Prediction Of Particle Moisture Content In Top-sprayed Fluidized Bed Granulation Process

With the development of science level,the fluidized bed granulation equipment which is produced by traditional way can not statisfy the requirements of automation and information production.In the granulation process,the moisture content information of particles can not be obtained in real time,and the influencing factors of the moisture content of particles in the granulation process can not be accurately grasped,which may lead to the drug efficacy not reaching the expectation or even the failure of granulation batch.Therefore,it is of great significance to obtain the moisture content of particles in real time and accurately.In view of the difficulty of data acquisition in the fluidized bed granulation process,the existing fluidized bed equipment is designed and modified,and the near-infrared probe and process parameter sensor are installed.In order to eliminate the influence of material powder on the near-infrared probe on the spectral data,a set of near-infrared probe purging device is designed,so as to establish the data acquisition platform in the fluidized bed granulation process.The results show that the data acquisition platform can collect process parameter data and near infrared spectrum data in real time.In view of the problems of not real-time monitoring of particle moisture data in granulation process and low signal-to-noise ratio of original near-infrared spectrum,the near-infrared spectrum is pretreated by the combination of normalization and SG convolution smoothing method;the band selection is carried out by random forest method,and 125 bands are reduced to 60;the processed spectrum data are modeled offline by PLS,PSO-SVR and pso-krr respectively The results show that the RMSE of pso-krr model is 0.210,the prediction result is the best,and the online transfer of the model is completed successfully.Aiming at the problem of not fully understanding the internal mechanism of the fluidized bed granulation process and the relatively expensive near-infrared probe equipment,the moisture data of particles are obtained in real time,and the relationship between the technological parameters of fluidized bed granulation and the moisture content of particles is analyzed in a visual way.The most important feature is the material temperature.BP neural network,xgboost and stacking model fusion algorithm are used to model the process parameters and particle moisture content respectively.The results show that the proposed stacking model fusion algorithm obtains the best results with RMSE of 0.101 and 0.104 on the verification set and prediction set respectively.The online verification of the stacking model fusion algorithm is carried out with RMSE of 0.261,which proves the replacement of process parameters The feasibility of near infrared probe to predict the moisture content of particles reduces the cost.