Artificial intelligence-based adaptive optimization method and system for nodular cast iron alloy composition

Abstract

The present application relates to the technical field of intelligent manufacturing, in particular to a nodular cast iron alloy component self-adaptive optimization method and system based on artificial intelligence, comprising the following steps: obtaining furnace temperature and component parameters and generating a stable contrast legend, screening offset abnormal combinations to form a usable set, extracting performance curve features to construct a trend chart, superimposing structure parameters and element distribution to locate segregation areas, and outputting feed correction parameters to generate an optimization scheme. In the present application, through normalization processing of furnace temperature, carbon-silicon ratio and other multi-factor offset values, a component offset atlas is constructed, the trend feature extraction and sorting of response curves under multiple working conditions are combined, stable component combinations are screened out, a mapping relationship is established using structure curvature, wall thickness slope and injection rate, a spatial coincidence atlas is generated through superimposed analysis, the feed direction and injection rate are calculated, the dynamic adaptation of components and structures is realized, and the uniformity of the structure and the consistency of the performance are improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent manufacturing technology, and in particular to an artificial intelligence-based adaptive optimization method and system for ductile iron alloy composition. Background Art

[0002] The field of intelligent manufacturing technology refers to a comprehensive technological system that achieves the digitization, networking, and intelligence of manufacturing processes through the integration of next-generation information technology and advanced manufacturing technologies. Its core issues include key aspects such as intelligent perception, real-time data collection, autonomous decision-making optimization, and precise control in manufacturing systems. It encompasses the deep integration and application of the Industrial Internet, smart factories, digital twins, intelligent equipment, and artificial intelligence in manufacturing. This technical field emphasizes the efficient and coordinated allocation of manufacturing resources and capabilities, promoting intelligent upgrades in all aspects, from product design and process planning to production execution and quality control, and achieving flexible and intelligent control of the entire manufacturing process. It is widely used in high-end equipment manufacturing, automotive, metallurgy, electronics, materials, and other industries. Traditional ductile iron alloy composition design involves manually adjusting the proportions of various elements based on empirical formulas or historical data to meet the mechanical properties, corrosion resistance, and casting suitability requirements under different working conditions. The technical issues involved primarily involve optimizing material ratios. This process typically uses empirical formula derivation, fixed process window selection, and trial-and-error experimental verification to adjust and select alloying elements.

[0003] Existing technologies rely on fixed ratio schemes and empirical rules to construct composition adjustment models. When dealing with castings with high structural complexity or significant changes in injection paths, it is often difficult to reflect the composition offset state and temperature change trend corresponding to the working conditions in real time. There is a lack of dynamic perception and data matching mechanisms for key physical quantities, making it difficult to accurately identify the differences in material responses under different working conditions, resulting in insufficient adaptability between the composition design and the actual structure. In scenarios where the injection rate and structural curvature are coupled, segregation accumulation and uneven organization are prone to occur, causing performance fluctuations to spread and the stability of the finished product to decrease, making it difficult to meet high-level requirements for response consistency and process control accuracy. Summary of the Invention

[0004] The purpose of the present invention is to solve the shortcomings of the prior art and propose an artificial intelligence-based adaptive optimization method and system for ductile iron alloy composition.

[0005] In order to achieve the above object, the present invention adopts the following technical solution: an artificial intelligence-based adaptive optimization method for ductile iron alloy composition, comprising the following steps:

[0006] S1: Obtain the current furnace temperature, carbon-silicon ratio, total amount of deoxidizer added, spheroidizing addition ratio, and incubation time in the smelting process, calculate the offset difference between the value and the upper and lower temperature limits of the thermal stability range, perform temperature span conversion, and generate a composition stability comparison chart;

[0007] S2: call the carbon-silicon ratio offset item and the furnace temperature offset item in the component stability contrast legend, match the data at the time of spheroidizing treatment stage, set the combination of carbon-silicon offset difference greater than the average amplitude of temperature zone expansion and the downward direction of furnace temperature offset as an abnormal item, exclude part of the combination, and generate a set of available component combinations;

[0008] S3: read the tensile, fatigue and impact curve records corresponding to the set of available component combinations, divide the curves into five sections, extract the numerical slope, extreme value quantity and curvature direction transformation number of each section, construct a trend integration record according to the stable order, and generate a set of stable trend response graphs;

[0009] S4: based on the component combination of the set of stable trend response graphs, collect the curvature change value, wall thickness slope and injection rate of the corresponding part of the mold structure, perform spatial mapping and superimposed analysis with the element deviation distribution graph, locate the coupling concentrated area, and generate a segregation overlap distribution area graph.

[0010] As a further scheme of the present application, the component stability contrast legend includes furnace temperature offset normalized value, carbon-silicon ratio offset normalized value, deoxidizing substance offset normalized value, spheroidizing additive ratio value offset normalized value, and inoculation time offset normalized value; the set of available component combinations includes carbon-silicon offset over-limit combination, temperature downward offset combination, and effective combination after abnormality elimination; the set of stable trend response graphs includes tensile trend graph, fatigue trend graph and impact trend graph; the graph set adopts slope change characteristics, extreme value distribution characteristics and curvature transformation characteristics; and the segregation overlap distribution area graph includes curvature change distribution, wall thickness slope distribution, injection rate distribution and element deviation superimposed layer.

[0011] As a further scheme of the present application, the specific steps of S1 are as follows:

[0012] S101: obtain the furnace temperature, carbon content, silicon content, deoxidizing substance dosage, spheroidizing agent dosage, metal weight and inoculation treatment time in the current smelting process, calculate the ratio based on the carbon-silicon content, calculate the additive ratio value based on the spheroidizing agent dosage and the metal weight, calculate the treatment time according to the inoculation start and end time, and generate a set of component basic parameters;

[0013] S102: according to the furnace temperature in the set of component basic parameters, call the upper and lower limits of the thermal stability interval, calculate the temperature offset coefficient, take the interval span as the reference, divide the deviation by the span and take the absolute value, and generate a temperature offset coefficient set;

[0014] S103: call the set of component basic parameters and the temperature offset coefficient set, normalize all parameters according to the temperature offset coefficient proportion to generate a contrast graph, draw a column chart and label the legend label, and obtain a component stability contrast legend.

[0015] As a further solution of the present invention, the specific steps of S2 are:

[0016] S201: calling the carbon-silicon ratio offset item and the furnace temperature offset item in the composition stability comparison diagram, extracting each corresponding data pair in chronological order based on the time node of the spheroidization treatment stage, and combining the two data items into a group to generate a carbon-temperature corresponding group set;

[0017] S202: Based on the carbon-temperature corresponding group set, conditional judgment is performed on each group of data to screen data whose carbon-silicon offset difference is greater than the average amplitude of the thermal stable range expansion and whose corresponding furnace temperature offset is negative, and the data that meets the conditions is marked as an abnormal group to generate an offset abnormality set;

[0018] S203: calling the carbon temperature corresponding group set and the offset anomaly set, removing the anomaly group from all combinations, retaining the combination and re-arranging the index order to obtain an available component combination set.

