【作者】 韩宁；

【导师】 刘庆瑞；

【作者基本信息】 河北科技大学 ， 电机与电器， 2011， 硕士

【摘要】 中频冶炼技术近年来作为一门新兴技术,在淬火、熔炼、透热工业领域中得到广泛运用。随着现代集成化感应加热技术日益趋于成熟,中频冶炼自动化水平的不断提高,中频冶炼系统的可靠、稳定、安全运行也越来越重要。本文主要是通过对中频冶炼系统主电路电压信号进行监测从而实现对变频器短路故障进行智能故障诊断的研究。发生短路故障的变频器电压信号是非平稳随机信号,以往的傅里叶变换难以满足故障特征向量提取要求,并且建立变频器故障数学模型比较复杂,基于以上原因本文提出了基于小波包变换和PSO神经网络的中频冶炼故障诊断方法,通过小波包提取故障特征向量,利用PSO神经网络进行故障识别,并输出系统相应故障类型。该故障诊断方法首先是通过利用MATLAB软件对中频冶炼主电路进行建模、仿真,模拟了主电路中晶闸管开路故障,获得了相应的故障波形。其次,当一个系统或电路发生故障时,测试点测得的电压或电流与正常情况下相同频带内能量值是不一样的,根据能量守恒定律,某一频带能量低,另一频带能量值可能就高,信号每个频带的能量值是体现故障状态一个有效信息。本文使用的小波包算法就是对信号进行系数分解,得到一组能量形式的故障特征向量;最后,本文主要对变频器三大类短路故障状态进行测试:无故障,一只晶闸管短路故障、两只晶闸管短路故障,设计了相应的故障信号采集系统和试验方案,利用已经得到的特征向量和三大类故障状态设计并训练PSO神经网络。通过MATLAB平台实现相应的算法。本文主要是通过仿真实验验证相应故障诊断算法的准确性和有效性。通过仿真模拟故障信号,再对信号分别进行采集和测试,使用小波包算法对故障信号进行特征提取得到故障特征向量;利用已知故障类型的样本对PSO神经网络进行训练,然后用测试信号对训练好的神经网络进行测试。试验结果显示输出故障类型与实际测试信号所对应的故障状态一致。结果证明了基于小波包变换和PSO神经网络智能故障诊断方法对中频冶炼这样随机性和模糊性很强的系统切实可行,收敛速度快准确度高,减少了企业的经济损失,具有很大的运用价值。更多还原

【Abstract】 In recent years, intermediate frequency smelting technology as an emerging technology has been widely used in quenching, melting, heat penetrating industrial areas. As the automation levels rising, reliable, stable and safe operation is becoming more and more important to intermediate frequency smelting system.The fault diagnosis of intermediate frequency smelting system is researched by monitoring main circuit voltage signals. The short-circuited fault signals of converter are random and unstable. Fourier Transform can’t meet the unsteady signal feature vector extraction requirements. At the same time, establishing a mathematical model of inverter fault is more complex. For this reason, the fault diagnosis method of inverter based on wavelet packet analysis and PSO neural network is put forward in this paper. Wavelet packet is used to extract the feature vector from fault signals. The PSO neural network is used to identify the feature vector and output the corresponding fault type of the system.First, the method of intermediate frequency smelting fault diagnosis obtains corresponding fault waveform through MATLAB software to build a simulation model.Secondly, when circuit in fault conditions, energy value of the voltage test point is different in the same band. This is the so-called law of conservation of energy. If a band energy is low ,another band energy may be high. Anyhow, various changes of band energy embody the fault characteristics. The algorithm of wavelet packet strikes the energy of frequency band and gains the open circuit fault characteristics from various changes in each energy frequency band. Finally, this paper focuses on testing five types of short-circuit fault state of the converter: no-power tube failure, single power tube failure, two power tube failure in the same bridge arm, two power tube failure in the same half bridge arm, cross two power tube failure. After finishing feature extraction, PSO neural network is established by setting initial value and training neural network. The algorithms in this paper are realized with MATLAB software.The simulation experiment verifies the accuracy and validity of the fault diagnosis algorithm. The samples and testing signals are collected separately by simulating. Fault feature vectors are obtained by wavelet packet analysis which are off-line analysis. Neural network based on PSO is trained by sample eigenvectors, and tested by testing eigenvectors after the successful training. Test results show that the output fault type accords with the corresponding actual fault state of the test signals. The simulation experiment results prove that PSO neural network intelligent fault diagnosis method is practicable and has the advantage of fast convergence and high accuracy更多还原