How steel is made with AI

Panzhihua Iron and Steel Group started the attempt of handicraft intelligence from the two scenarios of cold-rolled plate surface inspection and steelmaking desulfurization process optimization

Process optimization in the desulfurization process

Pain points:

Desulfurization is an important step in the steel production process to reduce the sulfur content in molten iron. In the desulfurization process, a large amount of metal material will be taken away due to slag removal with iron (the desulfurization slag produced after the reaction of the desulfurizer contains a large amount of iron). After calculation, the average amount of desulfurization slag in each furnace (220 tons) is 5 tons, and the iron loss in desulfurization slag accounts for about 40%-55%. It is assumed that the addition of desulfurizer can be reduced after parameter recommendation and optimization 10%, theoretically, it can reduce the consumption of steel materials by 0.8~1kg/ton of steel.

Purpose:

Through analysis and modeling, optimize the desulfurization process, recommend the optimal amount of desulfurization agent, improve the utilization rate of desulfurization agent, and reduce the iron loss in the desulfurization process.

process:

Collect desulfurization process flow data, obtain key factors for desulfurization process optimization through modeling analysis, combine expert knowledge, rely on desulfurization simulation model and parameter optimization model to find optimal parameters.

· Simulation model:

Based on historical data and real-time data, build a prediction model after desulfurization. By combining more than ten key parameters such as the amount of desulfurization agent added and the injection rate, the whole process of desulfurization is simulated, and the sulfur content after desulfurization is predicted, and the rationality and effectiveness of different sets of parameters are tested by cooperating with the parameter optimization model.

· Parameter optimization model:

Combining machine learning and the experience of the master in industrial control, identify the key factors in the desulfurization process (parameters that have the greatest impact on the desulfurization results), including the addition of passivated magnesium, the ratio of passivated lime, The average flow rate, injection duration, etc., identify the optimal relationship between parameters through the parameter optimization model. The multiple sets of optimization parameters provided by the optimization model are returned to the simulation model for repeated verification and optimization, and finally the optimal parameters are obtained—that is, the set of "recipes" that achieve the minimum desulfurization agent addition under the premise of satisfying the desulfurization effect

Income:

After the molten iron enters the station, the desulfurization optimization model will extract relevant data and perform parameter optimization calculations. The desulfurization operator will dynamically adjust the amount of desulfurization agent added according to the recommended parameters to reduce the consumption of desulfurization agent. According to actual calculations, through optimized parameter recommendations, about 1 kg of steel material can be saved for every ton of steel produced. For the Pansteel Xichang steel and vanadium base with an annual output value of 4 million tons of steel, the annual profit is expected to be 7 million yuan.

AI detection of cold-rolled sheet surface

Pain points:

After the cold-rolled steel strip undergoes continuous processing such as rolling and heat treatment, it will form a steel coil about a kilometer long. In the surface inspection process, quality inspectors usually identify as few as a dozen to as many as hundreds of defects in just 5 to 10 minutes, and the surface inspection instrument (surface inspection equipment) scans within 30 seconds The judgments of surface grade, sorting degree, main defect and eligibility etc. are given within. The nature of the high-intensity, repetitive and boring work that inspectors have been engaged in for a long time determines that the stability of their process output cannot be guaranteed. At the same time, differences in personal experience of inspectors in understanding and grasping product standards will inevitably lead to high judgment levels. Down, uneven. The final result is the deterioration of customer experience and satisfaction or the increase of quality costs. Undoubtedly, these constitute huge hidden cost losses for enterprises.

Purpose:

For the surface quality inspection scene of cold-rolled plates in the steel industry, an automatic surface quality judgment model is built to assist manual judgment of product defects, reduce manual dependence, and improve judgment accuracy.

Process:

First of all, the meter inspector structures the unstructured data (picture information), and summarizes thousands of product appearance defects into major categories on June 2, 2023, such as: flat spots, warped skin, scratches, embossing, hemp Dots, bubbles, etc. Secondly, the table inspection data is combined with MES data and input into the model for training. Through deep learning technologies such as clustering algorithm, rule engine (including self-learning reconstruction function for user demand identification), and defect severity quantification model, the most preliminary classification of rolled steel is carried out. determination. Finally, combined with further manual confirmation, the final judgment result is obtained.

Income:

Comparing the automatic grading results output by the algorithm model with the manual judgment results, the accuracy rate of table judgment codes is over 92%, the accuracy rate of sorting degree is over 80%, and the accuracy rate of main defects is over 60%.