[0019] As a further solution of the present invention, the specific steps of S3 are:

[0020] S301: Reading the curve response records of multiple groups of components in the available component combination set under three types of working conditions: tension, fatigue, and impact, dividing each curve into five segments according to the time axis, extracting the numerical slope, number of local extreme values, and number of curvature direction changes of continuous data in each segment, and generating a curve structure feature group;

[0021] S302: Based on the three indicator values ​​in the curve structure feature group, sort each type of feature by the fluctuation amplitude of the value, score the sorting stability and record the sequence position, and establish a multidimensional data set based on the three sorting orders to obtain a structural sorting trend set;

[0022] S303: calling the structural sorting trend set, mapping the operating condition type and structural characteristic result corresponding to the sorting sequence to the curve number, sorting and classifying them in the order of the numbers and constructing a graph group annotation to generate a stable trend response graph group.

[0023] As a further solution of the present invention, the specific steps of S4 are:

[0024] S401: Based on the component combinations listed in the stable trend response graph group, collecting curvature change values, wall thickness slope and injection rate data of corresponding parts in the current casting structure, establishing a corresponding table according to the part number, and respectively arranging the three groups of data into sequences of equal length, mapping them to spatial nodes, and generating a structural response coordinate set;

[0025] S402: calling the structural response coordinate set and the configured element deviation distribution map, performing two-dimensional area mapping processing on the three sets of structural response data and the deviation map, performing coordinate cross-matching based on node positions and marking overlap marks, counting the number of overlapping nodes and distribution boundaries, and obtaining an overlap area index set;

[0026] S403: Based on the overlap region index set, with the mold structure cross-section as the base map, the closed boundary regions are superimposed and marked at multiple nodes, and a polygonal outer frame is generated for display. The relationship between the drawing group labels and the spatial coordinates is output according to the numbering identifier to generate a segregation overlap distribution region map.

[0027] As a further embodiment of the present invention, the method further comprises:

[0028] S5: Based on the segregation overlap distribution area map, the channel injection rate, the structural curvature within the area, and the concentration gradient are called. If all three exceed the standard range line, the feed direction correction angle, the rate correction amount, and the injection sequence number combination are output to generate a self-matching optimization plan for the ductile iron alloy composition.

[0029] As a further solution of the present invention, the ductile iron alloy composition self-matching optimization solution includes a feed direction correction angle, an injection rate correction amount, and an injection sequence number.

[0030] As a further solution of the present invention, the specific steps of S5 are:

[0031] S501: Based on the spatial overlapping areas marked in the segregation coincidence distribution area map, three values ​​of channel injection rate, structural curvature, and concentration gradient are collected at the corresponding positions. The set standard range line is called to perform interval judgment on the multiple values ​​and the upper and lower limits of the multiple standards. The numbers of all out-of-bounds areas are filtered to generate an out-of-bounds area set.

[0032] S502: Based on the number in the set of out-of-bounds areas, extract the injection port direction, injection rate, and channel number of the corresponding area, calculate the current direction vector based on the original injection path, calculate the direction correction angle, extract the current rate deviation and calculate the correction amount, and generate an injection correction parameter set;

[0033] S503: Calling the injection correction parameter set, arranging the numbers in the same area in ascending order, constructing a channel list corresponding to the sequence, binding the correction angle, correction amount and sequence number and outputting them, and generating a self-matching optimization solution for the ductile iron alloy composition.

[0034] The AI-based adaptive optimization system for ductile iron alloy composition includes:

[0035] The temperature difference normalization module obtains the current furnace temperature, carbon-silicon ratio, total amount of deoxidized materials, spheroidization ratio, and incubation time, calculates the difference between the upper and lower temperature limits of the thermal stability zone, divides the offset difference by the temperature difference span, generates a normalized graph, and obtains a composition stability comparison chart.

[0036] The anomaly recognition module extracts the carbon-silicon offset value and the furnace temperature offset value of the composition stability comparison diagram, intercepts and compares them according to the spheroidization stage time, and screens out data combinations with offset differences greater than the average value of the temperature zone expansion and with a decrease in furnace temperature, and generates a usable composition combination set after eliminating them;

[0037] The response curve extraction module reads the tensile, fatigue and impact curves corresponding to the available component combination set, divides them into five segments according to the time axis, extracts the slope, the number of extreme values ​​and the number of curvature reversals, sorts them by stability and integrates them to generate a stable trend response graph group;

[0038] The mold matching module calls the component combination in the stable trend response diagram group, collects the curvature change value, wall thickness slope and injection rate of the corresponding structural parts, constructs a spatial distribution diagram of the three data, and overlays and analyzes it with the component stability comparison legend, marks the overlapping area, and generates a segregation coincidence distribution area map;

[0039] The path correction module extracts the injection rate, structural curvature and concentration gradient of the segregation overlap distribution area map, and compares them with the corresponding standard lines respectively. If all three items exceed the limit, the feed angle, rate and sequence number of the area are calculated to generate a self-matching optimization plan for the ductile iron alloy composition.

[0040] Compared with the prior art, the advantages and positive effects of the present invention are:

[0041] In the present invention, a component offset map is constructed by normalizing the offset values ​​of multiple factors such as furnace temperature and carbon-silicon ratio. Combined with the trend feature extraction and sorting of the response curves under multiple working conditions, component combinations with stable performance are screened out. A mapping relationship is established using the structural curvature, wall thickness slope and injection rate. The spatial overlap map is generated by superposition analysis, and the feed direction and injection rate are inversely calculated to achieve dynamic adaptation of the composition and structure, thereby improving the uniformity of the organization and the consistency of performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0043] Figure 1 Schematic diagram of the steps of the present invention;

[0044] Figure 2is a flow chart of the steps of S1 of the present invention;

[0045] Figure 3 This is a flow chart of the steps of S2 of the present invention;

[0046] Figure 4 This is a flow chart of the steps of S3 of the present invention;

[0047] Figure 5 This is a flow chart of the steps of S4 of the present invention;

[0048] Figure 6 This is a flow chart of the steps of S5 of the present invention;

[0049] Figure 7 It is a system module diagram of the present invention. DETAILED DESCRIPTION

[0050] The technical solution of the present invention is described below in conjunction with the accompanying drawings.

[0051] In the embodiments of the present invention, words such as "exemplarily" and "for example" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as an "exemplary" in the present invention should not be interpreted as being preferred or advantageous over other embodiments or designs. Rather, the use of the word "exemplary" is intended to present concepts in a concrete manner. Furthermore, in the embodiments of the present invention, "and / or" can mean both or either of the two.

[0052] In the embodiments of the present invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, when the distinction is not emphasized, the meanings they convey are the same. The terms "of," "corresponding," and "corresponding" may sometimes be used interchangeably. It should be noted that, when the distinction is not emphasized, the meanings they convey are the same.

[0053] In the embodiments of the present invention, sometimes a subscript such as W1 may be written as a non-subscript such as W1. When the difference is not emphasized, the meanings to be expressed are the same.

