Display screen production process management system and method based on artificial intelligence

Abstract

The invention discloses a display screen production process management system and method based on artificial intelligence, and belongs to the technical field of production management. According to the method, a component type library and a production database are constructed, a process flow decision tree for bearing component productivity parameters is created, trunk / cotyledon nodes are marked, and a production process topological graph is generated; tasks are divided according to production batches, two types of training sample sets with or without incomplete components are constructed, and a node comprehensive sample set is separated after standardization processing; and calculating a feature information gain ratio based on a C4.5 algorithm to screen an optimal feature set, constructing a C4.5 decision tree-based learning device through bootstrap resampling, integrating the C4.5 decision tree-based learning device into a double-branch random forest model by adopting a weighted voting method, and inputting real-time feature data of a current batch to output a productivity prediction result. According to the method, the productivity prediction accuracy and the process suitability of display screen production batches are improved, and the problem of productivity imbalance of multi-component and multi-batch production is effectively solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of production management, in particular to a display screen production process management system and method based on artificial intelligence. BACKGROUND

[0002] As a core component of electronic terminals, the production process of display screens has the characteristics of multiple component types, complex assembly process, and obvious batch production. With the acceleration of display technology iteration, display screen products are developing towards high resolution, flexibility, and integration. The required component models have reached dozens or even hundreds, and the production rhythm and consumption law of core components (such as display panels and driving chips) and auxiliary components (such as fixing brackets and connecting wires) differ significantly, making the requirement for production capacity matching even more stringent.

[0003] In the current display screen production process management, there are many problems in the existing technology that need to be solved: first, production capacity prediction mostly uses a single model to predict all components in the whole process, without distinguishing the functional priority of core and auxiliary components, leading to frequent imbalance problems such as core component capacity gap or auxiliary component overcapacity, seriously affecting production progress; second, after production batch division, no batch-specific process model and sample library is established, and cross-batch data interference leads to insufficient prediction accuracy, making it difficult to adapt to production needs of different types of batches such as mass production, trial production, and customization; third, the training samples only cover ideal production states, without considering sudden situations such as component defects and equipment failures, resulting in poor model robustness and large prediction errors in actual production; fourth, feature selection relies on manual experience or simple statistical methods, without quantifying feature importance with algorithms, leading to high model redundancy and low operation efficiency, and the inability to quickly respond to real-time production scheduling needs; fifth, production process modeling mostly uses linear programming, without reflecting the assembly relationship between components through the combination of decision trees and topological graphs, making it difficult to accurately map the actual production logic.

[0004] In addition, in the existing technology, the application of AI models such as random forests in production capacity prediction mostly stays at the global unified prediction level, without customized design for the batch characteristics and node function differences of display screen production, leading to insufficient model adaptation to production processes. At the same time, most schemes ignore the impact of component defect states on production capacity consumption, and the training samples cover only a single scenario, making it impossible to meet the dynamic production capacity adjustment needs in complex production environments. These problems lead to frequent component supply and demand imbalance, batch delivery delays, and production cost waste in the display screen production process, restricting the improvement of production efficiency and product quality. SUMMARY

[0005] The purpose of the present application is to provide an AI-based display screen production process management system and method to solve the problems raised in the background art, and the technical solution of the present application can be summarized as a full-link intelligent logic of "process modeling-feature optimization-batch adaptation-integrated prediction", which includes: Process modeling level (S1-S3): Based on the sequential logic of the display screen assembly process flow, a process flow decision tree is constructed, so that each tree node corresponds to a component model and carries capacity parameters, and then the main stem (core component) and leaf (auxiliary component) nodes are marked through cross-flow relationship to generate a visual production process topology map, realizing the digital mapping of the production process. Batch adaptation level (S4-S5): Tasks are divided according to production batches, an independent stage topology map and sample set are constructed for each batch to avoid cross-batch data interference, and batch-specific adaptation of the prediction model is realized according to the production characteristics of each batch (such as the scale of mass production batches and the individuality of customized batches). Feature optimization level (S6): C4.5 algorithm is used to quantify feature information gain ratio to select the most critical core features (yield demand real-time value and consumption speed parameters) for capacity prediction, eliminate redundant information, reduce model operation complexity, and ensure strong correlation between features and prediction targets. Integrated prediction level (S7-S8): Multiple C4.5 decision tree base learners are constructed by generating diversified training subsets through bootstrap resampling and integrating them into a double-branch random forest using weighted voting method, with the main stem branch focusing on core components and the leaf branch focusing on auxiliary components, and the node importance weight distribution is combined to improve the relevance and generalization ability of the prediction.

