Echtzeitfähige Modellprädiktive Regelung des Schusseintrags beim Luftdüsenweben

  • Real-time model predictive control of weft insertion during air-jet weaving

Wu, Tong; Abel, Dirk (Thesis advisor); Trimpe, Johann Sebastian (Thesis advisor)

Aachen : RWTH Aachen University (2023)
Dissertation / PhD Thesis

Dissertation, Rheinisch-Westfälische Technische Hochschule Aachen, 2023


The efficiency of the air-jet weaving machine and the quality of the fabric products depend mainly on the settings of the machine control system. The stochastic behavior of the weft thread during weft insertion poses a significant challenge for parameterization. Closed-loop control of the weft insertion that reacts to weft-specific behavior holds considerable potential for improving the air-jet weaving process. The braking process should be optimized using a highly dynamic yarn brake regarding its significant influence on fabric quality. Therefore, the present work targets an investigation of a closed-loop control system for braking during air-jet weaving. In addition to determining the optimal manipulated variable trajectory, model predictive control allows for the explicit consideration of constraints and can thus avoid the drop in yarn tension and backward movement of the yarn tip during braking. Because of the fast dynamics during braking, the control system’s real-time capability must be ensured. First, a real-time capable process model is developed, which describes the relevant physical relationships during braking. For this purpose, the weft thread is investigated concentrated-parametrically during the entire weft insertion process and distributed parametrically during braking. The aerodynamic forces, as well as the braking force, are modeled in this process. By applying model order reduction techniques appropriate to each of the two modeling methods, two process models are derived for use within the predictive controller. Furthermore, the current position of the yarn tip is required in the model predictive control. This position is measured at each point in time through a camera-based measurement system consisting of two high-speed cameras and a real-time capable image processing system. An automatic adjustment of the parameters in the image processing system enables a robust application of the designed image processing method. Finally, real-time model predictive controllers are developed based on the two process models and evaluated in closed-loop control. In addition to the real-time capable controllers, explicit model predictive control and approximated model predictive control using machine learning offers the potential to provide comparable control quality with reduced computation time. The validation on an air-jet weaving machine proves that the presented control approaches can control the speed of the yarn tip without a backward movement in real time. In this way, the uniformity of weft insertions can be increased, which offers further potential for avoiding machine downtime.