[0054] In order to make the technical problems, technical solutions and advantages to be solved by the present invention clearer, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0055] See also Figure 1 , an artificial intelligence-based adaptive optimization method for ductile iron alloy composition, comprising the following steps:

[0056] S1: Obtain the current furnace temperature, carbon-silicon ratio, total amount of deoxidizer added, spheroidizing addition ratio, and incubation time in the smelting process, calculate the offset difference between the value and the upper and lower temperature limits of the thermal stability range, use the temperature span as the denominator to perform a conversion operation on the offset difference, generate a normalized comparison chart, and obtain a composition stability comparison chart;

[0057] S2: Call the carbon-silicon ratio offset item and the furnace temperature offset item in the composition stability comparison diagram, and select points to match the two data items according to the time cutoff of the spheroidization treatment stage. The data combination where the carbon-silicon offset difference is greater than the average amplitude of the temperature zone expansion and the furnace temperature offset direction is downward is set as an abnormal item, and some combinations are excluded to generate a set of available composition combinations;

[0058] S3: Read the curve response records under tension, fatigue and impact conditions corresponding to each component in the available component combination set, divide the curve into five segments according to the time axis, extract the numerical slope, number of extreme values ​​and number of curvature direction changes of each segment, construct trend integration records for the three types of extraction results according to stability sorting, and generate a stable trend response graph group;

[0059] S4: Based on the associated component combinations in the stable trend response diagram group, the curvature change value, wall thickness slope and injection rate data of the corresponding parts in the current casting structure are collected, and a spatial mapping operation is performed to perform cross-region superposition analysis on the three sets of data and the element deviation distribution diagram, locate the coupling concentration area, and generate a segregation overlap distribution area map;

[0060] S5: Based on the spatial overlapping area in the segregation coincidence distribution area map, the three values ​​of channel injection rate, structural curvature in the area and concentration gradient are called to determine whether they all exceed the set standard range line. If they are all satisfied, the feed direction correction angle, rate correction amount and injection sequence number combination are output to generate a self-matching optimization plan for the ductile iron alloy composition.

[0061] The composition stability comparison chart includes the normalized value of furnace temperature offset, the normalized value of carbon-silicon ratio offset, the normalized value of deoxidizer offset, the normalized value of spheroidizing addition ratio offset, and the normalized value of incubation time offset. The available composition combination set includes the carbon-silicon offset out-of-limit combination, the temperature deviation combination, and the effective combination after anomaly elimination. The stable trend response chart group includes the tensile trend chart, the fatigue trend chart, and the impact trend chart. The chart group adopts slope change characteristics, extreme value distribution characteristics, and curvature transformation characteristics. The segregation overlap distribution area diagram includes curvature change distribution, wall thickness slope distribution, injection rate distribution, and element offset overlay layer. The ductile iron alloy composition self-matching optimization plan includes feed direction correction angle, injection rate correction amount, and injection sequence number.

[0062] See also Figure 2 , the specific steps of S1 are:

[0063] S101: Obtain the furnace temperature, carbon content, silicon content, deoxidizer dosage, spheroidizer dosage, metal weight, and inoculation treatment time in the current smelting process, calculate the proportion based on the carbon and silicon content, calculate the addition ratio based on the spheroidizer dosage and metal weight, calculate the treatment time based on the inoculation start and end time, and generate a basic composition parameter set;

[0064] First, extract the furnace temperature value recorded by the real-time monitoring system, obtain the current time node temperature of 1448℃ on the operation interface, then check the sampling results in front of the furnace, read the carbon content of this furnace as 3.7%, the silicon content as 1.8%, check the addition record table to obtain the addition amount of deoxidizer as 2.5kg, and obtain the total amount of spheroidizer added this time as 5.2kg through the ledger. Then, call the total weight of metal used in this furnace from the weighing module as 950kg, and record the start time of inoculant addition as 19:32:15 and the end time as 19:36:30. The processing time is 4 minutes and 15 seconds, that is, 255 seconds. Then, according to the proportional relationship between carbon content and silicon content, calculate the carbon content as 3.7 and silicon content as 1.8. The current carbon-silicon ratio is 2.06 after division, and is recorded as the ratio index for this furnace. The spheroidizer addition ratio is calculated to be 0.00547 based on the ratio of 5.2 kg of spheroidizer to 950 kg of metal, that is, 5.47 kg of spheroidizer is added per ton of metal. Subsequently, it is confirmed that the start and end time difference of the inoculation treatment is the inoculation treatment duration. The system automatically converts the timestamp to 255 seconds. Finally, all data parameters are organized and summarized into a set of basic component parameters containing 7 indicators, namely, furnace temperature 1448°C, carbon content 3.7%, silicon content 1.8%, deoxidizer 2.5 kg, spheroidizer 5.2 kg, metal 950 kg, and inoculation time 255 seconds, which are used for subsequent analysis.

[0065] S102: Based on the furnace temperature in the component basic parameter set, the upper and lower limits of the thermal stability range are called, the temperature offset index is calculated, the interval span is taken as the benchmark, the deviation is divided by the span and the absolute value is taken to generate a temperature offset coefficient group;

[0066] Temperature offset indicator formula:

[0067] ;

[0068] in, Represents the temperature offset index, Represents the temperature offset coefficient group that is ultimately used for control feedback, Represents the real-time furnace temperature during the current smelting process. Represents the lower limit temperature of the thermal stability range of this material, Represents the upper limit temperature of the thermal stability range of the material, Indicates that the furnace temperature data is divided into time intervals and then The sum of the differences between the mean temperature of a segment and the mean temperature of all segments, Representative The average temperature of the segment, represents the overall average value of all temperature segments, Represents the weight coefficient for temperature zone difference correction, Represents the total mass of metal added in the current heat, Represents the metal quality reference value preset by the system, represents the mass normalization correction factor, and n represents the total number of temperature sections;

[0069] The overall structure of the formula introduces addition ( Four types of calculation logic are used: weighted summation, multiplication (the product of the external offset effect and the mass normalization term), fraction (the ratio of offset to span), and absolute value (eliminating the effects of positive and negative offsets). The first part within the brackets represents the initial offset between the furnace temperature and the lower thermal stability limit, superimposed with the corrected cumulative value of the temperature difference between each temperature segment and the global temperature, and proportionally adjusted by the weight coefficient α. The second part is the ratio of the current heat metal weight to the reference value, correcting for the impact of flux load on temperature, with the overall contribution adjusted by the coefficient β. This structural logic eliminates the amplification of false anomalies in the offset coefficient calculation caused by short-term furnace temperature jumps or changes in furnace charge.

[0070] Parameter description and data acquisition logic:

[0071] :The stable temperature data of the current furnace center area is collected in real time by the high-temperature thermocouple and extracted after 5-second sliding average processing. The value is 1448K;

[0072] : By consulting the lower limit database of the thermal stability range, the corresponding temperature control standard for this batch of HT200 grade ductile iron is set to 1400K;

[0073] : The upper limit of the thermal stability range is also set by the database and is set to 1460K;

[0074] n: The total number of temperature segments obtained by dividing the furnace temperature data into equal time intervals, a total of 5 segments;

[0075] : After dividing the first 15 minutes of the furnace temperature curve into 5 equal time segments, the average temperature data of each segment is as follows: , a total of 5 sections;

[0076] :The average temperature of the above 5 sections is calculated as ;

[0077] :The difference sequence is , the sum is 0.0, indicating that the internal distribution of the segment has an average impact on the global distribution;

[0078] : Read the charging record of this furnace, the original added metal weight is 950kg;

[0079] :The standard reference weight is set to 1000kg, which is used as the average loading setting for conventional production lines;

[0080] : is the weight coefficient of the temperature segment deviation. Its setting is based on the statistical degree of influence of local fluctuations on thermal control in historical working conditions and is set to 0.1. This value ensures that the segment difference ratio does not exceed the overall furnace temperature deviation master control item;

[0081] : Metal mass normalization correction coefficient, set to 1.0. This value is based on the full load (1000kg) benchmark for multiplication processing and does not require additional scaling;

[0082] : It is the final calculated temperature deviation coefficient group, indicating the degree of temperature deviation of this furnace.

