基于健康评估和劣化趋势预测的水电机组故障诊断系统研究

【作者】 潘罗平；

【导师】 唐澍； 高忠信；

【作者基本信息】 中国水利水电科学研究院 ， 水利水电工程， 2013， 博士

【摘要】 水电机组运行状态的实时诊断直接关系到水电站的安全稳定运行、电力质量和电力生产成本等重要的经济效益指标,其社会效益巨大。随着电站规模和监测辅助系统的不断扩大,机组的控制和监测数据信息量越来越大,运行操作人员对机组运行状态的实时有效监控、对设备故障做出迅速而准确地判断变得越来越困难,因此,研究水电机组运行状态健康评估和性能劣化趋势预测是非常必要的。考虑到水电机组状态监测技术已得到广泛应用,但现阶段水电机组故障样本较少,现有的诊断技术无法满足工程应用等实际情况,提出基于健康评估和劣化趋势预测的水电机组故障诊断新思路。从研究机组运行正常特征入手,通过建立健全的监测特征量健康样本来实现水电机组的健康诊断,诊断方法侧重于设备运行状态的实时健康诊断,重点在于监测异常和预测异常,其诊断方法与传统的基于故障征兆的诊断方法有较大差异。鉴于水电机组实际运行中,出现的故障概率较小,基于该诊断理念开发的诊断系统工程实用性很强,且系统具备实时健康诊断和量化的性能退化趋势预测功能,既能实时监测异常,又能预测异常,可满足现阶段的工程应用需求。本文从运行状态特征提取、健康标准及健康样本的建立、基于特征样本的健康评估、基于时序分解模型的状态趋势预测、基于LS-SVM的参数性能退化评估以及集成化故障诊断系统的研究与应用六个方面,系统开展了基于健康评估和劣化趋势预测的水电机组故障诊断技术与应用研究。以水电机组的水轮机和发电机为研究对象,在归纳总结水电机组运行异常时可能出现的各种特征表现的基础上,提出了表征水电机组运行状态的特征参数,以及特征参数三种可以量化的特征指标：幅值、频率、波形形状,并给出了相应的计算方法。提出了可用于水电机组运行状态健康评估的三种评价标准：绝对评价标准、相对评价标准、类比评价标准,并给出了标准限值相应的取值计算方法。根据概率论与数理统计学的极限理论,以及休哈特控制图理论,提出了采用样本均值作为特征量的标准值,以3σ准则确定的X。=X±3σ为报警界限值的特征量健康标准。以机组前期正常运行条件下的振动监测样本为例,分析了机组运行条件(功率、水头)对监测参数特征量指标的影响,提出以控制样本方差的方法来对运行条件进行分区,建立分区健康样本的具体方法和步骤,这样既保证了样本的判异准确性,同时也减低了样本空间维数(样本个数)。建立了基于时间序列变化分解的水电机组特征量趋势预测模型,提出了基于时间序列分解模型的趋势预测和性能退化预测的算法。采用电站实际监测数据对分解模型和算法进行了验证,结果表明,预测趋势与实际监测趋势具有很好的吻合性,可满足水电机组监测特征量的趋势预测和性能退化预测,对早期预警机组潜在异常,具有很好的实用性。以水电机组振动为例,提出了基于LS-SVM的水电机组振动参数性能退化评估三维标准模型(功率-水头-振动量)。将机组实时运行的有功功率和工作水头代入训练好的模型,即可获取当前工况下机组振动量是否偏离正常状态,实现机组运行状态的健康评估。基于振动参数性能退化时间序列,提出了基于LS-SVM的水电机组振动参数性能退化预测模型,采用上导摆度和上机架振动参数现场状态监测数据对所提模型进行验证。结果表明,该模型能较好的对水电机组振动参数性能退化进行评估和预测。最后以三峡集团公司远程状态监测与故障诊断系统为例介绍了集成化故障诊断系统的研究与应用,提出了现地监测层—厂站集成层—中心诊断层分布式故障诊断系统的总体结构,以标准化的数据格式来实现不同监测设备之间的数据通信,通过标准化数据平台的集成,实现了状态信息的共享和多信息的融合诊断。通过建立机组运行状态健康样本库,实现了机组运行状态的实时健康评估和性能退化预测,达到了设备健康状态定量评估的目的,可为指导机组状态检修提供技术依据。更多还原

