Intelligent casting raw material feeding method based on industrial cloud platform

Abstract

The embodiment of the invention provides an intelligent casting raw material feeding method based on an industrial cloud platform, which is applied to the technical field of industrial information and data processing, and comprises the following steps: receiving an order generation component requirement through the industrial cloud platform, and calling a rule base and historical data to form an initial feeding scheme; the core is to perform multi-heat rolling optimization, establish a comprehensive cost target and multiple constraints, and solve and generate a charging strategy including an economic adjustment interval; after smelting, according to stokehole detection data, the compensation addition amount is calculated through feedback control in the strategy interval, and equipment is driven to perform precise execution; and dynamically updating and optimizing model parameters by analyzing the correlation of production data to form a closed-loop self-learning system. According to the method, global cost optimization, accurate process control and continuous self-optimization of the system are realized.

Description

Technical Field

[0001] This application relates to the field of industrial information and data processing technology, and in particular to a method for intelligent feeding of casting raw materials based on an industrial cloud platform. Background Technology

[0002] The raw material preparation in the foundry industry still relies heavily on manual experience and static calculations. Based on the composition requirements of each order, the batching operator performs quality balance calculations with reference to a fixed proportioning manual to generate an independent feeding plan. This method treats each furnace as an isolated unit, only considering the immediate procurement cost of materials such as pig iron and scrap steel. For example, the high-temperature melt remaining from the previous furnace can be used as the initial heat source and material for the next furnace. However, the traditional method cannot utilize this heat energy to reduce energy consumption, nor can it coordinate the recycling of return materials between multiple furnaces. This results in blind spots in the optimization of energy and material costs, leading to high overall production costs.

[0003] In the dynamic control phase of the smelting process, traditional methods rely on the on-site experience of operators for compensatory material addition. When furnace-side monitoring shows that the composition deviates from the target, operators need to estimate the yield based on experience and decide on the amount of material to be added. Due to the lack of precise algorithms and scientific boundaries, in order to ensure the composition is up to standard, operators often tend to add excessive amounts of alloy, which drives up raw material costs and can easily lead to secondary composition exceeding the standard due to excessive addition, affecting the final performance of the casting and making it difficult to reliably guarantee quality stability.

[0004] However, in actual production, factors such as batch fluctuations in raw material composition, changes in the condition of furnace linings in smelting equipment, and differences in process operation can all cause these parameters to drift. Due to the lack of effective feedback and self-updating mechanisms, the fixed parameters gradually deviate from actual production, causing the deviation between theoretical calculations and actual results to accumulate continuously. The system cannot adapt to changes in production conditions, which restricts the continuous improvement of production accuracy and efficiency. Summary of the Invention

[0005] The embodiments of this application provide an intelligent feeding method for casting raw materials based on an industrial cloud platform. While ensuring stable quality, it achieves optimal overall cost in casting production and continuous self-improvement in process control precision. To achieve the above objectives, this application adopts the following technical solution: A smart feeding method for casting raw materials based on an industrial cloud platform, characterized in that the method includes: Receive the target casting production order information and generate the corresponding composition requirements; Based on the aforementioned composition requirements, the initial charging scheme for the current heat is generated by calling the charging rule library and combining the current raw material composition data with the residual melt data from the previous heat obtained from the production process database. Based on the initial feeding scheme, rolling optimization calculations are performed for continuous production of multiple furnaces. An optimization objective is established that comprehensively considers raw material costs, energy costs, and waste disposal costs. Constraints including total element balance, inventory limits, and process ratios are set. The feeding strategy range for the current furnace is obtained by solving the optimization objective and constraints. After smelting according to the initial charging scheme, the melt composition data obtained by furnace front detection is received. Using the charging strategy range output by the rolling optimization calculation as the boundary condition, the feedback control algorithm is used to calculate the deviation from the target composition and generate composition deviation data. Based on the component deviation data and under the constraints of the feeding strategy range, the compensation addition amount of key raw materials is calculated; Integrate the basic addition amount and the compensation addition amount in the initial feeding scheme to generate a quantitative feeding instruction and control the feeding equipment to execute it; After the feeding equipment completes its operation, it collects the actual production data for this batch and compares it with the expected value of the optimization target set in the rolling optimization calculation to generate a production comparison result. Based on the production comparison results, the component deviation sequence in the historical production data is extracted, and the correlation between the sequence and the raw material component fluctuation and smelting process parameters is analyzed to generate correlation analysis results. The correlation analysis results are fed back to the rolling optimization calculation process as the basis for updating the element yield and loss rate parameters used, thus forming a closed-loop optimization.

[0006] In some possible implementations, the rolling optimization calculation based on the initial feeding scheme for multiple batches of continuous production establishes an optimization objective that comprehensively considers raw material costs, energy costs, and waste disposal costs, including: Acquire data from multiple consecutive production orders within a preset future period and generate an order sequence; Based on the order sequence, raw material cost components, energy cost components, and waste material disposal cost components are generated by calculating materials, estimating energy consumption, and estimating waste material disposal. The raw material cost component, energy cost component, and waste disposal cost component are weighted and fused according to preset weights, and the fused optimization objective function is output.

