Fire door retardant core material formula recommendation method based on knowledge graph

Abstract

The present application relates to the field of fireproof materials and intelligent information processing technology, in particular to a fire door retardant core material formula recommendation method based on knowledge graph, comprising: obtaining fire door retardant core material data, constructing a knowledge graph containing material component layer, flame-retardant mechanism layer, performance parameter layer and application scene layer; establishing the corresponding relationship of material components, flame-retardant mechanism and performance parameters, and constructing fire door grade performance index constraint subgraph; determining target performance parameters according to formula research and development requirements, and reversely looking up target flame-retardant mechanism; screening candidate material pool, extracting node feature vector and calculating the synergy coefficient of material components for target flame-retardant mechanism node; determining target material formula path based on the synergy coefficient and outputting recommended formula combination; the method realizes formula rapid recommendation for specific fire door project, reduces the iteration number of traditional manual trial and improves the test pass rate.

Description

Technical Field

[0001] This invention relates to the fields of fireproof materials and intelligent information processing technology, specifically a method for recommending formulations of fire-retardant core materials for fire doors based on knowledge graphs. Background Technology

[0002] Currently, the formulation development of flame-retardant core materials for fire doors is usually based on human experience, dispersion test data, and comparison of individual performance indicators. When multiple constraints such as flame retardancy, low smoke, low thermal conductivity, and environmental protection are imposed for specific application scenarios and fire door grades, researchers cannot efficiently reverse screen the relationship between material components, flame retardant mechanisms, and performance parameters. Each trial formulation may include a large number of invalid combinations that do not meet grade standards or environmental protection requirements, resulting in low accuracy of formulation screening and a cumbersome trial formulation process. Summary of the Invention

[0003] The purpose of this invention is to provide a knowledge graph-based method for recommending fire-retardant core material formulations for fire doors, addressing the following technical problems: Existing fire door fire-retardant core material formulation development technologies have significant shortcomings in efficiently reverse-engineering the relationships between material components, fire-retardant mechanisms, and performance parameters under multiple constraints, and in reducing ineffective combinations that do not meet grade standards or environmental requirements to improve formulation recommendation efficiency. There is an urgent need to propose a knowledge graph-based method for recommending fire door fire-retardant core material formulations that can more accurately handle reverse reasoning of material-performance relationships in specific scenarios, dynamically quantify synergistic effects, and optimize the recommended formulation path. The purpose of this invention can be achieved through the following technical solutions: The knowledge graph-based method for recommending fire-retardant core material formulations for fire doors includes the following steps: Obtain material data of fire door flame-retardant core material, construct entity node system of flame-retardant core material knowledge graph. The entity node system includes material component layer composed of multiple material component nodes, flame-retardant mechanism layer composed of multiple flame-retardant mechanism nodes, performance parameter layer composed of multiple performance parameter nodes, and application scenario layer composed of multiple application scenario nodes. In the physical node system, the correspondence between material composition nodes, flame retardant mechanism nodes and performance parameter nodes is established, and a fire door rating performance index constraint sub-graph is constructed based on the fire door rating standard. Receive formulation development requirements, and based on the application scenarios and fire door ratings in the formulation development requirements, select the corresponding application scenario nodes and the corresponding fire door rating performance index constraint subgraphs in the application scenario layer to determine the target performance parameters. Following the correspondence, we reverse-search from the target performance parameters to determine the target flame retardant mechanism nodes that satisfy the target performance parameters; Candidate material pools are obtained by screening material component nodes in the material component layer. The knowledge graph is represented by a graph embedding model to map material component nodes and flame retardant mechanism nodes into dense continuous vectors of fixed dimensions. The node feature vectors of material component nodes and target flame retardant mechanism nodes in the candidate material pool are extracted. Based on the node feature vectors, the synergy coefficients between material components in the candidate material pool for the target flame retardant mechanism node are calculated. Based on the synergy coefficient, the target material formulation path is determined in the candidate material pool, and the material components corresponding to the path are output as the recommended flame-retardant core material formulation combination.

[0004] Optionally, the specific composition of the physical node system includes: the material component layer includes a base resin node and a flame retardant node, and integrates environmental attribute information such as volatile organic compound content, halogen content, and environmental certification status in the material component node; the flame retardant mechanism layer includes condensed phase flame retardant nodes, gas phase flame retardant nodes, endothermic flame retardant nodes, and chemical flame retardant nodes; the performance parameter layer includes oxygen index nodes, vertical combustion rating nodes, fire resistance time nodes, smoke density rating nodes, and thermal conductivity nodes; the application scenario layer includes building fire door scenario nodes, industrial fire door scenario nodes, and residential entrance fire door scenario nodes.

[0005] Optionally, in the entity node system, a correspondence is established between material component nodes, flame retardant mechanism nodes, and performance parameter nodes. Specifically, this includes: establishing directed edges between material component nodes and flame retardant mechanism nodes to form a many-to-many correspondence from material component to flame retardant mechanism; and establishing associated edges from flame retardant mechanism nodes to performance parameter nodes to form a correspondence from flame retardant mechanism to performance parameter.

[0006] Optionally, the fire door rating performance index constraint subgraph is constructed as follows: taking the rating node corresponding to the fire door rating selected in the formula development requirements as the center, a constraint edge with a threshold attribute is established to the corresponding node in the performance parameter layer, and in the subsequent formula path determination process, material combination paths whose associated performance parameters do not meet the threshold condition are eliminated based on the threshold attribute.

[0007] Optionally, the material component nodes in the material component layer are screened to obtain a candidate material pool, specifically including: traversing the material component layer, and according to the environmental protection attribute information, removing material component nodes whose volatile organic compound content is greater than a preset volatile organic compound content threshold or whose halogen content is greater than zero, and using the remaining material component nodes as the candidate material pool.

[0008] Optionally, the synergy coefficients among the material components in the candidate material pool for the target flame retardant mechanism node are calculated according to the following formula: In the formula, Indicates material composition and material components For the target flame retardant mechanism The coefficient of synergy; , These represent the index numbers of the material components in the candidate material pool; Index number indicating the target flame-retardant mechanism; and These represent the material components extracted from the knowledge graph. and material components The node feature vector; assuming the extracted node feature vector has a dimension of d, then the result obtained after vector concatenation is... The dimension is 2d, and correspondingly, the action coefficient matrix The dimension is set to 1×2d to ensure that the result of linear computation between the two is a one-dimensional scalar; This represents a vector concatenation operation; Represents matrix multiplication operation; Indicates the target flame retardant mechanism The coefficient matrix of action; Indicates the target flame retardant mechanism The one-dimensional scalar correction term; This represents the Sigmoid activation function; This represents the physical-chemical compatibility parameter.

[0009] Optionally, the physicochemical compatibility parameter is determined based on the thermal decomposition temperature matching degree, and its calculation formula is as follows: In the formula, Indicates material composition The initial thermal decomposition temperature; Indicates material composition The initial thermal decomposition temperature; This represents the temperature-sensitive adjustment coefficient, whose dimension is the reciprocal of temperature, to ensure that the result after multiplying it by the temperature difference is dimensionless. This represents the absolute value of the temperature difference between the two values; Represented by natural constant An exponential function with base 0.

[0010] Optionally, the action coefficient matrix and corrections The parameter update method is as follows: Obtain the combustion test database of historical formulas, extract the performance increment data before and after combustion for each historical formula in the combustion test database; construct a loss function with the difference between the actual performance increment and the predicted performance increment as the objective; specifically, the loss function adopts the mean square error function, and the prediction residual is quantified by calculating the sum of the squares of the differences between the actual performance increment and the predicted performance increment of each performance parameter. The gradient descent method is then used to update the coefficient matrix. and corrections Update the parameters.

