Soft Sensor For Ball Mill Fill Level Based On Cloud Reasoning And Information Fusion

Ball mill is the key equipment of coal pulverizing which is widely used inthermal power plant. Whether it can operate normally and is running under theoptimum operating condition is an important factor that affecting the efficiencyof coal pulverizing systems. Therefore, an accurate measurement of fill level isessential for energy saving, optimization control and production safety in ballmill system.Due to that ball mill is always working in the rotation and closed state, filllevel of ball mill cannot be directly measured. The indirect method is taken tomeasure it. According to soft sensor method, the measured variables can beestimated through establishing model between auxiliary variables and dominantvariables. The researches show that the acoustic signals and vibration signals ofball mill are variables that are closely related to the fill levels. Thus, acousticsignals and vibration signals signal are selected as auxiliary variables forestablishing system model. Although power spectrum analysis method iscommonly used as feature extraction method for selecting the system input fromauxiliary variables, Mel Frequency Cepstrum Coefficient, which is based on thehuman auditory characteristics, is introduced in this paper as well. It can simulate the human ear identification from the ball mill noise signalssuccessfully and has advantages of strong practicality and convenientcalculation. Thus, it provides a reliable basis for the realization of reflecting ballmill fill levels through acoustic signals.Through analyzing the acoustic signals and vibration signals of ball mill, itis found that there exist strong random and uncertain characteristics. Therefore，cloud model， which has a strong uncertainty processing capabilities isintroduced in this paper. Cloud model system can realize the mapping frominput universe to the output universe. What’s more, virtual cloud andsynthesized cloud can solve the problem of the lack of rule and thesimplification of rule.Considering the limitations of single information, acoustic signals andvibration signals are selected as auxiliary variables to build a measurementsystem for ball mill fill level based on uncertainty reasoning of two-dimensionalcloud model and information fusion. The main research work is as follows.(1) Power spectrum analysis method and MFCC method are used to extractfeatures from bearing vibration signals and acoustic signals collected byacceleration sensor and audio sensor.(2) Backward cloud algorithm is adopted to obtain features for buildingantecedent clouds, then the consequent clouds as well as rule base of reasoningsystem can be acquired.(3) One-dimensional cloud reasoning is employed for modeling measurement system based on the individual acoustic signal or vibration signal.What’s more, virtual cloud algorithm is used to realize the sparse rule reasoningunder incomplete data sample.(4) When both acoustic signal and vibration signal are taken as the inputdata, two-dimensional cloud model is adopted to establish soft sensor model andcomplete formation fusion. After that, synthesized cloud algorithm is used tosimplify the rules.The experimental results show that comparing with one-dimensional cloudreasoning, the measuring accuracy of two-dimensional cloud reasoningexperiment measuring is higher, and has more advantages compared with otherinformation fusion algorithms. The research shows that the proposed methodcan meet the requirement of field measurement application.