基于云推理信息融合的球磨机料位软测量

发布时间：2018-03-31 03:23

本文选题：球磨机料位　切入点：梅尔频率倒谱系数　出处：《太原理工大学》2015年硕士论文

【摘要】：钢球磨煤机是国内广泛用于火力发电厂制粉系统的关键设备。其是否能够正常运行并且是否运行在最佳工况是影响制粉系统工作效率的重要因素。因此，球磨机料位的准确测量是实现优化控制、安全生产和节能降耗的关键。由于球磨机一般工作在旋转和密闭状态，无法直接进行料位测量，一般通过间接法进行检测。根据软测量思想，可以通过建立辅助变量及其特征参数与主导变量之间的模型来估计待测变量的值。研究发现，球磨机噪音信号及振动信号是与球磨机料位变化密切相关的变量，所以本研究以球磨机噪音信号及振动信号作为辅助变量，建立软测量模型。在对软测量模型输入辅助信号进行分析与处理时，常用的是功率谱分析方法，本文引入梅尔频率倒谱系数，它是基于人耳听觉特性提出的，能够更好的模拟人耳对球磨机噪音信号的识别，并且具有计算方便且实用性较强的优点，为实现通过球磨机噪音信号反映料位提供了可靠依据。通过对球磨机的振动和振声信号进行分析，发现其具有强随机和不确定性的特点，因此本文引入云模型，它对不确定性问题具有很强的处理能力。云模型系统能够实现输入论域到输出论域的函数映射，并且以云理论为基础的虚拟云、综合云算法可以解决规则缺失及规则精简的问题。针对单一信息反映料位的局限性，以球磨机噪音信号及振动信号作为系统输入，采用基于二维云模型不确定性推理信息融合的方法建立球磨机料位软测量系统。本文的主要研究工作包括： (1)针对加速度传感器和音频传感器采集到球磨机轴承振动信号和噪声信号，分别采用功率谱分析方法和MFCC方法进行信号处理与分析； (2)根据逆向云算法得到构建云推理系统前件的概念云模型数值特征，，并结合推理机制给出相应的后件云参数，完成云推理规则库的建立； (3)以单独的音频信号或振动信号作为辅助变量，采用一维云推理建立软测量模型，并利用虚拟云算法完成不充足样本类下的稀疏规则推理； (4)以两个传感器下的信号为辅助变量，结合二维云推理建立软测量系统，实现信息融合，并利用综合云算法进行规则精简。实验结果表明，二维云推理实验相对于一维云推理实验，其测量精度更高，并且与其他信息融合算法相比也具有一定的优势。本方法的测量精度能够满足现场测量应用的需求。
[Abstract]:Ball mill is the key equipment widely used in the pulverizing system of thermal power plant in our country. Whether it can run normally and in the best working condition is an important factor affecting the efficiency of pulverizing system. Accurate measurement of material level of ball mill is the key to realize optimal control, safe production and energy saving and consumption reduction. Because the ball mill generally works in the state of rotation and sealing, it can not measure the material level directly, and generally it is detected by indirect method. The values of the variables to be tested can be estimated by establishing a model between the auxiliary variables, their characteristic parameters and the dominant variables. It is found that the noise and vibration signals of the ball mill are closely related to the change of the material level of the ball mill. So in this study, the noise and vibration signals of ball mill are taken as auxiliary variables, and the soft sensor model is established. In the analysis and processing of the input auxiliary signal of the soft sensor model, the power spectrum analysis method is commonly used. In this paper, the Mel frequency cepstrum is introduced, which is based on the auditory characteristics of the human ear. It can better simulate the recognition of the noise signal of the ball mill by the human ear, and has the advantages of convenient calculation and strong practicability. It provides a reliable basis for reflecting the material level through the noise signal of ball mill. Through the analysis of vibration and vibration signal of ball mill, it is found that it has the characteristics of strong randomness and uncertainty, so the cloud model is introduced in this paper. It has a strong ability to deal with uncertain problems. The cloud model system can realize the functional mapping from input domain to output domain and the virtual cloud based on cloud theory. Synthesis cloud algorithm can solve the problem of rule deficiency and rule reduction. Aiming at the limitation of single information reflecting material level, the noise signal and vibration signal of ball mill are used as system input. The soft sensor system of ball mill material level is established by using the method of fusion of uncertain reasoning information based on two-dimensional cloud model. The main research work of this paper is as follows:. 1) aiming at the vibration signal and noise signal of ball mill bearing collected by accelerometer and audio sensor, the power spectrum analysis method and MFCC method are used to process and analyze the signal. (2) according to the reverse cloud algorithm, the numerical features of the conceptual cloud model of the former part of the cloud inference system are obtained, and the corresponding cloud parameters of the latter part are given in combination with the reasoning mechanism, and the cloud inference rule base is established. (3) taking individual audio signal or vibration signal as auxiliary variable, using one-dimensional cloud reasoning to establish soft sensor model, and using virtual cloud algorithm to complete sparse rule reasoning under insufficient sample class. Taking the signal under two sensors as the auxiliary variable and combining with two-dimensional cloud reasoning, a soft sensor system is established to realize information fusion, and the rules are reduced by using the synthetic cloud algorithm. The experimental results show that the measurement accuracy of two-dimensional cloud reasoning experiment is higher than that of one-dimensional cloud reasoning experiment. Compared with other information fusion algorithms, the measurement accuracy of this method can meet the needs of field measurement applications.
【学位授予单位】：太原理工大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TQ051.9

【参考文献】