In the hot strip rolling (HSR) process, accurate prediction of bending force can improve the control accuracy of the strip crown and flatness, and further improve the strip shape quality. In this paper, six machine learning models, including Artificial Neural Network (ANN), Support Vector Machine (SVR), Classification and Regression Tree (CART), Bagging Regression Tree (BRT), Least Absolute Shrinkage and Selection operator (LASSO), and Gaussian Process Regression (GPR), were applied to predict the bending force in the HSR process. A comparative experiment was carried out based on a real-life dataset, and the prediction performance of the six models was analyzed from prediction accuracy, stability, and computational cost. The prediction performance of the six models was assessed using three evaluation metrics of root mean square error (RMSE), mean absolute error (MAE), and coefficient of determination (R²). The results show that the GPR model is considered as the optimal model for bending force prediction with the best prediction accuracy, better stability, and acceptable computational cost. The prediction accuracy and stability of CART and ANN are slightly lower than that of GPR. Although BRT also shows a good combination of prediction accuracy and computational cost, the stability of BRT is the worst in the six models. SVM not only has poor prediction accuracy, but also has the highest computational cost while LASSO showed the worst prediction accuracy.

In the hot rolling process, the prediction of strip crown is the key factor to improve the flatness quality of the strip. However, the traditional prediction method can only provide prediction values, but does not quantitatively evaluate the prediction error and stability. While Gaussian process regression (GPR) provides full probability prediction and estimates the uncertainty in the prediction. Therefore, for the first time, GPR is applied to predict strip crown. Furthermore, considering the negative influence of unavoidable outliers in measurement data, this article proposes an improved local outlier factor (LOF) algorithm to calculate the weights. And a novel Weight-GPR based on improved LOF prediction model is established. The proposed model not only retains the effective information of outlier values, but also avoids the negative influence brought by outlier values. The prediction experiments based on the real world production line data show that the proposed model can be successfully applied to the prediction of the strip crown in hot rolling process. Also, the performance of the proposed model is compared with typical GPR, ANN and SVM, and the results demonstrate that the Weight-GPR based on the improved LOF model provides better prediction accuracy and stability.

Precise and rapid grasping recognition based on machine vision is one of the challenging problems for intelligent robots. As a nonlinear network for recognition, the stochastic configuration network (SCN) is considered as a promising method due to its universal approximation capability and fast modelling property, therefore, in this paper, SCN with depth image is applied to robotic grasping recognition for the first time. Moreover, in order to obtain better grasping performance, a novel autoencoder-based SCN (SCN-AE) is proposed to extract the higher level features of the captured image. Finally, a field programmable gate array (FPGA) framework for implementing SCN models is designed and analyzed, which can effectively balance the recognition accuracy and resource utilization. Experimental results on Cornell grasping dataset proved that the proposed SCN-AE method achieved higher accuracy and better stability in robotic grasping recognition task, and the designed FPGA framework implemented the SCN and SCN-AE recognition models successfully. The average recognition accuracy of the proposed SCN-AE based on FPGA reaches 91.935%.

The implementation of neural network regression prediction based on digital circuits is one of the challenging problems in the field of machine learning and cognitive recognition, and it is also an effective way to relieve the pressure of the Internet in the era of intelligence. As a nonlinear network, the stochastic configuration network (SCN) is considered to be an effective method for regression prediction due to its good performance in learning and generalization. Therefore, in this paper, we adapt the SCN to regression analysis, and design and verify the field programmable gate array (FPGA) framework to implement SCN model for the first time. In addition, in order to improve the performance of the SCN model based on the FPGA, the implementation of the nonlinear activation function on the FPGA is optimized, which effectively improves the prediction accuracy while considering the utilization rate of hardware resources. Experimental results based on the simulation data set and the real data set prove that the proposed FPGA framework successfully implements the SCN regression prediction model, and the improved SCN model has higher accuracy and a more stable performance. Compared with the extreme learning machine (ELM), the prediction performance of the proposed SCN implementation model based on the FPGA for the simulation data set and the real data set is improved by 56.37% and 17.35%, respectively.

In the hot strip rolling process, accurate prediction of bending force can improve the control accuracy of the strip flatness and further improve the quality of the strip. In this paper, based on the production data of 1300 pieces of strip collected from a hot rolling factory, a series of bending force prediction models based on an extreme learning machine (ELM) are proposed. To acquire the optimal model, the parameter settings of the models were investigated, including hidden layer nodes, activation function, population size, crossover probability, and hidden layer structure. Four models are established, one hidden layer ELM model, an optimized ELM model (GAELM) by genetic algorithm (GA), an optimized ELM model (SGELM) by hybrid simulated annealing (SA) and GA, and two-hidden layer optimized ELM model (SGITELM) optimized by SA and GA. The prediction performance is evaluated from the mean absolute error (MAE), root-mean-squared error (RMSE), and mean absolute percentage error (MAPE). The results show that the SGITELM has the highest prediction accuracy in the four models. The RMSE of the proposed SGITELM is 11.2678 kN, and 98.72% of the prediction data have an absolute error of less than 25 kN. This indicates that the proposed SGITELM with strong learning ability and generalization performance can be well applied to hot rolling production.