【作者】 乔宏；

【导师】 傅攀；

【作者基本信息】 西南交通大学 ， 测试计量技术与仪器， 2011， 硕士

【摘要】 数控机床作为典型的机电一体化设备,在现代机械制造行业占据了重要地位,它的拥有量已经成为衡量一个地区制造业能力的标准,并且其应用还在不断地推广,因此数控机床的故障诊断技术也就显得尤为重要。滚珠丝杠作为数控机床上常用的进给传动部件,对机床的加工性能具有很大的影响,丝杠的微小异常有可能会导致机床的加工零件不合格,甚至影响到产品的加工进度,所以对滚珠丝杠的状态进行监测,及时发现故障是保证数控机床正常工作的一个重要条件。小波分析凭借其在时域与频域的综合分析能力,成为信号分析领域中非常重要的一种分析技术。它采取变时窗方法克服了傅里叶变换的频域局部分析弱点,将时域与频域结合起来,更加准确地描述了信号的特征。小波包分析是对小波分析的延伸,通过对在小波分析中被忽略掉的高频信号部分进行补充分析来完善小波分析的结果。本文所研究的小波神经网络将小波分析引入到对信号进行模式识别与故障诊断的人工神经网络中,利用小波函数替换人工神经元中的激励函数,将小波变换的多尺度特性引入神经网络,从而使小波神经网络不但具有小波的多分辨率特性而且保留了神经网络的自学习与模式识别的能力。本文在BP神经网络的基础上引入morlet小波函数作为隐层神经元的激励函数,从而搭建起小波神经网络。利用正弦函数进行仿真验证后,证实了该小波神经网络的可行性,即本文所建立的小波神经网络可以用来进行模式识别。本文对采集到的滚珠丝杠的振动信号进行分析,并在时域、频域以及时频域等各个方面从信号中提取出能够体现丝杠状态的有效特征值,将这些有效特征值输入建立的小波神经网络进行训练及模式识别,证实了小波神经网络在数控机床故障诊断中的实用性。通过与BP神经网络的对比,突出了小波神经网络稳定性好、识别率高等特点。本文的重点在于通过对BP神经网络算法及应用的研究,成功地将小波函数引入其中,从而建立起新型的小波神经网络。并运用此网络成功地对实际工程中的数控机床滚珠丝杠振动数据进行模式识别,正确地诊断滚珠丝杠的故障状态。更多还原

【Abstract】 As a typical electromechanical integration equipment, CNC machine occupies an important position of modern machinery manufacturing industry, the number of its ownership has become a standard to measure the manufacturing capability of a regional, and its application is continuing to promote, so the fault diagnosis technology of CNC machine is especially important.As a conventional feed drive components of CNC machine, ball screw has a great impact on the processing performance of the machine. Its tiny abnormality may result in the fail of parts processed, and even affect the overall processing of the product, so it’s an important condition of ensuring CNC machine’s normal operation to monitor the condition of ball screw and detect its faults timely.With the comprehensive analysis capability in time domain and frequency domain, wavelet analysis is very important in the area of signal analysis. It overcomes the Fourier transform’s weakness of partial analysis in the frequency domain, it combines the time-domain with frequency-domain, so the signal characteristics can be accurate descried. Wavelet packet analysis is an extension of wavelet analysis, it analyzes the high-frequency signal which is ignored in wavelet analysis, so its result is more complete.The wavelet neural network in this paper introduced wavelet into neural network, it replaced the activation function of artificial neuron with wavelet function to lead the characteristics of multi-scale into neural network. Thereby the wavelet neural network not only has the multi-resolution characteristics of wavelet but also keep the self-learning and pattern recognition capabilities of neural network.In this paper, the wavelet neural network is based on BP neural network. It introduced morlet wavelet function as the activation function of hidden layer neuron, which erected wavelet neural network. The simulation using sine function has confirmed the feasibility of the wavelet neural network, that is the wavelet neural network established can be used for pattern recognition.In this paper, the vibration signal of the ball screw is collected and analyzed, and the effective features which can express the condition of the ball screw are extracted from the time domain, frequency domain and time-frequency domain, then put these features into the WNN established to be trained and pattern recognized. The results confirmed the practicability of fault diagnosis of CNC using WNN. And By comparison with the BP neural network, the advantages of high stability and recognition rate are highlighted.The most important part in this paper is that a new type of wavelet neural network is established by introduce the wavelet function into the BP network successfully based on the study of algorithm and application of BP neural network. And the vibration signal of the ball screw is recognized successfully by the wavelet neural network so that the fault state of the ball screw is diagnosed correctly.更多还原