Precision Controlled and Optimized Manufacturing of Sheet Metal Process by Perception and Prediction

To solve problems of the large dimensional and angular errors during robot-aided bending, a bending precision control and optimization method that combines visual detection and model prediction was proposed. The detection principle of positioning errors was analyzed, the training dataset was acquired, and an enhanced sparrow search algorithm (ESSA) based on Tent chaotic mapping was proposed. Then, an ESSA-Elman prediction model of positioning errors was built by optimizing the initial weight and threshold of the Elman neural network. Afterward, the machining strategies with error prediction and compensation were developed for sheet metal parts, and the bending effectiveness was verified. In addition, a machine vision detection and bending platform was established and calibrated. This was followed by compensation experiments of machining errors during bending. Results show that compared with the unoptimized Elman model, the mean absolute error (MAE), mean square error (mse), and root-mse (RMSE) of the ESSA-Elman model separately decrease by 62.8%, 66.7%, and 61.1%, while the coefficient of determination ( $R^{2}$ ) increases by 9.8%. Moreover, the dimensional and angular errors of sheet metal parts separately reduce by 79.7% and 68.1% after error compensation, thus realizing the high-precision bending of sheet metals.