Automatic material grading and classifying method based on large model

Abstract

The invention relates to the technical field of artificial intelligence, in particular to an automatic material grading and classifying method based on a large model, and the method comprises the following steps: obtaining material morphological characteristics and attribute distribution, analyzing association in combination with an original category, extracting stability and consistency indexes, screening a high-adaptation path, collecting a monitoring value, and performing performance evaluation. The method comprises the following steps: analyzing morphological characteristics and attribute distribution of materials, retrieving a current classification state comparison target adjustment strategy, calling an instruction set to compare delay and fluctuation, and outputting a classification recovery path and a synchronous node configuration table. Response monitoring and performance evaluation of classification nodes ensure high efficiency, classification strategies can be dynamically adjusted to cope with task cycle changes, adaptability and accuracy are improved, material grading and classification tasks can be adjusted in real time, unstable factors are reduced, higher-quality classification results and an accurate recovery path are ensured, and the classification efficiency is improved. And the flexibility and the intelligent level of material classification are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, in particular to a material automatic grading and classification method based on a large model. BACKGROUND

[0002] The technical field of artificial intelligence relates to the research and application of simulating human intelligence using computer systems, including natural language processing, computer vision, pattern recognition, knowledge reasoning, and training and inference of large-scale machine learning models. Through the acquisition, labeling and modeling of massive data, the technical field realizes the automatic understanding and processing of information, forming a systematic technical system from data collection, feature extraction, semantic understanding to intelligent decision-making. Among them, the traditional material automatic grading and classification method based on a large model refers to the classification of material objects in industrial production, logistics transportation or warehouse management, etc. through artificial rule setting or feature comparison based on shallow learning. The method relies on artificially extracted morphological features, physical properties or historical classification samples as the basis for judgment, and combines preset threshold comparison, pattern matching or linear discrimination to automatically grade and classify materials.

[0003] The prior art relies on artificial rule setting or feature comparison based on shallow learning to complete the grading and classification of materials. The process lacks flexibility and adaptability, is easily affected by external environmental changes or data fluctuations, and has poor classification result stability and is limited by artificially set thresholds, resulting in inaccurate classification. The material grading of the prior art mainly relies on historical classification samples or preset physical properties. In the face of complex and variable production or logistics environments, there is a lack of sufficient real-time adjustment mechanism, and the classification strategy cannot be optimized in a timely manner according to different task cycle requirements, resulting in a decrease in classification accuracy or a failure to meet actual requirements, affecting production efficiency and resource management effectiveness. SUMMARY

[0004] In order to solve the technical problems existing in the prior art, the embodiments of the present application provide a material automatic grading and classification method based on a large model, comprising the following steps: In order to achieve the above-mentioned purpose, the present application adopts the following technical scheme: a material automatic grading and classification method based on a large model, comprising the following steps: S1: Obtain morphological features and attribute distribution information of material objects, combine material original category distribution characteristics, analyze the correlation between material categories, and generate a material grading and classification atlas; S2: Call the material grading and classification atlas, extract the stability index and consistency index of each category, sort them according to the stability priority and exclude categories with consistency lower than the set threshold, and generate an adaptability classification path list; S3: calling the adaptability classification path list, collecting the response monitoring value and the classification performance evaluation value of the candidate classification node, comparing and screening the node meeting the classification condition, mapping through the logical connection relationship between the node and the classification path to obtain a node path association matching group; S4: according to the node path association matching group, searching the current classification state and the set target classification demand, judging the difference direction, adjusting the classification strategy of the classification node in the current task period according to the difference direction to obtain a material grading classification execution instruction set.

[0005] As a further scheme of the present application, the material grading classification atlas comprises a category identifier, an association feature label and a classification environment adaptability value, the adaptability classification path list comprises a stability parameter, a consistency limit value and a priority label, the node path association matching group comprises a logical connection relationship identifier, a response reference interval and a classification performance judgment value, and the material grading classification execution instruction set comprises a classification strategy code, a classification adjustment direction and an optimization parameter in a task period.

[0006] As a further scheme of the present application, the steps of the material grading classification atlas are specifically as follows: S101: obtaining the morphological feature and attribute distribution information of a material object, performing comparison operation on the category distribution feature and the original category distribution, identifying the material category association feature, and generating a category association sequence; S102: calling the category association sequence, performing period-by-period comparison according to the category association feature and the set balance threshold, screening a time period with a continuous feature value higher than the balance threshold, extracting the corresponding category number and time window, and obtaining a category association balance section index set; S103: according to the category association balance section index set, screening the category with a balance feature as an optimal grading classification path, marking the controllability interface and the predictable category distribution path, and generating a material grading classification atlas.

[0007] As a further scheme of the present application, the steps of the adaptability classification path list are specifically as follows: S201: calling the material grading classification atlas, extracting the region number, stability index and consistency index of the corresponding category, identifying the category number matching each data, and establishing a category operation parameter set; S202: according to the category operation parameter set, performing descending order sorting according to the stability index, screening a stability category sequence, and obtaining a stability sorting list; S203: calling the stability sorting list, comparing the category consistency index with the set consistency lower limit, eliminating the category with a consistency lower than the lower limit, and generating an adaptability classification path list.

[0008] As a further scheme of the present application, the step of the node path association matching group is specifically: S301: calling the adaptive classification path list, capturing the response monitoring value and classification performance evaluation value of the candidate classification node, summarizing the real-time response value and performance value of the node, and generating the node running state parameter; S302: according to the node running state parameter, comparing the response value, the performance value and the classification condition interval, screening the nodes meeting the classification condition, and obtaining the classification control available node set; S303: calling the classification control available node set, mapping the node number and the connection channel according to the logical connection relationship between the list node and the classification path, identifying the node path association degree index, and obtaining the node path association matching group.