[0083] Bring the actual example into the calculation:

[0084] Substitute all the above parameter values ​​into the formula and calculate step by step as follows:

[0085] Calculate the initial offset term for the furnace temperature:

[0086] ;

[0087] Calculate the sum of the piecewise correction terms:

[0088] ,therefore

[0089] ;

[0090] Correction offset term = initial offset + segment correction = 48 + 0 = 48;

[0091] Calculate the mass normalization term:

[0092] ;

[0093] ;

[0094] Offset product = 48 × 1.95 = 93.6;

[0095] Interval span = ;

[0096] Final offset coefficient calculation:

[0097] ;

[0098] The table shows the relevant parameter data:

[0099] Table 1 Monitoring section furnace temperature distribution and metal quality data

[0100]

[0101] As shown in Table 1, the relevant data are obtained through temperature sampling, historical process configuration tables and charge loading records, and have source certainty and statistical rationality.

[0102] Numerical significance and result interpretation:

[0103] The results show that the current furnace has an offset coefficient group of 1.56 after considering the overall furnace temperature offset and weighted consideration of the segmented temperature fluctuation stability and metal mass load difference, which exceeds the conventional offset control range of [0, 1.2]. This indicates that the temperature control range has deviated too much and a control submodule should be introduced to perform temperature correction or casting parameter fine-tuning. The temperature normalization factor used in the component normalization process in step S103 is directly involved in the normalization calculation.

[0104] The benefit of the formula is that by introducing the adjustment terms of the weighted sum of the temperature segment fluctuation values ​​and the metal mass normalization ratio, a multi-parameter dynamic offset correction mechanism is established, making the offset coefficient more sensitive and discriminatory in non-steady-state melting conditions; this method can make flexible adjustments based on the furnace temperature distribution and charging differences, effectively enhancing the dynamic adaptability during subsequent composition normalization processing.

[0105] S103: calling the component basic parameter set and the temperature offset coefficient group, normalizing all parameters according to the temperature offset coefficient ratio to generate a comparison chart, drawing a histogram and marking the legend label to obtain a component stability comparison legend;

[0106] Call the composition basic parameter set and temperature offset coefficient group, multiply each parameter by the temperature offset coefficient of the corresponding heat for numerical normalization. For example, when the offset coefficient is 0.8, the original carbon content of 3.7% is normalized to 2.96%, the silicon content of 1.8% is normalized to 1.44%, the deoxidizer of 2.5kg is normalized to 2.0kg, the spheroidizer of 5.2kg is normalized to 4.16kg, the metal of 950kg is normalized to 760kg, and the inoculation time of 255s is normalized to 204s. The normalization is completed by calling each original value and the offset coefficient of the heat for item-by-item multiplication operation. If the offset coefficient of another heat is 0.33, the carbon content of 3.7% is normalized to 2.96%, the silicon content of 1.8% is normalized to 1.44%, the deoxidizer of 2.5kg is normalized to 2.0kg, the spheroidizer of 5.2kg is normalized to 4.16kg, the metal of 950kg is normalized to 760kg, and the inoculation time of 255s is normalized to 204s. The content is the same as 3.7%, which is 1.22% after normalization. The stability performance can be analyzed after comparison. A bar chart is generated by normalizing the parameters. In the graph drawing, the horizontal axis is set to the parameter type and the vertical axis is set to the normalized value. Different colors are used to identify different furnaces, such as the current batch is marked in blue and the previous batch is marked in green. The legend clearly marks items such as "Carbon content-current furnace: 2.96%", "Carbon content-comparison furnace: 1.22%", etc., to ensure that each bar line has a clear value and category label. At the same time, a data integrity check is performed before drawing to ensure that there are no missing items or abnormal values. After outputting the graph, a component stability comparison legend that can be directly analyzed is formed.

[0107] See also Figure 3 , the specific steps of S2 are:

[0108] S201: calling the carbon-silicon ratio offset item and the furnace temperature offset item in the composition stability comparison diagram, extracting each corresponding data pair in chronological order based on the time node of the spheroidization treatment stage, and combining the two data items into a group to generate a carbon-temperature corresponding group set;

[0109] Call the carbon-silicon ratio offset item and furnace temperature offset item in the composition stability comparison legend. First, set several time nodes in the complete period of spheroidization treatment. For example, when the total treatment period is 300 seconds, divide it into 10 time nodes every 30 seconds, numbered from T1 to T10 in sequence, and then extract the carbon-silicon offset value and the corresponding furnace temperature offset value at that moment from the comparison legend in order of node time. For example, the carbon content corresponding to the T1 node is 3.8%, the silicon content is 1.7%, and the offset value is 2.1. At this time point, the furnace temperature is 1410℃, and the thermal stability range is 1400℃~1460℃. The current furnace temperature is lower than the median value, which is classified as a section with a smaller offset coefficient. The offset value can be The value of carbon-silicon offset is 1.9 at T2, the furnace temperature is 1442℃, and the furnace temperature offset value is recorded as 0.7. The group is (1.9, 0.7). In this way, all data values ​​are extracted node by node to form multiple component control groups. Each group contains two values ​​and corresponds to the time node one by one. Finally, the data groups generated by all nodes are recorded as carbon-temperature corresponding group sets to ensure that all data items are derived from the legend records and are arranged and combined in a logical time order during processing. All data groups are retained in the structure table for the next stage of data comparison and screening operations.

[0110] S202: Based on the carbon-temperature corresponding group set, conditional judgment is performed on each data set to screen data whose carbon-silicon offset difference is greater than the average amplitude of the thermal stability range expansion and whose corresponding furnace temperature offset is negative. Data that meets the conditions is marked as an abnormal group to generate an offset abnormality set;

[0111] According to all the data groups in the carbon temperature corresponding group set, the conditional judgment is performed, and the carbon-silicon offset value and the furnace temperature offset value in each group are called one by one. First, it is confirmed whether the carbon-silicon offset of this group is greater than the average expansion amplitude of the thermal stability range. The expansion average amplitude is set to 1.6 here. The source is the evaluation threshold set by taking the average between the median offset value 1.2 and the highest offset value 2.0 in the previous statistics of 100 furnace carbon-silicon offset values. Then, it is judged whether the furnace temperature offset value in the current data group is negative. The judgment standard is that the furnace temperature offset value recorded when it is lower than the median value of the thermal stability range of 1430℃ is less than 0.5, which means the offset is negative. If both conditions are met , then mark the group as an abnormal group. During the execution process, take group T4 as an example. The carbon-silicon offset is 1.85, and the furnace temperature offset value is 0.42. Since 1.85 is greater than 1.6 and 0.42 is less than 0.5, group T4 is judged to be abnormal and marked. If the carbon-silicon offset of group T5 is 1.9 and the furnace temperature offset value is 0.65, although the first condition is met, the second condition is not established, and it is not marked. This judgment process is executed group by group, and all data groups that meet the two conditions are summarized and recorded to form an offset abnormality set. In this set, only the original offset value, furnace temperature value and time index of the abnormal data are retained to form an identification set for subsequent elimination and calibration.

