Soft Sensor Of Wet Ball Mill Multimode Load Parameters Based On Transfer Learning

Wet ball mill is a kind of high energy consumption equipment which is widely used in these fields of coal,grinding,power,chemical and metallurgical industries.Its running process has the characteristics of time varying parameter,multi scale,high nonlinearity and strong coupling.Limited by physical conditions,technical level and complexity of process mechanism,how to control the optimal operation of the process has been the focus of academic and industrial research.It is one of the key factors to deal with the real-time and accurate detection of wet ball mill load parameters.These parameters include material to ball volume ratio(MBVR),pulp density(PD)and charge volume ratio(CVR).These load parameters represent the working state of the ball mill which can accurately reflect the ball mill load status and play an important role in the quality control,grinding efficiency and energy consumption reduction.Detecting these ball mill load parameters in real-time and accurately is of great significance to improve the grinding quality and productivity,while to reduce the energy consumption and to ensure the safe operation of the industrial process.However,these load parameters are lack of effective detection methods.This problem has seriously restricted the control level and production efficiency of wet ball mill.Soft sensor technology is usually used to solve the problem of process parameters which are difficult to measure directly in the process of industrial production.It is mainly through building the function mapping relation between measurable secondary variable and primary variable,then measuring and estimating those variables which need to be measured.But in the actual operation of wet ball mill,due to the change of the operation task and the setting value,the product type,the process load and so on,it will lead to change the production process conditions and result in the operation of the system showing a characteristic of multimodal.The wet ball mill will appear a lot of unknown modes under the complex multi working condition,and the data distribution and the spatial structure of the ball mill load parameters secondary variable and the primary variable which need to be measured is different under different modes.This is the transfer of the working conditions and the traditional modeling method of ball mill load parameters will not be applicable or can not get satisfactory results.Aiming at the problem of the transfer of the working condition of the wet ball mill,the bearing vibration signal of the experimental ball mill is used as the input secondary variable for the soft sensor model of ball mill load parameters in this paper.We proposed a modeling method of a soft sensor of wet ball mill multimode load parameters based on transfer learning.Transfer learning,as its name implies,is to learn the knowledge system from the source domain and apply it to the target domain to solve the similar problems in the target domain.In this paper,the transfer learning is applied to the soft sensor modeling of the load parameters of wet ball mill,which can solve the problem of the soft measurement of the load parameters of the unknown mode in complex working conditions.In this paper,based on the load monitoring of the grinding process of wet ball mill,with the aim of accurately predicting the wet ball mill load parameters,the soft measurement method of the mill load parameters is developed by the laboratory wet ball mill.The main work of this paper is summarized as follows:(1)The research status and development trend of the wet ball mill load parameters soft sensor are analyzed.We learned the operating principle and grinding process principle of wet ball mill.At the same time,the influence of wet ball mill load parameters on grinding productivity is described.(2)Aiming at the unknown transfer mode of wet ball mill in complex multi working conditions,soft sensor experiment of ball mill load parameters had been experimenting on the laboratorial wet ball mill.During the experiment,the wet ball mill bearing vibration signals were collected,and then used a certain method to preprocess the original signal data.Finally,we analyzed the distribution of the data modal and the difference between different working conditions.(3)The conventional soft sensor methods are used to verify the wet ball mill load parameters,such as the traditional soft sensor method and the real-time learning soft sensor method.(4)Aiming at the transformational working condition situation of the target field modal variable which need to be measured with a small number of labels during the operation of wet ball mill,soft sensing of wet ball mill load parameters by using the offset transfer learning method was proposed in the situation of transfer learning from source domain to target domain.The experimental result of this transfer learning method is compared with the experimental results of traditional soft sensor modeling and real-time learning modeling method.(5)Aiming at the transformational working condition situation of the target field modal variable which need to be measured without any labels during the operation of wet ball mill,soft sensing of wet ball mill load parameters by using the discriminative subspace transfer learning method was proposed in the situation of transfer learning from source domain to target domain.