[0009] As a further scheme of the present application, the response value is the monitoring value of the node in the classification path in real time; The performance value is the classification performance value evaluated by the node in the classification path; The comparison of the classification condition interval is to compare and screen through the preset performance evaluation threshold and the threshold of the response monitoring value, wherein the threshold is set according to the real-time performance and original performance data of the node.

[0010] As a further scheme of the present application, the step of the material grading classification execution instruction set is specifically: S401: according to the node path association matching group, collecting the real-time classification state value in the current task cycle of the path, and matching the target classification demand value corresponding to each path, merging the current value and the target value according to the path number, and generating the classification state difference data group; S402: calling the classification state difference data group, judging the difference direction between the current classification state and the target classification demand of each path, and according to the positive and negative of the difference value, distinguishing the state into over classification and under classification, and obtaining the classification difference direction identification set; S403: according to the classification difference direction identification set, positioning the list classification node connected to the corresponding path, adjusting the node classification strategy in the current task cycle, analyzing the node classification coordination, and obtaining the material grading classification execution instruction set.

[0011] As a further scheme of the present application, the real-time classification state value of the path in the current task cycle refers to the real-time throughput of the material of the path in the task cycle and the average residence time of the material at the path node; The target classification demand value corresponding to each path is to limit the expected throughput and expected residence time of the material through the preset numerical range according to the preset material classification efficiency standard of the path; The difference direction between the current classification state of each path and the target classification requirement is determined, that is, when the real-time classification state value is higher than the upper threshold of the target classification requirement value, the state is determined as over-classification, and when the real-time classification state value is lower than the lower threshold of the target classification requirement value, the state is determined as under-classification.

[0012] As a further scheme of the present application, the method further comprises a step S5: S5: calling the material hierarchical classification execution instruction set, comparing the classification delay data in the task time sequence with the fluctuation rate, performing matching and screening according to the path characteristics and the key time points, and outputting a classification recovery path and a synchronization node configuration table; The classification recovery path and the synchronization node configuration table comprise a path delay coefficient, a synchronization configuration time point, and a fluctuation response characteristic.

[0013] As a further scheme of the present application, the step of the classification recovery path and the synchronization node configuration table is specifically: S501: calling the material hierarchical classification execution instruction set, extracting the delay change of the path at the key time points according to the classification delay data in the task time sequence, and performing correlation analysis with the fluctuation rate to generate a path delay fluctuation correlation coefficient; S502: performing matching judgment operation according to the path delay fluctuation correlation coefficient and the path characteristic parameters, screening the path set satisfying the synchronization condition, and obtaining a synchronization path judgment value interval; S503: calling the synchronization path judgment value interval, comparing it with the synchronization node parameters in the fluctuation signal, determining the optimal path for classification recovery and the corresponding node parameters, and outputting a classification recovery path and a synchronization node configuration table.

[0014] Compared with the prior art, the present application has the following advantages and positive effects: In the present application, by analyzing the morphological characteristics and attribute distribution information of the material objects, a more accurate material hierarchical classification atlas can be generated, and the classification accuracy can be improved according to the correlation between the material categories. By sorting the stability and consistency indicators of each category and excluding categories with low consistency, the selection of the classification path is effectively optimized. The response monitoring and performance evaluation of the classification nodes not only ensure the efficiency of the classification process, but also dynamically adjust the classification strategy to cope with changes in different task cycles, improve the adaptability and accuracy of the classification, and the material hierarchical classification task can not only be adjusted according to real-time data, but also reduce unstable factors during the classification process, finally ensure higher quality and more accurate recovery path of the classification result, and significantly improve the flexibility and intelligent level of the material classification. BRIEF DESCRIPTION OF DRAWINGS

[0015] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings needed in the embodiments description. Obviously, the drawings in the following description only show some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained based on these drawings without any creative effort.

[0016] Figure 1 Schematic diagram of the step flow of the present application; Figure 2 S1 detailed schematic diagram of the present application; Figure 3 S2 detailed schematic diagram of the present application; Figure 4 S3 detailed schematic diagram of the present application; Figure 5 S4 detailed schematic diagram of the present application; Figure 6 S5 detailed schematic diagram of the present application. DETAILED DESCRIPTION

[0017] The technical solutions in the present application will be described below in combination with the drawings.

[0018] In the embodiments of the present application, the words such as "exemplarily", "for example" and the like are used to represent as an example, illustration or explanation. Any embodiment or design scheme described as "example" in the present application should not be interpreted as more preferred or more advantageous than other embodiments or design schemes. In fact, the word "example" is intended to present the concept in a specific way. In addition, in the embodiments of the present application, the meaning expressed by "and / or" can be both, or can be one of the two.

[0019] In the embodiments of the present application, "image" and "picture" can be used interchangeably at times. It should be pointed out that when the distinction is not emphasized, the meanings expressed are consistent. "Of", "corresponding" and "corresponding" can be used interchangeably at times. It should be pointed out that when the distinction is not emphasized, the meanings expressed are consistent.

[0020] In the embodiments of the present application, sometimes the subscript such as W1 can be written in the form of non-subscript such as W1. When the distinction is not emphasized, the meanings expressed are consistent.

