基于Hadoop2.0云平台的风力发电机组故障诊断技术研究

发布时间：2018-06-09 21:04

本文选题：风力发电机 + 故障诊断　；参考：《沈阳工业大学》2017年硕士论文

【摘要】：随着风能产业的高速发展,风力发电机的使用规模越来越大,如何维护机组的安全运转受到了越来越多人的关注。各类风机故障中,传动系统零部件的故障往往造成风力发电机组停机维修时间最长,也是致使风力发电机的维护成本高居不下的重要原因之一,对其进行及时准确的诊断就显得尤为重要。因此本文以风力发电机组传动系统轴承为例,提出了一种基于深度信念网络DBN的风机轴承故障诊断算法,并通过将该算法在云计算平台并行化处理来提高算法的处理速度。本文主要从两个方面着手研究:一方面来说,诊断技术采用深度学习网络作为故障数据的特征表征,利用深度信念网络(DBN)技术作为网络构建的深层结构,能够高度拟合任何非线性信号的特征,并从每个隐藏层中输入的数据中学习获得更高阶的相关性特征,最后连接一个特征提取器在输出端级联处理输出的信息。本文通过实验验证了深度信念网络DBN在数据处理方面的优异表现,利用该网络模型提出的风力发电机轴承故障诊断方法可以直接通过将原始振动数据归一化处理后进行无监督学习,当整个DBN训练完成后,利用标签层的信息在最顶端进行有监督故障识别,避免了繁杂的故障特征人为经验参与选取的情况,从而使得整个诊断更加智能。另一方面,传统的信号分析方法在处理数据规模持续增长的振动数据时,会遇到计算能力上的瓶颈,而云计算在大数据计算和网络存储方面有着卓越的性能表现,并且在Hadoop平台上,基于Spark实现对振动数据的并行处理方法,具有很好的容错性与自动均衡负载特性,能够有效解决风机组采集到的振动数据处理速度过慢,数据使用不完全,有效信息流失严重的问题。本文实现的基于Spark的风机组轴承故障诊断算法并行运算,在Hadoop2.0云平台上的实验表明,算法的各项并行指标良好,能够满足故障诊断处理任务需求。
[Abstract]:With the rapid development of wind energy industry, the scale of wind turbine is becoming larger and larger, and more people pay attention to how to maintain the safe operation of the unit. Among all kinds of fan faults, the failure of transmission system parts often results in the longest downtime of wind turbine generator, which is one of the important reasons for the high maintenance cost of wind turbine generator. It is very important to make timely and accurate diagnosis. This paper presents a fault diagnosis algorithm for wind turbine bearing based on deep belief network DBN, and improves the processing speed by parallelizing the algorithm in cloud computing platform. This paper mainly studies from two aspects: on the one hand, the deep learning network is used as the feature representation of the fault data, and the deep belief network (DBN) technology is used as the deep structure of the network construction. It can highly fit the features of any nonlinear signal and learn from the input data in each hidden layer to obtain higher-order correlation features. Finally, a feature extractor is connected to process the output information in cascade at the output end. In this paper, the excellent performance of DBN in data processing is verified by experiments. Using the network model, the bearing fault diagnosis method of wind turbine generator can be directly studied by normalizing the original vibration data, when the whole DBN training is completed. The information of label layer is used for supervised fault identification at the top, so that the complicated fault features can be selected by human experience, thus making the whole diagnosis more intelligent. On the other hand, the traditional signal analysis method will meet the bottleneck of computing power when dealing with the vibration data with increasing data scale, while cloud computing has excellent performance in big data computing and network storage. And on Hadoop platform, the parallel processing method of vibration data based on Spark is realized. It has good fault tolerance and automatic load balancing characteristics, and can effectively solve the problem of slow processing speed and incomplete use of vibration data collected by wind turbine. A serious problem with the loss of effective information. The parallel calculation of bearing fault diagnosis algorithm based on Spark is realized in this paper. The experiment on Hadoop 2.0 cloud platform shows that the parallel indexes of the algorithm are good and can meet the needs of fault diagnosis and processing task.
【学位授予单位】：沈阳工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TM315;TP393.09

【相似文献】