[0011] Optionally, based on the synergy coefficient, a target material formulation path is determined in the candidate material pool, specifically including: using a heuristic search algorithm to find a combination of material components in the candidate material pool that satisfies the target flame retardant mechanism node requirements and has the largest sum of synergy coefficients, as the target material formulation path.

[0012] Optionally, the material components corresponding to the path can be output as recommended flame-retardant core material formulation combinations. Specifically, this includes: outputting the recommended flame-retardant core material formulation combinations corresponding to the target material formulation path, and simultaneously displaying the correspondence between the recommended flame-retardant core material formulation combinations and the performance parameter nodes in the flame-retardant core material knowledge graph.

[0013] Compared with the prior art, the present invention has the following beneficial effects: This application presents a knowledge graph-based method for recommending fire-retardant core material formulations for fire doors. It acquires material data for these materials and constructs an entity node system including a material composition layer, a flame-retardant mechanism layer, a performance parameter layer, and an application scenario layer. It establishes the correspondence between material composition nodes, flame-retardant mechanism nodes, and performance parameter nodes. Based on fire door rating standards, it constructs a rating performance index constraint subgraph with threshold attributes. Combining application scenarios and fire door ratings, it performs a reverse lookup of target performance parameters. Furthermore, it filters material composition nodes based on volatile organic compound content, halogen content, and environmental certification status. Based on candidate... The synergy coefficient between materials is calculated using the feature vectors of material nodes and target flame retardant mechanism nodes. After determining the target material formulation path using a heuristic search algorithm, recommended material combinations and their correlation links from material components to flame retardant mechanisms and performance parameters are output. This unifies scattered material knowledge, flame retardant mechanisms, and standard constraints into a graph framework that allows for reversible reasoning and quantitative screening. It can quickly eliminate invalid combinations that do not meet grade standards or environmental protection requirements for specific application scenarios, improve the efficiency and accuracy of formulation recommendations, reduce the number of invalid iterations in the formulation trial and error process, and enhance the interpretability and engineering review adaptability of the recommendation results. Attached Figure Description

[0014] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0015] Figure 1 This is a flowchart illustrating the method for recommending fire door flame-retardant core material formulations based on knowledge graphs, as provided in an embodiment of this application. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0017] like Figure 1 As shown, the method for recommending fire-retardant core material formulations for fire doors based on knowledge graphs includes the following steps: Obtain material data of fire door flame-retardant core material, construct entity node system of flame-retardant core material knowledge graph. The entity node system includes material component layer composed of multiple material component nodes, flame-retardant mechanism layer composed of multiple flame-retardant mechanism nodes, performance parameter layer composed of multiple performance parameter nodes, and application scenario layer composed of multiple application scenario nodes. In the physical node system, the correspondence between material composition nodes, flame retardant mechanism nodes and performance parameter nodes is established, and a fire door rating performance index constraint sub-graph is constructed based on the fire door rating standard. Receive formulation development requirements, and based on the application scenarios and fire door ratings in the formulation development requirements, select the corresponding application scenario nodes and the corresponding fire door rating performance index constraint subgraphs in the application scenario layer to determine the target performance parameters. Following the correspondence, we reverse-search from the target performance parameters to determine the target flame retardant mechanism nodes that satisfy the target performance parameters; Candidate material pools are obtained by screening material component nodes in the material component layer. The knowledge graph is represented by a graph embedding model to map material component nodes and flame retardant mechanism nodes into dense continuous vectors of fixed dimensions. The node feature vectors of material component nodes and target flame retardant mechanism nodes in the candidate material pool are extracted. Based on the node feature vectors, the synergy coefficients between material components in the candidate material pool for the target flame retardant mechanism node are calculated. Based on the synergy coefficient, the target material formulation path is determined in the candidate material pool, and the material components corresponding to the path are output as the recommended flame-retardant core material formulation combination.