[0006] To solve the above technical problems, the present application provides the following technical solutions: The AI-based display screen production process management system comprises a component and task management module, a process flow decision tree construction module, a production process topology map generation module, a production batch division and sample set construction module, a feature processing module, a feature importance calculation module, a base learner construction module, and a random forest capacity prediction module. The component and task management module is used to build a display screen component type library to record information of each component model, build a display screen component production database to record component production data, create production tasks and encapsulate them into the production database, record the number of single component models, and issue the upper limit of yield demand to each production end. The process flow decision tree construction module creates a process flow decision tree based on the assembly process flow of finished display screens, so that each tree node corresponds to a component model and carries the yield demand real-time value and consumption speed parameters of the component. The production flow topology generation module is configured to establish cross-flow relationships between the process flow decision trees, mark trunk nodes and leaf nodes, and construct a complete production flow topology graph for the display screen production task based on the two types of nodes. The production batch division and sample set construction module is configured to divide the production task into production batches, construct a complete production flow topology graph for each batch, and realize independent storage and associated calling; for the topology graph of each batch, construct a first type of training sample set in the state of no missing components and a second type of training sample set in the state of missing components; The feature processing module is configured to extract two types of core feature parameter samples, namely yield demand real-time values and consumption speed parameters, from the two types of training sample sets, perform standardization processing on the core feature parameter samples to eliminate dimensional differences, and separate to obtain a trunk node comprehensive sample set and a leaf node comprehensive sample set. The feature importance calculation module is configured to construct independent C4.5 algorithm execution models for the trunk nodes and the leaf nodes respectively, take “whether the component production capacity meets the assembly demand” as a target classification label, quantify information entropy, conditional entropy, information gain, and information gain ratio of the target features, and sort and filter based on the information gain ratio to obtain a trunk node optimal feature set and a leaf node optimal feature set. The base learner construction module is configured to perform bootstrap resampling on the trunk node comprehensive sample set and the leaf node comprehensive sample set based on the trunk node optimal feature set and the leaf node optimal feature set, respectively, to generate k trunk node training subsets and k leaf node training subsets; for each training subset, a C4.5 decision tree model is constructed based on the corresponding optimal feature set, and m tree nodes are randomly selected from the optimal feature set to participate in node splitting decision-making during the construction process. The random forest production capacity prediction module is configured to use a weighted voting method, integrate the C4.5 decision tree models of the trunk nodes and the leaf nodes according to the importance proportion of the tree nodes in the production flow, and form a double-branch random forest model; receive the real-time feature data of each tree node after standardization processing under the current production batch, input the corresponding random forest branch respectively, and output the production capacity prediction results of each component.

[0007] The display screen production flow management method based on artificial intelligence includes the following steps: Step S1: Construct a display screen component type library and a production database, create a production task, record the number of component models, and release the upper limit of yield demand to the production end; Step S2: Create a process flow decision tree based on the display screen assembly process, so that the tree nodes carry the yield demand real-time values and consumption speed parameters of the corresponding components; Step S3: Establish cross-flow relationships between the process flow decision trees, mark trunk nodes and leaf nodes, and construct a complete production flow topology graph. Step S4: dividing the production batches and constructing the stage complete production process topology of each batch, and constructing two types of training sample sets of intact components and defective components for each batch; Step S5: extracting core feature parameters from the two types of training sample sets and performing standardization processing, and separating the comprehensive sample set of the main node and the leaf node; Step S6: calculating the feature information gain ratio of the main node and the leaf node based on the C4.5 algorithm respectively, and screening to obtain the optimal feature set; Step S7: bootstrap resampling the comprehensive sample set, constructing multiple C4.5 decision tree base learners based on the optimal feature set, and adopting a random feature selection strategy in the construction process; Step S8: integrating the C4.5 decision tree models of the main node and the leaf node through a weighted voting method to form a double-branch random forest model, inputting the real-time feature data of the current production batch, and outputting the production capacity prediction result of each component.

[0008] As a preferred scheme of the present application, the specific implementation process of step S1 comprises: A display screen component type library is constructed for recording the component model information of the assembled finished display screen, and one component model information corresponds to one display screen component type encapsulated as data; A display screen component production database is constructed for recording the production data of each display screen component; A display screen production task is created and encapsulated into the display screen component production database; Under the display screen production task, the quantity of a single component model of the assembled finished display screen is recorded; Under the display screen production task, a yield demand upper limit value is issued to each display screen component production end, and the yield demand upper limit value is the product of the finished assembly quantity of the display screen and the quantity of the single component model.

[0009] As a preferred scheme of the present application, the specific implementation process of step S2 comprises: Based on the assembly process flow of the finished display screen, a process flow decision tree is created, wherein one assembly process flow corresponds to the creation of one process flow decision tree, and each assembly process flow generated according to the assembly sequence logic between the single component model corresponds to the generation of each process flow decision tree, one tree node in the process flow decision tree corresponds to one flow node in the assembly process flow, and one flow node corresponds to one component model; In the created process flow decision tree, each tree node carries a yield demand real-time value of a single component model, and the yield demand real-time value is the difference between the yield demand upper limit value and the yield consumption value, and the yield consumption value is the number of completed assembly in the assembly state of the single component model; In the created process flow decision tree, each tree node also carries a consumption speed parameter of a single component model.

[0010] As a preferred scheme of the present application, the implementation process of step S3 includes: Establishing a cross-flow relationship between the created process flow decision trees, the cross-flow relationship being an assembly relationship between a tree node in one process flow decision tree and a tree node in another process flow decision tree; Based on the cross-flow relationship, marking the tree nodes having the assembly relationship as trunk nodes, and marking the tree nodes not having the assembly relationship as sub-leaf nodes; Based on the trunk nodes and the sub-leaf nodes, constituting a complete production flow topology under the display screen production task.