[0021] In order to make the technical problems, technical solutions and advantages to be solved by the present application more clear, the following will be described in detail in combination with the drawings and specific embodiments.

[0022] Please refer to Figure 1The embodiment of the application provides a material automatic grading and classifying method based on a large model, which comprises the following steps: S1: morphological features and attribute distribution information of a material object are acquired, correlation between material categories is analyzed in combination with material original category distribution characteristics, and a material grading and classifying graph is generated; S2: a material grading and classifying graph is called to extract stability indexes and consistency indexes of each category, categories with consistency lower than a set threshold are excluded in priority order of stability, and an adaptability classification path list is generated; S3: the adaptability classification path list is called to collect response monitoring values and classification performance evaluation values of candidate classification nodes, nodes meeting classification conditions are compared and screened, node path correlation matching groups are obtained through logical connection between nodes and classification paths; S4: according to the node path correlation matching groups, current classification states and set target classification requirements are searched, difference directions are judged, classification strategies of classification nodes in current task periods are adjusted according to the difference directions, and a material grading and classifying execution instruction set is obtained; S5: the material grading and classifying execution instruction set is called, classification delay data and fluctuation rates in a task time sequence are compared, matching and screening are performed according to path characteristics and key time points, and a classification recovery path and a synchronization node configuration table are output.

[0023] The material grading and classifying graph comprises a category identifier, a correlation characteristic label and a classification environment adaptability value, the adaptability classification path list comprises stability parameters, consistency limit values and priority labels, the node path correlation matching group comprises a logical connection relationship identifier, a response reference interval and a classification performance determination value, the material grading and classifying execution instruction set comprises a classification strategy code, a classification adjustment direction and optimization parameters in a task period, and the classification recovery path and the synchronization node configuration table comprise a path delay coefficient, a synchronization configuration time point and fluctuation response characteristics.

[0024] Please refer to Figure 2 The steps of the material grading and classifying graph are specifically as follows: S101: morphological features and attribute distribution information of a material object are acquired, comparison and operation of category distribution characteristics and original category distribution are performed, material category correlation characteristics are identified, and a category correlation sequence is generated; Data collection is performed on the material object to obtain morphological characteristics and attribute distribution information of the material object. For example, for lithium battery positive material, the particle size distribution is collected, such as 0.5 μm particle size accounting for 15%, 1.0 μm particle size accounting for 50%, 5.0 μm particle size accounting for 30%, and 10.0 μm particle size accounting for 5%. At the same time, morphological characteristic data such as specific surface area and porosity are obtained, and the content distribution of elements such as nickel element, cobalt element and manganese element is obtained, such as the content of nickel element in batch M1 is 81.2%, the content of cobalt element is 10.1%, and the content of manganese element is 8.9%. The collected data constitute the category distribution characteristics of the current batch of materials, and the category distribution characteristics are compared and operated item by item with the preset original category distribution. The original category distribution is a standard value formed based on historical stable production batch data statistics. For example, the standard of original category A is set as nickel element content 81.0% ± 0.3%, cobalt element content 10.0% ± 0.1%, and manganese element content 9.0% ± 0.1%. The absolute deviation of each characteristic value of the current batch M1 from the standard value of the original category A is calculated, for example, the absolute deviation of the nickel element content is |81.2%-81.0%|=0.2%, the absolute deviation of the cobalt element content is |10.1%-10.0%|=0.1%, and the absolute deviation of the manganese element content is |8.9%-9.0%|=0.1%. The material category correlation characteristics are identified, and the correlation judgment threshold is set, for example, the nickel element content deviation threshold is 0.25%, the cobalt element content deviation threshold is 0.15%, and the manganese element content deviation threshold is 0.12%. When the absolute deviation is less than or equal to the corresponding threshold, it is determined that the characteristics have strong correlation with the original category A, for example, the nickel element deviation of batch M1 is 0.2% which is less than the threshold of 0.25%, and it is determined that the correlation is strong. The cobalt element deviation is 0.1% which is less than the threshold of 0.15%, and it is determined that the correlation is strong. The manganese element deviation is 0.1% which is less than the threshold of 0.12%, and it is determined that the correlation is strong. If the deviation is greater than the threshold, the correlation is weak. Thus, a category correlation sequence is generated, which reflects the correlation degree of material M1 and each characteristic of category A.

[0025] S102: Call the category correlation sequence, compare the category correlation characteristics with the set balance threshold period by period, filter the time period with continuous characteristic value higher than the balance threshold, extract the corresponding category number and time window, and obtain the category correlation balance section index set; The association sequence of the material batch M1 and the category A is called, and the category association characteristics are compared with the set balance threshold value period by period, wherein the balance threshold value is defined as the characteristic association degree score which needs to be continuously maintained above 85 points, and the score is obtained by mapping the standardized processing of the characteristic deviation in the historical stable batch data to the 0-100 point interval, for example, if the nickel element content is continuously monitored in the period, its association degree score with the original category A is: period 1 is 92 points, period 2 is 90 points, period 3 is 88 points, and period 4 is 83 points, and the balance threshold value is set to 85 points, the time period of the continuous characteristic value higher than the balance threshold value is screened, for example, the association degree score of the nickel element content in the three continuous periods of period 1 to period 3 is higher than 85 points, and the 83 points of period 4 is lower than the threshold value, therefore, the continuous balance time period of the nickel element content characteristic for category A is extracted as [period 1, period 3], and the corresponding category number is A, and the time window is [period 1, period 3], if another characteristic such as particle size distribution is higher than 85 points in period 1 to period 5, then its time window is [period 1, 5], thus the category association balance section index set is obtained, the index set includes: { (category number A, nickel element content, [period 1, 3]), (category number A, particle size distribution, [period 1, 5]), (category number B, cobalt element content, [period 2, 4])}, the index set records the continuous balance state of each material category specific characteristic and the corresponding monitoring time window.