[0112] S203: calling the carbon temperature corresponding group set and the offset anomaly set, removing the anomaly group from all combinations, retaining the combination and re-arranging the index order to obtain a usable component combination set;

[0113] The carbon temperature corresponding group set and the offset abnormality set are called to start the data screening process. Each set of data in the carbon temperature corresponding group set is read one by one and compared with the offset abnormality set to determine whether it belongs to the abnormal group. If the current data group has been marked in the abnormal set, it will not be included in the retained group set and the data will be skipped. If it does not appear in the abnormal set, the current group data will be directly copied to the available data buffer area. For example, if the T2, T3, T5, T6, T9, and T10 data groups are not marked as abnormal, their carbon-silicon offset values ​​and furnace temperature offset values ​​will be retained to form an available data set. After the screening is completed, the data will be reused. The index order of the retained group is rebuilt and renumbered in order of time nodes, starting from G1 to ensure that the data structure is continuous and the index is consistent. After all elimination operations are completed, the retained carbon temperature data groups are obtained, such as G1 (1.9, 0.7), G2 (1.75, 0.82), G3 (1.6, 0.55), G4 (1.45, 0.88), G5 (1.5, 0.77), and G6 (1.3, 0.85). The above data are recorded as the available component combination set for subsequent parameter normalization and control factor calculation operations.

[0114] See also Figure 4 , the specific steps of S3 are:

[0115] S301: Reading the curve response records of multiple groups of components under three types of working conditions: tension, fatigue, and impact, from the available component combination set, dividing each curve into five segments according to the time axis, extracting the numerical slope, number of local extreme values, and number of curvature direction changes of the continuous data in each segment, and generating a curve structure feature group;

[0116] Read the curve response records of multiple groups of components in the available component combination set under three types of working conditions: tension, fatigue and impact. First, extract the stress-strain response curve or deformation-time response curve generated in the three types of working conditions one by one according to the combination number. Each response curve is divided into five segments equidistantly along the time axis. For example, a tensile curve with a total test time of 100 seconds is divided into five segments: 0–20s, 20–40s, 40–60s, 60–80s and 80–100s. Extract a sequence of continuous data points in each segment, calculate the numerical slope of the current segment, that is, the ratio of the stress change to the time change at adjacent time points, and then count the number of local maximum and minimum stress values ​​in the segment as the number of local extreme values. Determine whether the curvature direction switches between positive and negative directions. For example, if the slope first changes from positive to negative and then to positive in the 20-40s segment, the number of curvature direction changes is 2. After each calculation is completed, record the slope value, number of extreme values, and number of curvature changes. For example, the slope of the first segment is 2.5, the number of extreme values ​​is 3, and the number of changes is 1. Repeat this operation for all five segments of each curve and repeat the process under the three working conditions of tension, fatigue, and impact. Finally, generate three index values ​​for each response curve, a total of 15 indicators, which are combined to form the structural feature group of the curve. They are stored in the form of a structure table and associated with the corresponding working condition and combination number by number. After the summary is completed, they are uniformly archived as a curve structural feature group.

[0117] S302: Based on the three indicator values ​​in the curve structure feature group, sort each type of feature by the fluctuation amplitude of the value, score the sorting stability and record the sequence position, and establish a multidimensional data set based on the three sorting orders to obtain a structural sorting trend set;

[0118] According to the three index values in the curve structure feature group, the fluctuation amplitude of each type of structure index under each working condition is sorted, the maximum and minimum values of the five slopes in the tensile working condition are extracted under all combination numbers, the fluctuation amplitude is calculated and the sorting position is recorded, for example, the slope amplitude of G1 is 3.2, the slope amplitude of G2 is 1.9, and the slope amplitude of G3 is 2.6, then the sorting result is G1>G3>G2, and the sequence bit is 1, 2 and 3, and the extreme value number and the curvature direction transformation number are processed in the same way. The fluctuation amplitude of each index under each working condition is used as the basis for numerical comparison, and the judgment basis is that the larger the amplitude value, the higher the sorting position, and if the amplitude values are the same, the secondary comparison is made according to the frequency of occurrence. After the sorting is completed, the stability score of each group of sorting sequence in the three indexes is calculated, the scoring method is to calculate the difference between the maximum difference and the minimum difference of the current combination number in the three index sorting, for example, the sorting of the combination number G1 in the three indexes is 2, 1 and 3, and the stability score is 2, and the smaller the stability score, the more consistent the structure response is. The score is recorded under the number, and three-dimensional coordinate value set is constructed in the order of tensile, fatigue and impact according to the three types of sorting sequence, and the structure sorting trend data point of the number is recorded, and the structure sorting trend data of all numbers is generated, and the structure sorting trend set is summarized.

[0119] S303: Call the structure sorting trend set, map the working condition type corresponding to the sorting sequence to the curve number, and arrange and classify according to the number sequence and construct the group label to generate the stable trend response graph group;

[0120] Call the structure sorting trend set, and map the working condition type recorded in each sorting sequence to the corresponding structure feature result in sequence, retrieve the number, three types of index items and sorting data corresponding to the working conditions in the archival record of each curve, and assemble these contents as the label information of the structure graph group, for example, the number G2 has the most curvature transformation times in the tensile working condition, the lowest extreme value number in the fatigue working condition, and the most gentle slope in the impact working condition, and these features are written in the legend label in the form of text and numerical value. In the group construction, each number corresponds to a complete map, which contains three types of five-section structure feature labels, and labels such as "maximum slope change working condition: fatigue", "most intensive extreme value working condition: impact" and "most frequent curvature transformation working condition: tensile" are marked on the lower side of the map to ensure that the graph group completely depends on the number sequence in the structure sorting trend set, so that all the maps are arranged in sequence from 1 to N, and the stable trend response graph group with number, feature label and structure attribute is generated.

[0121] Please refer to Figure 5 , the specific steps of S4 are:

[0122] S401: Based on the combination of components listed in the stable trend response graph set, the curvature change value, wall thickness slope and injection rate data of the corresponding parts in the current mold structure are collected, a corresponding table is established according to the part number, and three groups of data are arranged into equal length sequences respectively, which are mapped to spatial nodes to generate a structure response coordinate set;

[0123] Based on the combination of components listed in the stable trend response graph set, for each numbered sample mold piece, after the actual molding process is completed, enter the mold structure detection stage, select the corresponding parts such as the gate vicinity, the riser area, the bottom thick wall corner, the cold separation prone position, etc. The preset detection point numbers P1 to P20 are selected, the surface curvature change value is measured by a three-dimensional scanner, and the curvature value is obtained by calculating the normal change amount per unit area, for example, the curvature of P3 point is 0.24, and the curvature of P8 point is 0.57. At the same time, the wall thickness gradient of the mold structure at this position is read, the wall thickness slope is obtained by dividing the thickness change by the distance difference, for example, the thickness at P3 changes from 7mm to 5.5mm, the distance is 15mm, and the slope is -0.1. The injection rate data comes from the record of the pouring start and completion time in the casting real-time data acquisition system, and the conversion data of the injection port cross-sectional area and the actual volume, such as the total injection time of the flow-through section at P3 point is 1.5s, the total volume is 230cm 3 , and the instantaneous injection rate is 153.3cm 3 / s. The curvature change value, wall thickness slope and injection rate are aligned with the same timestamp and number, and a number corresponding table is established according to P1 to P20, which ensures that the three types of data are aligned according to the same sequence length, and the curvature, slope and rate are arranged in sequence in each group of data points. Convert into equal length three-dimensional sequence, further map the spatial nodes according to the actual position of the detection points in the three-dimensional coordinate system of the mold, label each group of data values on the corresponding nodes, and construct a complete structure response coordinate set.