[0011] As a preferred scheme of the present application, the implementation process of step S4 includes: Dividing the display screen production task into production batches, each production batch corresponding to a stage complete production flow topology, and the stage complete production flow topology corresponding to the production batch being independently stored and associated with calling; For the stage complete production flow topology corresponding to each production batch, constructing two types of training sample sets, including: The first type of training sample set under the state of no defective components, used to record the yield demand real-time value and the consumption speed parameter of each tree node under the state of no defective components; The second type of training sample set under the state of defective components, used to record the yield demand real-time value and the consumption speed parameter of each tree node under the state of defective components.

[0012] As a preferred scheme of the present application, the implementation process of step S5 includes: From the first type of training sample set and the second type of training sample set, respectively extracting core feature parameter samples of the tree nodes, denoted as tree node samples, the core feature parameter samples including the yield demand real-time value, denoted as feature F1, and the consumption speed parameter, denoted as feature F2; Standardizing the core feature parameter samples in the two types of training sample sets to eliminate dimensional differences, to obtain a standardized feature sample set, including a first type of standardized feature sample set corresponding to the first type of training sample set, denoted as SS1, and a second type of standardized feature sample set corresponding to the second type of training sample set, denoted as SS2; Separating the tree node samples marked as trunk nodes from the first type of standardized feature sample set and the second type of standardized feature sample set to form a trunk node comprehensive sample set Smain; Separating the tree node samples marked as leaf nodes from the first and second sets of normalized feature samples to form a leaf node comprehensive sample set Sleaf.

[0013] As a preferred scheme of the present application, the implementation process of step S6 comprises: Respectively constructing independent C4.5 algorithm execution models for the trunk nodes and the leaf nodes; For the trunk node comprehensive sample set, taking “whether the component production capacity meets the assembly demand” as a target classification label, including a target meeting label and a target not meeting label, calculating the information entropy H(D) and the conditional entropy H(D|A) of each tree node sample (F1, F2), to obtain the information gain Gain(D, A) of the target feature A, wherein D is a sample universe formed by the trunk node comprehensive sample set, and the target feature A is the feature F1 or the feature F2; Based on the information gain Gain(D, A), the information gain ratio Gain_ratio(D, A) = Gain(D, A) / H_A(D) of the target feature A is obtained, wherein H_A(D) is the feature entropy of the target feature A, and the feature entropy is an intrinsic value; For the leaf node comprehensive sample set, the same calculation logic is adopted to calculate the information gain ratio of the features F1 and F2 with the same target classification label; Taking the numerical value of the information gain ratio as a quantitative indicator of the feature importance of the tree nodes, the feature importance of the trunk nodes and the leaf nodes is respectively sorted in descending order, and the tree node samples with the top N feature importance are screened to obtain the optimal feature set Fmain of the trunk nodes and the optimal feature set Fleaf of the leaf nodes, and the feature selection of the tree nodes is completed.

[0014] As a preferred scheme of the present application, the implementation process of step S7 comprises: Based on the optimal feature set Fmain of the trunk nodes and the optimal feature set Fleaf of the leaf nodes, bootstrap resampling is performed on the trunk node comprehensive sample set Smain and the leaf node comprehensive sample set Sleaf respectively: Randomly extracting k tree node samples from the trunk node comprehensive sample set Smain to form k trunk node training subsets; Randomly extracting k tree node samples from the leaf node comprehensive sample set Sleaf to form k leaf node training subsets; Wherein, k is the number of base learners of the random forest; For each trunk node training subset, a corresponding C4.5 decision tree model is constructed based on the optimal feature set Fmain of the trunk nodes; For each sub-leaf node training subset, a corresponding C4.5 decision tree model is constructed based on the optimal feature set Fleaf of the sub-leaf node; In each base learner construction process, a random feature selection strategy is adopted to randomly select m tree node samples from the optimal feature set Fmain of the main node or the optimal feature set Fleaf of the sub-leaf node to participate in the splitting decision of the base learner.

[0015] As a preferred scheme of the present application, the specific implementation process of step S8 includes: The optimal feature set Fmain of the main node is integrated with the optimal feature set Fleaf of the sub-leaf node to form a double-branch random forest model by adopting a weighted voting method based on the importance proportion of the tree node in the production process, wherein the main node branch is responsible for predicting the capacity matching degree of the core assembly component, and the sub-leaf node branch is responsible for predicting the capacity matching degree of the auxiliary assembly component. The real-time feature data of each tree node after data standardization processing under the current production batch is input into the corresponding random forest branch, and the capacity prediction result of each component is output.

[0016] Compared with the prior art, the present application has the following beneficial effects: The "main / sub-leaf node + double-branch random forest" collaborative design is adopted to deeply integrate the component function priority with the ensemble learning model, thereby solving the technical problem of core and auxiliary component capacity prediction confusion in the prior art, and realizing differentiated and accurate capacity prediction. The "production batch-topology graph-sample set" trinity adaptation scheme is proposed, and the stage topology graph and the exclusive sample set are independently constructed for each batch, which not only avoids cross-batch data interference, but also realizes batch-level dynamic model optimization, solves the problem of unstable prediction accuracy in multi-batch production, and is helpful for improving the batch capacity matching degree compared with the existing cross-batch mixed prediction scheme. Through the construction of two types of training sample sets with or without missing components, combined with the feature quantization screening of the C4.5 algorithm, the robustness of the model to complex production environments is improved, the model can still maintain stable prediction under sudden conditions such as component failure and supply shortage, and the pain point of "feature redundancy leading to low efficiency" of the existing model is solved, which helps to improve the prediction accuracy while reducing the complexity of the model, avoids the confusion of core and auxiliary component capacity prediction, and avoids the spread of core component capacity gap prediction error and the waste of auxiliary components. BRIEF DESCRIPTION OF DRAWINGS

[0017] The accompanying drawings are included to provide a further understanding of the present application, and constitute a part of the specification, together with the embodiments of the present application, for explaining the present application, and do not constitute a limitation on the present application.

