智慧制造环境下感知数据驱动的加工作业主动调度方法研究

发布时间：2018-07-04 20:06

本文选题：智慧制造 + 复杂事件处理　；参考：《华南理工大学》2016年博士论文

【摘要】：随着云计算、物联网、大数据、信息物理融合系统、企业2.0、工业4.0等的提出,信息技术与先进制造技术深度融合,孕育出基于社会信息物理系统的智慧制造,形成一种面向服务、基于知识运用的人机物协同制造模式。在智慧制造环境下,物联网覆盖整个生产车间,部署于车间的各种传感器(如RFID、加速度计等)实时监测整个生产过程,并通过网络将数据传送到处理中心。由于各种不确定因素,导致生产过程容易发生异常事件,造成生产过程的信息复杂且不易控制。需要对各种传感器数据实时处理,挖掘出生产现场的异常事件,并预测将要发生的异常状况,进而基于实时与预测的异常事件,实现生产车间设备主动调度,避免由于异常事件而给生产系统造成的危害。为此,本论文研究基于机械加工的工件异常事件监测和刀具剩余寿命预测的主动调度,包括如下主要内容:(1)新型智慧制造模式分析与总结智慧装备的特征,探讨网络融合与社会信息物理系统视角下的智慧制造模式,研究实现智慧制造的社会环境与关键共性技术问题。(2)基于RFID的工件异常事件监测构建智慧制造车间的感知环境,定义各类RFID事件模型,包括标签事件、简单事件和复杂事件等;给出复杂事件处理系统的框架,提出综合的RFID数据清洗方法,实现面向实时的工件异常事件监测,最后实验验证数据清洗方法和异常事件监测的有效性。(3)基于无线加速度计的刀具状态监测给出刀具状态监测系统的框架,并搭建刀具状态监测的实验装置;应用小波变换去除振动信号噪声,用不同的方法提取信号在时域、频域和时频域的特征,并依据皮尔森相关系数选择关键特征;建立神经模糊网络(Neuro-Fuzzy Networks,NFN)预测模型,编写刀具磨损与剩余寿命预测的人机接口程序,并与反向传播神经网络、径向基函数网络相比较,验证NFN预测效果。(4)基于深度学习的刀具状态监测比较5种深度学习模型的结构与训练方法,提出基于深度卷积神经网络的刀具状态监测方法;并且搭建卷积神经网络学习平台,比较卷积神经网络不同模型的执行效果,同时与传统神经网络的预测性能进行对比,验证所建立的模型有效性。(5)智慧车间加工作业的主动调度给出调度模型的分类,构建智慧车间加工作业的感知环境;提出一种主动调度方案,具体研究包括加工作业调度数学模型、主动调度框架、策略和多目标双层编码双级进化双重解码遗传算法(MD3GA);搭建智慧车间加工作业的原型平台,实现加工机器与AGV的集成调度,用实验加以验证所提出的主动调度方法。
[Abstract]:With the development of cloud computing, Internet of things, big data, information physics fusion system, enterprise 2.0, industry 4.0 and so on, the deep integration of information technology and advanced manufacturing technology gives birth to intelligent manufacturing based on social information physics system. A service-oriented and knowledge-based collaborative manufacturing model for human-machine is formed. In the intelligent manufacturing environment, the Internet of things covers the whole workshop, and all kinds of sensors (such as RFID-accelerometers) deployed in the workshop monitor the whole production process in real time, and transmit the data to the processing center through the network. Due to various uncertain factors, the production process is prone to abnormal events, resulting in the production process information complex and difficult to control. It is necessary to process all kinds of sensor data in real time, mine out the abnormal events in the production site and predict the abnormal situation that will happen, and then realize the active scheduling of the production workshop equipment based on the real-time and the predicted abnormal events. Avoid damage to production system due to abnormal events. Therefore, this paper studies the active scheduling of workpiece abnormal event monitoring and tool residual life prediction based on machining. The main contents are as follows: (1) the characteristics of intelligent equipment are analyzed and summarized in the new intelligent manufacturing mode. This paper discusses the intelligent manufacturing model from the perspective of network fusion and social information physics system, and studies the social environment and key common technical problems to realize intelligent manufacturing. (2) the perceptual environment of intelligent manufacturing workshop is constructed based on the abnormal event monitoring of workpiece based on RFID. Various RFID event models are defined, including tag events, simple events and complex events, the framework of complex event processing system is given, and a comprehensive RFID data cleaning method is proposed to realize real-time workpiece anomaly event monitoring. Finally, the validity of data cleaning method and abnormal event monitoring is verified. (3) based on wireless accelerometer tool condition monitoring, the framework of tool condition monitoring system is given, and the experimental device of tool condition monitoring is built. Wavelet transform is used to remove vibration signal noise, different methods are used to extract the signal features in time domain, frequency domain and time frequency domain, and the key features are selected according to Pearson correlation coefficient, and a neurofuzzy networks (NFN) prediction model is established. The man-machine interface program for tool wear and residual life prediction is written and compared with backpropagation neural network and radial basis function network. (4) the structure and training methods of five depth learning models are compared, and the tool condition monitoring method based on deep convolution neural network is proposed, and the learning platform of convolution neural network is built. The performance of different models of convolution neural network is compared, and the prediction performance of traditional neural network is compared to verify the validity of the established model. (5) the active scheduling of intelligent job shop gives the classification of scheduling model. The perceptual environment of intelligent job shop is constructed, and an active scheduling scheme is proposed, including the mathematical model of processing job scheduling and the active scheduling framework. The strategy and multi-objective double-level evolutionary double decode genetic algorithm (MD3GA) are used to build the prototype platform of intelligent workshop to realize the integrated scheduling of machining machine and AGV. The proposed active scheduling method is verified by experiments.
【学位授予单位】：华南理工大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：TH186

【相似文献】