[0018] This embodiment provides a knowledge graph-based recommendation mechanism for fire-retardant core material formulations for fire doors. Specifically, a unified business scenario is set as follows: A fire door manufacturing company undertakes a project for a large general hospital inpatient building and needs to complete the screening of new formulations for fire-retardant core materials for building fire doors within a preset R&D cycle. The project requires the core material to not only meet the fire door rating indicators, but also take into account low smoke, low thermal conductivity, and environmental constraints, so as to enter the subsequent pilot testing and testing process. Specifically, the system extracts material data from the company's internal formula database, raw material testing reports, combustion test records, and third-party environmental certification materials, and constructs a knowledge graph of flame-retardant core materials based on this data. This graph contains at least four layers: a material composition layer, a flame-retardant mechanism layer, a performance parameter layer, and an application scenario layer. The material composition layer can include nodes such as basic resin, intumescent flame retardant, inorganic filler, and char formation promoter. The flame-retardant mechanism layer can include nodes such as condensed phase char formation, gas phase free radical capture, endothermic decomposition, and cross-linking promotion. The performance parameter layer can include nodes such as oxygen index, vertical burning rating, fire resistance time, smoke density rating, and thermal conductivity. The application scenario layer can include nodes such as building fire doors, industrial fire doors, and residential entrance fire doors. In the graph relationship construction stage, the system establishes a directed relationship between material components and flame retardant mechanisms, and then establishes a correlation between flame retardant mechanisms and performance parameters. For example, material node M1 ammonium polyphosphate can point to mechanism node R1 condensed phase flame retardancy, material node M2 ​​aluminum hydroxide can point to mechanism node R2 endothermic flame retardancy, R1 can be further correlated to performance nodes P1 oxygen index increase and P2 smoke density decrease, and R2 can be correlated to P3 fire resistance time extension and P4 thermal conductivity reduction. Based on this, the system then establishes a performance index constraint subgraph according to the fire door rating standard. This subgraph can be centered on the fire door rating node G1, connecting constraint edges to multiple performance nodes, with threshold attributes on the edges. For ease of explanation, a simplified data extrapolation can be used. Assume the hospital project specifies a fire door scenario with a target rating of G1. Thresholds associated with G1 include: oxygen index not less than 32, fire resistance time not less than 1.5 hours, smoke density not higher than 75, and thermal conductivity not higher than 0.12. Upon receiving this R&D requirement, the system first locks the application scenario node A1 (fire door), then reads the performance constraints associated with G1 to obtain the target performance set {P1, P3, P4, P2}. The system then reverse-engineers the graph to find mechanism nodes that can simultaneously support these performance targets. For example, P1 and P2 can both be traced back to R1 condensed phase flame retardancy, and P3 and P4 can be traced back to R2 endothermic flame retardancy. If a certain performance also corresponds to R3 gas phase flame retardancy, the system retains multiple mechanism candidates instead of just a single path. After obtaining the target mechanism, the system begins to screen the candidate material pool. Assuming there are originally six material nodes in the material component layer, namely M1, M2, M3, M4, M5, and M6, M4 is eliminated due to its unacceptable environmental properties, and M6 is marked as low priority because the variance of its historical thermal stability parameters exceeds a preset fluctuation threshold. The final candidate material pool retains M1, M2, M3, and M5. The system extracts the node feature vectors of these material nodes and the node feature vectors of the target mechanism node from the graph embedding model. The graph embedding model uses either a graph translation embedding model or a graph space sampling and aggregation model. By learning the representation of triples in the knowledge graph, the material component nodes and flame retardant mechanism nodes are mapped into dense, continuous vectors of fixed dimensions, thus obtaining the node feature vectors. For illustration, let the vector of M1 be [0.8, 0.2, 0.6], M2 be [0.7, 0.5, 0.4], M3 be [0.3, 0.9, 0.2], and R1 be [0.9, 0.1, 0.7]. The system does not make decisions directly based on the individual material feature scores, but further calculates the synergy coefficients of the pairwise combinations of materials for the target mechanism. For example, if the synergy coefficient of M1 and M2 on R1 is 0.82, that of M1 and M3 is 0.41, and that of M2 and M3 is 0.56, then the system will tend to place M1 and M2 in the same path to continue expanding. After the synergy coefficient matrix is ​​generated, the system searches for material formulation paths in the candidate material pool. These paths can be understood as starting from a base resin and gradually connecting several functional materials to form a material combination chain that satisfies the target mechanism and performance. For example, path 1 can represent base resin B1 → ammonium polyphosphate M1 → aluminum hydroxide M2 ​​→ char formation accelerator M5; path 2 can represent base resin B1 → vapor phase flame retardant M3 → aluminum hydroxide M2. The calculation rule for the sum of the synergy coefficients of each path is as follows: obtain all material components included in the material combination path, calculate the synergy coefficients between all non-repeating material components for the target flame retardant mechanism node, and sum all the calculated pairwise synergy coefficient values. After comparing the sum of the synergy coefficients, constraint satisfaction, and historical test feasibility of each path, the system outputs the material combination corresponding to the optimal path as the recommended formulation. In supplementary scenarios, if a missing performance threshold exists in the subgraph corresponding to a node in an application scenario, the system marks that performance as needing manual completion and performs initial screening only on the remaining known thresholds to prevent the entire search from being interrupted due to information gaps. If no mechanism combination that can cover all performance targets is found after reverse searching, the system automatically switches to maximum coverage mode, prioritizing the retention of core indicators, such as ensuring fire resistance time and vertical combustion rating first, and then prompting R&D personnel to supplement special performance materials. If the candidate material pool is empty, the system returns a list of reasons for rejection, such as exceeding environmental standards, abnormal thermal stability, or lack of test data, and triggers the data entry process. If the sum of the synergy coefficients of multiple formulation paths is close, for example, the difference is less than the preset threshold of 0.03, the system can output the first two paths in parallel for subsequent experimental verification. In this comprehensive hospital project, the R&D department submitted requirements for fire doors, Class A fire doors, low smoke, halogen-free, and compliance with the hospital's environmental protection acceptance standards. The system identified the building scene and Class A constraint subgraph from the graph, and reverse-engineered to identify the need to consider both condensed phase flame retardancy and endothermic flame retardancy mechanisms. After screening, it retained four candidate components: ammonium polyphosphate, aluminum hydroxide, pentaerythritol, and modified resin. Through collaborative analysis and path search, the system recommended the combination of modified resin + ammonium polyphosphate + aluminum hydroxide + pentaerythritol as the preferred flame retardant core material formulation. The system also simultaneously presented the correlation link between material composition → flame retardant mechanism → performance parameters → grade constraint on the interface for R&D personnel to review its causes. The purpose of this step is to unify the scattered material knowledge, flame retardant mechanisms and standard constraints into a reversible reasoning and quantitative screening framework, so as to achieve rapid formulation recommendation for specific fire door projects and reduce the number of formulation trial and error iterations caused by relying solely on experience-based formulation. In this embodiment, the specific composition of the physical node system includes: a material component layer comprising a base resin node and a flame retardant node, and integrating environmental attribute information such as volatile organic compound content, halogen content, and environmental certification status in the material component node; a flame retardant mechanism layer comprising condensed phase flame retardant nodes, gas phase flame retardant nodes, endothermic flame retardant nodes, and chemical flame retardant nodes; a performance parameter layer comprising oxygen index nodes, vertical combustion rating nodes, fire resistance time nodes, smoke density rating nodes, and thermal conductivity nodes; and an application scenario layer comprising building fire door scenario nodes, industrial fire door scenario nodes, and residential entrance fire door scenario nodes.

[0019] This embodiment provides a detailed modeling mechanism for node systems. Specifically, in the aforementioned general hospital inpatient building project, if only two types of nodes with unrefined material properties are set, although basic retrieval can be completed, the recommended result data granularity is likely to be larger than the preset precision during environmental review and scenario migration. For example, two flame retardants that can improve the oxygen index may be different; one may be suitable for hospital wards, while the other may not be able to enter the site for construction due to its high volatile organic compounds. Specifically, to avoid the aforementioned defects, the system performs a more granular breakdown of the entity node system; the material component layer includes at least a basic resin node and a flame retardant node, and environmental attribute information is attached to each material node; the environmental attributes here are not unstructured text data, but can be converted into structured fields; For example, a basic resin node B1 stores volatile organic compounds (VOCs) of 18 g / L, halogens of 0, and is certified as having passed environmental protection certification; a flame retardant node F2 stores VOCs of 42 g / L, halogens of 0, and is under testing for environmental protection certification; the flame retardant mechanism layer clearly distinguishes between condensed phase flame retardancy, gas phase flame retardancy, endothermic flame retardancy, and chemical flame retardancy, so that the system can identify the differences in the pathways through which the materials function during reverse reasoning; the performance parameter layer includes at least the curing oxygen index, vertical burning rating, fire resistance time, smoke density rating, and thermal conductivity; the application scenario layer maintains three scenarios: building fire doors, industrial fire doors, and residential fire doors, to accommodate subsequent project switching; In a specific implementation scenario, suppose there are three material nodes: B1 is the base resin, F1 is a halogen-free intumescent flame retardant, and F2 is a nitrogen-containing endothermic flame retardant. B1 can form a potential formulation together with F1 and F2, but F1 mainly connects the condensed phase flame retardant node, and F2 mainly connects the endothermic flame retardant node. If the key indicators for a hospital project are fire resistance time and smoke density, the system will pay more attention to the contribution chain of F1 and F2 to the corresponding performance nodes, rather than simply ranking them based on whether they are flame retardants. For example, in the same building scenario, hospital inpatient buildings and ordinary office buildings are both mapped to the building fire door scenario node, but hospital projects usually use the environmental protection attributes in the material nodes more strictly during subsequent screening. This is the significance of embedding environmental protection fields in the node system. As an exception handling mechanism, if a material node lacks environmental certification status, the system will not immediately remove the material, but will mark it as lacking certification, allowing it to enter mechanism analysis but not the final recommended list; if a material corresponds to multiple mechanisms, such as both promoting char formation and having certain endothermic decomposition characteristics, the system allows a material node to be connected to multiple mechanism nodes, without forcing a single classification; if a new performance parameter is not yet included in the existing performance layer, such as a project temporarily adding a hygrothermal cycling resistance index, the system can first attach it as an extended performance node, and then merge it into the standard layer after sufficient data is available. During the advancement of the general hospital project, the R&D team discovered that although a traditional flame retardant had good fire resistance, its environmental certification had not been updated. If it continued to participate in the recommendation, subsequent submissions would face the risk of missing data. The system relied on the certification field in the material node to mark it as a restricted node during the spectrum construction stage, making the recommendation results more in line with the actual access requirements of the hospital engineering scenario. At the same time, another halogen-free intumescent flame retardant received a higher priority in the ranking because it simultaneously connected the two links of condensed phase flame retardancy and smoke density reduction. The purpose of this mechanism is to incorporate the environmental friendliness of materials, differences in mechanisms, and differences in scenarios into knowledge representation through refined entity modeling, thereby achieving a more auditable and engineering-compliant basis for formula recommendation. In this embodiment, in the entity node system, the correspondence between material component nodes, flame retardant mechanism nodes and performance parameter nodes is established. Specifically, this includes: establishing directed edges between material component nodes and flame retardant mechanism nodes to form a many-to-many correspondence between material components and flame retardant mechanisms; and establishing associated edges between flame retardant mechanism nodes and performance parameter nodes to form a correspondence from flame retardant mechanisms to performance parameters.

