Material quality detection method and device fusing knowledge graph and numerical simulation

Abstract

The invention discloses a knowledge graph and numerical simulation fused material quality detection method and device, and the method comprises the steps: obtaining defect elimination detection text data, carrying out the entity recognition based on an entity recognition model of fusion language preprocessing, and constructing a defect elimination knowledge graph; and carrying out detection item search from the defect elimination knowledge graph based on the target material type to obtain a target detection item, carrying out numerical simulation on the target material, giving and updating an initial damage condition of the target material, and giving an accumulated damage condition of the target material so as to complete quality detection of the target material. Through analysis of defect elimination detection text data, a defect elimination knowledge graph is constructed, a relationship among different material detection data is known, and based on the relationship, a target material is analyzed to obtain an accumulated damage condition of the target material, and quality detection of the target material is completed. And manpower, material resources and time required in the quality detection process are reduced, and meanwhile, the working efficiency of material quality detection is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of material quality detection, and specifically relates to a material quality detection method and device integrating knowledge graph and numerical simulation. Background Art

[0002] As an important control method for power material management, power grid material quality inspection connects the material production site and the project construction site. It is related to the safety and smooth operation of the power grid project and is of great significance to power construction, power grid safety, operation and maintenance.

[0003] With the development of the economy and society, the demand for electricity continues to increase, the power grid continues to grow, and the supply of materials continues to increase. The corresponding workload of material testing is also increasing year by year. The existing manpower and material resources are difficult to meet the current testing work needs. Therefore, it is urgent to improve the efficiency and quality of material testing.

[0004] Patent CN118798710A discloses a material quality detection method and device based on semantic tags of power material ledgers, the method includes: obtaining power material ledger data, preprocessing the power material ledger data, and giving material ledger preprocessing data; determining evaluation indicators, and screening the material ledger preprocessing data to give target data of each material; normalizing the target data to form data to be evaluated; combining preset evaluation rules, evaluating the data to be evaluated, and giving material quality detection results. By integrating semantic tags when preprocessing power material ledger data, considering the additional information contained in the semantic tags, more contextual information can be provided, and the information that is usually ignored in numerical analysis can be fully utilized, which helps to better understand the internal structure of the data, and can also help identify and correct deviations, thereby reducing the impact on the final analysis results. Although it realizes the quality detection of materials through the analysis of data, and reduces the workload of technical personnel for data analysis to a certain extent, it still requires staff to inspect each device to obtain material ledger data, which has limited improvement in efficiency.

[0005] How to improve the efficiency and quality of material testing to meet the needs of testing work is a problem that needs to be solved at present. Summary of the invention

[0006] In view of the defects existing in the above-mentioned prior art, the present invention provides a material quality detection method and device integrating knowledge graph and numerical simulation, the method comprising: obtaining defect detection text data and performing entity recognition based on the entity recognition model of fusion language preprocessing, and constructing a defect elimination knowledge graph; searching for detection items from the defect elimination knowledge graph based on the target material type, obtaining the target detection items, and performing numerical simulation on the target material, giving the initial damage of the target material and updating it, and giving the cumulative damage of the target material to complete the quality detection of the target material. By analyzing the defect elimination text data, a defect elimination knowledge graph is constructed, the relationship between different material detection data is understood, and based on this, the target material is analyzed to obtain the cumulative damage of the target material, and the quality detection of the target material is completed, which reduces the manpower, material resources and time required in the quality detection process while improving the work efficiency of material quality detection.

[0007] In a first aspect, the present invention provides a material quality detection method integrating knowledge graph and numerical simulation, which specifically comprises the following steps: Acquire defect elimination and detection text data, wherein the defect elimination and detection text data includes detection text data of various types of materials; Based on the entity recognition model that integrates language preprocessing, entity recognition is performed on the defect detection text data and a defect elimination knowledge graph is constructed; Based on the target material type, the inspection items are searched from the defect elimination knowledge graph to obtain the target inspection items; Based on the target detection project, numerical simulation is performed on the target material to give the initial damage of the target material; Based on the initial damage situation, the initial damage situation of the target material is updated, and the cumulative damage situation of the target material is given to complete the quality inspection of the target material.

