Development Of Online Real-time Soft-sensing Model Of Tissue Quality Based On Data-driven Method

With the increase in the production and quality requirements of tissue paper,how to produce paper products in a timely,efficient and stable manner,to ensure and improve the quality qualification rate,and to reduce the cost of control is an issue that enterprises need to solve at present.Due to the enterprise existing quality inspection mode(pre-production and post-inspection based on offline instruments and sampling inspections),workers have time lags in the process adjustment of unqualified paper products,and the quality problems of a large number of paper products cannot be resolved in time,causing huge economic losses.Therefore,this study takes a tissue paper mill as an example to establish an online real-time soft-sensing model of tissue paper quality.After collecting a large amount of external data to verify the validity and application value of the model,based on the existing MES system(Manufacturing Execution System)of the enterprise,the online development and industrial application of the model are realized,and the closed-loop refined intelligent production management is realized for the enterprise.Aiming at the problem that the quality of tissue paper cannot be measured online and the quality is delayed,this paper first analyzes and soft-measures the key variables that affect the quality of tissue paper,such as the pulp fiber shape after refining,and the accuracy meets the application requirements.For the direct and indirect factors that affect the physical indicators of base paper quality,the univariate correlation analysis and multivariate feature analysis based on the gradient boosting decision tree(GBT algorithm)are performed,and the important modeling variables for each key physical quality indicator are selected.Then,based on the GBT algorithm,RF algorithm(Random Forest)and Ada Boost algorithm in machine learning,a quality online soft-sensing model is established,and the accuracy of each algorithm is compared through the test error.It is found that the RF algorithm is the best when establishing the online soft-sensing model of tensile strength.MRE(Mean Relative Error)is 6.85%,using the GBT algorithm to build the softness online soft-sensor model with the best accuracy,MRE is 5.44%,using the Ada Boost algorithm to build the bulk online soft-sensor model,the accuracy is the best,MRE is 2.53%,using the RF algorithm to establish the water absorption online The soft measurement model has the best accuracy,MRE is 3.46%,and the RF algorithm has the best accuracy when establishing the online soft measurement model of longitudinal elongation,the MRE is 9.18%.The RF algorithm has the best accuracy when establishing the online soft measurement model of the lateral elongation.the MRE is 6.89%.It shows that the prediction effect of each model is good and meets the error demand of real-time quality inspection.Finally,a large number of production data of more production lines are collected to verify the generality of the model.The results show that the quality of the online soft-sensing model established is still good under different production line structures,has good applicability,and can provide a practical value alternative for the quality inspection of enterprise products.In order to realize the online and real-time soft-quality measurement and industrial application of the model of the paper,to provide enterprises with stable product quality,timely and online detection means,optimize production guidance and stabilize product quality assurance.This article first selects a production line to develop and deploy the established online real-time soft-sensing model of tissue quality.The online test of the model runs well and stably.Then went to the site to debug and verify.After actual operation verification,the model running time was less than 1 minute.The main paper type A bulk thickness soft measurement MRE was 1.19%,the softness MRE was 3.93%,and the main paper type B water absorption MRE was 0.59%,Softness MRE was 3.34%.The results show that most indicators of the main production paper grades have good accuracy,and a few indicators perform very well.They can meet the timeliness requirements and error requirements in production.The model is suitable for online real-time soft measurement,has good industrial application value,and can provide guidance for enterprises to monitor abnormal production and optimize paper machine technology.Finally,for the developed model,a set of online real-time soft-sensing general model process methods for industrial paper quality are proposed.This method expands the scope of application of the model and further realizes the built model by encapsulating and applying the general model.Reuse value and industrial value.