[0026] S103: According to the category association balance section index set, the categories of balanced characteristics are screened as the optimal classification path, the controllable interface and the predictable category distribution path are marked, and the material classification map is generated; According to the category correlation balanced section index set, for example, the index set indicates that the nickel element content and the particle size distribution characteristics of category A are balanced within a long time window, and the categories with balanced characteristics are screened as the optimal classification path, and the determination standard is set as: when at least 75% of the core characteristics (for example, the nickel cobalt manganese content and particle size of the positive electrode material) in a material category remain balanced state in the past 10 consecutive monitoring periods, the category is determined to be stable and suitable as an optimal classification path, for example, 4 out of 5 core characteristics (80%) of category A have been balanced in the past 12 periods, so category A is screened as an optimal classification path, and the controllability interface and the path of predictable category distribution are marked, where the controllability interface refers to the process parameter points that can be adjusted on the production line, for example, the stirring speed in the mixing stage, the temperature curve in the sintering stage, and the grinding time in the grinding stage in the production of positive electrode materials, and the interface is marked as a key point that can be used to adjust the classification strategy, and the predictable category distribution refers to that after the path processing, the indicators of the material can be stably within the expected range, for example, the nickel element content of the category A positive electrode material processed by the optimization path is expected to be controlled within the narrow range of 81.0%±0.1%, and the particle size distribution D50 value is expected to be stable at 1.0 μm±0.05 μm, and accordingly a material classification and sorting map is generated, which presents each optimal classification path in a graphical manner, and the corresponding controllability interface and the predictable category distribution range of the material after path processing are marked on the path.

[0027] Please refer to Figure 3 The steps of the adaptive classification path list are as follows: S201: Call the material classification and sorting map, extract the region number, stability index and consistency index of the corresponding category, identify the category number matching each item of data, and establish a category running parameter set; Call the material classification atlas, for example, extract the optimal classification path information corresponding to category A from the atlas, the path contains the area number of category A "P-LCO-001" (high nickel ternary material production line), stability index and consistency index, wherein the stability index is calculated according to the average length of the balanced period and the balanced feature coverage rate of each core feature in S102, for example, if the average balanced period length of 5 core features of category A is 10 periods, and the balanced feature coverage rate is 80%, the stability index score is calculated as 0.8, the consistency index is calculated according to the average absolute deviation of category distribution characteristics and the inverse of the original category distribution in S101, for example, if the average absolute deviation of each feature of category A is 0.1%, the consistency index score is 0.9, match each data with the category number, match the above area number, stability index and consistency index with category A, establish a category running parameter set, for example, for category A, its running parameter set is {area number: P-LCO-001, stability index: 0.8, consistency index: 0.9}, the set describes the running characteristics of a specific material category in classification.

[0028] S202: According to the category running parameter set, sort the stability categories in descending order according to the stability index, and obtain a stability sorting list; According to the category running parameter set, for example, the running parameters of category A in the set are {area number: P-LCO-001, stability index: 0.8, consistency index: 0.9}, the running parameters of category B are {area number: P-NCM-002, stability index: 0.75, consistency index: 0.85}, and the running parameters of category C are {area number: P-LFP-003, stability index: 0.9, consistency index: 0.95}, sort the stability index in descending order, the sorting process is to compare the stability index values of all categories, and arrange the category with the largest stability index value in the first place, the category with the second largest stability index value in the second place, and so on, for example, among categories A, B and C, the stability index of category C is 0.9, which is the largest, the stability index of category A is 0.8, which is the second largest, and the stability index of category B is 0.75, which is the smallest, the sorting result is C, A and B, filter the stability category sequence, which is composed of categories with stability index higher than the set threshold value 0.78, for example, the stability index threshold value is set to 0.78, then category C (0.9) and category A (0.8) are filtered out, and category B (0.75) is excluded, to obtain a stability sorting list, which arranges the material categories with strong stability.

[0029] S203: Call the stability sorting list, compare the category consistency index with the set consistency lower limit, and remove the categories with consistency lower than the lower limit to generate an adaptability classification path list; Call the stability ranking list, for example, the list is [category C (0.9), category A (0.8)], compare the category consistency index with the set consistency lower limit, wherein the consistency lower limit is an acceptable standard set based on historical production experience and product quality requirements, for example, the consistency lower limit value is set to 0.82, compare the actual consistency index of each category in the stability ranking list, for example, the consistency index of category C is 0.95, which is higher than the consistency lower limit of 0.82, so category C is retained, the consistency index of category A is 0.9, which is higher than the consistency lower limit of 0.82, so category A is also retained, eliminate categories with consistency lower than the lower limit, for example, if there is category D (stability 0.85, consistency 0.80), category D will be eliminated because its consistency 0.80 is lower than the lower limit of 0.82, generate an adaptive classification path list, which contains stable and consistent material categories, for example, the adaptive classification path list is [category C (stability 0.9, consistency 0.95), category A (stability 0.8, consistency 0.9)], the path represents the current optimal classification selection.

