基于多源信号融合技术的球磨机负荷预测方法研究

发布时间：2018-05-10 07:38

本文选题：磨机负荷 + 特征提取　；参考：《江西理工大学》2017年硕士论文

【摘要】：球磨机具有操作简单、制造成本低、破碎比大、既可用于湿磨又可用于干磨等诸多优点,广泛应用在玻璃、陶瓷、水泥、化工、矿山等领域。但是,球磨机磨矿过程存在多变量相互制约、强耦合、滞后时间长等缺点,造成其筒体内部负荷参数无法显性描述和实时控制,难于充分发挥磨机的实际效能。因此,实现磨机内部负荷的有效预测,使球磨机运行在最佳工况状态,是提高磨矿效率、降低生产成本的根本任务之一。本文以试验球磨机为研究对象,通过经验分析、实验探究、信号处理相结合的方法,采用多种传感器分别检测球磨机轴承振动信号、筒体磨音信号和主电机电流信号,应用最优融合集和D-S证据理论的多源信息融合技术,对磨机负荷的多源信号特征提取与预测方法进行了深入研究,实现了磨机内部负荷状态参数的有效预测。主要研究结果为:首先,针对球磨机耗能高、产量低、噪音大等问题,通过经验分析得出磨矿过程的主要影响因素和信号检测方法;搭建了球磨机多源信号检测系统,采用单因素变量法进行了磨矿实验,以加球量、给料量、入料粒度分布、球配比为输入参数,以能耗和-200目产率为评价指标,相关实验结果表明,不同磨机负荷参数可划分为欠负荷、正常负荷、过负荷三种状态。其次,为了对三种状态的多源信号进行特征提取和识别,采用了小波变换技术分别对球磨机振动、磨音信号进行特征提取,得到振动特征信息为信号的均值、方差和频率段能量值,磨音特征信息为信号的A计权总声压级和A计权倍频程声压级;通过对比分析不同工况下信号特征信息值的欧氏距离,结果表明与单一信号相比,多源信号能更准确、更快速的对磨机负荷进行识别;通过对不同时间段的电流信号进行均值化处理,得出随着磨机负荷的增加,电流值呈先增加后减小的趋势。最后,针对磨机负荷预测中的检测信号存在高冲突、强突变、低相关的问题,采用了改进后的最优融合集算法,对同类信号在不同时间段的检测数据进行融合,结果表明该方法能有效剔除高冲突信息;采用改进后的D-S证据理论融合规则,提出了一种磨机负荷的多源异类信号特征层融合方法,并通过实例验证和不同算法对比分析,表明该方法应用于磨机负荷预测时,得到的融合结果置信度更高、收敛速度更快、稳定性更强。综上所述,通过单因素变量磨矿试验和多源信号特征提取与识别,采用最优融合集和D-S证据理论建立多源信号特征层融合方法,对磨机负荷预测具有较强的实用性及可靠性,也可为其它选矿设备的节能降耗提供设计新思路。
[Abstract]:Ball mill has many advantages, such as simple operation, low manufacturing cost, large crushing ratio, and can be used in wet grinding and dry grinding. It is widely used in glass, ceramics, cement, chemical industry, mine and so on. However, the grinding process of ball mill has many disadvantages, such as multi-variable mutual restriction, strong coupling, long lag time and so on. As a result, the internal load parameters of ball mill can not be explicitly described and real-time controlled, and it is difficult to give full play to the actual efficiency of the mill. Therefore, it is one of the fundamental tasks to improve the grinding efficiency and reduce the production cost to realize the effective forecasting of the internal load of the mill and to make the ball mill run in the best working condition. This paper takes the test ball mill as the research object, through the experience analysis, the experiment inquiry, the signal processing method, uses the many kinds of sensors separately detects the ball mill bearing vibration signal, the cylinder body grinding sound signal and the main motor current signal. Based on the optimal fusion set and the multi-source information fusion technique of D-S evidence theory, the multi-source signal feature extraction and prediction method of the mill load is deeply studied, and the effective prediction of the internal load state parameters of the mill is realized. The main results are as follows: firstly, aiming at the problems of high energy consumption, low output and high noise, the main influencing factors and signal detection methods of grinding process are obtained by empirical analysis, and the multi-source signal detection system of ball mill is built. The grinding experiments were carried out by single factor variable method. The parameters of ball addition, feed rate, particle size distribution, ball ratio, energy consumption and -200 mesh yield were used as input parameters. Different mill load parameters can be divided into three states: underload, normal load and overload. Secondly, in order to extract and recognize the features of multi-source signals in three states, wavelet transform technology is used to extract the feature of ball mill vibration and grinding sound signal respectively, and the characteristic information of vibration is obtained as the mean value of the signal. Variance and energy value of frequency band, A weighted total sound pressure level of signal and A weighted frequency doubling range sound pressure level, Euclidean distance of signal characteristic information under different working conditions are analyzed, the results show that compared with single signal, The multi-source signal can identify the mill load more accurately and quickly. Through the average processing of the current signal in different time period, it is concluded that the current value increases first and then decreases with the increase of mill load. Finally, aiming at the problems of high conflict, strong mutation and low correlation in the detection signal of mill load forecasting, the improved optimal fusion set algorithm is adopted to fuse the detection data of the same kind of signal in different time periods. The results show that this method can effectively eliminate the high conflict information, adopt the improved D-S evidence theory fusion rule, propose a multi-source and heterogeneous signal feature layer fusion method for mill load, and verify it by an example and compare different algorithms. It is shown that the proposed method has higher confidence, faster convergence speed and stronger stability when it is applied to mill load forecasting. To sum up, through single-factor variable grinding test and multi-source signal feature extraction and recognition, the optimal fusion set and D-S evidence theory are adopted to establish multi-source signal feature layer fusion method, which has strong practicability and reliability for mill load forecasting. It can also provide a new design idea for energy saving and consumption reduction of other mineral processing equipment.
【学位授予单位】：江西理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TD453

【参考文献】