[0007] In some possible implementations, the settings include constraints such as total element balance, inventory limits, and process proportions, including: Based on the types of raw materials involved in the optimization objective, the inventory system is queried to obtain the available inventory of each raw material, and a first constraint boundary limiting the total amount of raw materials is generated. Based on the types of raw materials and the first constraint boundary, a set of input-output balance equations for key elements is constructed according to metallurgical principles, and a second constraint boundary is generated that defines the element allocation relationship. Based on the raw material type, the first constraint boundary and the second constraint boundary, a preset process rule library is called to parse and obtain the ratio range between raw materials that meets the process requirements, and a third constraint boundary is generated. The first constraint boundary, the second constraint boundary, and the third constraint boundary are combined to form a set of constraints for rolling optimization calculations.

[0008] In some possible implementations, the step of solving for the current furnace charge strategy range based on the optimization objective and constraints includes: Solve the optimization objective and output a preliminary solution set that satisfies all constraints; From the preliminary solution set, extract the cost sensitivity analysis results for the key raw materials to determine the economic range of addition of the key raw materials; Based on the aforementioned economic addition range, the lower and upper limits of the allowable addition amount for each key raw material are determined; The lower limit and the upper limit of the allowable addition amount constitute the range of addition amount of key raw materials; Based on the aforementioned addition range, an executable feeding strategy range is generated, including the addition range of each key raw material.

[0009] In some possible implementations, receiving melt composition data obtained from furnace pre-detection after smelting according to the initial charging scheme includes: Receive component detection data and melt temperature change data; The component detection data and the melt temperature change data are integrated into a multi-source detection dataset; The multi-source detection dataset is timestamped and aligned. Data calibration is performed on the time-aligned multi-source detection dataset based on a confidence assessment algorithm; Output melt composition data from furnace pre-detection that has been calibrated by data.

[0010] In some possible implementations, the step of using the feeding strategy range output by the rolling optimization calculation as boundary conditions, employing a feedback control algorithm to calculate the deviation from the target component, and generating component deviation data includes: Extract the actual content of each element from the melt composition data; The actual content is compared with the target component requirements to generate a comparison result; The deviation values ​​of the content of each element are calculated based on the comparison results; The types of key elements that need to be adjusted are determined based on the aforementioned deviation values; Based on the aforementioned deviation values ​​and the types of key elements, component deviation data, including adjustment parameters, is generated.

[0011] In some possible implementations, calculating the compensatory addition amount of key raw materials based on the component deviation data and under the constraints of the feeding strategy range includes: Identify key elements in the component deviation data that exceed a preset threshold, and generate a set of key elements; For the elements in the set of key elements, determine the compensation raw materials and generate a set of candidate raw materials; Within the constraints defined by the feeding strategy range, the amount of each raw material in the candidate raw material set is optimized to generate a raw material compensation scheme.

[0012] In some possible implementations, the integration of the base addition amount and the compensation addition amount in the initial feeding scheme to generate a quantitative feeding command and control the feeding equipment to execute it includes: Extract the basic addition amount data of each raw material from the initial feeding scheme; The compensation addition amount is arithmetically superimposed with the basic addition amount data of the corresponding key raw materials, while the basic addition amount data of non-key raw materials remains unchanged. Based on the superimposed calculation results and the basic addition amount data, a list of feeding instructions including the precise addition amount of each raw material is generated; The list of feeding instructions is converted into precise quantitative feeding instructions for controlling the feeding execution equipment.

[0013] In some possible implementations, the step of extracting a component deviation sequence from historical production data based on the production comparison results, analyzing the correlation between this sequence and raw material component fluctuations and smelting process parameters, and generating correlation analysis results includes: Collect data on raw material composition fluctuations during historical production processes, and extract process parameter records including melting temperature, holding time, and stirring intensity; The component deviation sequence is correlated with the raw material component fluctuation data and process parameter records to calculate the contribution and correlation coefficient of each influencing factor to the component deviation. Correlation analysis results are generated based on the contribution and correlation coefficient.

[0014] In some possible implementations, feeding back the correlation analysis results to the rolling optimization calculation process includes: The element yield in actual production can be deduced from the aforementioned component deviation sequence. By comparing the element recovery rate in actual production with the preset recovery rate in the optimization calculation, recovery rate difference data is generated. Update the element's harvest rate parameter based on the harvest rate difference data; Based on the component deviation sequence or production comparison results, the element loss rate in actual production can be inferred. Update the loss rate parameter based on the element loss rate obtained by reverse calculation; The updated yield and loss rate parameters are fed back into the multi-furnace rolling optimization calculation process.

[0015] As can be seen from the above technical solution, this application has the following beneficial effects: 1. This method takes the order sequence over a future period as the object, constructs a comprehensive optimization objective that integrates raw material costs, estimated energy consumption, and waste disposal costs, and sets multi-dimensional constraints, including inventory, element balance, and process ratio, for collaborative solution. It outputs a dynamic feeding strategy range, which sets an economically reasonable fluctuation range for the addition amount of key raw materials in a single furnace while ensuring the overall cost optimization. This provides a scientific boundary for production execution that combines strategic planning and tactical flexibility, thereby achieving a fundamental improvement from local cost optimization to overall cost optimization.