[0030] See Figure 4 The steps of associating and matching the node path group are as follows: S301: Call the adaptive classification path list, capture the response monitoring value and classification performance evaluation value of the candidate classification node, and summarize the real-time response value and performance value of the node to generate the node running state parameter; Call the adaptive classification path list, for example, the list contains category C and category A as high adaptive classification paths, capture the response monitoring value and classification performance evaluation value of the candidate classification node, wherein the response monitoring value is the monitoring data output by the node in the classification path in real time, for example, the real-time throughput (tons / hour) and material residence time (seconds) of a screening node on a path are response monitoring values, the classification performance evaluation value is the classification efficiency, accuracy and other indicators evaluated by the node in the classification path according to the preset standard, for example, the material purity (%) and misclassification rate (%) of the screening node are classification performance evaluation values, summarize the real-time response value and performance value of the node, for example, screening node X, in the last task cycle, its real-time throughput monitoring value is 50 tons / hour, the material residence time is 15 seconds, the material purity evaluation value is 98.5%, and the misclassification rate is 1.0%, the data is collected to form the node running state parameter, for example, the running state parameter set of node X is {real-time throughput: 50 tons / hour, residence time: 15 seconds, material purity: 98.5%, misclassification rate: 1.0%}, which reflects the current running condition of the node.

[0031] S302: According to the node running state parameter, compare the response value, performance value and classification condition interval, and select the nodes that meet the classification conditions to obtain the classification control available node set; The response value is the monitoring value of the node output in real time in the classification path; The performance value is the classification performance value evaluated by the node in the classification path; The comparison of the classification condition interval is the comparison and screening by the preset performance evaluation threshold and the threshold of the response monitoring value, wherein the threshold is set according to the real-time performance and the original performance data of the node; According to the node running state parameters, for example, the running state parameter set of node X is {real-time throughput: 50 tons / hour, residence time: 15 seconds, material purity: 98.5%, misclassification rate: 1.0%}, according to the comparison of the response value, the performance value and the classification condition interval, the node meeting the classification condition is screened, wherein the response value is the monitoring value of the node output in real time in the classification path, the performance value is the classification performance value evaluated by the node in the classification path, and the classification condition interval is compared and screened by the preset performance evaluation threshold and the threshold of the response monitoring value, for example, the high-efficiency classification condition is set as: the real-time throughput needs to be greater than or equal to 45 tons / hour (response value threshold), the material residence time needs to be less than or equal to 20 seconds (response value threshold), the material purity needs to be greater than or equal to 98% (performance evaluation threshold), and the misclassification rate needs to be less than or equal to 1.5% (performance evaluation threshold), the running parameters of node X are compared with the condition interval item by item, for example, the throughput 50 tons / hour≥45 tons / hour, the residence time 15 seconds≤20 seconds, the purity 98.5%≥98%, and the misclassification rate 1.0%≤1.5%, all the parameters of node X meet the high-efficiency classification condition, so node X is screened as a node meeting the classification condition, and a classification control available node set is obtained, which contains nodes in good running condition and capable of meeting the high-efficiency classification demand, for example, the classification control available node set is {node X, node Y, node Z}, and the nodes all have the ability to perform high-adaptability classification tasks.

[0032] S303: Call the classification control available node set, map the node number and the connection channel according to the logical connection relationship of the inventory node and the classification path, identify the node path association degree index, and obtain a node path association matching group; The classification control available node set, for example, the available node set includes node X, node Y, and node Z, is mapped to the connection channel according to the logical connection relationship of the list node and the classification path, wherein the list node refers to a physically existing classification device or processing unit, for example, a screening machine, a magnetic separator, a grinding machine, etc., and the logical connection relationship refers to the connection of nodes in series or in parallel in a specific order and manner in a classification path, for example, in a positive material production path, a grinding node is connected to a screening node, and the screening node is connected to a magnetic separation node, the node number is mapped to the connection channel, for example, the number of node X is ND-001, and the downstream connection channel is CH-001, which leads to node Y (ND-002), and the downstream connection channel of node Y is CH-002, which leads to node Z (ND-003), the node path correlation degree index is determined by calculating the proportion of the amount of material processed by the node in a specific path, the contribution degree to the overall performance of the path, etc., for example, if node X processes 90% of the total amount of material in the path, and the contribution weight of its classification purity to the overall purity is 0.6, then the correlation degree index is 0.54, the node path correlation matching group is obtained, the matching group binds the available classification nodes, the classification paths they are in, and the logical connection relationship between them, and the correlation degree index, for example, the matching group includes {node X (ND-001), path P-LCO-001, downstream channel CH-001, correlation degree 0.54}, {node Y (ND-002), path P-LCO-001, upstream channel CH-001, downstream channel CH-002, correlation degree 0.70}, {node Z (ND-003), path P-LCO-001, upstream channel CH-002, correlation degree 0.65}, which provides a basis for subsequent classification execution instruction generation.