[0020] This embodiment provides a graph edge relationship construction mechanism; specifically, in a general hospital project, if only the final test results corresponding to the materials are recorded, without recording the intermediate process through which the materials affect which performance, then once a certain performance fails to meet the standard, it is difficult for the R&D personnel to determine whether it is due to material selection deviation, insufficient mechanism coverage, or excessive scenario constraints, thereby reducing the interpretability of the graph. Specifically, the system establishes relationships using a hierarchical directed edge approach. The first type of edge points from material component nodes to flame retardant mechanism nodes, expressing that a material can trigger or enhance a certain type of flame retardant effect. The second type of edge points from flame retardant mechanism nodes to performance parameter nodes, expressing that a certain type of flame retardant effect tends to improve a certain performance. Among these, many-to-many relationships are key. For example, material M1 can simultaneously point to both condensed phase flame retardant node R1 and chemical flame retardant node R4, indicating that the material possesses both types of effects. Mechanism node R1 can further point to oxygen index node P1 and smoke density node P2, indicating that the mechanism simultaneously affects two types of performance. Conversely, a performance node can also be supported by multiple mechanisms. For example, the fire resistance time node P3 can accept edges from both R1 and R2 (endothermic flame retardancy). In specific implementation scenarios, we can assume that the material-mechanism relationship matrix is ​​3×4, with three rows corresponding to M1, M2, and M3, and four columns corresponding to R1, R2, R3, and R4. If the first row and first column of the matrix is ​​1, it means that M1 is connected to R1; if the first row and fourth column is 1, it means that M1 is also connected to R4; if the second row and second column is 1, it means that M2 is connected to R2; and if the rest are 0, it means that no relationship is established yet. The mechanism-performance relationship matrix can be set to 4×5, with four rows corresponding to R1 to R4 and five columns corresponding to P1 to P5. If R1 acts on both P1 and P2, then the first row and first column and the first row and second column are both 1. In this way, when the system searches backward from P3, it can first find R1 and R2, and then trace back to M1 and M2 associated with them, thus forming a clear reverse reasoning chain. As an anomaly handling mechanism, if the mechanism of action of a material is not sufficiently supported by experiments, the system allows the creation of candidate edges instead of formal edges. Candidate edges are only provided for reference during the search phase and do not participate in the final high-confidence ranking. If two mechanisms have opposite effects on the same performance, for example, an additive that can improve fire resistance time may increase smoke density, the direction of influence and confidence can be attached to the edge to prevent the system from treating it as an indiscriminate positive contribution. If a performance node cannot be traced back to any mechanism node for the time being, the system marks the performance as an isolated performance and prompts for supplementary mechanism research data, rather than discarding it without recording it. In the general hospital project, ammonium polyphosphate was connected to two types of nodes: condensed phase flame retardant and chemical flame retardant. Aluminum hydroxide was connected to the endothermic flame retardant node, and modified phosphorus-containing resin was connected to the condensed phase flame retardant node. The system then connected condensed phase flame retardant to oxygen index improvement, smoke density reduction, and fire resistance time extension, and endothermic flame retardant to fire resistance time extension and thermal conductivity reduction. In this way, when the R&D team found that the smoke density of the first batch of samples was too high, they could quickly identify the char-forming components that needed to be enhanced by following the smoke density node → condensed phase flame retardant → specific material. The purpose of this mechanism is to explicitly express the interaction chain between materials, mechanisms and performance, thereby achieving a knowledge organization method that allows for reversible reasoning and traceability. In this embodiment, the fire door rating performance index constraint subgraph is constructed as follows: taking the rating node corresponding to the fire door rating selected in the formula development requirements as the center, a constraint edge with a threshold attribute is established to the corresponding node in the performance parameter layer, and in the subsequent formula path determination process, material combination paths that do not meet the threshold condition are eliminated based on the threshold attribute.

[0021] This embodiment provides a mechanism for constructing a fire door rating constraint subgraph. Specifically, after the aforementioned graph already has a material-mechanism-performance main chain, if rating constraints are not further introduced, the system may recommend multiple combinations of predicted performance parameters that meet the conventional target requirements but do not meet the legal threshold. This still does not have a direct basis for use in strongly compliant scenarios such as hospital projects. Specifically, the system constructs a constraint subgraph around the fire door rating node; each rating node serves as a central node, connected to multiple nodes in the performance parameter layer via constraint edges; threshold attributes are recorded on the constraint edges, which can be either lower or upper limits; for example, the Class A building fire door node GA can be connected to the fire resistance time node with an edge ≥1.5h, to the oxygen index node with an edge ≥32, to the smoke density node with an edge ≤75, and to the thermal conductivity node with an edge ≤0.12; thus, when the system evaluates a material path, it does not only consider whether it can improve performance, but also whether its predicted result crosses the corresponding threshold. In a specific implementation scenario, assume there are two candidate paths: Path 1 predicts an oxygen index of 33, a fire resistance time of 1.6 hours, a smoke density of 72, and a thermal conductivity of 0.11; Path 2 predicts an oxygen index of 35, a fire resistance time of 1.3 hours, a smoke density of 70, and a thermal conductivity of 0.10. Although Path 2 is better than Path 1 in some individual indicators, it is eliminated during the constraint subgraph comparison stage because its fire resistance time is below the 1.5-hour threshold. The system can represent this elimination as follows: when checking the constraint edge from GA to P3, Path 2 does not meet the edge attributes, so the entire path stops being output downstream. As an exception handling mechanism, if the same level differs in different regions or different implementation versions, the system can add a standard version label to the level node, such as version V1 or version V2, and automatically match it by region and contract date when receiving project requirements; if a performance indicator only has an interval suggestion instead of a statutory threshold, the system can set it as a soft constraint and use it for sorting instead of directly triggering the elimination mechanism; if a material path lacks a predicted value for a certain performance, such as no thermal conductivity model result, the system will put it in the list of paths to be tested, neither directly passing nor eliminating it, and make a decision after small-scale testing; In the general hospital project, the company originally planned to use a flame-retardant core material combination that had been used in ordinary office buildings. After the system mapped the combination to the Class A node selected for the hospital project, it was found that its oxygen index and smoke density were qualified, but the fire resistance time was only predicted to be 1.42 hours, so it was directly blocked by the constraint subgraph. Based on this, the R&D personnel abandoned the old formula reuse plan and instead introduced endothermic inorganic flame-retardant components to improve the fire resistance time margin. The purpose of this mechanism is to embed standard thresholds into the recommendation process as pre-constraints, thereby eliminating non-compliant paths at an early stage and reducing the waste of subsequent sample and testing resources. In this embodiment, the candidate material pool is obtained by screening the material component nodes in the material component layer. Specifically, this includes: traversing the material component layer, removing material component nodes whose volatile organic compound content is greater than a preset volatile organic compound content threshold or whose halogen content is greater than zero based on environmental attribute information, and using the remaining material component nodes as the candidate material pool.