[0124] S402: Call the structure response coordinate set and the configured element deviation distribution graph, perform two-dimensional area mapping processing on the three groups of structure response data and deviation graph, perform coordinate cross matching based on node position and mark the overlapping identifier, count the number and distribution boundary of overlapping nodes, and obtain the index set of the overlapping area;

[0125] Call the structural response coordinate set and the configured element deviation distribution map. First, extract the element concentration abnormal area marked on the deviation map, extract the two-dimensional coordinate points in each area one by one, and then extract the corresponding spatial node coordinates from the structural response coordinate set. Match the horizontal and vertical coordinate values ​​according to the XY axis. The judgment standard is that when the difference between the two coordinate points in the XY axis direction is less than 2mm, the nodes are considered to coincide. Set the 2mm as the spatial registration threshold, which comes from the error tolerance set for the maximum allowable deviation of scanning positioning in the previous error analysis. During the matching process, read the X, Y position of each structural response node one by one and compare it with each coordinate point on the deviation distribution map. Perform position difference calculation. If the difference is less than the threshold at the same time, the point is set as a coincidence point and marked. The marking method is to add the "⊗" symbol to the node coordinates and record the node number in the data set in the temporary coincidence list. After processing all nodes, summarize the list and count the total number. For example, among nodes P1 to P20, there are five points P3, P6, P12, P15 and P18 that meet the coincidence condition. Record their numbers, spatial positions and corresponding three types of structural response values. Then, based on the spatial distribution range of these coincidence points, calculate the maximum and minimum values ​​of their XY coordinates to form a distribution bounding box. The diagonal of the bounding box is the limit envelope area. The above data constitutes the coincidence area index set.

[0126] S403: Based on the coincident region index set, using the mold structure cross-section as the base map, superimpose and mark the boundary closed regions at multiple nodes, generate polygonal outer frames for display, output the relationship between the drawing group labels and the spatial coordinates according to the numbering identifiers, and generate a segregation coincident distribution region map;

[0127] According to the coincidence area index set, the casting structure profile is loaded as the base map, and the identified coincidence nodes are mapped to the corresponding positions of the actual profile by layer overlay. First, a marker point is drawn at each node position, and the distribution boundary to which it belongs is read. All coincidence points on the boundary are connected point by point to form a closed path line. In the coincidence set with more than 3 nodes, point by point connection is performed and the end and end are closed to form a closed polygonal area. For example, the numbers P3, P6, and P12 constitute the coincidence triangle area, and P15 and P18 are added to form a pentagonal closed area. Then, the outer frame line is drawn along the path boundary and filled with grayscale texture to mark the closed area. At the same time, a number mark is added at the center of the boundary, such as "R1", "R2", etc., and the corresponding coincidence point index is recorded. Then, for each group of polygon structures, the coordinate point position, the corresponding node number, and the structural response value are recorded and output as a map group structure. Finally, all coincidence closed areas are marked and attached to the atlas base map annotation layer according to the sequential number to generate a segregation coincidence distribution area map.

[0128] See also Figure 6 , the specific steps of S5 are:

[0129] S501: Based on the spatial overlap areas marked in the segregation coincidence distribution area map, three values ​​of channel injection rate, structural curvature, and concentration gradient are collected at the corresponding positions. The set standard range line is called to perform interval judgment on the multiple values ​​and the upper and lower limits of the multiple self-standards. The numbers of all out-of-bounds areas are filtered to generate an out-of-bounds area set.

[0130] According to the spatial overlapping areas marked in the segregation coincidence distribution area map, the position coordinates corresponding to the area number are read one by one, and the three types of data, namely, the channel injection rate, structural curvature and concentration gradient, at the actual casting mold or numerical simulation platform are collected. The injection rate is calculated by extracting the local filling volume change per unit time from the casting data record. For example, the channel flow rate per unit time in the R3 area record is 125 cm 3 / s, the structural curvature is obtained by the rate of change of the local surface tangent and normal directions in the 3D scanning data. If the curvature value in the R3 area is 0.38, the concentration gradient is obtained by the spectrometer to obtain the ratio of the change of the element content along a certain length direction. For example, if the element concentration changes by 0.27% within 10mm from the center to the edge, the concentration gradient of this section is 0.027 / mm. After obtaining all the data, they are compared with the standard range line set in the system. The standard is obtained by statistics of 200 batches of qualified samples in the past. The normal range of injection rate is 80–110cm 3 / s, the curvature reference interval is 0.15–0.35, and the concentration gradient allowable range is 0.015–0.025 / mm. Exceeding either side of the upper and lower limits is considered out of bounds. The judgment process performs numerical judgment on each data item and the corresponding upper and lower limits in turn. For example, the R3 injection rate 125 is greater than the upper limit 110 and the curvature 0.38 is greater than the upper limit 0.35. Both items are out of bounds. The R3 number is recorded, and the judgment process is executed to all regional nodes. The numbers of all data points where any item is out of bounds are classified into the out-of-bounds region set. For example, the final record is R2, R3, R7, and R11, a total of four region numbers, to form a number set for subsequent path correction.

[0131] S502: Based on the number in the out-of-bounds area set, extract the injection port direction, injection rate, and channel number of the corresponding area, calculate the current direction vector based on the original injection path, calculate the direction correction angle, extract the current rate deviation and calculate the correction amount, and generate an injection correction parameter set;

[0132] The calculation formula for the direction correction angle is as follows:

[0133] ;

[0134] in, A comprehensive expression representing the direction correction angle, Represents the final direction angle correction value output as the control parameter, unit direction vector representing the current injection path of the cross-boundary region, reference direction vector representing the original design of the injection channel, dot product between two vectors, and respectively the length of two vectors, arccosine value representing the included angle, rate-area perturbation weight coefficient, instantaneous injection rate of the injection port numbered the difference between and the average rate in the region , and the product of the offset of the cross-sectional area of the injection port relative to the average cross-sectional area summed, area offset

[0135] correction coefficient, the number of injection ports in the current cross-boundary region;

[0136] The formula consists of two main parts: the first term constructs the angular deviation between the current direction and the design direction by the inverse value of the cosine of the included angle, and the second term is the sum of the weighted products of the difference between the instantaneous injection rate and the average rate , multiplied by the normalized difference between the cross-sectional area of the corresponding injection port and the average area , and the modified angle is introduced after the product term is summed, weighted, and square rooted, which represents the complex effect of multiple source physical field parameters on direction correction. Addition is used for total offset superposition, multiplication is used for local offset coupling, fraction is used for area normalization conversion, summation is used for spatial overall aggregation response, absolute value is used to eliminate the directional influence of direction bias, and root conversion converts nonlinear growth into linear weight accumulation.

