Research On Online Soft Measurement Method Of Sizing Rate Based On Incremental Learning

The sizing progress improves the fabrics of warp,such as weaving and style.During the process of textile,the sizing percentage affects the sizing quality and weaving production.If the rate is too small,the undersized yarn will lose strength and the yarn's breakage will be increased.Conversely,the warp's elasticity will be damaged with higher sizing percentage,and it will cause too much waste of material.In order to measure the rate in real time,this paper has been studied in the following aspects.According to analyze the process of sizing,the major factor which can report the sizing percentage is determined accurately.In most of textile factory,yarn usually sizes by double dip and double pressure.The eight main factors is given by the analysis of yarn's characteristics,sizing device,slurry properties and drying device,such as tension of yarn,cover factor of yarn,speed of sizing machine,pressure of grouting roller,concentration,viscosity and temperature of slurry and drying temperature.The sizing soft measurement models which based on ELM and AdaBoost.RT algorithms have been built.The model can predict the sizing percentage with the help of eight auxiliary variables which influence sizing progress.Through the simulation experiments,it can be found that the prediction of ELM model is more or less fit on the real data with a few errors.The AdaBoost.RT model's performance is better than ELM model and the number of error data is less.However,both of them can not realize the online measurement of sizing percentage,and the accuracy of measurement cannot fit the actual production requirements.A new algorithm ILMELM is proposed to establish sizing percentage soft sensor model for online and real-time measurement.The algorithm's integration is divided into two parts.The inner one introduces error judgment coefficients to select multiple ELMs which meet the accuracy of requirement.Then use them for integrated sub models.Meanwhile,the external integration is based on incremental learning.When the new data is added into system,the new model will be integrated and updated without the learning of old data.The results of experiments demonstrate that the soft sensor model based on ILMELM has the best accuracy,good generalization and the ability of update online.A new incremental online algorithm is proposed to establish and improve anti-noise performance of online soft sensor model with the dynamic data of sizing percentage.The algorithm includes three parts:redundant data filtering,noise data filtering and soft sensor modeling.Firstly,in order to improve the efficiency of algorithm,incremental learning method is introduced to reduce the redundant.Secondly,the improved mountain method was used to delete the noisy data and increases the accuracy of model.Finally,the algorithm is suitable for the online update of a variety of intelligent algorithms.The experiments show that the new algorithm's MAE and RMSE is lower than the others' and has the better accuracy and anti-noisy performance.That is to say,it can fit the sizing percentage's online measurement.