【作者】 刘震宇；

【导师】 王福利；

【作者基本信息】 东北大学 ， 控制理论与控制工程， 2021， 硕士

【摘要】 电熔镁砂是重要的工业材料,应用广泛。利用电熔镁炉的电极和炉料间的电弧产生的热量使炉料熔化,再经过冷却,破碎,分拣等操作,从而获得高纯度的电熔镁砂晶体。电熔镁炉的熔炼过程机理复杂,三相电极之间具有强耦合性,电弧之间并且存在很强的非线性,所以很难建立准确的数学模型。除此之外,工业现场环境恶劣,目前对异常工况的识别主要依靠操作员的经验,主观性强,识别率低,往往不能及时准确的调节控制电极位置,有可能导致生产过程资源浪费严重、产品品位不佳、频繁发生异常工况,严重情况下会危害到操作人员的安全。由于安全生产在电熔氧化镁熔炼过程中具有重要意义,因此近年来异常工况识别的研究已成为人们关注的热点。在电熔镁熔炼过程中,会生成三种类型的异构信息,包括图像,声音和电流信息,并且不同的异常情况会反映在不同类型的信息中。为了提高特征表示和异常工况识别的准确性,本文提出了一种结合多源异构信息和领域知识的异常工况识别方法。基于领域知识分析与每种异常工况相关的特征,然后针对每种类型的异构信息通过适当的特征提取方法进行提取。此外,通过融合针对不同异常情况的不同类型的异构信息,建立了四个二分类异常工况识别网络,并根据实际生产过程中每种异常工况的发生频率进行集成,得到最终的异常工况识别网络,从而可以显着改善识别结果的准确性。仿真结果表明,与其他方法相比,本文所提方法能够准确地识别异常工况,并具有较好的性能。在此基础上,进一步提出了基于异构信息的电熔镁炉自愈控制方法。本文的主要工作如下:(1)针对电熔镁炉熔炼过程中的图像信息,使用深度学习方法中的卷积神经网络进行特征提取。(2)针对电熔镁炉熔炼过程中的声音信息,使用短时傅里叶变换进行特征提取。(3)针对电熔镁炉熔炼过程中的电流信息,将电流信号按周期分割,计算电流周期的均值,方差,跟踪误差,最大值和最小值作为电流信号的特征。(4)结合电熔镁炉领域知识,分析了电熔镁炉运行过程中常发生的三种异常工况的特点,针对每种异常工况,融合不同种类的信息特征并且建立特定的二分类识别神经网络。(5)基于图像、声音和电流三种异构信息以及深度神经网络,提出了基于异构信息的电熔镁炉异常工况识别方法。(6)在该异常工况识别方法的基础之上,提出了基于异构信息的电熔镁炉自愈控制方法。更多还原

【Abstract】 The electro-fused magnesium is an important industrial material with a wide range of applications.The fused magnesia furnace is the main equipment for the production of the electro-fused magnesium.The raw materials are melted by the heat generated by the arc between the electrode and the raw materials,and then cooled,crushed,and sorted to obtain high-purity electro-fused magnesium crystal.In the electro-fused magnesia smelting process,there is a strong coupling between the three-phase electrodes,and the AC arc has a strong nonlinearity,which makes the smelting process mechanism very complicated,so it is difficult to establish an accurate mathematical model.In addition,the industrial site environment is harsh,the current identification for abnormal conditions mainly relies on the experience of the operator,which is highly subjective and has a low recognition rate,and they often cannot adjust the control variables in time and accurately,which ultimately leads to high energy consumption in the production process,poor product quality,and frequent abnormal working conditions.In severe cases,it may even pose a great threat to the safety of operators.Since the safe production is of great significance in the electro-fused magnesia smelting process,researches on abnormal condition identification have been paid much attentions in recent years.During the electrofused magnesia smelting process,three types of heterogeneous information are generated,including image,sound and current information,and different abnormal conditions are reflected in different types of information.In order to improve the accuracy of feature representation and abnormal condition identification,a new abnormal condition identification method is presented by combining multi-source heterogeneous information and domain knowledge in this thesis.The features related to each abnormal condition are analyzed based on domain knowledge and then extracted by appropriate feature extraction methods with respect to each type of heterogeneous information.Furthermore,four binary abnormal condition identification networks are established by fusing different types of heterogeneous information for different abnormal conditions,and they are integrated according to the frequency of each condition in actual production process to form the final abnormal condition identification network,which can significantly improve the accuracy of recognition results.The simulation results show that the proposed method can identify abnormal conditions accurately and acquire better performance than other methods.On the basis of this method,a self-healing control method based on heterogeneous information is further proposed.The main contents of this thesis are as follows:(1)For the image information in the electro-fused magnesia smelting process,the convolutional neural network in the deep learning method is used for feature extraction.(2)For the sound information in the electro-fused magnesia smelting process,Short-time Fourier transform is used for feature extraction.(3)For the current information in the electro-fused magnesia smelting process,it is divided into periods,and the average value,variance,tracking error,maximum value and minimum value of each current period are calculated as the features of the current signal.(4)Combining the field knowledge of fused magnesia furnace,this thesis analyzes the characteristics of three abnormal conditions that often occur during the operation of fused magnesia furnace.For each abnormal condition,different types of information features are combined and a specific binary identification neural network is established.(5)A multi-source information fusion method for identifying abnormal conditions in the electro-fused magnesia smelting process based on image,sound and current information and deep neural network is proposed.(6)On the basis of the abnormal condition identification method,a self-healing control method based on heterogeneous information is proposed.更多还原