[0008] Furthermore, based on the entity recognition model of fusion language preprocessing, entity recognition is performed on the defect detection text data and a defect elimination knowledge graph is constructed, which specifically includes: The entity recognition model of fusion language preprocessing is used to perform entity recognition on the defect detection text data to obtain multiple entities and the attributes of each entity; Using the spectral clustering algorithm, multiple entities are divided according to the attributes of each entity to obtain multiple entity subsets; Based on the attributes of each entity in each entity subset, analyze the subset attributes of each entity subset and give the subset attribute value; According to the subset attribute values ​​of each entity subset, the entity subsets are fused to obtain the missing knowledge graph.

[0009] Furthermore, an entity recognition model fused with language preprocessing is used to perform entity recognition on the defect detection text data to obtain multiple entities and the attributes of each entity, including: Segment the defect detection text data after text cleaning to obtain a first entity set; Based on the stop word set, the first entity set is screened, the stop words in the first entity set are removed, and a second entity set is obtained; Filling or truncating each text sequence in the second entity set to obtain a sequence vector set, wherein the lengths of each sequence vector in the sequence vector set are the same; The attention mechanism layer is used to process each sequence vector in the sequence vector set, analyze the correlation between each sequence vector, and obtain multiple entities and the attributes of each entity.

[0010] Furthermore, based on the attributes of each entity in each entity subset, the subset attributes of each entity subset are analyzed, and subset attribute values ​​are given, including: Based on the attributes of each entity in each entity subset, the embedding vector of each entity is obtained; According to the embedding vector of each entity, the norm of the embedding vector is given; For each entity subset, the mean of the norms of the embedding vectors corresponding to each entity in the entity subset is taken as the subset attribute value of the corresponding entity subset.

[0011] Furthermore, based on the target material type, the inspection items are searched from the defect elimination knowledge graph to obtain the target inspection items, including: Based on the target material type, search from the deficiency elimination knowledge graph to obtain the entity subset corresponding to the target material type; Based on the entity subset corresponding to the target material type, the historical occurrence frequency of each entity in the entity subset is analyzed, and the priority of each entity in the entity subset is given based on the mapping relationship between the defective parts of the material and the inspection items; Determine the target detection items based on the priority of each entity.

[0012] Furthermore, based on the entity subset corresponding to the target material type, the historical occurrence frequency of each entity in the entity subset is analyzed, and the priority of each entity in the entity subset is given in combination with the mapping relationship between the defective parts of the material and the inspection items, including: Based on the entity subset corresponding to the target material type, the historical occurrence frequency of each entity in the entity subset is obtained; Based on the historical occurrence frequency of each entity in the entity subset, the state occurrence frequency of different entities corresponding to different states is given; Based on the state occurrence frequencies of different entities corresponding to different states, a state transition frequency matrix is ​​constructed, where the elements in the state transition frequency matrix are the number of occurrences of the target material from one state to another; Analyze the state transition frequency matrix and give the state transition probability matrix; Based on the initial state analysis and combined with the state transition probability matrix, the future state distribution is obtained; Prioritize each entity in the entity subset based on the probability of occurrence and / or severity of the future state distribution.

[0013] Furthermore, based on the target detection project, the target material is numerically simulated to give the initial damage of the target material, including: Construct a finite element model based on the target material type; Combined with the target detection project, the load is added to the finite element model to obtain the structural time domain response of the target material; The response results are extracted and analyzed from the structural time-domain response of the target material to give the initial damage status of the target material.

[0014] Furthermore, combined with the target detection project, the load is added to the finite element model to obtain the structural time domain response of the target material, including: Determine the load vector of the finite element model based on the target detection items; Based on the load vector, the finite element model is load adjusted to simulate different states of the target material and obtain the material dynamics equation; Combined with the finite element model, the material dynamics equation is solved, the data of different nodes of the target material are analyzed, and the structural time domain response of the target material is given.

[0015] Furthermore, based on the initial damage situation, the initial damage situation of the target material is updated, and the cumulative damage situation of the target material is given to complete the quality inspection of the target material, specifically including: Obtain the prior probability of the target material in different states; Combined with the initial damage situation, the likelihood probability of the target material in different states is given; Based on the prior probability and likelihood probability of the target material, the posterior probability is given; Determine the relationship between the prior probability and the posterior probability, and update the initial damage status of the target material; The posterior probability is used as the new prior probability, the posterior probability is updated, and combined with the updated initial damage situation, the cumulative damage situation of the target material is given to complete the quality inspection of the target material.