[0018] Figure 1 is a step schematic diagram of the display screen production process management method based on artificial intelligence of the present application. DETAILED DESCRIPTION

[0019] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0020] In the first embodiment of the present application: a display screen production process management system based on artificial intelligence is provided, which comprises: A component and task management module is configured to build a display screen component type library to record component model information, build a display screen component production database to record component production data, create production tasks and encapsulate them to the production database, record the number of single component models, and publish the upper limit value of production demand to each production end; A process flow decision tree construction module is configured to create a process flow decision tree based on the assembly process flow of the finished display screen, so that each tree node corresponds to a component model and carries the real-time value of the production demand of the component and the consumption speed parameter; A production process topology graph generation module is configured to establish cross-flow relationships among the process flow decision trees, mark the main nodes and the sub-leaf nodes, and form a complete production process topology graph based on the two types of nodes under the display screen production task; A production batch division and sample set construction module is configured to divide the production task into production batches, build a complete production process topology graph for each batch and realize independent storage and associated calling, and construct a first type of training sample set under the state of no missing components and a second type of training sample set under the state of missing components for the topology graph of each batch; A feature processing module is configured to extract two types of core feature parameter samples, i.e., the real-time value of production demand and the consumption speed parameter, from the two types of training sample sets, standardize the core feature parameter samples to eliminate dimensional differences, and separate to obtain a main node comprehensive sample set and a sub-leaf node comprehensive sample set; A feature importance calculation module is configured to build independent C4.5 algorithm execution models for the main nodes and the sub-leaf nodes respectively, take "whether the component capacity meets the assembly demand" as the target classification label, quantify the information entropy, conditional entropy, information gain and information gain ratio of the target features, and sort and select the main node optimal feature set and the sub-leaf node optimal feature set based on the information gain ratio; The base learner construction module performs bootstrap resampling on the comprehensive sample sets of the main nodes and the comprehensive sample sets of the leaf nodes, based on the optimal feature sets of the main nodes and the optimal feature sets of the leaf nodes, to generate k training subsets of the main nodes and k training subsets of the leaf nodes. For each training subset, a C4.5 decision tree model is constructed based on the corresponding optimal feature set. During the construction process, m tree nodes are randomly selected from the optimal feature set to participate in the node splitting decision. The Random Forest Capacity Prediction Module uses a weighted voting method to integrate the C4.5 decision tree model of the trunk node and the leaf node into a two-branch random forest model based on the importance ratio of the tree node in the production process. It receives the real-time feature data of each tree node in the current production batch after standardization, inputs them into the corresponding random forest branches, and outputs the capacity prediction results of each component.

[0021] Please see Figure 1 In this second embodiment, an AI-based display screen production process management method is provided to be applicable to the first embodiment. This embodiment takes the mass production of a 65-inch 4K LCD display screen from a display screen manufacturer as an example. The production task is 10,000 finished products per month, divided into 10 production batches of 1,000 units per batch, involving 28 types of components (including 6 core components: display panel, driver chip, backlight module, control circuit board, power module, touch sensor; and 22 auxiliary components: fixing bracket, connecting wires, dustproof stickers, etc.). The method includes the following steps: Step S1: Build a display component type library and production database, create production tasks and register component models and quantities, and publish the upper limit of production demand to the production end; Specifically, a display component type library is constructed to record the model information of each component in the assembled finished display screen, and the data corresponding to one component model information is encapsulated into one display component type; Build a production database for display screen components to record production data for each display screen component; Create a display screen production task and encapsulate it into the display screen component production database; Under the display screen production task, record the quantity of a single component model for assembling the finished display screen; Under the display production task, the upper limit of production demand is issued to each display component production end. The upper limit of production demand is the product of the number of finished display assembly units and the number of single component models. For example, the finished product assembly quantity is 1000 units / batch, the quantity of a single core component (such as a display panel) is 1 unit / set, and the quantity of auxiliary components (such as connecting wires) is 4 pieces / set; Upper limit of yield demand = 1000 units x 1 = 1000 (core components), 1000 units x 4 = 4000 (auxiliary components).

[0022] Step S2: Create a process flow decision tree based on the display screen assembly process flow, so that the tree nodes carry the yield demand real-time value and the consumption speed parameter of the corresponding components; Specifically, based on the assembly process flow of the finished display screen, a process flow decision tree is created, wherein one assembly process flow corresponds to the creation of one process flow decision tree, and each assembly process flow generated according to the assembly sequence logic between single component models corresponds to the generation of each process flow decision tree. A tree node in the process flow decision tree corresponds to a flow node in the assembly process flow, and a flow node corresponds to a component model; In the created process flow decision tree, each tree node carries the yield demand real-time value of a single component model, which is the difference between the upper limit of yield demand and the yield consumption value, and the yield consumption value is the number of completed assemblies of a single component model in the assembly state; In the created process flow decision tree, each tree node also carries the consumption speed parameter of a single component model; Exemplarily, according to the process flow of "display panel → backlight module → drive chip → control circuit board → power module → touch sensor → auxiliary component assembly", 1 main process flow decision tree is created, and each tree node corresponds to 1 component model; The tree nodes update the yield demand real-time value (such as 1000 for the display panel initially, and 500 after assembling 500 units) and the consumption speed parameter (0.5 units per minute on average for core components, and 2 units per minute on average for auxiliary components) in real time.

