基于信号特征提取的设备健康状态预测与评估

发布时间：2018-03-03 16:04

本文选题：故障预测　切入点：设备健康状态评估　出处：《中国矿业大学》2015年硕士论文　论文类型：学位论文

【摘要】：井下机械机电设备复杂度较高,导致其故障特征呈现非线性、时变性、并发性、不确定性等特征,而且井下噪声干扰大、设备数量多、分布时而分散时而聚集等,给设备的故障预测带来了很大的麻烦。本文在前人信号处理、分类算法、故障预测、机器学习的基础上,针对井下设备健康状态评估准确度、适用范围、实用性等问题进行了研究。主要内容如下:(1)设备健康状态预测与评估。井下的复杂环境导致通过直接预测其故障类型变得非常困难,而且对于突发故障来说,预测设备的故障类型意义不大,因此提出通过健康状态评价设备的运行状况。实验表明,对于出现渐变故障的设备该方法具有很高的应用价值。(2)基于特征提取的井下设备特性研究。设备的运行声音能够反映出当前设备的健康状态,研究了声音对设备健康状态的敏感度,提出了对所提取声音进行分帧、去噪等预处理,并且对信号的短时能量、倒频谱、Mel倒谱系数特征进行对比分析。实验表明,基于特征提取的设备健康状态预测与评估是可行的,而且对于同一信号,提取不同的特征,其预测精度也有所不同。(3)机器学习理论在健康状态预测中的应用研究。支持向量机(Support Vector Machine,SVM)有唯一的全局最优解与出色的机器学习能力,能够很好的解决小样本、非线性、高维化等问题。本文将评价设备健康状态的问题转化为对设备特征分类的模型进行处理,提出了基于特征提取和SVM的设备健康状态预测与评估方法。(4)设备健康状态评估研究及实验分析。提出了设备健康度的概念,通过在井下水泵处安装的拾音器(或振动传感器)获得设备运行的声音信号,验证所提方法的合理性与准确度。结合设备健康度,实验研究了本文所述方法的预测精度影响因素。实验结果表明,基于Mel倒谱系数特征提取和SVM的设备健康状态预测方法具有更高的预测精度。
[Abstract]:The complexity of underground mechanical and electrical equipment is high, which results in its fault features being nonlinear, time-varying, concurrency, uncertainty and so on. Moreover, the downhole noise interference is large, the number of equipment is large, the distribution is sometimes scattered and sometimes aggregated, etc. On the basis of previous signal processing, classification algorithm, fault prediction and machine learning, this paper aims at the accuracy and scope of application for evaluating the health status of underground equipment. The main contents are as follows: 1) Prediction and evaluation of equipment health status. The complex underground environment makes it very difficult to predict the type of failure directly, and for sudden failure, It is of little significance to predict the fault type of the equipment, so it is proposed to evaluate the operation condition of the equipment through the health condition. This method has high application value for equipment with gradual fault. It has high application value. (2) the characteristic of underground equipment based on feature extraction is studied. The sound of equipment running can reflect the health state of current equipment. In this paper, the sensitivity of sound to the health state of the equipment is studied, and the preprocessing of the extracted sound, such as framing and de-noising, is proposed. The characteristics of the short time energy of the signal and the Mel cepstrum coefficient of the cepstrum are compared and analyzed. The experimental results show that, It is feasible to predict and evaluate the health status of equipment based on feature extraction, and different features are extracted for the same signal. The application of machine learning theory in health state prediction is also different. The support vector machine support Vector machine has a unique global optimal solution and excellent machine learning ability, which can solve the problem of small sample size and nonlinearity. In this paper, the problem of evaluating the health status of equipment is transformed into a model of equipment feature classification. Based on feature extraction and SVM, the research and experimental analysis of equipment health state evaluation are presented, and the concept of equipment health degree is put forward. The sound signal of the equipment running is obtained by the pick-up (or vibration sensor) installed in the underground water pump, and the rationality and accuracy of the proposed method are verified. The experimental results show that the method based on Mel cepstrum coefficient feature extraction and SVM has higher prediction accuracy.
【学位授予单位】：中国矿业大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TD611

【参考文献】