【Abstract】 The real-time diagnosis of hydropower units operating state is directly related to the important economic and social benefits of the hydropower station, such as security and stability running, power quality and electricity production costs, etc.. With the continuous expansion of power plant scale and monitoring auxiliary systems, information of unit’s control and monitoring data is growing significantly, making it more and more difficult to the real-time monitoring of unit operating status and the quick and accurate determination of equipment failure for the operating personnels. Therefore, it’s of great necessarity for us to research on health assessment of operating status and performance degradation trend forecast of hydropower units.Taking into account the actual situation that hydropower units condition monitoring technology has been widely used while hydropower units fault samples are lack currently, and existing diagnostic technology can not meet the project applications, a new idea for hydropower units fault diagnosis based on health assessment and deterioration trend forecastideas was proposed. Starting from the normal feature research of unit operation, we achieved the health diagnosis of hydropower units by establishing health samples with sound monitoring features. Diagnostic methods focus on real-time health diagnosis of the equipment operating status, with emphasis on monitoring anomalies and predicting abnormal. This diagnostic method is quite different from traditional fault symptom-based methods. Since the probability of failure is small in the actual operation of hydropower units, diagnostic system developed based on this diagnostic concept is very practical. Moreover, the system with functions of real-time health diagnosis and quantified degradation trend forecasting, can not only monitor abnormal real-time, but also predict abnormalities, which meets the engineering applications at the present stage. Research on fault diagnosis of hydropower units based on health assessment and deterioration trend predicted was carried out in the paper from the study of the following six aspects including the running state feature extraction, establishment of health standards and health samples, feature sample based health assessment, state trend forecast based on timing decomposition model, LS-SVM based parameters performance degradation assessment and the research and application of integrated fault diagnosis system.With hydraulic turbine and generator as the object of study, and on the basis of summarizing various feature performances in operation abnormal of hydropower units, the paper put forward characteristic parameters to characterize the operational status of the hydropower units, as well as its three characteristic indexes that could be quantified: amplitude, frequency and waveform shape, and the corresponding calculation methods were given.Three evaluation criteria for operating state health assessment of hydropower units were proposed:absolute evaluation criteria, relative evaluation standard and analog evaluation criteria, and the corresponding value calculation methods of the standard limit given. According to Probability Theory and Limit theory of mathematical statistics, also the Shewhart control charts theory, we suggested the health standards for feature quantity that take the sample mean as the feature’s standard value, and Xc=X±3σ determined by3a criteria to as alarm limit value.In the paper, we analyzed the influences from unit operating conditions (power, head) to monitoring parameters characteristic indices illustrated by the case of vibration monitoring samples under normal operating conditions in unit’s prophase, proposed to partition the operating condition s adopting the method to control the sample variance, and established specific methods and procedures for health samples of each part. Thus to ensure the accuracy of the sample abnormal judgment, and to reduce the sample space dimension (the number of samples) at the same time.Also, characteristics trend forecasting model of hydropower units based on time series variation decomposition was established, and algorithms of trend forecast and performance degradation prediction based on this model were proposed as well. We verified this decomposition model and algorithms on power station monitoring data, results showing that the forecasting trends are in good agreement with monitoring trends, can meet the needs of trend forecasting and performance degradation prediction of monitor characteristic quantities in hydropower units, and are with good usability on the early warning of unit potential abnormalities.Taking vibration of hydropower units as an example,3D standard model (power-head-vibration) for vibration parameters performance degradation assessment of hydropower unit based on LS-SVM was proposed. We can get the judgement whether the unit vibration under current conditions has deviated from normal state once the run-time active power and water head are written into the trained model, and achieve health assessment of unit’s operating status. On the basis of performance degradation time series of vibration parameters, we established LS-SVM based vibration parameters performance degradation prediction model of hydropower unit, and employed the on-site monitoring data of upper guide swing and upper bracket vibration parameters to validate the proposed model. The results showed that the model could assess and predict vibration parameters performance degradation of hydropower units well.In the end, the research and application of integrated fault diagnosis system was introduced illustrated by the case of remote status monitoring and fault diagnosis system of Three Gorges Corporation, and proposed the overall structure of the distributed fault diagnosis system, composed of on site monitoring layer-plant and substation integration layer-the central diagnostic layer. To achieve data communication between different monitoring devices, we standardized the data format. Through the integration of standardized data platform, state information sharing and multi-information fusion diagnosis were achieved. Real-time health assessment and performance degradation prediction of unit operating status can be achieved through the establishment of its healthy sample library, realizing the purpose of the quantitative assessment of equipment health status, and can provide technical basis for the guidance of condition-based maintenance of the unit.更多还原