Classification and rating of steel scrap using deep learning

Conclusion

This paper addresses the problems of many types of steel scrap, complex actual inspection scenarios, and the difficulty of manual system interface, and transforms the current method of determining scrap grades in most steel enterprises, which is mainly determined by visual inspection by quality managers and caliper measurement together, into intelligent grading. How to achieve openness, fairness, and equity in scrap grade inspection and solve the problems of human factors interference and inefficiency, we propose a deep learning model of CSBFNet scrap quality inspection based on attention mechanism, in which, the SE attention mechanism is added to the feature extraction network and the feature fusion network is replaced with EfficientDet module (BiPFN), which makes the feature extraction in scrap category and multi-scale feature fusion with significant advantages, which improves the overall performance of the network.

The experiments were conducted on the laboratory dataset HK_L and the field-collected dataset HK_T for model training and optimization. The results showed that the mAP of the CSBFNet model reached 90.7%, and the average accuracy rate for all types of steel scrap reached 92.4%. The evaluation index was higher than that of the model without the SE attention mechanism and BiPFN module, as well as mainstream target detection models such as Faster R-CNN. The detection time for a single image is only 11 ms, which can meet the real-time requirements of scrap identification and grading in actual production scenarios. This indicates that the CSBFNet model has high performance and excellent detection and grading capabilities. Compared with the traditional manual scrap inspection, the CSBFNet steel scrap inspection model has obvious advantages in accuracy, safety, and fairness.

Future Work

The future research direction mainly includes three aspects. First, to reduce the influence of the complex background of the scrap steel image on the classification and rating effect, image segmentation technology could be employed to establish a compartment segmentation model and accurately determine the scrap steel classification and rating area. Second, a quality prediction model could be applied to estimate the quality of steel scrap after classification and grading. Finally, the steel scrap after quality prediction could be fed into the electric furnace to predict the tapping rate based on the predicted steel scrap quality data.

In future experiments, the dataset will be expanded and balanced to reduce the problem of inaccurate ratings for individual categories of scrap due to insufficient samples, while the model will be further optimized and improved for more complex steel scrap recycling quality inspection scenarios in the future. In addition, the dataset will provide a finer breakdown of the steel scrap classification and add a quality assessment model to the subsequent work to deduct weight from impurities in scrap recycling to make the inspection more complete. Finally, the system will be integrated into the real-time system of scrap inspection, which will help reduce the problem of inaccurate scrap rating and get more value in scrap recycling.

Furthermore, intelligent scrap rating not only affects cost settlement, but also dovetails with electric furnace charging, which has a catalytic effect on improving the steelmaking process regime.

Abstract

To address the issues of high human interference and low efficiency in traditional manual methods for classifying and rating steel scrap, we propose the development of CSBFNet, a deep learning-based model for multi-category steel scrap classification and rating. Firstly, we built a 1:3 physical model of steel scrap quality inspection to simulate the unloading of a truck. We used a high-resolution vision sensor to capture the morphological characteristics of various steel scraps. Next, we trained the CSBFNet model using this data to obtain characteristic information for classifying and judging various types of scrap steel. Finally, we tested and improved the CSBFNet model at a Chinese steel mill. The results demonstrate that the model can effectively determine the automatic rating for different grades of scrap. The average accuracy rate of all types of steel scrap reaches 92.4% for the full category, with an mAP of 90.7%. Compared to traditional artificial quality detection methods, it has clear advantages in accuracy and fairness. This model solves the problem of evaluating the quality of steel scrap in the recycling process.