基于在线序贯极限学习机的传感器状态监测系统研究

发布时间：2018-03-19 12:13

本文选题：传感器　切入点：状态分类　出处：《西安科技大学》2017年硕士论文　论文类型：学位论文

【摘要】：传感器作为自动化测试与控制系统的基础设备,为系统采集和提供数据,其输出结果对于后续的数据分析、数据处理、系统运行等很多方面的监控与决策都有着举足轻重的影响。并且一旦传感器发生了异常,轻则会使系统丧失监控能力,重则会造成难以估计的损失。因此,传感器故障与否,其输入输出特性是否发生变化,对于整个系统而言至关重要。本文针对传感器使用过程中可能产生的异常状况开展了以下研究:首先,搭建实验平台。以水厂的水泵系统为应用背景,选取电动机常用的压电式三轴振动传感器作为研究对象,采集复杂环境下多组有代表性的振动加速度信号。并对其进行清洗、采样、整合处理,存入数据库以备查询和后续分析。其次,建立传感器状态分类模型。本文根据传感器可能出现的异常,提出将传感器状态归纳分类为安全状态、干扰状态、冲击状态和偏差状态四种情况,直观地描述和反应传感器状态;采用小波包分解的特征提取方法进行三层分解,在时域和频域上将振动信号以能量特征的形式进行描述。然后,利用支持向量机、极限学习机和在线序贯极限学习机算法分别建立了振动传感器状态分类模型。针对支持向量机“过拟合”现象,结合交叉验证法,改善其泛化能力;对比极限学习机算法隐含层神经元个数和激活函数的不同选择对分类准确率的影响,寻找模型最优参数;从贴近实际工程使用角度出发,建立了改进的在线序贯极限学习机分类模型。并分别对三种分类模型进行实验验证,对比分析表明,构造的在线序贯极限学习机模型分类结果优于前两种。最后,设计实现振动传感器在线监测系统。将虚拟仪器LabVIEW与DAQmx、MATLAB软件相结合,实现了数据实时采集与存储、传感器当前状态显示、异常状态预警和历史数据查询功能。结果表明,本文所设计的振动传感器状态分类模型及在线监测系统为传感器实时监测、日常维护和管理提供理论参考及设计指导,具有深远的应用推广价值。
[Abstract]:As the basic equipment of the automatic test and control system, the sensor collects and provides the data for the system. Monitoring and decision-making in many aspects, such as system operation, play an important role. And once the sensor is abnormal, the light will make the system lose the monitoring ability, and the heavy will cause incalculable losses. Therefore, the sensor failure or not, It is very important for the whole system whether the input and output characteristics change. In this paper, the following researches have been carried out on the abnormal conditions that may occur during the use of the sensor: first of all, Taking the water pump system as the application background, the piezoelectric three-axis vibration sensor, which is commonly used in the motor, is selected as the research object, and many representative vibration acceleration signals are collected under the complex environment, and the vibration acceleration signals are cleaned. Sampling, integrating processing, storing in database for query and subsequent analysis. Secondly, the sensor state classification model is established. According to the possible anomalies of sensor, the sensor state is classified into security state and interference state. The state of sensor is described and reacted intuitively under the four conditions of shock state and deviation state. The feature extraction method of wavelet packet decomposition is used to decompose the vibration signal into three layers, and the vibration signal is described in the form of energy feature in time domain and frequency domain. The state classification model of vibration sensor is established by using support vector machine (SVM), extreme learning machine (LLM) and online sequential LLM algorithm. The overfitting phenomenon of SVM is combined with cross verification method to improve its generalization ability. Comparing the influence of the number of hidden layer neurons and the activation function of the LLM algorithm on the classification accuracy, the optimal parameters of the model are found. An improved online sequential extreme learning machine classification model is established, and three classification models are verified by experiments. The comparison and analysis show that the classification results of the online sequential limit learning machine model are better than those of the first two models. The on-line monitoring system of vibration sensor is designed and implemented. The virtual instrument LabVIEW and DAQmxX MATLAB software are combined to realize real-time data acquisition and storage, sensor current state display, abnormal state warning and historical data query function. The vibration sensor state classification model and on-line monitoring system designed in this paper provide theoretical reference and design guidance for sensor real-time monitoring, daily maintenance and management, and have far-reaching application and popularization value.
【学位授予单位】：西安科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP212;TP274

【参考文献】