[0016] 2. This method constructs a complete data-driven closed-loop optimization mechanism. With the strategy range of the rolling optimization output as a constraint, a feedback control algorithm is used to accurately calculate the composition deviation and compensation amount before the furnace, achieving precise fine-tuning. By continuously collecting actual production data and comparing it with the optimization expectation, the system can automatically analyze the correlation between composition deviation and raw material fluctuations and process parameters, and dynamically update the yield and loss rate parameters in the optimization model accordingly. It has the ability to adaptively correct parameters and can continuously track and adapt to the dynamic changes in production conditions, thereby significantly improving the control accuracy, stability and autonomous evolution capability of the system in long-term production. Attached Figure Description

[0017] The invention will now be further described with reference to the accompanying drawings.

[0018] Figure 1 A first flowchart of a smart feeding method for casting raw materials based on an industrial cloud platform provided in this application embodiment; Figure 2 A second flowchart of a smart feeding method for casting raw materials based on an industrial cloud platform provided in this application embodiment; Figure 3 The third flowchart of the intelligent feeding method for casting raw materials based on an industrial cloud platform provided in this application embodiment; Figure 4 The fourth flowchart of the intelligent feeding method for casting raw materials based on an industrial cloud platform provided in this application embodiment. Detailed Implementation

[0019] The terms "first," "second," and "third," etc., used in this application specification, claims, and drawings are for distinguishing different objects, not for specifying a particular order.

[0020] In the embodiments of this application, the words "exemplary" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the words "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0021] Research has revealed that traditional casting charging methods employ isolated calculations for each furnace, considering only raw material procurement costs while ignoring energy consumption and material interaction across multiple furnaces, resulting in higher overall costs. Furnace adjustments rely on manual experience, which can easily lead to over- or under-charging, affecting composition accuracy and quality stability. Furthermore, key process parameters remain fixed over time, failing to adapt to changes in production conditions, and the system lacks self-learning and continuous optimization capabilities.

[0022] To address the aforementioned issues, this application provides a smart feeding method for casting raw materials based on an industrial cloud platform: Example 1 To solve the above problems, such as Figure 1 - Figure 4 As shown, this embodiment is applied to the smelting workshop of a large-scale modern foundry enterprise. The enterprise receives diverse casting orders from different customers through an industrial cloud platform and produces products of various materials, including ductile iron and various alloy steels. The workshop is equipped with multiple medium-frequency induction furnaces, which can realize continuous production of multiple furnaces. The production site is equipped with automatic feeding systems, furnace-front rapid spectrometers, temperature sensors, and other equipment. All equipment and production process data are connected to the industrial cloud platform deployed by the enterprise in real time. This platform integrates system modules such as order management, manufacturing execution, inventory management, and advanced planning and scheduling, which constitute the core digital infrastructure for implementing this method.

[0023] For a long time, this workshop has faced a series of typical challenges in the raw material preparation process: First, cost control is crude, only considering the cost of raw materials for a single furnace, ignoring the linkage between raw material use, energy consumption differences, and the hidden costs of recycled materials, such as risers and waste castings; Second, dynamic adjustment is lagging, after the composition is tested before the furnace, the addition of materials relies on the operator's experience, lacking scientific boundary guidance and precise compensation algorithms, which easily leads to excessive composition or waste of materials; Third, process parameters are rigid, and key process parameters such as element yield and burn-off rate used for calculation remain unchanged for a long time, which cannot adapt to actual production disturbances such as batch fluctuations of raw materials and changes in equipment status, resulting in a gradual accumulation of deviations between theoretical calculations and actual results.

[0024] Step 1: Receiving orders and generating ingredient requirements When the industrial cloud platform's order management system receives a new target casting production order, such as a batch of ductile iron gearbox housings, the system parses the product code in the order. This product code is linked to the material technical specification library in the cloud platform's central database. Based on this specification, the system automatically generates precise composition requirements for the order. These requirements are not fixed values, but rather a range conforming to national or industry standards, for example: C: 3.6%–3.8%, Si: 2.4%–2.6%, Mn: <0.4%, P: <0.05%, S: <0.02%, Mg (residual): 0.03%–0.05%. This range is the ultimate target for all subsequent calculations and controls.

[0025] Beneficial effects: The traditional approach involves manually consulting drawings or standards, which is prone to errors and inefficient. This embodiment uses an industrial cloud platform to automate and standardize the mapping of order information to chemical composition requirements. The principle is to use the data integration capabilities of the cloud platform to seamlessly connect the order flow with the technical standard flow. This ensures the digitization and accuracy of production targets from the source, avoids incorrect material feeding caused by human interpretation bias, provides a precise input starting point for subsequent intelligent calculations, and lays the first cornerstone for full-process digital driving.

[0026] Step 2: Generate the initial feeding plan After generating the component requirements, the system will start the batching calculation. The system calls the feeding rule base, which is a knowledge set built on metallurgical principles and the factory's historical best practices. It defines basic rules and conventional proportioning experience such as using pig iron to provide basic carbon and silicon, using scrap steel to adjust carbon equivalent, using ferrosilicon alloy to increase silicon content, and using spheroidizing agents for spheroidization treatment.

