混沌演化下基于深度学习的数控机床精度衰退大数据早期预示方法

Early prediction of the decline of motion accuracy is of great significance to improve the accuracy and reliability of CNC machine tools. Deep learning has strong ability of modeling and representation, and is a hot spot of interdisciplinary frontier. The mass data in the operation of the machine tool contains all the useful information about the evolution of the motion precision, and it is difficult to identify due to the features of weak, strong noise, mutual coupling, unstructured only by existing methods. The team’s previous research has proved that the decline of the motion precision of the CNC machine tool is a space-time evolution process of chaotic dynamical system. Therefore, take the advantage of the sensitivity of initial value of chaotic system, a new method based on deep learning for accuracy degradation’s early prediction of machine tool’s motion data is proposed. The basic idea is that, through converting the weak signal into the chaotic phase space which based on the chaotic dynamics of the machine tool precision, to build large data model based on deep learning for early state identification and early trend prediction. Thus such key issues as deep learning modeling method, network topology, training and optimization, temporal and spatial representation, will all be solved. The purpose of this project is to realize the large data mining and early prediction of the machine tool’s motion accuracy. The project is expected to explore the law based on "Big data PLUS Deep learning", which is not easy to show past. This will be a valuable innovation, and will provide new ideas and theoretical support for improving the accuracy and reliability of machine tools.

运动精度衰退早期预示对提高数控机床精度保持性、可靠性有重大意义。深度学习具有强大的建模与表征能力，是多学科跨领域前沿热点。机床运行过程海量数据蕴含着运动精度演化的全部有用信息，早期多源大数据微弱、强噪声、互耦合、非结构化，现有方法难识别。前期证明，机床运动精度衰退是一个混沌动力学系统的时空演化过程。鉴于此，利用混沌系统的初值敏感性，项目提出混沌演化下基于深度学习的机床运动精度衰退大数据早期预示新方法。基本设想是：以机床精度混沌动力学演化规律为研究基点，通过混沌演化将微弱信号映射于多源混沌相空间下，建立基于深度学习的运动精度衰退早期状态辨识与早期趋势预示大数据模型，突破网络拓扑结构、训练与优化、时空特征映射等关键技术，实现精度衰退前兆信息大数据挖掘及早期预示。项目基于大数据+深度学习探索过去不易昭示的精度动态规律，是一项极具价值的创新工作，可为提高机床精度保持性和可靠性提供新思路和理论支撑。

课题围绕运动精度衰退早期预示微弱信号序列重构与映射等三个关键科学问题开展了以下研究，并取得了相应的成果：.（1）研究了基于混沌动力学演化的运动精度衰退动态行为特征及时变规律。通过求解多源大数据序列的最优延迟时间和最小嵌入维数，将运动精度衰退早期微弱信号重构到高维混沌相空间，在高维相空间下揭示了相空间拓扑结构特征与运动精度衰退因素的映射关系，掌握了混沌多维相点轨迹和运动精度衰退演化趋势的映射规律，为建立大数据深度学习模型来探索运动精度动力学演化规律提供了理论依据。.（2）研究实现了基于混沌相空间重构与深度学习大数据模型的数控机床运动精度早期劣化发现及溯源方法。提出将数控机床运动误差微弱信号引入混沌振子系统，利用混沌系统具有的参数敏感性、蝴蝶效应及噪声免疫特性，将精度劣化信号重构到具有不同相图的混沌相空间。提出了深度学习特征距离判别方法，以发现混沌系统相图的状态变化。建立了基于CNN的深度学习大数据溯源模型，以识别不同劣化因素对应的混沌相空间形态，实现了误差因素的准确溯因。.（3）研究实现了基于深度学习和多源混沌序列的运动精度衰退趋势大数据模型及早期定量预示方法。引入多元相空间技术，将多个精度特征量序列映射到高维相空间，结合电流、振动等多源大数据，建立基于LSTM的运动精度深度学习大数据模型，以数据驱动自动提取高维精度相空间的时空特征，对运动精度变化趋势进行了准确的早期预测。鉴于热误差恶化是导致数控机床精度衰退的主要因素，研究了温升过程的混沌特性及规律，定义温升过程的圆运动重复定位误差，建立了基于热误差混沌演化和LSTM网络的预测模型，实现了数控机床运动精度衰退的及早预示，在不同条件下均有较高的预测精度和泛化能力。.课题在机床运动精度劣化预示的共性基础理论和关键技术上取得了突破，为机床精度可靠性领域提供了一种新的理论思路和新的建模方法；成果在构建数控机床精度状态智能监控体系方面具有良好的应用前景。