[0022] This embodiment provides a candidate material pool screening mechanism based on environmental protection attributes. Specifically, after a general hospital project enters the substantive screening stage, if all materials with flame retardant potential are still sent to the collaborative calculation, two types of problems are likely to occur: First, computing resources are wasted on obviously non-compliant materials; second, even if a high-performance solution is finally recommended, it may not be able to enter the hospital project's procurement list due to excessive halogen content or volatile organic compounds. Specifically, the system sets up an environmental pre-screening step before the candidate material pool is formed; when traversing the material components and layers, it reads the volatile organic compound content, halogen content and certification status recorded in each node, and eliminates them according to the project's preset environmental threshold; a simple threshold rule can be used for deduction: if the volatile organic compound content of a material is greater than or equal to the preset upper limit threshold, such as 30g / L, it is judged as exceeding the standard and is eliminated; if it is less than the threshold and the halogen content is equal to zero, it is judged as qualified and is retained; For example, M1 has 18 g / L of volatile organic compounds and is halogen-free, so it is retained; M2 has 36 g / L of volatile organic compounds, so it is removed; M3 has 12 g / L of volatile organic compounds but halogens are detected, so it is removed; M4 has 22 g / L of volatile organic compounds and is halogen-free, so it is retained; after traversal, only M1 and M4 from the original 4 material nodes enter the candidate material pool. The screening here is not limited to a single static judgment; if the company maintains multiple batches of test records, the system can prioritize the most recent valid test value; if the same material has different attributes from different suppliers, the material name + supplier code can be used together as a node identifier to avoid mixing materials from different sources; in this way, the subsequent collaborative analysis will face a limited set that has already passed the initial screening of environmental thresholds. As an exception handling mechanism, if a material lacks volatile organic compound or halogen test data, the system will not default it to qualified, but will place it in the verification pool. When the number of candidate materials in the pool is too small, for example, if there are less than two flame retardants after elimination, the system can prompt the R&D personnel to relax non-core secondary conditions, such as allowing materials that have been submitted for certification but have passed actual testing to participate in the trial calculation. However, for materials that are clearly halogen-containing or obviously exceed the standard, the system should not relax the requirements. If the project itself allows the use of certain halogen-containing materials in specific industrial scenarios, the screening template can be switched when selecting the application scenario, and different environmental protection thresholds can be used for different scenarios. In the general hospital project, the R&D team originally considered a traditional high-efficiency flame retardant, but its halogen content was not zero. The system eliminated it during the environmental pre-screening stage, so subsequent collaborative calculations only focused on halogen-free materials. Another flame retardant with better performance has not yet completed third-party certification, but its internal measured volatile organic compounds meet the standards and it is halogen-free. The system put it into the verification pool and marked it separately on the interface, so that the R&D personnel can decide whether to verify it simultaneously in the pilot stage. The purpose of this step is to filter out materials that clearly do not meet the hospital's engineering access requirements in advance, thereby achieving candidate set convergence and improving the effectiveness of subsequent collaborative computing and formula search. In this embodiment, the synergy coefficients among the material components in the candidate material pool for the target flame retardant mechanism node are calculated according to the following formula: In the formula, Indicates material composition and material components For the target flame retardant mechanism The coefficient of synergy; , These represent the index numbers of the material components in the candidate material pool; Index number indicating the target flame-retardant mechanism; and These represent the material components extracted from the knowledge graph. and material components The node feature vector; assuming the extracted node feature vector has a dimension of d, then the result obtained after vector concatenation is... The dimension is 2d, and correspondingly, the action coefficient matrix The dimension is set to 1×2d to ensure that the result of linear computation between the two is a one-dimensional scalar; This represents a vector concatenation operation; Represents matrix multiplication operation; Indicates the target flame retardant mechanism The coefficient matrix of action; Indicates the target flame retardant mechanism The one-dimensional scalar correction term; This represents the Sigmoid activation function; This represents the physical-chemical compatibility parameter.

[0023] This embodiment provides a material synergy coefficient quantification calculation mechanism. Specifically, after the aforementioned pre-screening, if the system still only sorts the materials based on the correlation strength between individual materials and the target mechanism, it is easy to overlook situations where a single material is effective but the mixture fails or a single material is ordinary but the combination is synergistic. For hospital projects, which require a high hit rate in a single trial, the individual score is insufficient to support the actual formulation decision. Therefore, it is necessary to explicitly quantify the synergistic relationship between material pairs. Specifically, the system extracts the feature vector of each candidate material node from the graph embedding model, and then calculates the pairwise synergy coefficient of materials for each target flame retardant mechanism node; here, materials With materials The vectors are first concatenated to form a joint representation, and then combined with the representation targeting the mechanism. The action coefficients obtained from training are linearly calculated, and a correction term is added. After activation, the basic synergy coefficient is obtained, which is then multiplied by the materialization compatibility parameter. The result of this process is that the synergy coefficient not only reflects the semantic proximity of the two in the knowledge graph, but also takes into account whether the two are suitable for joint use in terms of actual materialization attributes. To illustrate the model calculation process, let's assume that for the condensed phase flame retardant mechanism R1, the nodal eigenvectors of material M1 are [0.8, 0.2], and those of material M2 are [0.6, 0.5]. Concatenating these two vectors yields [0.8, 0.2, 0.6, 0.5]. If the corresponding mechanism's coefficient can be simplified to [0.9, 0.4, 0.7, 0.3], and the correction term is -0.2, then the linear calculation result is approximately 0.8 × 0.9 + 0.2. ×0.4+0.6×0.7+0.5×0.3-0.2=1.17; After activation and compression, the basic synergy coefficient is approximately 0.76; If the physical compatibility parameter of both is 0.90, then the final synergy coefficient is approximately 0.684; Calculating the results for M1 and M3, if the basic synergy coefficient is 0.72, but the physical compatibility parameter is only 0.40, then the final synergy coefficient drops to 0.288; Based on this, the system tends to prioritize the combination of M1 and M2; This calculation mechanism can also be extended to situations where multiple target mechanisms exist in parallel; for example, a hospital project requires both condensed phase flame retardancy and endothermic flame retardancy, so the same pair of materials can obtain two synergy coefficients for R1 and R2 respectively, and the system then summarizes them according to the weight of different mechanisms in the project; if fire resistance time is the primary indicator, the synergy coefficient corresponding to endothermic flame retardancy can be given a higher weight; if smoke density control is more critical, the synergy coefficient corresponding to condensed phase flame retardancy can increase its influence ratio. As an anomaly handling mechanism, if a candidate material node lacks a stable feature vector, for example, if the material is a newly added component with few associated edges, the system can first use the average vector of similar materials as a temporary filling value and mark it as low confidence; if the synergy coefficient of a material pair is greater than the preset upper limit threshold for synergy but its historical test count is lower than the preset lower limit, the system can introduce a minimum sample threshold and add a prompt of insufficient validation on the sorting interface; if the synergy effect of a pair of materials is positive in one mechanism, but the synergy coefficient calculated in another key mechanism is lower than the preset lower limit threshold for synergy, the system should not only retain its highest single synergy coefficient, but should perform balanced screening according to the comprehensive objective; In the general hospital project, the system calculated the pairwise combinations of halogen-free intumescent flame retardant, aluminum hydroxide, pentaerythritol, and modified resin. The results showed that the intumescent flame retardant and pentaerythritol had a high synergistic coefficient in condensed phase flame retardancy, while aluminum hydroxide and modified resin had a moderate synergistic coefficient in the endothermic flame retardant path but good physicochemical compatibility. Therefore, the system did not select the material with the highest single synergistic coefficient, but instead prioritized the combination that could jointly form a stable synergistic chain. The purpose of this mechanism is to quantify the synergistic relationship of material combinations into comparable values, thereby realizing the shift from single material selection to combined path selection. In this embodiment, the physicochemical compatibility parameter is determined based on the thermal decomposition temperature matching degree, and its calculation formula is as follows: In the formula, Indicates material composition The initial thermal decomposition temperature; Indicates material composition The initial thermal decomposition temperature; This represents the temperature-sensitive adjustment coefficient, whose dimension is the reciprocal of temperature, to ensure that the result after multiplying it by the temperature difference is dimensionless. This represents the absolute value of the temperature difference between the two values; Represented by natural constant An exponential function with base 0.