[0033] Please refer to Figure 5 The steps of the material classification execution instruction set are as follows: S401: According to the node path correlation matching group, the real-time classification state value in the current task cycle of the path is collected, and the target classification demand value corresponding to each path is matched, the current value and the target value are merged according to the path number, and the classification state difference data group is generated; The real-time classification state value in the current task cycle of the path refers to the real-time throughput of the collected path material in the task cycle and the average residence time of the material at the path node; The target classification demand value corresponding to each path is limited by the expected throughput and expected residence time of the material through the preset numerical range according to the preset material classification efficiency standard of the path; According to the node path association matching group, for example, the matching group contains path P-LCO-001 and its associated nodes X, Y, Z, the real-time classification state value in the current task cycle of the path is collected, wherein the real-time classification state value in the current task cycle of the path refers to the real-time throughput of the collected path in the task cycle and the average residence time of the material at the path node, for example, in the current task cycle (September 24, 2025 8:00-9:00), the real-time throughput monitoring value of path P-LCO-001 is 1000 kg / hour, and the average residence time at the path node is 30 seconds, and the target classification demand value corresponding to each path is matched, wherein the target classification demand value corresponding to each path is limited by the expected throughput and expected residence time of the material through a preset numerical range according to the preset material classification efficiency standard of the path, for example, the target classification demand of path P-LCO-001 is set as: the expected throughput range is kg / hour, and the expected residence time range is seconds, the current value and the target value are integrated according to the path number, for example, the current value of path P-LCO-001 is {throughput: 1000 kg / hour, residence time: 30 seconds}, and the target value is {throughput range: kg / hour, residence time range: seconds}, and a classification state difference data set is generated, which contains the comparison between the current running state and the expected target of each path, for example, the difference data set contains {path P-LCO-001, real-time throughput 1000, target throughput, real-time residence time 30, target residence time}.

[0034] S402: Call the classification state difference data set to determine the difference direction between the current classification state and the target classification demand of each path, and according to the positive and negative of the difference value, the state is divided into over-classification and under-classification to obtain a classification difference direction identification set; The difference direction between the current classification state and the target classification demand of each path is determined, which means that when the real-time classification state value is higher than the upper limit threshold of the target classification demand value, the state is determined as over-classification, and when the real-time classification state value is lower than the lower limit threshold of the target classification demand value, the state is determined as under-classification; Call the classification state difference data set, for example, the difference data set contains path P-LCO-001, the real-time throughput is 1000 kg / hour, the target throughput range is kg / hour, the real-time residence time is 30 seconds, the target residence time range is seconds, and the difference direction between the current classification state of each path and the target classification demand is determined, wherein the difference direction is determined to be over-classification when the real-time classification state value is higher than the upper threshold value of the target classification demand value, and the state is determined to be under-classification when the real-time classification state value is lower than the lower threshold value of the target classification demand value, for example, for path P-LCO-001, the real-time throughput of 1000 kg / hour is within the target range of kg / hour, and the real-time residence time of 30 seconds is within the target range of seconds, and the state is determined to be normal classification, if the real-time throughput is 1100 kg / hour, it is determined to be over-classification because it is higher than the target upper limit of 1050 kg / hour, if the real-time throughput is 900 kg / hour, it is determined to be under-classification because it is lower than the target lower limit of 950 kg / hour, according to the positive and negative of the difference value, the state is divided into over-classification and under-classification, if the real-time value is higher than the upper limit of the target range, the difference is positive, and is marked as over-classification, if the real-time value is lower than the lower limit of the target range, the difference is negative, and is marked as under-classification, if the real-time value is within the target range, it is marked as normal, a classification difference direction identification set is obtained, the identification set indicates the specific direction of the current classification state of each path deviating from the target demand, for example, the identification set is {path P-LCO-001, throughput: normal, residence time: normal}, if another path P-LFP-002, the real-time value of the throughput is 800 kg / hour, and the target range is kg / hour, then the identification is {path P-LFP-002, throughput: under-classification}.

[0035] S403: According to the classification difference direction identification set, locate the corresponding path connection list classification node, adjust the node classification strategy in the current task period, analyze the node classification coordination, and obtain a material grading classification execution instruction set; According to the classification difference direction identification set, for example, the identification set shows that the throughput and residence time of path P-LCO-001 are in a normal classification state, and the throughput of path P-LFP-002 is in an under-classification state, the list of classification nodes corresponding to the path connection is located, for example, for the under-classification state of path P-LFP-002, the key nodes on the path are traced back, for example, the grinding node (ND-M-001) located at the beginning of the path and the subsequent screening node (ND-S-002), the node classification strategy in the current task cycle is adjusted, for example, for the throughput under-classification problem of path P-LFP-002, the grinding parameters of grinding node ND-M-001 are adjusted, for example, the grinding time is adjusted from 120 seconds to 100 seconds to improve the material throughput efficiency, and the screen amplitude of screening node ND-S-002 is fine-tuned, for example, the amplitude is increased from 5 millimeters to 6 millimeters, the node classification coordination is analyzed, the influence of the adjustment on other nodes on the path and the overall classification performance is evaluated, for example, whether the adjustment of the grinding time and the screen amplitude affects the uniformity of the particle size distribution of the material or increases the load of the downstream magnetic separation node is analyzed, the material grading classification execution instruction set is obtained, the instruction set lists specific adjustment measures for specific paths and nodes, for example, the instruction set is {path P-LFP-002, node ND-M-001: grinding time is adjusted to 100 seconds, node ND-S-002: screen amplitude is adjusted to 6 millimeters}, and the instruction is executed down to the production line.

