基于阳极电流分布的铝电解槽异常诊断

发布时间：2018-05-18 03:08

本文选题：铝电解 + 异常诊断系统　；参考：《北方工业大学》2017年硕士论文

【摘要】：现代铝生产的主要方法是基于氧化铝—冰晶石的电解法。该方法由于内部复杂的物理、化学变化,以及外部各种场的存在,形成了复杂的槽况特征。建立有效的故障诊断系统不但可以提高生产过程中铝的质量和产量,同时也能降低电能消耗,对铝的生产有重要意义。目前,国内外都对铝电解槽槽况诊断技术进行了大量研究,提出了基于解析模型的故障诊断方法和基于知识的故障诊断方法。由于铝生产过程中数据获取比较困难并且我国铝电解工艺和整流设备与国外相比还有一定差距,使一些方法在我国无法推广。针对这一问题,本文研究了一种基于阳极电流和集成神经网络的故障诊断方法。阳极电流中包含了大量的槽况信息,通过对阳极电流的分析,可以为铝电解槽槽况的诊断提供依据。在本文中,首先对阳极电流信号进行频谱分析,通过计算每一段频谱上的香浓熵提取其特征值,作为子神经网络1的输入,然后计算阳极电流的均值、方差、偏度和峰度作为子神经网络2的输入,对子神经网络1和子神经网络2的输出的加权平均值作为决策融合神经网络的输入。由于单神经网络的使用效果和使用者的经验有很大关系,同时当出现新的故障或者有新的特征值时很难扩展,所以本文采用集成神经网络来提高诊断系统的泛化能力。在本文中,首先对子神经网络1和子神经网络2进行并联,使他们之间的诊断相互独立,然后将并联后的网络和决策融合神经网络进行串联来完成对槽况的诊断。本文还通过python建立了一个铝电解槽槽况诊断系统,首先通过网络对采集到的阳极电流进行读取,并存入sqlite数据库,然后对数据进行预处理,提取相应的特征值,最后经过集成神经网络对槽况进行诊断。通过测试,该系统可以完成对铝电解槽槽况的诊断。
[Abstract]:The main method of modern aluminum production is based on alumina-cryolite electrolysis. Because of the complex physical and chemical changes in the interior and the existence of various external fields, the method forms complex channel characteristics. The establishment of an effective fault diagnosis system can not only improve the quality and output of aluminum in the production process, but also reduce the power consumption, which is of great significance to the production of aluminum. At present, a great deal of research has been done on the condition diagnosis technology of aluminum reduction cell at home and abroad, and a fault diagnosis method based on analytical model and a method based on knowledge are put forward. Because of the difficulty of data acquisition in aluminum production and the gap between China's aluminum electrolysis process and rectifier equipment compared with foreign countries, some methods can not be popularized in China. To solve this problem, a fault diagnosis method based on anode current and integrated neural network is studied in this paper. The anodic current contains a great deal of information about the status of the cell. The analysis of the anodic current can provide the basis for the diagnosis of the condition of the aluminum reduction cell. In this paper, the anodic current signal is analyzed by spectrum analysis, and the eigenvalue of the anodic current signal is extracted by calculating the aroma entropy of each section of the spectrum, which is used as the input of the subneural network 1, and then the mean value and variance of the anode current are calculated. Skewness and kurtosis are the input of subneural network 2, and the weighted average of the output of subneural network 1 and subneural network 2 is the input of decision fusion neural network. Since the use effect of single neural network is closely related to the user's experience and it is difficult to extend it when new faults or new eigenvalues occur, the integrated neural network is used to improve the generalization ability of diagnostic system. In this paper, the sub-neural network 1 and the sub-neural network 2 are connected in parallel to make the diagnosis independent of each other, and then the parallel network and the decision fusion neural network are connected in series to complete the diagnosis of the slot condition. In this paper, an aluminum reduction cell condition diagnosis system is established by python. Firstly, the anode current collected is read through the network, then stored in the sqlite database, and then the data is pretreated to extract the corresponding characteristic value. Finally, the slot condition is diagnosed by integrated neural network. Through testing, the system can complete the diagnosis of aluminum reduction cell condition.
【学位授予单位】：北方工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TF821;TP277

【参考文献】