[0024] This embodiment provides a mechanism for determining physicochemical compatibility parameters. Specifically, in the aforementioned synergy coefficient calculation, if the semantic synergy between materials is judged only based on the spectral features without considering whether the thermal decomposition behavior is matched, there may be a problem where the materials appear to be complementary on the spectral map but the thermal decomposition reaction stages are mismatched during actual heating. For example, one material decomposes earlier at a lower temperature, while another material only begins to function at a higher temperature, and the two cannot form an effective linkage in the same combustion stage. Specifically, to address the aforementioned shortcomings, the system employs thermal decomposition temperature matching to characterize physicochemical compatibility parameters. The core idea is that the closer the initial thermal decomposition temperatures of two materials are, the more likely they are to function simultaneously or sequentially during firing, resulting in higher compatibility parameters; conversely, the greater the temperature difference, the lower the compatibility parameters. This can be illustrated through specific numerical calculation examples. Assuming a temperature sensitivity adjustment coefficient of 0.02, an initial thermal decomposition temperature of 290℃ for material M1, and 300℃ for M2, then with a temperature difference of 10℃, the compatibility parameters are approximately... The result is approximately 0.82; if the initial thermal decomposition temperature of M3 is 380℃, then the temperature difference between M1 and M3 is 90℃, and the compatibility parameter is approximately... The result is approximately 0.17; it can be seen that even if M1 and M3 have some complementarity in the semantics of the graph, they will be significantly reduced in weight due to the large difference in the thermal decomposition time sequence. Furthermore, this parameter has direct engineering significance in hospital projects; flame-retardant core materials often undergo continuous stages such as heating, decomposition, charring, and heat insulation under the action of flame; if the decomposition windows of the two types of materials are similar, the products released by one type of material are more easily utilized by the other type of material, forming synergistic effects such as thickening of char, delayed heat absorption, or enhanced smoke suppression; conversely, if one type of material decomposes and is exhausted too early, and the other type of material takes effect later, the combined technology will be less effective than expected. As an anomaly handling mechanism, if there are multiple test values ​​for the initial thermal decomposition temperature of a material, the system can use the median or the stable batch average to reduce the impact of accidental measurement deviations. If the material is a compound masterbatch and a single decomposition starting point cannot be obtained directly, the main peak starting temperature or a representative value in the segmented temperature window can be used. If the difference between the initial thermal decomposition temperatures of two materials is greater than the preset temperature matching threshold, but historical tests have confirmed that there is still a significant synergistic effect under a specific ratio, the system can allow manual input of compensation coefficients to correct the compatibility parameters, but the source of the correction must be retained for auditing. In the general hospital project, the R&D team initially favored the combination of a certain intumescent flame retardant and a certain mineral filler, as they correspond to the char formation and endothermic pathways, respectively. However, after reading the thermogravimetric analysis results, the team found that the initial thermal decomposition temperatures of the two differed by nearly 100°C, and the compatibility parameters were low, resulting in a significant decrease in the overall synergy score. Conversely, another halogen-free flame retardant had a decomposition temperature range that was closer to that of pentaerythritol, and obtained higher compatibility parameters in the model. Therefore, it was given priority in the recommended pathway. The purpose of this mechanism is to introduce the actual thermal response matching into collaborative computation, thereby achieving joint constraints between the semantic relationships of the knowledge graph and the thermochemical behavior of materials. In this embodiment, the action coefficient matrix and corrections The parameter update method is as follows: Obtain the combustion test database of historical formulas, extract the performance increment data before and after combustion for each historical formula from the combustion test database; construct a loss function with the difference between the actual performance increment and the predicted performance increment as the objective; specifically, the loss function adopts the mean square error function, and the prediction residual is quantified by calculating the sum of the squares of the differences between the actual performance increment and the predicted performance increment of each performance parameter. The gradient descent method is then used to update the coefficient matrix. and corrections Update the parameters.

[0025] This embodiment provides a collaborative model parameter update mechanism. Specifically, prior to the hospital project, the company had accumulated combustion test results for multiple office buildings, industrial plants, and residential door core formulations. If the action coefficient matrix and correction terms remain fixed for a long period, the model may not be able to reflect the true effects of adding new materials, process fluctuations, or changes in standard requirements, thus causing the recommended results to gradually deviate from actual production. Specifically, the system establishes a historical formula combustion test database, recording the performance changes of each historical formula before and after combustion tests. For example, for a certain formula, the oxygen index can be recorded as 24 before the addition of the flame retardant system and 33 after the addition, resulting in a performance increase of 9; the smoke density decreases from 88 to 71, resulting in a performance increase of -17; the fire resistance time increases from 1.0h to 1.55h, resulting in an increase of 0.55h. The system maps the material combinations in historical formulas to a spectrum and uses the current parameters to predict the corresponding performance increase. Specifically, the calculation logic for predicting performance increment is as follows: The system extracts the basic performance contribution value of each material component in the historical formula when acting alone, and calculates the synergy coefficient between each material component for the relevant flame retardant mechanism based on the current action coefficient matrix and correction term. The preset transfer weight is set. The preset transfer weight is a constant value on the associated edge, which is pre-statistically calculated and weighted based on the average contribution of the corresponding single flame retardant mechanism to the specific performance parameter in past real tests in the historical formula combustion test database. The synergy coefficient is multiplied by the preset transfer weight on the edge connecting the flame retardant mechanism to the corresponding performance parameter node to obtain the synergy gain term of the multi-material combination. The basic performance contribution value of each material is linearly added to the synergy gain term to output the predicted performance increment corresponding to the historical formulation combination. After obtaining the above prediction results, the system uses the difference between the predicted increment and the actual increment as the source of error to construct a loss function. Then, it uses the gradient descent method to backpropagate and iteratively update the action coefficient matrix and correction term for each mechanism to minimize the prediction residual. In a specific implementation scenario, assuming a historical formula H1 consists of M1 and M2, the system predicts an oxygen index increment of 7, while the actual test result is 9, resulting in an error of 2; the predicted fire resistance time increment is 0.40h, while the actual result is 0.55h, resulting in an error of 0.15h. The system weights and sums multiple errors to form the total loss. For example, if a hospital project places greater emphasis on fire resistance time and smoke density, these two properties can be given higher weights during training. After an update, if the role of M1 and M2 in condensed phase flame retardancy is underestimated, the weights of the corresponding action coefficient matrix will be adjusted to improve the predictive synergy coefficient of this combination, thereby enhancing the final predictive performance increment. Through multiple iterations, the model parameters gradually approach the actual experimental patterns of enterprises. The parameter update can also be done in batches; for example, the historical database can be updated incrementally with new samples added quarterly to avoid training from scratch every time; or sub-models can be built according to scenarios, and when there is sufficient data on fire doors in buildings, the mechanism parameters related to the building scenario can be updated first, so that the recommendations for hospital projects are closer to the experience of similar projects. As an anomaly handling mechanism, if some records in the historical database have missing values, such as missing smoke density results, the record can still be used to update parameters for other performance items without having to be discarded entirely. If a batch of historical data is suspected of being abnormal, such as the same formula showing significant differences in results between two tests, the system can first screen for abnormal samples and reduce the weight of records that exceed the statistical range. If the number of new samples is insufficient, the system can retain the parameters from the previous round to avoid model oscillations caused by a small number of occasional data. Before the general hospital project was launched, the company imported the combustion test data of 20 sets of halogen-free flame retardant core materials from the past year into the update process. After the update, the system found that the actual contribution of a certain type of endothermic flame retardant to the fire resistance time was higher than the old model prediction for a long time. As a result, the coefficient of the relevant mechanism was adjusted upward. As a result, when the hospital project implemented the recommendation again, the ranking of this type of material in the candidate path moved forward, which ultimately helped the R&D team to lock in the solution that met the Class A requirements more quickly. The purpose of this mechanism is to continuously calibrate the collaborative model by utilizing feedback from real-world enterprise experiments, bridging the computational gap between purely theoretical collaborative coefficients and actual engineering increments, thereby enabling the recommendation results to dynamically improve accuracy as historical experience accumulates. In this embodiment, the target material formulation path is determined in the candidate material pool based on the synergy coefficient. Specifically, this includes using a heuristic search algorithm to find the combination of material components in the candidate material pool that meets the target flame retardant mechanism node requirements and has the largest sum of synergy coefficients, which is then used as the target material formulation path.

