设备健康监测分析与预测方法研究

论文总字数：32659字

摘 要

在现代工业中，大型设备长期工作在高温、高压等恶劣的工况下，其性能退化和故障发生不可避免，而故障一旦发生将会带来巨大的经济损失与人员伤亡，因此对设备进行运行状态监测在现代工业生产中极其重要，对避免严重事故，保障人身财产安全起了巨大贡献，设备健康预测能够提前预报故障，进一步保障了人身安全减少财产损失。

本文以往复式压缩机为研究对象，研究其健康监测与预测方法。将超球模型引入到压缩机的健康监测中，训练分辨设备故障与健康状态的二分类超球模型，并得出反映设备健康程度的健康度指标；将AR自回归预测算法用于健康状态的预测，训练自回归时间序列预测模型，对健康度指标变化趋势进行预测，有效地预测了健康度指标的上升与下降，对故障做出警报。具体工作内容包括以下几个方面

（1）本文针对往复式压缩机结构复杂，故障种类多样的问题，从往复式压缩机振动加速度信号中提取绝对均值、均方根值、峭度、峰值等时域特征和频率均值、重心频率等频域特征，根据压缩机常见故障提取活塞杆沉降量特征，提取气缸进气排气阀门温度反映压缩机做功情况，并且对进气排气阀门温度做差削弱环境因素的影响，较为全面地提取了对应压缩机多种故障的特征。

（2）本文针对特征过多造成的计算量过大的问题，通过计算皮尔逊系数和斯皮尔曼系数进行去除重复特征的方法，设定阈值为0.8，认为系数绝对值大于阈值的两特征互为重复特征并只保留其中一个特征，利用该基于相关系数的特征选择方法进行特征选择，选择特征峰值、偏斜度、峭度、活塞杆沉降量和进排气阀门温差等13种特征构建训练和测试样本。降低了数据的维度，有效的减少了超球模型计算量。

（3）针对难以直接用单一特征表示设备健康度的问题，选定健康训练样本与测试样本，基于高斯核函数训练超球模型，该超球模型将健康样本与故障样本分开，计算测试样本距超球中心的长度，对样本距超球中心的距离进行处理，得到反映设备健康程度的无量纲量健康度。健康度在0-1范围变化，健康度趋近1表示设备故障，健康度趋近0表示设备健康。直观简洁的反映了设备健康程度。

（4）设备状态变化复杂，难以直接得出变化规律，根据健康度序列的自相关系数和偏相关系数选用自回归模型，根据贝叶斯信息准则确定自回归模型阶数，使用最小二乘法确定自回归模型系数，进行模型适应性检测并对健康度时间序列进行预测，以相对误差绝对值为指标对预测结果进行评价。有效的预测了设备健康短期变化。

关键词：超球模型，健康监测，健康预测，时间序列预测，自回归预测

ABSTRACT

In modern industry, large-scale equipment works in harsh conditions such as high temperature and high pressure for a long time, and its performance degradation and failure are inevitable. Once the failure occurs, enormous economic losses and casualties will be brought. Therefore, it is extremely important to monitor the operation status of equipment in modern industrial production, which has made great contributions to avoid serious accidents and ensure the safety of personal and property. Faults could be predicted by equipment health prediction in advance, personal safety would be further protected and property losses will be reduced too.

In this paper, the reciprocating compressor is taken as the research object, and its health monitoring and prediction methods are studied. The hypersphere model is introduced into the compressor health monitoring, and the two-class hypersphere model is trained to distinguish the equipment fault and health status, and the health index reflecting the equipment health level is obtained. The AR autoregressive prediction algorithm is used to predict the health status, and the autoregressive time series prediction model is trained to predict the change trend of the health index and effectively predict the health level. Up and down of the target, alarm the failure. The specific work includes the following aspects

(1)Aiming at the complex structure of reciprocating compressor and various kinds of faults, time domain characteristics such as absolute mean value, root mean square value, kurtosis, peak value, frequency mean value and center of gravity frequency from the vibration acceleration signal of reciprocating compressor are extracted in this paper. According to the common faults of compressor, the sedimentation characteristics of piston rod are extracted, and the temperature reversal of intake and exhaust valve of cylinder is extracted. Compressor performance is reflected, and the influence of environmental factors is weakened by the difference of inlet and exhaust valve temperature. The characteristics of various faults of the corresponding compressor are comprehensively extracted. Then, the training set is established and the Gaussian kernel function is used to train the hypersphere model. The distance between the sample and the center of the supersphere is processed to obtain the dimensionless quantity health that reflects the health of the equipment.

(2)Aiming at the problem of too much computation caused by too many features, a threshold of 0.8 is set in this paper by calculating Pearson coefficient and Spielman coefficient to remove duplicate features. It considers that the two features whose absolute value of coefficients is greater than the threshold are duplicate features and only retain one of them. The feature selection method based on correlation coefficient is used to select features. The training and testing samples were constructed by selecting 13 characteristics, such as characteristic peak value, skewness, kurtosis, piston rod settlement and temperature difference between intake and exhaust valves. The dimension of data and the computational complexity of hypersphere model are reduced effectively.

(3) Aiming at the problem that it is difficult to express the equipment health directly with a single feature, health training samples and test samples are selected. Based on the Gauss kernel function, the hypersphere model is trained. Health samples are separated from the fault samples by the hypersphere model. The length of the test samples from the hypersphere center are processed, so as to reflect the equipment health. The degree of health is in the range of 0-1. The health of equipment is expressed by the health degree approaching 0, and the failure of equipment is expressed by the health degree approaching 1. The health of the equipment is shown intuitively and concisely.

(4) The state of the equipment is complex, and it is difficult to get the change rule directly. The autoregressive model is selected according to the autocorrelation coefficient and partial correlation coefficient of health degree sequence, the order of autoregressive model is determined according to Bayesian information criterion, the coefficient of autoregressive model is determined by least square method, the model adaptability is tested and the time series of health degree is predicted. The absolute value of relative error is taken as the index. The prediction results are evaluated by the benchmark. The short-term changes of equipment health were predicted effectively.

Keywords: hypersphere model, health monitoring, health prediction, time series prediction, autoregressive prediction

目 录

摘 要 I

ABSTRACT III

第一章 绪论 1

1.1课题背景及研究意义 1

1.2设备健康监测及预测研究现状 1

1.2.1故障特征提取选择研究现状 1

1.2.2超球模型分类研究现状 2

1.2.3预测模型研究现状 2

1.3本文研究内容与结构安排 3

1.3.1研究内容 3

1.3.2结构安排 3

第二章 设备运行状态特征提取方法 4

2.1往复式压缩机常见故障 4

2.2时域特征提取 5

2.3频域特征提取 7

2.4进出气阀温度差特征提取 8

2.5本章小结 11

第三章 基于相关系数的特征选择方法 12

3.1特征数据预处理 12

3.2基于相关系数的特征选择方法 13

3.3数据分析 13

3.4本章小结 18

第四章 基于超球的设备运行状态健康监测方法 19

4.1超球模型 19

4.1.1超球模型 19

4.1.2核函数 20

4.2基于超球的设备运行状态健康监测模型 21

4.3数据分析 23

4.4本章小结 26

第五章 基于自回归模型的设备健康预测方法 27

5.1自回归模型（AR模型） 27

5.2基于自回归模型的设备健康预测方法 27

5.2.1算法原理 27

5.2.2模型识别 28

5.2.3模型定阶 29

剩余内容已隐藏，请支付后下载全文，论文总字数：32659字