.png){kind=icon}
.png){kind=icon}
Return
A Hybrid Machine Learning Model for Initial Radius Estimation on Slitter Machine
.png){kind=icon}
PRE
AI
DOI：10.1002/cpe.70429.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Slitter machine is to slit wide film into multi-narrower film through roll-to-roll technology, in which the radius of the unwinding roll is critical for the control system to maintain proper film tension. Automatically obtaining the radius without sensors, especially the initial radius at the start of the transportation, is the main concern of this study, in which getting the initial radius sooner and more precisely is the goal for smoother film transportation. This paper presents a novel hybrid model including variational mode decomposition (VMD), eXtreme Gradient Boosting (XGBoost) and the Grey Verhulst model for the evaluation of the initial radius from raw radius data. From VMD, filtered time series (FTS) of the raw radius data are obtained and then inputted into the XGBoost model and Grey Verhulst model to get two groups of evaluated radius data series, namely, R_XGB (Real XGBoost) and R_GV (Real Grey Verhulst); the synthesized value of R_XGB and R_GV is then the final evaluated radius. Parameters of VMD are optimized using the Q-learning algorithm of reinforcement learning; and the NFQ algorithm of reinforcement learning is applied to find the optimized weights for the composition of R_XGB and R_GV values. The performance of the proposed hybrid model is conducted through two groups of experimental data, and the results show that the proposed model gives the radius estimation in one turn of the unwinding roll with drastically improved precision; concretely, the mean squared error of estimation is reduced by 99.15% and 99.67% compared with the Grey Verhulst and XGBoost models used alone.
Keywords:
eXtreme gradient boosting
Grey Verhulst model
reinforcement learning
slitter machine
stationary stochastic process
variational mode decomposition
Journal
C
IF:
1.5
Papers:
473
Citations:
0