[0027] The system retrieves the residual melt data from the previous batch from the production process database on the cloud platform. In continuous production, after the previous batch of iron is tapped, a small amount of melt, known as bottom molten iron, is usually left in the furnace. Its composition is known. At the same time, the system retrieves the current raw material composition data from the raw material inventory management module, such as the nominal composition of the current batch of pig iron, scrap steel, and various alloys, as well as the actual composition of the most recent sampling inspection.

[0028] The system takes the composition requirements, feeding rules, residual material data, and raw material composition data as inputs. Through a preliminary calculation model based on mass balance, it generates an initial feeding scheme for the current furnace. This scheme is theoretical and assumes that the yield of all elements is fixed. For example, the yield of silicon in ferrosilicon is preset to 90%. It also calculates the basic addition amount of each material such as pig iron, scrap steel, recycled material, and ferrosilicon to meet the target composition median.

[0029] Beneficial effects: This step transforms estimations relying on experienced technicians into rule-based, systematic preliminary calculations. The principle lies in applying the law of conservation of mass in metallurgical processes. Specifically, it incorporates data from the previous batch of residual melt, providing a precise model of the continuous production scenario. Ignoring this residual melt is equivalent to ignoring the initial material already present in the reactor, inevitably leading to inaccurate calculations of the new material addition amount and potentially causing a systematic shift in the final composition. By integrating real-time data, the starting point of the calculation is made more precise, ensuring the initial plan is closer to actual production and providing a reliable baseline for subsequent optimization, rather than starting from scratch.

[0030] Step 3: Multi-furnace rolling optimization and generation of feeding strategy range This is the core step in achieving global cost optimization. The system does not perform isolated optimal calculations for the current round, but adopts a forward-looking rolling optimization strategy.

[0031] Generate optimization objectives: The system obtains multiple consecutive production order data within a future preset period, such as the next 8 hours or a shift, from the advanced planning and scheduling module, forming an order sequence. For each furnace in this sequence, the system performs cost accounting simulation.

[0032] Raw material cost breakdown: calculated based on the amount of material added and the real-time unit price.

[0033] Energy cost component: The system's built-in energy consumption model estimates the electrical energy required to melt different raw material ratio schemes based on the melting point of different raw materials, the order of addition, the theoretical heat required for melting, and the electric furnace efficiency. The energy consumption of starting the furnace using all cold materials, pig iron, and scrap steel is higher than the energy consumption of using recycled materials as the main feed when there is still hot iron at the bottom of the furnace.

[0034] Residual material handling cost component: The system will estimate the types and quantities of return materials such as risers and gatings and scrap that may be generated after this batch of production, and make an estimate based on its handling cost model.

[0035] The system assigns different preset weights to these three cost components according to the enterprise's management strategy. For example, raw material costs have the highest weight, followed by energy costs. These weighted components are then combined into a general optimization objective function, which aims to minimize the total comprehensive cost of multiple future furnace cycles.

[0036] Set constraints: The first constraint boundary, namely the inventory limit: The system queries the inventory system to obtain the available inventory of each raw material, and the sum of the usage of all raw materials in the optimization calculation cannot exceed its inventory.

[0037] The second constraint boundary, namely the total element balance, is the mathematical expression of metallurgical principles. For key elements such as C, Si, and Mn, a set of balance equations is established between the input (the amount of various raw materials brought in × the yield) and the output (the required composition of the target casting). For example, the total silicon input (the silicon content from pig iron, scrap steel, ferrosilicon, recycled materials, etc. multiplied by their respective yields) must be equal to the silicon content required by the target melt.

[0038] The third constraint boundary, namely the process ratio, calls upon more detailed restrictions in the process rule library. For example, in order to ensure the graphite morphology of cast iron, the ratio of pig iron to scrap steel must be within a certain range; in order to control oxidation during the smelting process, the proportion of recycled material added cannot be too high; certain alloying elements must maintain a certain proportional relationship. These rules are resolved into mathematical inequality constraints.

[0039] By combining these three types of boundaries, we can form an optimization problem model with multiple constraints.

[0040] Solve and obtain the feeding strategy range: The system uses optimization algorithms such as linear programming or mixed integer programming to solve the above optimization objective under the set of constraints. The solution result is first a preliminary solution set, which gives a detailed batching scheme for each batch that satisfies the overall optimality of the future multi-batch production sequence.

[0041] For the current furnace batch, the system will conduct in-depth analysis from the solution set. By performing cost sensitivity analysis on the optimization objective function, it can identify which raw material cost changes have the greatest impact on the total cost. For example, under the current order sequence, it may be found that the amount of scrap steel used is very sensitive to cost. Based on this, the system will determine the economic addition range of these key raw materials, that is, the amount of raw material that can be flexibly changed without significantly increasing the total cost.

[0042] The system outputs a charging strategy range for the current furnace. This range includes not only the suggested addition amount of each raw material, but also the lower and upper limits of the allowable addition amount for each raw material, especially the key raw materials. For example, the initial plan suggests adding 2,000 kg of scrap steel, but the optimization results show that, under the premise of ensuring overall optimality, the amount of scrap steel can fluctuate between 1,800 kg and 2,200 kg. This range is the flexible decision-making space reserved for subsequent furnace fine-tuning.