[0023] Step S3: Establish cross-flow relationships between the process flow decision trees, mark the trunk nodes and the leaf nodes, and form a complete production process topology map; Specifically, cross-flow relationships are established between the created process flow decision trees, and the cross-flow relationships are assembly relationships between a tree node in one process flow decision tree and a tree node in another process flow decision tree; Based on the cross-flow relationships, the tree nodes with assembly relationships are marked as trunk nodes, and the tree nodes without assembly relationships are marked as leaf nodes; Based on the trunk nodes and the leaf nodes, a complete production process topology map under the display screen production task is formed; Exemplarily, the drive chip node and the control circuit board node have assembly dependency (cross-flow relationship) and are marked as core trunk nodes; the fixed support node has no cross-tree assembly relationship and is marked as a leaf node; a batch-specific topology map containing 6 trunk nodes and 22 leaf nodes is generated, and the component assembly correlation is visualized and displayed.

[0024] Step S4: dividing production batches and constructing stage complete production process topology graphs of each batch, and constructing two types of training sample sets for each batch; Specifically, the production batches are divided for the display screen production task, and each production batch corresponds to a stage complete production process topology graph, and the stage complete production process topology graphs corresponding to the production batches are independently stored and associated with each other. For each stage complete production process topology graph corresponding to the production batch, two types of training sample sets are constructed, including: The first type of training sample set in the state of no defective components, which is used to record the yield demand real-time value and consumption speed parameter of each tree node in the state of no defective components; The second type of training sample set in the state of defective components, which is used to record the yield demand real-time value and consumption speed parameter of each tree node in the state of defective components; Exemplarily, 1000 units per batch, a total of 10 batches, and each batch topology graph is independently stored; each batch constructs 1000 data of the first type of sample set (no defect), 300 data of the second type of sample set (defect, simulating 3% component failure), and records node F1 (yield demand real-time value) and F2 (consumption speed parameter).

[0025] Step S5: extracting core feature parameters from the two types of training sample sets and performing standardization processing, and separating the comprehensive sample set of the main node and the sub-leaf node; Specifically, the core feature parameter samples of the tree nodes are extracted from the first type of training sample set and the second type of training sample set, respectively, which are denoted as tree node samples. The core feature parameter samples include the yield demand real-time value, denoted as feature F1, and the consumption speed parameter, denoted as feature F2. The core feature parameter samples in the two types of training sample sets are standardized to eliminate dimensional differences, and a standardized feature sample set is obtained, including a first type of standardized feature sample set corresponding to the first type of training sample set, denoted as SS1, and a second type of standardized feature sample set corresponding to the second type of training sample set, denoted as SS2. The tree node samples marked as main nodes are separated from the first type of standardized feature sample set and the second type of standardized feature sample set to form a main node comprehensive sample set Smain. The tree node samples marked as sub-leaf nodes are separated from the first type of standardized feature sample set and the second type of standardized feature sample set to form a sub-leaf node comprehensive sample set Sleaf. Exemplarily, the Z-score standardization formula is used for data standardization processing to eliminate dimensional differences, and the sample set is classified according to the node type.

[0026] Step S6: Calculate the feature information gain ratio of the main node and the sub-leaf node respectively based on the C4.5 algorithm, and screen the optimal feature set; Specifically, independent C4.5 algorithm execution models are constructed for the main node and the sub-leaf node respectively; For the main node comprehensive sample set, taking whether the "component production capacity meets the assembly demand" as the target classification label, including the target meeting label and the target not meeting label, the information entropy H(D) and the conditional entropy H(D|A) of each tree node sample (F1, F2) are calculated to obtain the information gain Gain(D, A) of the target feature A, wherein D is the total sample domain composed of the main node comprehensive sample set, and the target feature A is the feature F1 or the feature F2; Based on the information gain Gain(D, A), the information gain ratio Gain_ratio(D, A) = Gain(D, A) / H_A(D) of the target feature A is obtained, wherein H_A(D) is the feature entropy of the target feature A, and the feature entropy is an intrinsic value; For the sub-leaf node comprehensive sample set, the same calculation logic is adopted to calculate the information gain ratio of the features F1 and F2 with the same target classification label; The numerical size of the information gain ratio is taken as the quantification index of the feature importance of the tree node, and the feature importance of the main node and the sub-leaf node is sorted in descending order respectively, and the top N tree node samples in the feature importance are screened out to obtain the optimal feature set Fmain of the main node and the optimal feature set Fleaf of the sub-leaf node, and the feature selection of the tree node is completed; Exemplarily, the C4.5 constraint condition is set as: the minimum information gain ratio threshold is 0.1, the termination condition is that the number of node samples < 5% of the total sample number; the information gain ratio of the main node F1 is 0.18, the information gain ratio of F2 is 0.12, the information gain ratio of the sub-leaf node F1 is 0.15, and the information gain ratio of F2 is 0.09.

