Abstract:

Objective In order to explore the influence of fingers compression force and sewing speed on the sewing deviations during long straight-line sewing process, the collaborative control of the robot and the sewing machine during this process was studied. This contributes to improving the theoretical research on long straight-line intelligent sewing, expanding the convenitional quilting sewing theory and production methods, and would provide certain theoretical references for the research on unmanned sewing methods for clothing.

Method This study focused on nine-square grid quilting, using three common pure cotton fabrics. The long straight-line sewing path was divided into three segments with three repeated sewing. Two soft fingers served as the UR5 arm's end-effector, with force sensors monitoring their normal contact force and feeding speed during sewing. L₂₅ (5²) orthogonal experiments were carried out to investigate the influences of force and speed combinations applied by soft fingers on the sewing deviations. A support vector regression model optimized by grid search (GS), Bayesian optimization (BO), and particle swarm optimization (PSO) was established to predict the sewing deviations.

Results When the sewing speed was set to the gear 2, the optimal combination of force and speed for the soft fingers of the fabric 1^# in the three sewing segments is (24 N, 9.5 mm/s), (24 N, 10.5 mm/s), and (24 N, 10.5 mm/s); for fabric 2^#, the optimal combination is (28 N, 8 mm/s), (22 N, 10.5 mm/s), and (26 N, 10 mm/s); and for fabric 3^#, the optimal combination is (24 N, 9.5 mm/s), (22 N, 10 mm/s), and (22 N, 10 mm/s). For the first sewing segment of fabric 1^#, the influence of the soft finger speed on the sewing deviation is the greatest, followed by the interaction effect. For the second and third sewing segments of fabric 1^#, the soft finger force, speed, and their interaction all have an impact on the sewing deviation. Among them, the interaction has the strongest impact on the second segment, while soft finger speed dominates the third segment. Before optimization, the support vector regression (SVR) model only had a small portion of the predicted values close to the actual values on both the training set and the test set. When the actual values were large, the corresponding predicted value distribution was relatively scattered, deviating significantly from the ideal line. After optimization using GS, BO, and PSO, the performance of the SVR model on the training set and the test set was significantly better than that of the unoptimized model. Among them, the model obtained by optimizing with GS had the best generalization ability and optimization efficiency (root-mean-square error of 0.039 7 mm, optimization time of 26.52 s), and most of the predicted values were concentrated near the ideal line, meaning they were close to the actual values.

Conclusion At the same sewing speed, the optimal combinations of soft finger compression force and sewing speed differ among the three-cycle sewing segments for the same fabric, and also vary across different fabrics for the same sewing segment. For the same fabric, the influence of soft finger force, speed, and their interaction on sewing deviations vary with sewing speed and cyclic sewing segment, with significant differences observed between different fabrics. Therefore, the optimal combinations of robotic arm parameters for different sewing segments of various fabrics under different sewing speeds should be dynamically matched according to the specific fabric, sewing segment, and sewing speed. This approach can effectively reduce sewing deviations and improve sewing accuracy. The optimized support vector regression models outperform the unoptimized ones significantly in both training and test sets. Among them, the model optimized by GS exhibits the best generalization ability and optimization efficiency. Thus, the support vector regression model optimized by GS can relatively accurately and efficiently predict the absolute value of sewing deviation for different sewing segments and robotic arm parameter combinations under varying sewing speeds.

Key words: sewing robot, soft finger, sewing speed, long straight-line sewing, sewing deviation, orthogonal experiment, support vector regression