[0043] Beneficial Effects: Traditional single-heater optimization is like taking it one step at a time. Although the cost of a single step may be the lowest, it may lead to a sharp increase in energy consumption for starting up subsequent heats or an accumulation of recycled materials. This step introduces rolling optimization calculations for continuous production across multiple heats. Its principle originates from the dynamic programming and model predictive control concepts in operations research. It treats the production task over a period of time as a whole. Through overall planning, it sacrifices the theoretical optimum in the local area in exchange for the overall cost optimum. For example, it may suggest using more preheated recycled materials that are slightly more expensive but can reduce energy consumption in the current heat, thereby reserving more hot iron at the bottom of the furnace for subsequent heats and reducing the energy consumption for starting up subsequent heats. The generated charging strategy range is the key output. It is not a rigid instruction but a flexible guide. Its beneficial effect is that it effectively transmits and empowers the real-time production process in the form of boundary conditions, allowing the site to respond dynamically within a controllable range. It achieves an organic combination of strategic planning and tactical execution and is the core means of overall cost control.

[0044] Beneficial Effects: In casting scenarios, the inclusion of energy cost components is crucial. For example, the difference in power consumption between starting the furnace with all cold pig iron and adding a large amount of rapidly melting recycled material to the molten iron at high temperatures is significant. This method quantifies this difference through modeling and incorporates it into the optimization objective, enabling the system to consciously select energy-saving material combinations during batching. For instance, it prioritizes using a high recycled material ratio during peak electricity price periods, directly reducing electricity costs, which constitute a significant portion of variable costs.

[0045] In the context of casting batching, process proportioning constraints (the third boundary) are the lifeline for ensuring metallurgical quality. For example, insufficient residual magnesium in ductile iron leads to poor spheroidization, while excessive magnesium easily causes slag porosity. This method transforms these constraints into mathematical constraints through a rule base, solving them in conjunction with inventory and quality balance constraints. This ensures that any optimization scheme automatically meets all process and quality requirements, fundamentally eliminating unethical schemes that sacrifice quality for cost reduction, and achieving synergistic optimization of cost and quality.

[0046] In furnace operation scenarios, operators often face a dilemma when encountering compositional deviations: not adding any leads to substandard composition; adding too much may result in cost overruns or exceed acceptable limits for other elements. The feeding strategy range provides a scientific basis for solving this problem; for example, if the spectrometer shows low silicon levels, ferrosilicon needs to be added. When calculating the amount to be added, the system uses the upper limit of the range as a hard constraint, ensuring that even with addition, the total ferrosilicon usage will not exceed the economically permissible range, keeping dynamic adjustments within a safe and economical range.

[0047] Step 4: Pre-furnace inspection and data calibration Workshop workers carry out smelting operations according to the initial feeding plan. After the smelting reaches the predetermined temperature and the preliminary refining is completed, molten iron samples are taken and sent to the rapid spectrometer in front of the furnace for testing. Temperature sensors installed in the furnace body continuously record the changes in the molten temperature.

[0048] The detection data, such as spectral composition and temperature, is uploaded to the industrial cloud platform. After receiving this component detection data and melt temperature change data, the platform integrates them into a multi-source detection dataset. Since there may be slight time asynchrony or error between the detection equipment and sensors, the system will perform timestamp alignment to ensure that the melt state is analyzed at the same moment.

[0049] The system initiates a data calibration process. For example, the detection of carbon by the spectrometer may fluctuate at high temperatures. The system will combine temperature data and use a confidence assessment algorithm trained on historical data to perform reasonableness verification and smoothing correction on the original detection values, and finally output reliable melt composition data that has been calibrated.

[0050] Beneficial effects: Furnace-front detection data is the sole basis for feedback control, and its accuracy is crucial. This step utilizes multi-source data fusion and intelligent calibration. The principle is to improve information quality by leveraging data redundancy and correlation. Single detection data may be affected by equipment status and environmental interference, but by combining it with related data such as temperature change curves, cross-validation and correction can be performed. This is equivalent to putting calibration lenses on the eyes of quality control, greatly improving the reliability of on-site perception data and providing a solid data foundation for subsequent accurate compensation calculations, avoiding the problem of "garbage in, garbage out".

[0051] Step 5: Calculate the component deviation The system compares the calibrated melt composition data with the target composition requirements generated in the first step. For example, if the target Si is 2.5% and the measured Si is 2.35%, the deviation is -0.15%.

[0052] The system automatically determines which elements' deviations exceed the preset process adjustment thresholds based on the deviation values, thereby identifying the key elements requiring adjustment. Assuming only Si's deviation exceeds the threshold, the system generates compositional deviation data for Si, including the deviation direction (insufficiency), deviation amount (0.15%), and adjustment parameters such as the element requiring compensation (Si). This calculation process uses the feeding strategy range output in the third step as an insurmountable boundary condition, ensuring that no adjustment attempt will break through the economic framework set by the optimization calculation.

[0053] Beneficial effects: This step transforms manual judgment based on intuition into quantitative and standardized deviation identification. Its principle is an automatic triggering mechanism that sets thresholds. Compensation is only triggered for significant deviations, avoiding overreaction to minor fluctuations in composition during the production process. This meets the robustness requirements of industrial production. By clearly defining the strategy range as the boundary, it means that the starting point for deviation calculation and subsequent compensation has been limited by global optimization, ensuring the consistency between local adjustments and global goals. It is a key bridge connecting optimization and execution.