[0027] Step S7: Bootstrap resampling is performed on the comprehensive sample set, and a plurality of C4.5 decision tree base learners are constructed based on the optimal feature set, and a random feature selection strategy is adopted in the construction process; Specifically, based on the optimal feature set Fmain of the main node and the optimal feature set Fleaf of the sub-leaf node, bootstrap resampling is performed on the main node comprehensive sample set Smain and the sub-leaf node comprehensive sample set Sleaf respectively: k tree node samples are randomly extracted from the main node comprehensive sample set Smain to form k main node training subsets; k tree node samples are randomly extracted from the sub-leaf node comprehensive sample set Sleaf to form k sub-leaf node training subsets; Wherein, k is the number of base learners of the random forest; For each main node training subset, the corresponding C4.5 decision tree model is constructed based on the main node optimal feature set Fmain; For each leaf node training subset, the corresponding C4.5 decision tree model is constructed based on the leaf node optimal feature set Fleaf; In each base learner construction process, a random feature selection strategy is adopted to randomly select m tree node samples from the main node optimal feature set Fmain or the leaf node optimal feature set Fleaf to participate in the splitting decision of the base learner; For example, set k = 50 (number of base learners), extract 50 training subsets from Smain and 50 training subsets from Sleaf; construct 100 C4.5 decision tree base learners (50 for main nodes and 50 for leaf nodes).

[0028] Step S8: Integrate the C4.5 decision tree models of the main nodes and the leaf nodes by weighted voting method to form a double-branch random forest model, input the real-time feature data of the current production batch, and output the production capacity prediction results of each component; Specifically, the weighted voting method is adopted to integrate the C4.5 decision tree models constructed based on the main node optimal feature set Fmain and the leaf node optimal feature set Fleaf according to the importance proportion of the tree nodes in the production process, to form a double-branch random forest model, wherein the main node branch is responsible for predicting the production capacity matching degree of the core assembly component, and the leaf node branch is responsible for predicting the production capacity matching degree of the auxiliary assembly component; Input the real-time feature data of each tree node after data standardization processing under the current production batch into the corresponding random forest branch, and output the production capacity prediction results of each component; For example, the main node branch weight can be set to 0.7 and the leaf node branch weight can be set to 0.3.

[0029] It is worth mentioning that the C4.5 algorithm is a classic decision tree generation algorithm, and the core goal is to construct an optimal decision tree by quantifying the contribution of features to the classification result, to realize accurate classification or prediction; In this embodiment, two types of training sample sets, "no defective components" and "defective components", are constructed to cover normal and abnormal scenarios of display screen production, solving the problem of dependence on "single state samples" of the C4.5 algorithm, making the algorithm training data more suitable for actual production of component failures, supply shortages and other sudden situations; Two types of comprehensive sample sets (Smain, Sleaf) are separated according to "main nodes (core components)" and "leaf nodes (auxiliary components)", providing "classification training data" for the C4.5 algorithm, avoiding confusion of the production capacity features of core and auxiliary components, and adapting to the characteristics of the "single node single splitting logic" of the C4.5 algorithm; The yield demand real-time value (F1) and the consumption speed parameter (F2) are standardized to eliminate dimensional differences (e.g., F1 is in units of "pieces", and F2 is in units of "pieces / minute"), so as to ensure that the feature weights are balanced when the C4.5 algorithm calculates the information entropy, and to avoid feature selection bias caused by inconsistent dimensions; In addition, the embodiment extracts two core features (F1 and F2) that are strongly related to capacity prediction, rather than blindly inputting multi-dimensional data, adapts to the operation logic of the C4.5 algorithm that "feature quality is prioritized over quantity", and reduces the impact of redundant features on algorithm operation efficiency; Independent C4.5 algorithm execution models are respectively constructed for the main nodes and the leaf nodes, with "whether the capacity of the main component meets the assembly demand" as the unified classification label, and the information gain ratios of F1 and F2 are respectively calculated. This "independent calculation of nodes" method adapts to the characteristics of the C4.5 algorithm that "a single sample set has a single classification target", and quantifies the importance of features through the information gain ratio, so as to ensure that the optimal feature set (Fmain and Fleaf) accurately matches the capacity rules of the node type (core / auxiliary component); During the construction of the C4.5 decision tree, the importance of the tree nodes in the production process is taken into account (the main node weight is 0.7, and the leaf node weight is 0.3), so that the algorithm prioritizes the feature contribution of the core component when splitting the nodes, and adapts to the actual demand of "prioritizing the capacity of core components" in display screen production. The termination condition of the C4.5 algorithm is set as "the number of tree node samples is less than 5% of the total number of samples", which, combined with the sample size of the display screen production batch, ensures the depth of the decision tree and avoids algorithm operation timeout caused by excessive sample size.

[0030] It should be noted that, in this document, relational terms such as first and second and the like can only be used to distinguish one entity or action from another entity or action, without necessarily requiring or implying that there is any such actual relationship or order between these entities or actions. Moreover, the terms "comprises", "comprising", or any other variant thereof are intended to cover non-exclusive inclusions, so that a process, method, article, or apparatus that includes a list of elements does not only include those elements, but also includes other elements not explicitly listed, or inherent to such a process, method, article, or apparatus.

