基于神经网络的电机故障诊断-控制工程专业论文.docx

上海交通大学工程硕士学位论文摘 上海交通大学工程硕士学位论文 摘 要 万方数据 万方数据 基于神经网络的电机故障诊断 摘 要 随着科学技术的飞速发展和工厂自动化程度的不断提高，作为机 电系统中主要的动力设备，电机越来越广泛地应用在工业生产的各个 领域中。电机故障会对生产秩序造成影响，甚至会对人身财产安全造 成危害，因此确保电机在正常状态下运行就显得尤为重要。国内外对 这方面的研究一直都非常重视，如何在故障早期发现先兆一直都是非 常重要的课题。 通常在电机运行时，电机故障和部件缺陷都会以某些特定的方 式，如电流的波动、壳体表面的振动变化、部件不同位置的异常温升、 或音频幅值和频率的变化等显示出来。如果能对这些信息进行收集记 录，然后使用各类处理方法进行信号变换，结合专家知识和各类分析 方法，就能实现电机故障判别。 本文提出了使用小波包能量谱和神经网络结合的方法，分析电机 的音频信号，从而达到辨识特定电机故障和部件缺陷的目的。首先采 用快速傅里叶变换，对电机的音频信号进行处理，得到相同型号多个 电机在单独运行时的声音频谱曲线，同时根据经验做出初步的判别。 接着凭借小波变换对频谱低频段进行聚焦，结合专家知识对该频段进 行观察，并对这些电机是否存在故障做出判断。然后按照故障种类和 阶段对其进行分析和归类，对特征频段做筛选。随后利用改进后的小 第 I 页 波包变换对音频信号进行处理，根据特征频段进行数据采集。最后， 使用选取特征数据对神经网络进行训练和验证，使其实现对特定电机 故障的判别功能。 关键词：电机故障，音频，快速傅里叶变换，小波包变换，神经网络 第 II 页 上海交通大学工程硕士学位论文AB 上海交通大学工程硕士学位论文 ABSTRACT MOTOR FAULT DIAGNOSIS BASED ON NEURAL NETWORK ABSTRACT As an important device in the electrical and mechanical system, motors are widely used in various fields of industrial production with the rapid development of science and technology as well as the continuous improvement of factory automation. Motor failure will has large influence on the order of production, and even cause damage to the safety of persons and property. Therefore, it is particularly important to ensure that the motor running in the normal state. Our country and abroad have always attached great importance to this aspect. How to find the fault in early stage has always been very important research topic. Generally, when the motor is running, the faults will display in some ways, such as the change of the current of the motor, vibration on the surface of the machine, the abnormal temperature rising, the changes of amplitude and frequency in audio and etc. These signals are acquired and processed in some ways. And then, with the application of the expert knowledge and various analytical methods, we can get the conclusion whether the motor faults exist. This dissertation presents an idea based on the combination of 第 I 页 wavelet packet and neural network in order to identify