[0137] Parameter description and numerical acquisition method:

[0138] : Obtain the actual flow direction vector in the cross-boundary region by three-dimensional scanning of the structure, construct a three-point fitting vector for each channel segment, and the example data is ;

[0139] : Extract the theoretical direction vector of the channel from the casting structure design model, which is the centerline guide vector of the injection port, and the example is ;

[0140] : Collect the instantaneous flow rate data of the injection port, which is calculated from the volume flow rate by the pouring monitoring system, and the example is:

[0141] ;

[0142] Take the average of the above flow rates, calculated as ;

[0143] : The injection port cross-sectional area is obtained from CAD model section measurement, example is

[0144] ;

[0145] : The average area of the zone injection port, calculated as ;

[0146] : The area relative offset reference coefficient, set as , which is set based on the area offset affecting linearized distribution interval;

[0147] : The disturbance response weight factor, set as , determined by historical 60 batches of error inverse adjustment value fitting value;

[0148] : The number of injection ports participating in summation, the implementation is ;

[0149] : The final angle correction amount, as the output control parameter value.

[0150] The table shows the relevant parameter data:

[0151] Table 2 Out-of-range area direction correction calculation parameter table

[0152]

[0153] As shown in Table 2, the product item sum of the data of 3 injection ports after quantization is , which will be brought into the formula later.

[0154] The actual example is completely substituted:

[0155] Calculate the direction vector angle item:

[0156] ;

[0157] ;

[0158] Calculate the correction disturbance item:

[0159] ;

[0160] Calculate the final correction angle:

[0161] ;

[0162] Result interpretation:

[0163] The results show that the current cross-border region number comprehensive direction correction angle is 28.81°, which exceeds the maximum allowed direction offset threshold (15°) set by the system, so it is determined as "serious direction offset state", and the injection path direction reconstruction or delay injection strategy should be executed; The angle value will be passed as a parameter into the next injection path correction module, and the electromagnetic valve or steering joint in the control channel will adjust the current injection vector direction.

[0164] Innovative explanation of the formula:

[0165] The beneficial effect of the formula is that by introducing the coupling correction structure of the direction vector cosine difference term and the instantaneous speed-channel area structure parameter, an integrated offset angle index based on the linkage of geometric physics and fluid transport state is constructed, which can identify the angle error caused by channel shape and working condition disturbance, realize quantitative identification and angle feedback expression of complex mold injection inlet multi-source disturbance, and effectively improve the precision dynamic response ability of injection process control.

[0166] S503: Call the injection correction parameter set, arrange the region numbers in ascending order, construct the channel list corresponding to the order, bind the correction angle, correction amount and order number, and output the nodular cast iron alloy composition self-matching optimization scheme;

[0167] Call the injection correction parameter set, first perform ascending order operation on all region numbers in the set to ensure that the numbers are arranged in order for process path construction, such as cross-border region numbers R2, R3, R7, R11, which are arranged in order as R2, R3, R7, R11 after sorting, then read the channel numbers corresponding to each number such as CH05, CH06, CH09, CH12, and form a channel list by arranging the channel numbers in the same order, construct the sequential channel path structure, and bind the correction angle and correction rate value corresponding to each channel number, ensure that each correction parameter can be directly positioned to the node position in the channel path, for example, CH05 binds the correction angle -12°, the correction amount -17cm / s, CH06 binds the correction angle -8°, the correction amount -22cm / s, and marks the correspondence between the number, parameter item and path order in the list, forms a channel number-parameter binding table, and finally summarizes all the binding data to form a complete nodular cast iron composition injection path adjustment list as the composition self-matching optimization scheme output. The structure field in the scheme list is fixed as: region number, channel number, correction angle, correction flow rate, serial number, to ensure that the scheme has channel control node positioning ability and adjustment parameter execution guidance. 3 3 ​​

[0168] Please refer to Figure 7 The artificial intelligence-based nodular cast iron alloy composition self-adaptive optimization system comprises:

[0169] The temperature difference normalization module obtains the current furnace temperature, carbon-silicon ratio, total amount of deoxidizing substance, spheroidizing ratio and inoculation time, calculates the difference value with the upper and lower limit temperature of the thermal stability zone, divides the offset difference by the temperature difference span, generates a normalized graph column, and obtains a composition stability comparison legend.

[0170] The abnormality recognition module extracts the carbon-silicon offset value and furnace temperature offset value of the composition stability comparison legend, compares them according to the spheroidizing stage time, screens data combinations with offset difference greater than the average value of the temperature zone expansion and furnace temperature drop, and generates a set of available composition combinations after elimination.

[0171] The response curve extraction module reads the tensile, fatigue and impact curves corresponding to the available composition combination set, divides them into five segments according to the time axis, extracts the slope, extreme value number and curvature inversion number, integrates them after stable sorting, and generates a stable trend response graph group.

[0172] The mold matching module calls the composition combination in the stable trend response graph group, collects the curvature change value, wall thickness slope and injection rate corresponding to the structure part, constructs a spatial distribution graph of the three data, and analyzes it in combination with the composition stability comparison legend, marks the overlapping area, and generates a segregation overlap distribution area graph.

[0173] The path correction module extracts the injection rate, structure curvature and concentration gradient of the segregation overlap distribution area graph, respectively compares them with the corresponding standard line, and calculates the feeding angle, rate and sequence number of the region if all three exceed, to generate a nodular cast iron alloy composition self-matching optimization scheme.

[0174] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto. Any skilled person in the art can easily think of changes or replacements within the technical scope disclosed by the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. An artificial intelligence-based adaptive optimization method for ductile iron alloy composition, characterized in that: The following steps are involved: S1: Obtain the current furnace temperature, carbon-silicon ratio, total amount of deoxidizer added, spheroidizing addition ratio, and incubation time in the smelting process, calculate the offset difference between the value and the upper and lower temperature limits of the thermal stability range, perform temperature span conversion, and generate a composition stability comparison chart; S2: calling the carbon-silicon ratio offset item and the furnace temperature offset item in the composition stability comparison diagram, cutting the matching data at the spheroidization stage, setting the combination where the carbon-silicon offset difference is greater than the average amplitude of the temperature zone expansion and the furnace temperature offset direction is downward as an abnormal item, excluding some combinations, and generating a set of available composition combinations; S3: reading the tensile, fatigue, and impact curve records corresponding to the available component combination set, dividing the curve into five segments, extracting the numerical slope, number of extreme values, and number of curvature direction changes of each segment, constructing trend integration records according to stability sorting, and generating a stable trend response graph group; S4: Based on the component combination of the stable trend response diagram group, collecting the curvature change value, wall thickness slope, and injection rate of the corresponding part of the mold structure, performing spatial mapping and superimposing analysis with the element deviation distribution diagram, locating the coupling concentration area, and generating a segregation overlap distribution area diagram; S5: Based on the segregation overlap distribution area map, the channel injection rate, the structural curvature within the area, and the concentration gradient are called. If all three exceed the standard range line, the feed direction correction angle, the rate correction amount, and the injection sequence number combination are output to generate a self-matching optimization plan for the ductile iron alloy composition; The composition stability comparison diagram includes the normalized value of furnace temperature offset, the normalized value of carbon-silicon ratio offset, the normalized value of deoxidizer offset, the normalized value of spheroidizing addition ratio offset, and the normalized value of incubation time offset. The available composition combination set includes the carbon-silicon offset exceeding limit combination, the temperature deviation combination, and the effective combination after abnormal elimination. The stable trend response diagram group includes the tensile trend diagram, the fatigue trend diagram, and the impact trend diagram. The diagram group adopts the slope change feature, the extreme value distribution feature, and the curvature transformation feature. The segregation overlap distribution area diagram includes the curvature change distribution, the wall thickness slope distribution, the injection rate distribution, and the element offset superposition layer. The ductile iron alloy composition self-matching optimization scheme includes the feed direction correction angle, the injection rate correction amount, and the injection sequence number.

