基于深度学习的刀具磨损状态监测技术的研究

发布时间：2018-01-06 12:41

本文关键词：基于深度学习的刀具磨损状态监测技术的研究　出处：《中国科学技术大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着机械加工过程向自动化、智能化的不断发展,机械故障智能化监测技术的研究就至关重要。刀具是机械加工中最重要的加工要素之一,刀具磨损不但直接影响了工件的尺寸精度和表面质量,也会间接影响加工效率和生产成本等。如何通过更加有效的方法去监测刀具磨损已成为大批学者的研究重点,所以关于刀具磨损状态监测技术的研究具有巨大的潜力和应用价值。本文以高速铣削过程中刀具的磨损状态为研究对象,利用刀具在不同磨损状态下的铣削力信号对不同监测技术进行研究。具体研究内容主要有:(1)建立了高速铣削刀具磨损状态监测的试验系统,通过试验的压力传感器采集了大量的力信号数据;(2)通过对铣削力信号在时域、频域以及时频域上进行特征提取和特征选择,选择13个典型特征作为后续神经网络的输入向量;(3)建立了基于BP神经网络的刀具磨损状态监测模型;(4)把深度学习理论引入到刀具磨损监测领域,提出了一种基于深度学习的高速铣削刀具磨损状态监测方法。本文的主要贡献点就是提出了应用深度学习方法来实现刀具磨损的智能监测。通过小波包变换提取铣削力信号在不同频段上的能量分布作为初始特征向量,采用无监督学习对稀疏自编码网络进行特征学习,并将单层网络堆栈构成深度神经网络,最后利用有监督学习对整个深度网络进行微调训练,建立铣削刀具磨损的智能监测模型。实验结果表明,与传统的浅层模型进行对比分析,本文提出的基于深度学习理论刀具磨损监测新方法可以自适应提取刀具的磨损信息,并具有更高的监测准确度。
[Abstract]:With the machining process automation, continuous development of intelligent monitoring technology, it is crucial to study the mechanical fault. The tool is one of the most important factors in the machining process, tool wear not only directly affects the dimensional accuracy and surface quality of the workpiece, but also will directly affect the processing efficiency and production cost. How to monitor the tool wear has become a research focus of many scholars through more effective methods, so the research on cutting tool wear condition monitoring has great potential and application value. Based on the high speed milling tool wear state in the process as the research object, the milling force signal using the tool in different wear conditions were studied on different monitoring technology specific contents are: (1) established a test system of state monitoring of tool wear in high speed milling, the pressure sensor test For a large amount of data acquisition signal; (2) based on the milling force signal in the time domain, frequency domain and time-frequency domain on feature extraction and feature selection, choose 13 typical features as input vector of subsequent neural network; (3) a loss condition monitoring model of grinding cutter based on BP neural network; (4) the deep learning theory into the field of tool wear monitoring, based on deep learning condition monitoring method of tool wear in high speed milling. The main contribution of this paper is to put forward the application of deep learning method to realize the intelligent tool wear monitoring. The extraction of milling force signal in different frequency band energy distribution as the initial feature vector by wavelet packet transform, using unsupervised learning of sparse features from the encoding network, and the single-layer network stack constitutes depth of neural network, finally using supervised learning on the whole Fine tune the depth of network training, the establishment of intelligent monitoring model of tool wear in milling. The experimental results show that compared with the traditional model of shallow and deep learning theory based on adaptive information extraction can be a new method for tool wear monitoring of tool wear monitoring is put forward in this paper, and has higher accuracy.

【学位授予单位】：中国科学技术大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TG714;TP18

【参考文献】