Research And Application Of Type2-fuzzy Logic System In Soft Sensor Modeling Of Ball Mill Fill Level

As a high energy consumption industrial grinding device, ball mill iswidely used in the field of power plant, ore dressing, ceramics, chemical andother industries. The level of the ball mill can’t be measured accurately for therotating and sealing characteristics of the ball mill. The operators often keep theball mill running with a low fill level to avoid the problem of over load causedby higher fill level. Therefore,the efficiency of the ball mill is turned down. Soit’s very important to measure the fill level accurately for improving theefficiency and stability of the ball mill.However, there are many uncertain factors in the ball mill, the traditionalsoft measurement methods such as PLSR, PCR, ELM can’t achieve thedesired effect. So finding an effective method which can deal with randomnessand uncertainty is very important. People began to pay attention to the graysystem, artificial neural network,wavelet theory and fuzzy logic systemand artificial intelligence methods. Among them, the type2-fuzzy logic systemhas the strong ability to deal with uncertainty, and provides a new direction forthe prediction of ball mill.In this paper, the soft model of ball mill was studied. As the type2-fuzzy sets are more powerful than ordinary fuzzy sets in dealing with uncertainty, thetype-2fuzzy logic was introduced to represent the concepts of fill level in ballmill. The main research work is as follows.(1) In the research, the feature of the ball mill’s motion, and the connectionbetween the ball mill level and the operational mechanism was defined.Summarized and compared various methods applied into soft-sensoringtechnology. According to the existence of the strongly nolinear and randomcharacteristics in the process of measuring fill level of ball mill by analyzingvibration signals, the type-2fuzzy logic was introduced to represent the conceptsof fill level in ball mill.(2) Firstly, a fuzzy C–means clustering algorithm was used to partition thespace of the data and compute the parameters of the antecedents. Then we used aleast square method to identify the parameters of the consequents.Simultaneously, the back-propagation algorithm was used for tuning parametersof the antecedents. At last, the soft sensor of the fill level was realized byuncertainty reasoning based on interval Type-2T-S fuzzy logic system.(3) For the difficulty of the interval type-2fuzzy logic systemidentification, the particle swarm optimization algorithm was intorduced for theparameter identification, and built an interval type-2T-S fuzzy logic model topredict the fill level of the ball mill. (4) For the powerful abstraction ability of DBN and the ability of intervaltype2-fuzzy systems, we built an deep fuzzy belief network to predict the filllevel of the ball mill.In order to validate effectiveness of the three models, experiments werecarried out on a lab-scale ball mil. The experiment results show that theforecasting model based on it has better performance than other methods,because it has higher prediction accuracy and better tracking characteristics forreal fill level curve.