[0031] Finally, it should be noted that the above only describes the preferred embodiments of the present application and is not intended to limit the present application, and although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent replacements to some technical features. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. An artificial intelligence-based display panel production process management method, characterized by, The method comprises the following steps: Step S1: constructing a display screen component type library and a production database, creating a production task and recording the number of component models, and issuing a production demand upper limit value to the production end; Step S2: creating a process flow decision tree based on the display screen assembly process flow, so that the tree nodes carry the real-time value of the production demand and the consumption speed parameter of the corresponding component; Step S3: establishing cross-flow relationships among the process flow decision trees, marking the main nodes and the sub-leaf nodes, and forming a complete production process topology graph; Step S4: dividing production batches and constructing a complete production process topology graph for each batch, and constructing two types of training sample sets for each batch, namely, intact component and defective component; Step S5: extracting core feature parameters from the two types of training sample sets and performing standardization processing, and separating the comprehensive sample set of the main nodes and the sub-leaf nodes; Step S6: calculating the feature information gain ratio of the main nodes and the sub-leaf nodes based on the C4.5 algorithm, and screening to obtain the optimal feature set; Step S7: bootstrap resampling the comprehensive sample set, constructing multiple C4.5 decision tree base learners based on the optimal feature set, and adopting a random feature selection strategy during the construction process; Step S8: integrating the C4.5 decision tree models of the main nodes and the sub-leaf nodes through a weighted voting method to form a double-branch random forest model, inputting the real-time feature data of the current production batch, and outputting the production capacity prediction results of each component. 2.The artificial intelligence-based display production flow management method of claim 1, wherein, The specific implementation process of step S1 comprises: Constructing a display screen component type library to record the component model information of the assembled display screen, and one type of component model information corresponds to one type of display screen component; Constructing a display screen component production database to record the production data of each display screen component; Creating a display screen production task and encapsulating it into the display screen component production database; Under the display screen production task, record the number of single component models of the assembled display screen; Under the display screen production task, issue a production demand upper limit value to each display screen component production end, and the production demand upper limit value is the product of the number of assembled display screens and the number of single component models. 3.The artificial intelligence-based display production flow management method of claim 1, wherein, The specific implementation process of step S2 comprises: Based on the assembly process flow of the finished display screen, a process flow decision tree is created, wherein one assembly process flow corresponds to one process flow decision tree, and each assembly process flow generated according to the assembly sequence logic between single component models corresponds to each process flow decision tree, one tree node in the process flow decision tree corresponds to one flow node in the assembly process flow, and one flow node corresponds to one component model; In the created process flow decision tree, each tree node carries the real-time value of the production demand of the single component model, and the real-time value of the production demand is the difference between the production demand upper limit value and the production consumption value, and the production consumption value is the number of completed assemblies in the assembly state of the single component model; In the created process flow decision tree, each tree node also carries the consumption speed parameter of the single component model. 4.The AI-based display production flow management method of claim 3, wherein, The specific implementation process of step S3 comprises: establishing a cross-flow relationship between the created process flow decision trees, the cross-flow relationship being an assembly relationship between a tree node in one process flow decision tree and a tree node in another process flow decision tree; based on the cross-flow relationship, marking the tree nodes having the assembly relationship as trunk nodes and marking the tree nodes not having the assembly relationship as sub-leaf nodes; based on the trunk nodes and the sub-leaf nodes, constituting a complete production flow topology under the display screen production task. 5.The AI-based display production flow management method of claim 4, wherein, The specific implementation process of the step S4 includes: dividing the display screen production task into production batches, each production batch corresponding to a stage complete production flow topology, and the stage complete production flow topology corresponding to each production batch being independently stored and associatedly called; for each stage complete production flow topology corresponding to each production batch, constructing two types of training sample sets, including: a first type of training sample set under a non-defective component state, used for recording a yield demand real-time value and a consumption speed parameter of each tree node under the non-defective component state; a second type of training sample set under a defective component state, used for recording a yield demand real-time value and a consumption speed parameter of each tree node under the defective component state. 6.The AI-based display production flow management method of claim 5, wherein, The specific implementation process of the step S5 includes: extracting core feature parameter samples of the tree nodes from the first type of training sample set and the second type of training sample set, respectively, and marking the core feature parameter samples as tree node samples, the core feature parameter samples including a yield demand real-time value, marked as a feature F1, and a consumption speed parameter, marked as a feature F2; standardizing the core feature parameter samples in the two types of training sample sets to eliminate dimensional differences, to obtain a standardized feature sample set, including a first type of standardized feature sample set corresponding to the first type of training sample set, marked as SS1, and a second type of standardized feature sample set corresponding to the second type of training sample set, marked as SS2; separating the tree node samples marked as the trunk nodes from the first type of standardized feature sample set and the second type of standardized feature sample set to form a trunk node comprehensive sample set Smain; separating the tree node samples marked as the sub-leaf nodes from the first type of standardized feature sample set and the second type of standardized feature sample set to form a sub-leaf node comprehensive sample set Sleaf. 7.The AI-based display production flow management method of claim 6, wherein, The specific implementation process of the step S6 includes: respectively constructing independent C4.5 algorithm execution models for the trunk nodes and the sub-leaf nodes; for the trunk node comprehensive sample set, taking "whether a component capacity meets an assembly demand” as a target classification label, including a target meeting label and a target not meeting label, calculating an information entropy H(D) and a conditional entropy H(D|A) of each tree node sample (F1, F2), to obtain an information gain Gain(D, A) of a target feature A, wherein D is a sample universe constituted by the trunk node comprehensive sample set, and the target feature A is the feature F1 or the feature F2; based on the information gain Gain(D, A), obtaining an information gain ratio Gain_ratio(D, A) = Gain(D, A) / H_A(D) of the target feature A, wherein H_A(D) is a feature entropy of the target feature A, and the feature entropy is an intrinsic value; For the cotyledon node comprehensive sample set, the same calculation logic is adopted to calculate the information gain ratio of features F1 and F2 with the same target classification label; The numerical size of the information gain ratio is taken as a quantitative indicator of the feature importance of the tree node, and the feature importance of the main node and the cotyledon node is sorted in descending order, respectively, to filter out the top N tree node samples in terms of feature importance, thereby obtaining the main node optimal feature set Fmain and the cotyledon node optimal feature set Fleaf, and completing the feature selection of the tree node. 8.The AI-based display production flow management method of claim 7, wherein, The specific implementation process of step S7 includes: Based on the main node optimal feature set Fmain and the cotyledon node optimal feature set Fleaf, bootstrap resampling is performed on the main node comprehensive sample set Smain and the cotyledon node comprehensive sample set Sleaf, respectively: k tree node samples are randomly selected from the main node comprehensive sample set Smain to form k main node training subsets; k tree node samples are randomly selected from the cotyledon node comprehensive sample set Sleaf to form k cotyledon node training subsets; Wherein, k is the number of base learners of the random forest; For each main node training subset, a corresponding C4.5 decision tree model is constructed based on the main node optimal feature set Fmain; For each cotyledon node training subset, a corresponding C4.5 decision tree model is constructed based on the cotyledon node optimal feature set Fleaf; In the construction process of each base learner, a random feature selection strategy is adopted to randomly select m tree node samples from the main node optimal feature set Fmain or the cotyledon node optimal feature set Fleaf to participate in the splitting decision of the base learner. 9.The AI-based display production flow management method of claim 8, wherein, The specific implementation process of step S8 includes: Using the weighted voting method, the main node optimal feature set Fmain and the cotyledon node optimal feature set Fleaf are integrated to form a double-branch random forest model based on the importance proportion of the tree node in the production process, wherein the main node branch is responsible for predicting the capacity matching degree of the core assembly component, and the cotyledon node branch is responsible for predicting the capacity matching degree of the auxiliary assembly component; The real-time feature data of each tree node after data standardization processing under the current production batch is input into the corresponding random forest branch, and the capacity prediction result of each component is output.

