基于稳态子空间分析的大型数控铣床刀具状态在线识别与预测方法研究

Tool damnification is the primary factor affecting the quality of large-scale CNC milling machine. Recognize and predict the tool’s state promptly and effectively are significant for improving the service performance of large-scale CNC milling machine. Aim to the problem of tool’s online recognition and prediction of large-scale CNC milling machine, this project researches the feature extraction of tool states, and provides a new method for tool’s online recognition and prediction based on Stationary subspace analysis in order to improve recognition accuracy and predict performance of tool state. The main contents include : 1) study the mechanism of tool wear (multi-level) and damage of large-scale CNC milling, obtain the tool states characterized sample data sets and sensitive parameter set; 2) study a new feature extraction method for tool states in small sample condition, explore a non-line Stationary subspace analysis method combining the kernel function technology, study a data-driven kernel generation algorithm, and provide a high effective feature extraction method for the stationary feature of tool’s states; 3) study the regular pattern of stationary feature in tool’s different states in order to explore the way of development and transfer of tool damnification, and provide a high-performance prediction algorithm with tool state based on incremental kernel regression. This project will help to reveal the nature of tool damnification’s germination, development and transfer in large-scale CNC milling machine, and provide comprehensive information to support its online recognition and prediction.

刀具损伤是影响大型数控铣床加工质量的首要因素，及时有效地识别与预测刀具损伤状态对提高大型数控铣床的服役性能具有重要意义。本项目针对大型数控铣床刀具状态在线识别与预测问题，以提高刀具状态的识别精度与预测性能为目标，以刀具状态特征提取研究为基础，提出基于稳态子空间分析的大型数控铣床刀具状态在线识别与预测新方法。主要内容包括：1）研究大型数控铣床刀具磨损(多级)和破损的损伤机理，获取刀具状态样本数据集和敏感特征参数集；2）研究小样本下刀具损伤特征提取方法，结合核函数技术探索非线性稳态子空间分析方法，研究数据驱动的核函数生成算法，提出高效的刀具损伤状态本质特征提取新方法；3）研究刀具损伤状态本质特征之间的规律，探索刀具损伤的发展和转移途径，提出基于增量核回归的高性能刀具损伤状态预测算法。本项目的研究有助于揭示大型数控铣床刀具损伤发生、发展和转移本质，为其在线识别与预测提供全面的信息支持。

本项目针对大型数控铣床刀具状态在线识别与预测问题，研究有限样本信息下的刀具损伤识别与预测方法。结合稳态子空间分析(SSA)方法不需要大量样本数据进行训练，也不要求数据的各维度分量相互独立，对数据要求较低等特点，本项目研究了基于稳态子空间分析的铣床刀具损伤特征提取方法，建立了基于一维振动信号的SSA-LSSVM刀具状态诊断模型；针对SSA方法假设稳态子空间与非稳态子空间线性叠加之弊端，结合核函数技术探索非线性稳态子空间分析方法，研究了数据驱动的核函数生成算法，提出高效的刀具损伤状态本质特征提取新方法；研究了基于Nadaraya-Watson核回归的大型铣床刀具状态预测算法，对其损失函数与带宽参数进行了优化。本项目的研究有助于揭示大型数控铣床刀具损伤发生、发展和转移本质，为其在线识别与预测提供全面的信息支持。