[0054] Step 6: Calculate the compensation amount of key raw materials The system identifies a set of key elements in the composition deviation data, such as only Si. Then, it determines the compensating raw materials for the elements in this set. According to the feeding rule base, the compensating raw material for Si is usually ferrosilicon (FeSi), thus generating a set of candidate raw materials ({ferrosilicon}).

[0055] Then, under the strict constraints of the feeding strategy range, such as the total amount of ferrosilicon not exceeding the upper limit of 2200 kg in the initial plan, the system starts a fast optimization solution with the goal of adjusting the composition to the target range with the least amount of compensation raw materials. It takes into account the yield of the compensation raw materials themselves and their dilution or enrichment effect on other elements. For example, adding ferrosilicon will increase the iron content in the molten iron. The solution result generates a raw material compensation plan, which accurately indicates how many kilograms of a specific grade of ferrosilicon need to be added.

[0056] Beneficial Effects: In emergency furnace replenishment scenarios, operators often rely on experience to estimate the amount, tending to over-add alloys rather than under-add them. This not only wastes resources but can also introduce new compositional issues, such as excessive silicon affecting toughness. This step addresses this by performing a secondary rapid optimization solution within the optimized boundary conditions. Its principle is optimization under constraints. It ensures that the replenishment amount is both sufficient and economical, without exceeding the overall economic limit. For example, calculations show that only 15 kg of ferrosilicon needs to be added to increase the silicon content to 2.48%, avoiding the waste and potential risks caused by empirically adding 25 kg. This represents a leap from empirical coarse-grained adjustment to model-based fine-tuning.

[0057] Step 7: Generate and execute quantitative feeding instructions The system extracts the basic addition amount data of each raw material from the initial feeding plan. Then, it arithmetically adds the compensation addition amount calculated in step six to the basic addition amount of the corresponding raw material (ferrosilicon). For non-critical raw materials, their addition amount remains unchanged.

[0058] Based on the results of the superposition calculation, the system generates a list of feeding instructions accurate to the kilogram or even gram. This list is converted into control instructions that the equipment can recognize. For example, 15.5 kg of ferrosilicon is precisely added from silo No. 3 via weighing belt No. 2. The precise quantitative feeding instruction is sent to the automatic feeding system, which controls the corresponding silo, weighing equipment and conveying mechanism to execute the feeding and complete the compensation feeding.

[0059] Beneficial effects: This step achieves a digital and automated closed loop from decision-making to action. Its principle lies in the accurate generation and reliable transmission of instructions. Traditionally, replenishment amounts require manual calculation, input, or visual estimation, resulting in large errors and low efficiency. This method automatically integrates and generates digital instructions to directly drive the equipment, eliminating errors and delays caused by human intervention. This ensures the original execution of the compensation plan, allowing all the results of previous intelligent calculations to be seamlessly translated into on-site actions.

[0060] Step 8: Production Data Collection and Comparison After the charging process is completed, the smelting continues until the furnace is tapped and cast. All actual production data for this furnace is collected, including: the spectral composition confirmed before final tapping, the actual weight of each raw material consumed, feedback from the charging system, the actual electrical energy consumed, the meter readings, and the types and weights of the actual return material generated.

[0061] The system compares these actual data with the expected values ​​of the optimization targets set for this furnace batch during rolling optimization calculations, as well as the predicted composition, cost, and energy consumption. For example, the expected silicon content is 2.5%, while the actual content is 2.52%; the expected energy consumption is 5000 kWh, while the actual energy consumption is 5100 kWh; and the expected cost is a certain amount, while the actual cost is a certain amount. Through comparison, production comparison results are generated, quantifying the difference between prediction and reality.

[0062] Beneficial effects: After traditional production ends, data is often archived, making it difficult to use for continuous improvement. This step, through a systematic comparison of expected and actual results, proactively and quantitatively reveals the degree of alignment between the planned and executed processes by analyzing the discrepancy between planning and execution. It links production results with optimization predictions, providing clear data clues and problem directions for subsequent analysis of why deviations occurred, and serves as a data source driving the updating of process parameters.

[0063] Step 9: Correlation Analysis of Historical Data The system is not limited to comparison of a single furnace. It extracts historical production data, such as the composition deviation sequence of the most recent 100 furnaces under similar process conditions to the current furnace, i.e. the deviation between the actual value and the target value of Si in each furnace, collects corresponding raw material composition fluctuation data, such as the fluctuation range of silicon content in each batch of pig iron, and records of smelting process parameters, such as the smelting temperature curve, holding time, and electromagnetic stirring intensity setting for each furnace.

[0064] The system employs statistical analysis algorithms, such as multiple linear regression and correlation analysis, to perform correlation analysis between the composition deviation sequence and raw material fluctuations and process parameters. The calculation results show, for example, that the correlation coefficient between silicon deviation and the fluctuation of silicon content in pig iron raw materials is 0.6, and the correlation coefficient with the highest temperature in the refining stage is -0.3. This indicates that fluctuations in raw material silicon content are the main influencing factor, while higher refining temperatures are beneficial for stabilizing silicon yield.

[0065] Based on these contributions and correlation coefficients, the system generates correlation analysis results, clearly indicating which factors and to what extent they affect the control precision of key components.

