A Bayesian-based Approach To Drug Coating Utilization

Nowadays,with the rapid development of next-generation information and communication technologies such as mobile Internet,the Internet of Things,and big data,as well as the continuous update of advanced manufacturing technologies,industrial changes with intelligent manufacturing as the core have emerged worldwide.It is hoped to solve the pain points of manufacturing production through industrial intelligent manufacturing,improve production efficiency,reduce labor costs,and bring production efficiency to the manufacturing industry.At present,the intelligent manufacturing of China’s pharmaceutical industry is still under development,which includes two major aspects of drug research and development and drug production process,including drug coating,which is also an important production process in the pharmaceutical industry.The research content of this article is the study on the utilization rate of materials reported by drugs.Most existing studies have optimized the production process from the perspective of experimental design and analysis of variance.However,in the actual production process of drug coatings,the method of manually controlling production parameters is adopted,which is entirely dependent on manual experience.Therefore,this paper intends to build a prediction model of production coating utilization rate based on observational data,rather than experimental data.The basic idea of this article is based on the time series data of observational data during production.First,I explore the production raw data and check the quality of the data,and I use the time series feature extraction method based on the logic of the actual production process.And I visualize the relationship between the production stage and various production parameters to compress time-dependent data to construct feature variables.Next,a Bayesian model and a linear regression model were established to predict the drug production coating rate.Then,I compare the empirical results of the two models and analyze the advantages and disadvantages of the two models,because the model involves clustering,this paper also tried the method of clustering and then regression to ensure the richness of the article.Finally,the deficiencies and further research directions of this paper are put forward.There are three innovations in this article.The first is analysis innovation is the analysis perspective innovation.The existing literature on coating process optimization uses the method of experimental design to obtain orthogonal experimental data sets,and then analyzes by means of analysis of variance and multiple regression;in this paper,the observation data is used for analysis and modeling to avoid experimental costs And practical difficulties.The second is to build features based on time series data.This article needs to reduce the production characteristics from the time series data of process parameters,while compressing the time series data dimensions while retaining important features.This article mainly adopts the idea of piecewise aggregation and linear fitting method to complete the feature extraction of each process parameter in the production stage.The third is innovation in modeling ideas.In this paper,a Bayesian Collaborative Model is established based on Bayesian ideas,which is an interdisciplinary application of the Bayesian Collaborative Model.In the field of image segmentation,the Markov random field prior is used to encourage information sharing,and the Gibbs sampling method is used to sample the posterior distribution to obtain more accurate and robust results.In this paper,the algorithm is migrated to the field of pharmaceutical coating industry production.In the case of a small number of drug production records,the goal of separate modeling by drug category is completed,which encourages information sharing in the drug production process and has important research value And practical significance.The following conclusions can be obtained through model demonstration: First,the hyperparameter K of the Bayesian model is equal to three which is selected based on the BIC criterion.Secondly,through 10-fold cross-validation,it is found that the linear regression fitting results are unstable and have large differences,which indicates that linear regression is not suitable for the research data in this paper.Third,the residuals after fitting the Bayesian model are less volatile than the linear residuals,and the regression parameters are closer to the true distribution.