2. The method for adaptive optimization of ductile iron alloy composition based on artificial intelligence according to claim 1, characterized in that: The specific steps of S1 are: S101: Obtain the furnace temperature, carbon content, silicon content, deoxidizer dosage, spheroidizer dosage, metal weight, and inoculation treatment time in the current smelting process, calculate the proportion based on the carbon and silicon content, calculate the addition ratio based on the spheroidizer dosage and metal weight, calculate the treatment time based on the inoculation start and end time, and generate a basic composition parameter set; S102: Based on the furnace temperature in the composition basic parameter set, the upper and lower limits of the thermal stability interval are called, the temperature offset coefficient is calculated, the interval span is taken as a reference, the deviation is divided by the span and the absolute value is taken to generate a temperature offset coefficient group; S103: calling the component basic parameter set and the temperature offset coefficient group, normalizing all parameters according to the temperature offset coefficient ratio to generate a comparison chart, drawing a histogram and marking a legend label to obtain a component stability comparison legend.

3. The method for adaptive optimization of ductile iron alloy composition based on artificial intelligence according to claim 1, characterized in that: The specific steps of S2 are: S201: calling the carbon-silicon ratio offset item and the furnace temperature offset item in the composition stability comparison diagram, extracting each corresponding data pair in chronological order based on the time node of the spheroidization treatment stage, and combining the two data items into a group to generate a carbon-temperature corresponding group set; S202: Based on the carbon-temperature corresponding group set, conditional judgment is performed on each group of data to screen data whose carbon-silicon offset difference is greater than the average amplitude of the thermal stable range expansion and whose corresponding furnace temperature offset is negative, and the data that meets the conditions is marked as an abnormal group to generate an offset abnormality set; S203: calling the carbon temperature corresponding group set and the offset anomaly set, removing the anomaly group from all combinations, retaining the combination and re-arranging the index order to obtain an available component combination set.

4. The method for adaptive optimization of ductile iron alloy composition based on artificial intelligence according to claim 1, characterized in that: The specific steps of S3 are: S301: Reading the curve response records of multiple groups of components in the available component combination set under three types of working conditions: tension, fatigue, and impact, dividing each curve into five segments according to the time axis, extracting the numerical slope, number of local extreme values, and number of curvature direction changes of continuous data in each segment, and generating a curve structure feature group; S302: Based on the three indicator values ​​in the curve structure feature group, sort each type of feature by the fluctuation amplitude of the value, score the sorting stability and record the sequence position, and establish a multidimensional data set based on the three sorting orders to obtain a structural sorting trend set; S303: calling the structural sorting trend set, mapping the operating condition type and structural characteristic result corresponding to the sorting sequence to the curve number, sorting and classifying them in the order of the numbers and constructing a graph group annotation to generate a stable trend response graph group.

5. The method for adaptive optimization of ductile iron alloy composition based on artificial intelligence according to claim 1, characterized in that: The specific steps of S4 are: S401: Based on the component combinations listed in the stable trend response graph group, collecting curvature change values, wall thickness slope and injection rate data of corresponding parts in the current casting structure, establishing a corresponding table according to the part number, and respectively arranging the three groups of data into sequences of equal length, mapping them to spatial nodes, and generating a structural response coordinate set; S402: calling the structural response coordinate set and the configured element deviation distribution map, performing two-dimensional area mapping processing on the three sets of structural response data and the deviation map, performing coordinate cross-matching based on node positions and marking overlap marks, counting the number of overlapping nodes and distribution boundaries, and obtaining an overlap area index set; S403: Based on the overlap region index set, with the mold structure cross-section as the base map, the closed boundary regions are superimposed and marked at multiple nodes, and a polygonal outer frame is generated for display. The relationship between the drawing group labels and the spatial coordinates is output according to the numbering identifier to generate a segregation overlap distribution region map.

6. The method for adaptive optimization of ductile iron alloy composition based on artificial intelligence according to claim 1, characterized in that: The specific steps of S5 are: S501: Based on the spatial overlapping areas marked in the segregation coincidence distribution area map, three values ​​of channel injection rate, structural curvature, and concentration gradient are collected at the corresponding positions. The set standard range line is called to perform interval judgment on the multiple values ​​and the upper and lower limits of the multiple standards. The numbers of all out-of-bounds areas are filtered to generate an out-of-bounds area set. S502: Based on the number in the set of out-of-bounds areas, extract the injection port direction, injection rate, and channel number of the corresponding area, calculate the current direction vector based on the original injection path, calculate the direction correction angle, extract the current rate deviation and calculate the correction amount, and generate an injection correction parameter set; S503: Calling the injection correction parameter set, arranging the numbers in the same area in ascending order, constructing a channel list corresponding to the sequence, binding the correction angle, correction amount and sequence number and outputting them, and generating a self-matching optimization solution for the ductile iron alloy composition.

7. The artificial intelligence-based adaptive optimization system for ductile iron alloy composition is characterized by: The method for adaptive optimization of ductile iron alloy composition based on artificial intelligence according to any one of claims 1 to 6, wherein the system comprises: The temperature difference normalization module obtains the current furnace temperature, carbon-silicon ratio, total amount of deoxidized materials, spheroidization ratio, and incubation time, calculates the difference between the upper and lower temperature limits of the thermal stability zone, divides the offset difference by the temperature difference span, generates a normalized graph, and obtains a composition stability comparison chart. The anomaly recognition module extracts the carbon-silicon offset value and the furnace temperature offset value of the composition stability comparison diagram, intercepts and compares them according to the spheroidization stage time, and screens out data combinations with offset differences greater than the average value of the temperature zone expansion and with a decrease in furnace temperature, and generates a usable composition combination set after eliminating them; The response curve extraction module reads the tensile, fatigue and impact curves corresponding to the available component combination set, divides them into five segments according to the time axis, extracts the slope, the number of extreme values ​​and the number of curvature reversals, sorts them by stability and integrates them to generate a stable trend response graph group; The mold matching module calls the component combination in the stable trend response diagram group, collects the curvature change value, wall thickness slope and injection rate of the corresponding structural parts, constructs a spatial distribution diagram of the three data, and overlays and analyzes it with the component stability comparison legend, marks the overlapping area, and generates a segregation coincidence distribution area map; The path correction module extracts the injection rate, structural curvature and concentration gradient of the segregation overlap distribution area map, and compares them with the corresponding standard lines respectively. If all three items exceed the limit, the feed angle, rate and sequence number of the area are calculated to generate a self-matching optimization plan for the ductile iron alloy composition.

Images