[0066] Beneficial effects: In casting production, the causes of compositional fluctuations are complex and varied, traditionally attributed to poor conditions or unstable raw materials, but lacking quantitative evidence. Through big data correlation analysis, the principle is to mine potential causal relationships from historical data. It can reveal, for example, that when using pig iron with high volatility from a certain ore source, the silicon yield will systematically decrease; or when the smelting temperature is below a certain threshold, the manganese burn-off will abnormally increase. This data-based insight is more comprehensive and objective than human experience, providing a direct basis for accurately adjusting process model parameters.

[0067] Step 10: Feedback updates form a closed-loop optimization. Updated elemental yield: Based on historical composition deviation sequences, the system can extrapolate the elemental yield in actual production. For example, if the final silicon yield is lower than the theoretical value after multiple consecutive furnace additions of ferrosilicon, the system can calculate a new average silicon yield that better reflects current production conditions, such as lowering it from 90% to 88%. The system then compares the actual yield with the preset values ​​in the optimization model, generating yield difference data and updating the elemental yield parameters in the model accordingly.

[0068] Update element loss rate: Similarly, by analyzing compositional deviations, such as the loss of Mn, or information such as energy consumption and slag volume in the production comparison results, the system can back-calculate the element loss rate and burn-off rate in actual production, and update the corresponding parameters in the optimization model.

[0069] These updated, more realistic yield and loss rate parameters are immediately fed back and applied to the next multi-furnace rolling optimization calculation.

[0070] Beneficial Effects: Traditional process parameters, once set, often remain unchanged for a long period, failing to adapt to dynamic changes in production conditions and causing the model to gradually become ineffective. This step constructs a complete closed loop, based on data-driven adaptive parameter correction. The system is no longer a static, open-loop instruction generator, but an adaptive system that learns from each production result and continuously corrects its internal cognitive parameters. For example, when the system learns that the current scrap steel source is causing a decrease in carbon yield, it will automatically lower the preset value of carbon yield in subsequent optimization calculations. The newly generated feeding scheme will naturally increase the addition of carbon source to compensate for the expected loss, thereby achieving more precise control in the next round of production. This gives the entire intelligent feeding system the ability to continuously evolve, tracking and adapting to slow variables such as changes in the raw material market, equipment aging, and changes in operating habits, maintaining high-precision feeding guidance and cost optimization capabilities, truly achieving a leap from automation to intelligence.

[0071] This embodiment details the application of an intelligent raw material feeding method for casting based on an industrial cloud platform in a specific industrial scenario. This method constructs an intelligent decision-making and control system covering the entire production process through a series of closely linked and progressive steps: order receiving, initial rule calculation, rolling optimization, interval generation, furnace pre-detection, deviation calculation, compensation solution, precise execution, data comparison, correlation analysis, and parameter updates. Its core value lies in: minimizing the overall global cost through multi-furnace rolling optimization; providing a scientific boundary for dynamic adjustment through generating feeding strategy intervals; achieving precise fine-tuning of composition through furnace pre-feedback control; and finally, enabling the system to have adaptive and continuous optimization capabilities through data-driven closed-loop learning. The entire method deeply integrates the data processing capabilities of cloud computing, optimization theory of operations research, the physicochemical principles of metallurgical processes, and automatic control technology, providing a systematic and implementable intelligent solution to address the long-standing problems of cost control difficulties, poor dynamic adjustment, and parameter rigidity in the raw material batching process of the casting industry.

[0072] The foregoing has shown and described the basic principles, main features, and advantages of this application. Those skilled in the art should understand that this application is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of this application. Various changes and modifications can be made to this application without departing from the spirit and scope thereof, and all such changes and modifications fall within the scope of this application as claimed. The scope of protection of this application is defined by the appended claims and their equivalents.

Claims

1. A casting raw material intelligent feeding method based on an industrial cloud platform, characterized in that, The method comprises: receiving target casting production order information, generating corresponding component requirements; based on the component requirements, calling a charging rule library, combining current raw material component data and previous heat residual melt data obtained from a production process database, generating an initial charging scheme for the current heat; based on the initial charging scheme, performing rolling optimization calculation for multi-heat continuous production, establishing an optimization target considering raw material cost, energy cost and residual material treatment cost, setting constraint conditions including element total amount balance, inventory limit and process ratio, solving based on the optimization target and constraint conditions to obtain a charging strategy interval for the current heat; after smelting according to the initial charging scheme, receiving melt component data obtained by furnace detection, taking the charging strategy interval output by the rolling optimization calculation as a boundary condition, calculating the deviation from the target component using a feedback control algorithm to generate component deviation data; based on the component deviation data and under the constraint of the charging strategy interval, calculating the compensation addition amount of key raw materials; integrating the base addition amount in the initial charging scheme and the compensation addition amount to generate quantitative charging instructions to control the charging equipment to execute; after the charging equipment executes, collecting actual production data of the current heat and comparing it with the expected value of the optimization target set by the rolling optimization calculation to generate a production comparison result; based on the production comparison result, extracting a component deviation sequence from historical production data, analyzing the correlation between the sequence and raw material component fluctuation and smelting process parameters to generate a correlation analysis result; feed the correlation analysis result back to the rolling optimization calculation process as a basis for updating the element recovery rate and loss rate parameters used by it to form a closed-loop optimization.