[0036] Please refer to Figure 6 The steps of classification recovery path and synchronization node configuration table are as follows: S501: Call the material grading classification execution instruction set, extract the delay change of the path at the key time point according to the classification delay data in the task time sequence, and perform correlation analysis with the fluctuation rate to generate a path delay fluctuation correlation coefficient; The material classification execution instruction set is called, for example, the instruction set includes the grinding time adjustment instruction of the grinding node ND-M-001 on the path P-LFP-002, the delay change of the path at the key time point is extracted according to the classification delay data in the task time sequence, wherein the classification delay data in the task time sequence refers to the deviation between the actual time and the standard time of the material from entering the path to completing classification, for example, the average classification delay of the path P-LFP-002 before the execution of the instruction is + 50 seconds (50 seconds slower than the standard), and the delay of the path P-LFP-002 at the 1st hour, the 2nd hour and the 3rd hour after the execution of the instruction is + 30 seconds, + 10 seconds and-5 seconds respectively, the fluctuation rate is analyzed in association, the fluctuation rate is calculated as the ratio of the absolute value of the delay change of the continuous time points to the time interval, for example, the fluctuation rate from the 1st hour to the 2nd hour is | 30 seconds-10 seconds | / 1 hour = 20 seconds / hour, and the fluctuation rate from the 2nd hour to the 3rd hour is | 10 seconds- (-5 seconds) | / 1 hour = 15 seconds / hour, the path delay fluctuation correlation coefficient is generated, the correlation coefficient is obtained by comprehensively evaluating the delay change and the fluctuation rate thereof, for example, if the delay change shows a continuous decreasing trend and the fluctuation rate is stable at a low level (for example, less than 10 seconds / hour), the correlation coefficient is high, otherwise the correlation coefficient is low, for example, the correlation coefficient of the path P-LFP-002 is 0.75.

[0037] S502: According to the path delay fluctuation correlation coefficient, in combination with the path characteristic parameters, the matching judgment operation is performed on the performance of the path at the key time point, the path set satisfying the synchronization condition is screened, and the synchronization path judgment value interval is obtained. According to the path delay fluctuation correlation coefficient, for example, the correlation coefficient of path P-LFP-002 is 0.75, combined with path characteristic parameters, wherein the path characteristic parameters include the material handling capacity of the path, the equipment load rate, the historical stability performance, etc., for example, the material handling capacity of path P-LFP-002 is 800 kg / hour, the equipment load rate is 70%, the historical stability score is 0.8, and the performance of the path at the key time point is performed, wherein the key time point refers to the link that significantly affects the classification efficiency in the production task cycle, for example, the grinding completion time, the screening completion time, etc., and the judgment standard is set as: when the path delay fluctuation correlation coefficient is greater than 0.7, and the equipment load rate in the path characteristic parameters is less than 85%, and the historical stability score is higher than 0.75, the path is determined to meet the synchronization condition, the path set meeting the synchronization condition is screened, for example, the correlation coefficient of path P-LFP-002 is 0.75, which is greater than 0.7, the load rate is 70%, which is lower than 85%, and the stability score is 0.8, which is higher than 0.75, so path P-LFP-002 is screened as meeting the synchronization condition, and a synchronization path judgment value interval is obtained, which describes the reasonable range of each parameter of the path meeting the synchronization condition, for example, the interval is {delay fluctuation correlation coefficient: [0.7, 1.0], equipment load rate: [0%, 85%], historical stability score: [0.75, 1.0]}, which defines which paths have the potential for synchronization recovery.

[0038] S503: Call the synchronization path judgment value interval, compare it with the synchronization node parameters in the fluctuation signal, determine the optimal path for classification recovery and the corresponding node parameters, and output the classification recovery path and the synchronization node configuration table; The call synchronization path determination value interval, for example, the interval is {delay fluctuation correlation coefficient: [0.7, 1.0], device load rate: [0%, 85%], historical stability score: [0.75, 1.0]}, is compared with the synchronization node parameters in the fluctuation signal, wherein the synchronization node parameters in the fluctuation signal are the running state fluctuation conditions monitored in real time from different nodes on the production line, for example, the motor speed fluctuation range of node ND-M-001, the temperature sensor fluctuation range, the vibration frequency fluctuation range of node ND-S-002, the parameters reflect the dynamic stability of the node, the optimal path of classification recovery and the corresponding node parameters are determined, according to the comparison result, the path with the highest delay fluctuation correlation coefficient, the smallest fluctuation of the synchronization node parameters and in the stable interval is selected as the optimal recovery path, for example, if the correlation coefficient of path P-LFP-002 is 0.9, the motor speed fluctuation amplitude of its associated node ND-M-001 is ±2 RPM, the temperature fluctuation amplitude is ±0.5℃, and both are in the set stable interval, while the correlation coefficient of another path P-NCM-003 that meets the condition is 0.85, and the fluctuation amplitude of the node is slightly larger, then path P-LFP-002 is determined as the optimal recovery path, and the classification recovery path and the synchronization node configuration table are output, the table details the number of the selected optimal recovery path and the specific configuration parameters of the key nodes on the path that need to be synchronized and adjusted, for example, the configuration table is {optimal recovery path number: P-LFP-002, node configuration: [node ND-M-001 (motor speed: 1500 RPM, temperature: 80℃), node ND-S-002 (vibration frequency: 30 Hz, amplitude: 6 mm)]}, the table provides the basis for classification recovery operation.

[0039] 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 range 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. A method for automatic material grading and classification based on a large model, characterized in that, Includes the following steps: S1: Obtain the morphological characteristics and attribute distribution information of material objects, combine them with the original category distribution characteristics of materials, analyze the correlation between material categories, and generate a material classification map; S2: Call the material classification map, extract the stability index and consistency index of each category, sort them according to stability priority and exclude categories with consistency below the set threshold, and generate a list of adaptability classification paths. S3: Call the adaptive classification path list, collect the response monitoring value and classification performance evaluation value of the candidate classification nodes, compare and filter the nodes that meet the classification conditions, and map them through the logical connection relationship between the nodes and the classification paths to obtain the node path association matching group. S4: Based on the node path association matching group, retrieve the current classification status and the set target classification requirements, determine the direction of difference, adjust the classification strategy of the classification node in the current task cycle according to the direction of difference, and obtain the material classification execution instruction set.

