摘要
为了在强背景噪声下有效识别齿轮故障,提出了一种多传感器信息融合的识别方法。首先计算多传感器振动信号的小波相关特征尺度熵,并以此作为强噪声背景下齿轮故障特征信息;以各尺度信息熵作为SOM神经网络的输入层,用标准训练样本训练神经网络,齿轮故障类型在竞争层聚类;为了提高识别过程的准确性及完整性,采用多传感器决策层融合技术,构造D-S证据理论识别框架,建立基于统计SOM神经网络识别率的基本信任函数分配方法。每个传感器的子决策作为一条子证据,根据D-S证据理论合成规则及各传感器的基本信任函数分配完成融合识别。试验结果证明,齿轮多传感器信息融合识别方法可以实现强噪声背景下齿轮故障的准确识别,消除识别的不确定性,识别率可达90%以上,是一种有效的齿轮故障识别方法。
To effectively identify the gear fault at strong noise background, a gear fault identification method was proposed based on multi-sensor information fusion. Firstly, the calculated wavelet correlation feature scale entropy(WCFSE) of the acquired multi-sensor signal was taken as the gear fault characteristic information at strong noise background; WCFSE was taken as the input layer of SOM neural network, and the standard training sample was applied to train the neural network, the gear fault type was gathered in competitive layer; to improve the accuracy and completeness in identification process, the multi-sensor fusion technology was introduced to establish the recognition framework of D-S evidence theory and the basic belief function allocation method based on the recognition rate of statistical SOM neural network. The sub-decision of each sensor served as a sub-evidence, and the recognition fusion was completed according to the combination rules of D-S evidence theory and the basic belief function of each sensor. The test results showed that the proposed gear multi-sensor information fusion recognition method realized accurate identification of gear failure at strong noise background, eliminated the uncertainty of recognition, recognition rate reached more than 90%. It was a kind of effective gear fault identification method.
出处
《矿山机械》
2015年第11期125-130,共6页
Mining & Processing Equipment
基金
国家高技术研究发展计划(863计划)"面向煤矿灾害现场救援机器人研究开发与应用示范"(2012AA041504)