[0026] This embodiment provides a target material formulation path search mechanism. Specifically, after completing the pairwise collaborative calculation of materials, if several highest-scoring materials are simply selected for splicing, problems such as internal combination conflicts, incomplete mechanism coverage, or excessive total amount may still occur. Therefore, under the premise of meeting the target mechanism requirements, it is necessary to search for an overall optimal combination path from the candidate material pool. Specifically, the system can use a heuristic search algorithm to solve the path; the search starting point is usually the basic resin node, and the expansion node is selected according to the coverage gain of the target mechanism after the material is added and the synergistic benefit of the newly added edge; for example, the target mechanisms locked in the hospital project include condensed phase flame retardant R1 and endothermic flame retardant R2; in the candidate material pool, the basic resin B1 must be selected; M1 mainly supports R1, M2 mainly supports R2, and M5 can both assist R1 and improve smoke density; During the system search, the synergy coefficient value of a single edge is comprehensively evaluated, as well as whether the combination simultaneously covers the target flame-retardant mechanism and meets the fire door rating constraints. For example: the current total synergy coefficient of B1+M1 is 0.62; the current total synergy coefficient of B1+M2 is 0.58; after adding M5 to B1+M1, the total synergy coefficient increases to 1.18 and covers R1; after adding M2, the total synergy coefficient increases to 1.76 and simultaneously covers R1 and R2; another path, B1+M3+M2, although having a total synergy coefficient of 1.70, has a weaker improvement on smoke density; therefore, the system prioritizes the former path. In heuristic search, an evaluation function can be set, such as the sum of the current synergy coefficients plus the maximum potential synergy coefficient increment of the remaining mechanisms. In this way, when a certain intermediate path has a low current synergy coefficient but still has a high probability of supplementing the key mechanism through subsequent materials, the system will not prune it prematurely. For a candidate pool with a large number of materials, this method can reduce the computational complexity of exhaustive search. As a supplementary scenario, if the sum of the synergy coefficients of multiple paths is close during the search process, the path length, material acquisition difficulty, supply stability, or historical test success rate can be further compared. If all paths cannot simultaneously satisfy all target mechanisms, the system can degenerate into a core mechanism priority search, for example, first ensuring the flame retardancy of the condensed phase, and then increasing the endothermic flame retardancy contribution within feasible limits. If a material has a high synergy coefficient, but it leads to excessive repetition of similar components in the formulation, the system can set category proportion constraints to prevent path imbalance. In the general hospital project, the system searched for five candidate components: modified resin, halogen-free intumescent flame retardant, aluminum hydroxide, pentaerythritol, and smoke suppressant filler. Ultimately, it was found that the path of modified resin → halogen-free intumescent flame retardant → pentaerythritol → aluminum hydroxide could cover both condensed phase flame retardancy and endothermic flame retardancy, and the total synergy coefficient was higher than that of the alternative path of introducing smoke suppressant filler. It also met the fire resistance time threshold of Class A doors, so it was determined as the target material formulation path. The purpose of this mechanism is to further screen out the globally better combination structure from the material relationships with high local scores, so as to realize the determination of a feasible formulation path. In this embodiment, the material components corresponding to the path are output as recommended flame-retardant core material formulation combinations. Specifically, this includes: outputting the recommended flame-retardant core material formulation combinations corresponding to the target material formulation path, and simultaneously displaying the correspondence between the recommended flame-retardant core material formulation combinations and the performance parameter nodes in the flame-retardant core material knowledge graph.

[0027] This embodiment provides a recommendation result output and link display mechanism; specifically, in a comprehensive hospital project, R&D personnel do not only need a list of recommended materials, but also need to know why the system recommends the combination, what mechanism it relies on to meet which performance requirements, and how it crosses the level threshold; if there is a lack of interpretable display, the recommendation results are not easy to be adopted in R&D review, procurement communication, and testing decisions; Specifically, when outputting the target material formulation path, the system simultaneously generates a graph-based link view. The output includes at least three parts: First, the recommended material combination itself, such as base resin, main flame retardant, synergistic flame retardant, and auxiliary filler; Second, path scoring information, such as comprehensive synergy value, key performance prediction value, and level constraint satisfaction status; Third, the link display, which is the complete chain starting from each material node, passing through the corresponding flame retardant mechanism node, reaching the performance parameter node, and then mapping to the project level node. When using simplified data examples, if the recommended result is B1+M1+M2+M5, the system can display the following link logic: M1 and M5 both point to the condensed phase flame retardant node, and M2 points to the endothermic flame retardant node; the condensed phase flame retardant node connects to the increase in oxygen index and the decrease in smoke density, and the endothermic flame retardant node connects to the extension of fire resistance time and the decrease in thermal conductivity; these performance nodes are then compared one by one with the thresholds on the Class A building fire door rating nodes, and it is shown that they have been met; on the interface, the link can be presented using a highlighted or layered view, so that researchers can intuitively see the role of each component in the entire recommended path; Furthermore, the output results can also include alternative solutions; if the scores of the first recommended path and the second recommended path are similar, the system can simultaneously display the differences between the two, for example, the first solution is better in terms of fire resistance time, while the second solution is better in terms of acquisition difficulty or smoke density; in this way, the R&D department can make a secondary decision based on the subsequent process conditions and material supply of the hospital project. As a supplementary scenario, if a performance prediction value in a recommended path still comes from a low-confidence model, the system will mark the node as a risk in the link display and remind users to prioritize targeted testing; if there are supply chain fluctuations in the recommended materials, the system can automatically switch to the suboptimal alternative path and retain the reason for the replacement; if the user only has production position permissions and no model viewing permissions, the system can hide parameter details, but still retain the basic correlation link of material-mechanism-performance, taking into account both interpretability and data security; At the technical review meeting for the comprehensive hospital project, the system recommended a preferred combination of modified resin, halogen-free intumescent flame retardant, pentaerythritol, and aluminum hydroxide. It was simultaneously demonstrated that the halogen-free intumescent flame retardant and pentaerythritol together enhance the flame retardancy of the condensed phase, thereby increasing the oxygen index and reducing smoke density; aluminum hydroxide contributes endothermic flame retardancy, thereby increasing fire resistance time and helping to reduce thermal conductivity; all key performance indicators have exceeded the thresholds corresponding to Class A building fire doors; based on this, researchers can directly determine the sample preparation sequence without needing to manually disassemble the recommendation rationale. The purpose of this mechanism is to transform the recommendation results from an uninterpretable output list into reviewable conclusions with causal chains, thereby enabling information flow between the research and development, review, and testing stages.

