Elevating Prediction Performance for Mechanical Properties of Hot-Rolled Strips by Using Semi-Supervised Regression and Deep Learning

Abstract

In the present work, to solve the problem of the lacking enough labeled training data for deep learning, a safe semi-supervised regression supporting Bayesian optimization deep neural network (SAFER-BODNN) model was proposed to establish mechanical property prediction model of hot-rolled strips. The Pearson correlation coefficient was applied to reduce the data dimension. The safe semi-supervised regression was implemented to add the pseudo labels to the unlabeled data for training dataset expansion. The deep neural network was trained with Bayesian optimization to determine the optimal hyper-parameters of the network. The results show that the SAFER-BODNN model achieves good accuracy for mechanical property prediction of hot-rolled strips with R of 0.9610 for yield strength, 0.9682 for tensile strength, and 0.8619 for elongation, respectively. Compared with the deep neural network trained on the labeled dataset, the SAFER-BODNN model obtains stable smaller predicted errors. Among all the variables, C content and Mn content have large influence on the yield strength and tensile strength, coiling temperature has the largest influence on the elongation. The investigation makes full use of unlabeled data to elevate the accuracy of the deep neural network, and also provides a way for deep learning modeling when the data are insufficient.

Advancements in Individual Tree Detection and Forest Structural Attributes Estimation From LiDAR Data: MSITD and SAFER Approaches

“…Although the results, in general, showed the superiority of SVM over the RF and CB algorithms, at certain study areas, on average, the RF and CB algorithms performed better. The better performance of the semi‐supervised algorithms compared to conventional algorithms in the face of the limited training data has also been reported in other studies (Han et al., 2018; Levatić et al., 2018; Wu et al., 2020). This can be explained by the main feature of these algorithms, which is to do with the use of the unlabeled data to capture a better representation of the data, leading to improved performance.…”

“…In this study, given the limited size of the reference ground sample data, we utilized a semi‐supervised algorithm called SAFER (SAFE semi‐supervised Regression), which has specifically been designed to address regression problems with limited sample data (Li et al., 2017; Syed et al., 2021; Wu et al., 2020). The SAFER algorithm aims to mitigate the issue of the mislabeled predictions within the automated learning process of the semi‐supervised regression models (Li et al., 2017).…”

Online Prediction of Mechanical Properties of the Hot Rolled Steel Plate Using Time-series Deep Neural Network

“…Since the 1950s, physical metallurgical models have been designed to calculate the mechanical properties of steels. 6) For example, the microstructure and mechanical properties have been successfully precited and controlled in the case of C-Mn steel in the hot rolling and cooling. 7,8) Meanwhile, the equations, based on the volume fraction and the grain size of the second phase are also developed.…”

“…Among them, the deep neural networks method is capable to map highly complex nonlinear correlated data 11) and offers the alternative to predict the mechanical properties of hot-rolled products with higher prediction accuracy than that of classical physical metallurgical models. 6) Several ML methods have been implemented in steel production, such as the support vector machine, artificial neural networks, separate artificial neural networks, and deep neural networks. The common prediction scenarios include (1) optimizing the production parameters of hot-rolled 510L steel, 12) (2) mechanical properties of hotrolled plain carbon steel Q235B, 13) (3) toughness, hardness, and thermal strength of different steel grades, 14) (4) UTS, YS, and EL of hot rolled mild steel coils and implemented in a hot rolling mill, 15) (5) mechanical property of hot-rolled sheets based on large sample size (11 101 data), which has been applied in the actual production in steel mills.…”

Research on Rice Yield Prediction Model Based on Deep Learning

“…It is mainly divided into encoding and decoding structure and expansion of convolution structure; The network representatives of encoding and decoding structure include U-net [ 18 ], Seg Net [ 19 ], Refinet [ 20 ], etc., where an encoder is used to extract image features and dimension reduction, and a decoder is used to recover image dimension and spatial information. The representative networks of expansive convolution are Deep Labv1 [ 21 ], V2 [ 22 ], V3 [ 23 ], V3+ [ 24 ],and PSPNet [ 25 ] which can increase the size of the input image even if no pooling layer is used so that each convolution can contain more information when outputting. In addition, the networks with good effect in the field of target detection have also been applied to the field of instance segmentation, and achieved good segmentation results, such as regional convolution network (R–CNN) [ 26 ], FAST R–CNN [ 27 ], Faster R–CNN [ 28 ], Maskr-CNN [ 29 ], and so on.…”