2. The method of claim 1, wherein, The rolling optimization calculation based on the initial charging scheme for multi-heat continuous production establishes an optimization target that considers raw material cost, energy cost and residual material treatment cost, including: obtain multiple continuous production order data within a future preset period to generate an order sequence; based on the order sequence, calculate material, estimate energy consumption and estimate residual material treatment to generate raw material cost components, energy cost components and residual material treatment cost components; weight and fuse the raw material cost components, energy cost components and residual material treatment cost components according to a preset weight to output a fused optimization target function.

3. The method of claim 2, wherein, The constraint conditions including element total amount balance, inventory limit and process ratio include: based on the raw material types involved in the optimization target, query the inventory system to obtain the available inventory of each raw material, generate a first constraint boundary limiting the total charging amount; based on the raw material types and the first constraint boundary, construct the input-output balance equation set of key elements according to metallurgical principles to generate a second constraint boundary defining the element distribution relationship; based on the raw material types, the first constraint boundary and the second constraint boundary, call a pre-set process rule library to analyze and obtain the raw material ratio range that meets the process requirements to generate a third constraint boundary; The first constraint boundary, the second constraint boundary, and the third constraint boundary are combined to form a set of constraints for rolling optimization calculations.

4. The method of claim 3, wherein, The process of solving the optimization objective and constraints to obtain the current furnace charging strategy range includes: Solve the optimization objective and output a preliminary solution set that satisfies all constraints; From the preliminary solution set, extract the cost sensitivity analysis results for the key raw materials to determine the economic range of addition of the key raw materials; Based on the aforementioned economic addition range, the lower and upper limits of the allowable addition amount for each key raw material are determined; The lower limit and the upper limit of the allowable addition amount constitute the range of addition amount of key raw materials; Based on the aforementioned addition range, an executable feeding strategy range is generated, including the addition range of each key raw material.

5. The method of claim 1, wherein, The step of receiving melt composition data obtained from furnace-front detection after smelting according to the initial charging scheme includes: Receive component detection data and melt temperature change data; The component detection data and the melt temperature change data are integrated into a multi-source detection dataset; The multi-source detection dataset is timestamped and aligned. Data calibration is performed on the time-aligned multi-source detection dataset based on a confidence assessment algorithm; Output melt composition data from furnace pre-detection that has been calibrated by data.

6. The method of claim 5, wherein, The step of using the feeding strategy range output by the rolling optimization calculation as boundary conditions, employing a feedback control algorithm to calculate the deviation from the target component, and generating component deviation data includes: Extract the actual content of each element from the melt composition data; The actual content is compared with the target component requirements to generate a comparison result; The deviation values ​​of the content of each element are calculated based on the comparison results; The types of key elements that need to be adjusted are determined based on the aforementioned deviation values; Based on the aforementioned deviation values ​​and the types of key elements, component deviation data, including adjustment parameters, is generated.

7. The method of claim 4, wherein, The calculation of the compensation addition amount of key raw materials based on the component deviation data and under the constraints of the feeding strategy range includes: Identify key elements in the component deviation data that exceed a preset threshold, and generate a set of key elements; For the elements in the set of key elements, determine the compensation raw materials and generate a set of candidate raw materials; Within the constraints defined by the feeding strategy range, the amount of each raw material in the candidate raw material set is optimized to generate a raw material compensation scheme.

8. The method of claim 1, wherein, The process of integrating the basic addition amount and the compensation addition amount in the initial feeding scheme to generate a quantitative feeding command and control the feeding equipment to execute it includes: Extract the basic addition amount data of each raw material from the initial feeding scheme; The compensation addition amount is arithmetically superimposed with the basic addition amount data of the corresponding key raw materials, while the basic addition amount data of non-key raw materials remains unchanged. Based on the superimposed calculation results and the basic addition amount data, a list of feeding instructions including the precise addition amount of each raw material is generated; The list of feeding instructions is converted into precise quantitative feeding instructions for controlling the feeding execution equipment.

9. The method of claim 1, wherein, Based on the production comparison results, the component deviation sequence is extracted from historical production data, and the correlation between this sequence and raw material component fluctuations and smelting process parameters is analyzed to generate correlation analysis results, including: Collect data on raw material composition fluctuations during historical production processes, and extract process parameter records including melting temperature, holding time, and stirring intensity; The component deviation sequence is correlated with the raw material component fluctuation data and process parameter records to calculate the contribution and correlation coefficient of each influencing factor to the component deviation. Correlation analysis results are generated based on the contribution and correlation coefficient.

10. The method of claim 1, wherein, The step of feeding back the correlation analysis results to the rolling optimization calculation process includes: The element yield in actual production can be deduced from the aforementioned component deviation sequence. By comparing the element recovery rate in actual production with the preset recovery rate in the optimization calculation, recovery rate difference data is generated. Update the element's harvest rate parameter based on the harvest rate difference data; Based on the component deviation sequence or production comparison results, the element loss rate in actual production can be inferred. Update the loss rate parameter based on the element loss rate obtained by reverse calculation; The updated yield and loss rate parameters are fed back into the multi-furnace rolling optimization calculation process.

Images