fusion. This paper evaluates the F-test test value and based on mutual information (MI) based on the evaluation value for the feature selection, in order to improve the speed and accuracy of model fitting. In this paper, respectively using decision regression tree (DTR) and support vector regression (SVR) model is verified by experiments and analysis of the multi information fusion characteristics and main influencing factors of tool wear. The results show that the multi information fusion is generally better than single information, and that the milling force signal characteristics and X axis characteristics are the main factors influencing the tool wear. Finally, this paper introduces the integration method in the field of machine learning as much model integration strategy, and regression tree as the basis of integrated method of learning, the evaluation and prediction of tool wear. In this paper based on the multi model fusion tool wear assessment process, from the mold The three aspects of accuracy, stability and applicability, through experimental verification and comparative analysis method to analyze the multi model fusion method and single model method in prediction of tool wear performance. The results show that the multi model integration method fusion strategy in the above three indicators have better effect than single model based on thus, the model integration method of fusion strategy can be effectively used for evaluation and prediction of tool wear based on.

【学位授予单位】：中国科学院大学(中国科学院沈阳计算技术研究所)
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TG71;O212.1

【相似文献】