2. The automatic material classification and grading method based on a large model according to claim 1, characterized in that, The material classification map includes category identifiers, related feature labels, and classification environment adaptation values. The adaptive classification path list includes stability parameters, consistency limit values, and priority labels. The node path association matching group includes logical connection relationship identifiers, response benchmark intervals, and classification performance judgment values. The material classification execution instruction set includes classification strategy codes, classification adjustment directions, and optimization parameters within the task cycle.

3. The automatic material classification and grading method based on a large model according to claim 1, characterized in that, The specific steps for creating the material classification and grading map are as follows: S101: Obtain the morphological features and attribute distribution information of material objects, perform a comparison operation between the category distribution features and the original category distribution, identify the material category correlation features, and generate a category correlation sequence; S102: Call the category correlation sequence, compare the category correlation characteristics with the set balance threshold period by period, filter the time period where the continuous feature value is higher than the balance threshold, extract the corresponding category number and time window, and obtain the category correlation balance segment index set. S103: Based on the category-related balanced segment index set, filter categories with balanced characteristics as optimizable hierarchical classification paths, mark the controllability interface and the path of predictable category distribution, and generate a material hierarchical classification map.

4. The automatic material classification and grading method based on a large model according to claim 3, characterized in that, The specific steps for creating the adaptive classification path list are as follows: S201: Call the material classification map, extract the corresponding category area number, stability index and consistency index, identify the category number to match each data item, and establish a category operation parameter set; S202: Based on the set of operating parameters for each category, sort them in descending order according to stability indicators, filter the stability category sequence, and obtain a stability sorting list; S203: Call the stability sorting list, compare the category consistency index with the set consistency lower limit, remove categories with consistency below the lower limit, and generate an adaptive classification path list.

5. The automatic material classification and grading method based on a large model according to claim 4, characterized in that, The specific steps for associating and matching node paths are as follows: S301: Call the adaptive classification path list, capture the response monitoring value and classification performance evaluation value of the candidate classification node, summarize the real-time response value and performance value of the node, and generate node running status parameters; S302: Based on the node operating status parameters, and by comparing the response value, performance value, and classification condition range, select nodes that meet the classification conditions to obtain a set of available nodes for classification control; S303: Call the set of available nodes for classification control, map the node number and connection channel according to the logical connection relationship between the list nodes and the classification path, identify the node path correlation index, and obtain the node path correlation matching group.

6. The automatic material classification and grading method based on a large model according to claim 5, characterized in that, The response value is the monitoring value that the node outputs in real time in the classification path; The performance value is the classification performance value evaluated by the node in the classification path; The comparison of the classification condition intervals is carried out by comparing and filtering the preset performance evaluation threshold with the response monitoring value threshold, wherein the threshold is set based on the real-time performance and original performance data of the node.

7. The automatic material classification and grading method based on a large model according to claim 5, characterized in that, The specific steps of the material classification and categorization instruction set are as follows: S401: Based on the node path association matching group, collect the real-time classification status value of the path in the current task cycle, match the target classification requirement value corresponding to each path, merge the current value and the target value according to the path number, and generate a classification status difference data group. S402: Call the classification state difference data group, determine the difference direction between the current classification state of each path and the target classification requirement, and classify the state into overclassification and underclassification according to the sign of the difference value to obtain the classification difference direction identifier set; S403: Based on the classification difference direction identifier set, locate the list classification node connected by the corresponding path, adjust the node classification strategy within the current task cycle, analyze node classification coordination, and obtain the material classification execution instruction set.

8. The automatic material classification and grading method based on a large model according to claim 7, characterized in that, The real-time classification status value of the path within the current task cycle refers to the real-time throughput of materials in the collection path and the average dwell time of materials at path nodes within the task cycle. The target classification requirement value corresponding to each path is determined by limiting the expected throughput and expected dwell time of materials within a preset numerical range based on the material classification efficiency standard preset for the path. The determination of the difference between the current classification status of each path and the target classification requirement refers to determining the status as overclassified when the real-time classification status value is higher than the upper threshold of the target classification requirement value, and as underclassified when the real-time classification status value is lower than the lower threshold of the target classification requirement value.

9. The automatic material classification and grading method based on a large model according to claim 1, characterized in that, The method also includes step S5: S5: Call the material classification and sorting execution instruction set, compare the classification delay data and fluctuation rate in the task time sequence, perform matching and filtering according to path characteristics and key time points, and output the classification recovery path and synchronization node configuration table. The classification recovery path and synchronization node configuration table includes path delay coefficient, synchronization configuration time point, and fluctuation response characteristics.

10. The automatic material classification and grading method based on a large model according to claim 9, characterized in that, The specific steps for configuring the classification recovery path and synchronization node table are as follows: S501: Call the material classification and sorting execution instruction set, extract the path delay changes at key time points based on the classification delay data in the task time sequence, and perform correlation analysis with the fluctuation rate to generate the path delay fluctuation correlation coefficient. S502: Based on the path delay fluctuation correlation coefficient and combined with the path characteristic parameters, perform a matching judgment operation on the path performance at key time points, filter the set of paths that meet the synchronization conditions, and obtain the synchronization path judgment value range. S503: Call the synchronization path determination value range, compare it with the synchronization node parameters in the fluctuation signal, determine the optimal path and corresponding node parameters for classification recovery, and output the classification recovery path and synchronization node configuration table.

Images