基于蝙蝠仿生耳的远距离非接触式设备故障检测与诊断技术问题研究

Long-distance non-contact device failure detection and diagnosis based on the bat Bionic Ear.Fault diagnosis based on the audio signal is a non-contact fault diagnosis method that can achieve a secure online detection. However,the acoustic sensors laid in a fixed position, and cables of data transmission are also needed to be laid, which will bring the limitation of utility. Because of higher security requirements in the site of power, mining and other similar places,it is not facilitate to installate detection devises there. if the highly directional acoustic sensors installed in the distance can also receive the sound signal, great convenience will be shown and will promote the application of the devices.With the rapid development of information technology, high sensitivity, strong anti-jamming, low-power antenna has become the focus of research in communications. for which people put forward the concept of biomimetic antenna.in the case bat sonar system is an important example current attentioned . Some relevant research results have been utilized and has shown good effect. This research try to apply this technic in non-contact devices of the sound signal receiving device, .and improve the audio pattern recognition algorithm to explore the analysis theory and method for monitoring the running of Large-scale equipment online by the use of remote mobile audio receiver.

基于音频信号的故障诊断是一种非接触式的故障诊断方法，该方法能够实现安全的在线检测，然而该方法需要在固定位置布置声音传感器，并采用有线方式传送数据，这就给该方法的利用带来了限制，因为类似电力、矿山等场所对安全性要求较高，不便于进行固定式的安装。如能在远处利用高指向性声音传感器获取信号并进行处理将对这种音频信号故障诊断的应用带来极大便利。随着信息技术的快速发展，高灵敏、强抗干扰性、低功耗的天线成为当前通信领域的研究热点，为此，人们提出了仿生天线的概念，蝙蝠声纳系统是当前被人们关注的重要范例。一些成果已经在走向实用阶段。本项目借助于蝙蝠声纳系统中的研究成果，试图应用在非接触的设备音频信号的接收装置上，并改进当前音频模式识别的算法，探索利用远距离可移动式音频接收设备对设备故障进行在线监视和模式识别的分析理论与方法。

基于音频信号的故障诊断是一种非接触式的故障诊断方法，该方法能够实现安全的在线检测，在自然基金资助下，我们针对这一问题展开了研究。首先，为了克服干扰对变压器音频故障检测的影响，开展了蝙蝠耳及蝙蝠面部声道等与声纳发射有关的生理结构的研究，目的是找到就有较好指向性的结构使生接收器在接收有效信号的同时，抑制干扰声的影响。其次，针对变电站声频正常音和故障音的特点，并针对系统实时性的要求，分别设计了基于KNN和大数判决的方法，基于PCA、2DPCA结合SVM的方法，基于HMM和黎曼流形特征的方法，分别进行了识别实验，并进行了比较，结果证明2DPCA结合SVM的方法识别效果最好，通用也最强，即使是基于HMM的连续声频的识别中，也是2DPCA的效果好，基于此，我们改进了我们之前设计的声频监测平台。