10. A system for performing the artificial intelligence-based display production process management method according to claim 1, characterized by, The system includes a component and task management module, a process flow decision tree construction module, a production process topology generation module, a production batch division and sample set construction module, a feature processing module, a feature importance calculation module, a base learner construction module, and a random forest capacity prediction module; wherein: The component and task management module is used to build a display screen component type library to record the component model information, build a display screen component production database to record the component production data, create a production task and package it to the production database, record the number of single component models, and publish the upper limit value of the output demand to each production end. The process flow decision tree construction module creates a process flow decision tree based on the assembly process flow of the finished display screen, so that each tree node corresponds to a component model and carries the real-time value of the yield demand of the component and the consumption speed parameter; The production flow topology generation module is configured to establish cross-flow relationships between the process flow decision trees, mark the main nodes and the sub-leaf nodes, and form a complete production flow topology graph based on the two types of nodes under the display screen production task; The production batch division and sample set construction module is configured to divide the production task into production batches, construct a complete production flow topology graph for each batch, and realize independent storage and associated calling; for the topology graph of each batch, a first type of training sample set under the condition of no defective components and a second type of training sample set under the condition of defective components are constructed; The feature processing module is configured to extract two types of core feature parameter samples, i.e., the yield demand real-time value and the consumption speed parameter, from the two types of training sample sets, perform standardization processing on the core feature parameter samples to eliminate dimensional differences, and separate to obtain a main node comprehensive sample set and a sub-leaf node comprehensive sample set; The feature importance calculation module is configured to construct independent C4.5 algorithm execution models for the main nodes and the sub-leaf nodes, respectively, take "whether the component capacity meets the assembly demand" as the target classification label, quantify the information entropy, conditional entropy, information gain, and information gain ratio of the target feature, and sort and filter the main node optimal feature set and the sub-leaf node optimal feature set based on the information gain ratio; The base learner construction module is configured to perform bootstrap resampling on the main node comprehensive sample set and the sub-leaf node comprehensive sample set based on the main node optimal feature set and the sub-leaf node optimal feature set, respectively, to generate k main node training subsets and k sub-leaf node training subsets; for each training subset, a C4.5 decision tree model is constructed based on the corresponding optimal feature set, and m tree nodes are randomly selected from the optimal feature set to participate in node splitting decision-making during the construction process; The random forest capacity prediction module is configured to use a weighted voting method, integrate the C4.5 decision tree models of the main nodes and the sub-leaf nodes according to the importance proportion of the tree nodes in the production flow, form a double-branch random forest model, receive the real-time feature data of each tree node after standardization processing under the current production batch, input the corresponding random forest branch, and output the capacity prediction result of each component.

Images