[0028] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for recommending formulations of flame-retardant core materials for fire doors based on knowledge graphs, characterized in that, Includes the following steps: Obtain material data of fire door flame-retardant core material, construct entity node system of flame-retardant core material knowledge graph. The entity node system includes material component layer composed of multiple material component nodes, flame-retardant mechanism layer composed of multiple flame-retardant mechanism nodes, performance parameter layer composed of multiple performance parameter nodes, and application scenario layer composed of multiple application scenario nodes. In the physical node system, the correspondence between material composition nodes, flame retardant mechanism nodes and performance parameter nodes is established, and a fire door rating performance index constraint sub-graph is constructed based on the fire door rating standard. Receive formulation development requirements, and based on the application scenarios and fire door ratings in the formulation development requirements, select the corresponding application scenario nodes and the corresponding fire door rating performance index constraint subgraphs in the application scenario layer to determine the target performance parameters. Following the correspondence, we reverse-search from the target performance parameters to determine the target flame retardant mechanism nodes that satisfy the target performance parameters; Candidate material pools are obtained by screening material component nodes in the material component layer. The knowledge graph is represented by a graph embedding model to map material component nodes and flame retardant mechanism nodes into dense continuous vectors of fixed dimensions. The node feature vectors of material component nodes and target flame retardant mechanism nodes in the candidate material pool are extracted. Based on the node feature vectors, the synergy coefficients between material components in the candidate material pool for the target flame retardant mechanism node are calculated. Based on the synergy coefficient, the target material formulation path is determined in the candidate material pool, and the material components corresponding to the path are output as the recommended flame-retardant core material formulation combination.

2. The method for recommending fire door flame-retardant core material formulations based on knowledge graphs according to claim 1, characterized in that, The specific composition of the physical node system includes: the material component layer includes a basic resin node and a flame retardant node, and integrates environmental attribute information such as volatile organic compound content, halogen content, and environmental certification status in the material component node; the flame retardant mechanism layer includes condensed phase flame retardant nodes, gas phase flame retardant nodes, endothermic flame retardant nodes, and chemical flame retardant nodes; the performance parameter layer includes oxygen index nodes, vertical burning rating nodes, fire resistance time nodes, smoke density rating nodes, and thermal conductivity nodes; the application scenario layer includes building fire door scenario nodes, industrial fire door scenario nodes, and residential entrance fire door scenario nodes.

3. The method for recommending fire door flame-retardant core material formulations based on knowledge graphs according to claim 1, characterized in that, In the aforementioned entity node system, a correspondence is established between material component nodes, flame retardant mechanism nodes, and performance parameter nodes. Specifically, this includes: establishing directed edges between material component nodes and flame retardant mechanism nodes to form a many-to-many correspondence from material component to flame retardant mechanism; and establishing associated edges from flame retardant mechanism nodes to performance parameter nodes to form a correspondence from flame retardant mechanism to performance parameter.

4. The method for recommending fire door flame-retardant core material formulations based on knowledge graphs according to claim 1, characterized in that, The construction method of the fire door rating performance index constraint subgraph is as follows: taking the rating node corresponding to the fire door rating selected in the formula development requirements as the center, establish constraint edges with threshold attributes to the corresponding nodes in the performance parameter layer, and in the subsequent formula path determination process, eliminate material combination paths whose associated performance parameters do not meet the threshold conditions based on the threshold attribute.

5. The method for recommending fire door flame-retardant core material formulations based on knowledge graphs according to claim 2, characterized in that, The process of filtering material component nodes in the material component layer to obtain a candidate material pool specifically includes: traversing the material component layer, and based on the environmental attribute information, removing material component nodes whose volatile organic compound content is greater than a preset volatile organic compound content threshold or whose halogen content is greater than zero, and using the remaining material component nodes as the candidate material pool.

6. The method for recommending fire door flame-retardant core material formulations based on knowledge graphs according to claim 1, characterized in that, The synergy coefficients among the material components in the candidate material pool for the target flame-retardant mechanism node are calculated using the following formula: In the formula, Indicates material composition and material components For the target flame retardant mechanism The coefficient of synergy; , These represent the index numbers of the material components in the candidate material pool; Index number indicating the target flame-retardant mechanism; and These represent the material components extracted from the knowledge graph. and material components The node feature vector; assuming the extracted node feature vector has a dimension of d, then the result obtained after vector concatenation is... The dimension is 2d, and correspondingly, the action coefficient matrix The dimension is set to 1×2d to ensure that the result of linear computation between the two is a one-dimensional scalar; This represents a vector concatenation operation; Represents matrix multiplication operation; Indicates the target flame retardant mechanism The coefficient matrix of action; Indicates the target flame retardant mechanism The one-dimensional scalar correction term; This represents the Sigmoid activation function; This represents the physical-chemical compatibility parameter.

7. The method for recommending fire door flame-retardant core material formulations based on knowledge graphs according to claim 6, characterized in that, The physicochemical compatibility parameters are determined based on the thermal decomposition temperature matching degree, and their calculation formula is as follows: In the formula, Indicates material composition The initial thermal decomposition temperature; Indicates material composition The initial thermal decomposition temperature; This represents the temperature-sensitive adjustment coefficient, whose dimension is the reciprocal of temperature, to ensure that the result after multiplying it by the temperature difference is dimensionless. This represents the absolute value of the temperature difference between the two values; Represented by natural constant An exponential function with base 0.

8. The method for recommending fire door flame-retardant core material formulations based on knowledge graphs according to claim 6, characterized in that, The action coefficient matrix and corrections The parameter update method is as follows: Obtain the combustion test database of historical formulas, extract the performance increment data before and after combustion for each historical formula in the combustion test database; construct a loss function with the difference between the actual performance increment and the predicted performance increment as the objective; specifically, the loss function adopts the mean square error function, and the prediction residual is quantified by calculating the sum of the squares of the differences between the actual performance increment and the predicted performance increment of each performance parameter. The gradient descent method is then used to update the coefficient matrix. and corrections Update the parameters.

9. The method for recommending fire door flame-retardant core material formulations based on knowledge graphs according to claim 1, characterized in that, Based on the synergy coefficient, the target material formulation path is determined in the candidate material pool, specifically by using a heuristic search algorithm to find the combination of material components in the candidate material pool that meets the target flame retardant mechanism node requirements and has the largest sum of synergy coefficients, as the target material formulation path.

10. The method for recommending fire door flame-retardant core material formulations based on knowledge graphs according to claim 1, characterized in that, The material components corresponding to the path are output as recommended flame-retardant core material formulation combinations. Specifically, this includes: outputting the recommended flame-retardant core material formulation combinations corresponding to the target material formulation path, and simultaneously displaying the correspondence between the recommended flame-retardant core material formulation combinations and the performance parameter nodes in the flame-retardant core material knowledge graph.

Images