[0016] In a second aspect, the present invention further provides a material quality detection device integrating knowledge graph and numerical simulation, which adopts any one of the above-mentioned material quality detection methods integrating knowledge graph and numerical simulation, including: A data acquisition module is used to acquire defect elimination and detection text data, wherein the defect elimination and detection text data includes detection text data of various types of materials; A graph construction module is used to perform entity recognition on the defect detection text data and construct a defect elimination knowledge graph based on the entity recognition model of the fusion language preprocessing; The project determination module is used to search for inspection projects from the defect elimination knowledge graph based on the target material type to obtain the target inspection project; The damage determination module is used to perform numerical simulation on the target material based on the target detection project and give the initial damage status of the target material; The damage update module is used to update the initial damage condition of the target material based on the initial damage condition, and provide the cumulative damage condition of the target material to complete the quality inspection of the target material.

[0017] The present invention provides a material quality detection method and device integrating knowledge graph and numerical simulation, which has at least the following beneficial effects: (1) By analyzing the previous inspection data, namely the defect inspection text data, a defect inspection knowledge graph is constructed to understand the relationship between the inspection data of different materials. Based on this, the target materials are analyzed to obtain the cumulative damage of the target materials and complete the quality inspection of the target materials. This reduces the manpower, material resources and time required in the quality inspection process and improves the work efficiency of material quality inspection.

[0018] (2) According to the target material type, the target detection items are obtained from the defect elimination knowledge graph, and the load is added to the finite element model according to the target detection items, so that the initial damage condition of the target material is more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 A flow chart of a material quality detection method integrating knowledge graph and numerical simulation provided in an embodiment of the present invention; Figure 2 A flowchart of constructing a knowledge graph for eliminating deficiencies provided by an embodiment of the present invention; Figure 3 A flowchart for determining attributes of each entity provided by an embodiment of the present invention; Figure 4 A flowchart for determining target detection items provided by an embodiment of the present invention; Figure 5 A flowchart of analyzing the priority of each entity in an entity subset provided by an embodiment of the present invention; Figure 6 A flowchart for determining an initial damage condition provided by an embodiment of the present invention; Figure 7 A flowchart for determining cumulative damage conditions provided by an embodiment of the present invention; Figure 8 A structural block diagram of a material quality detection device that integrates knowledge graph and numerical simulation provided in an embodiment of the present invention.

[0020] Among them, 201 is a data acquisition module; 202 is a graph construction module; 203 is a project determination module; 204 is a damage determination module; 205 is a damage update module. DETAILED DESCRIPTION

[0021] In order to better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0022] The terms used in the embodiments of the present invention are only for the purpose of describing specific embodiments, and are not intended to limit the present invention. The singular forms "a", "said" and "the" used in the embodiments of the present invention and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings, and "multiple" generally includes at least two.

[0023] It should also be noted that the term "includes", "comprising" or any other variation thereof is intended to cover non-exclusive inclusion, so that a commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprising a ..." do not exclude the existence of other identical elements in the commodity or device including the elements.

[0024] If electric power companies want to build themselves into world-class companies, they must continuously strengthen and improve the quality control of power grid materials and strictly control the quality of equipment and materials to provide strong guarantees for power grid security.

[0025] Grid material quality inspection usually refers to the inspection of power equipment and materials supplied to the State Grid Corporation through test witnessing, sampling and third-party inspection. However, affected by the different operating conditions, technical levels, environment, climate and other factors in different regions, the quality inspection of power materials can only be carried out based on their own regional characteristics, and a standard and unified production management method has not yet been formed. At the same time, the power material quality inspection centers of various provincial grid companies are usually located in the local power science institutes, without special laboratories, instruments and personnel support.

[0026] In recent years, with the construction of smart grids and the continuous expansion of power grids, the overall supply of power materials has become larger and larger, which has increased the workload of quality inspection year by year. At the same time, in order to ensure the quality of materials entering the network, the requirements for quality inspection of power grid materials have been continuously improved, which requires more time and manpower. How to improve the efficiency of material quality inspection is a problem that needs to be solved at present.

[0027] To this end, an embodiment of the present invention provides a material quality inspection method that integrates knowledge graph and numerical simulation. The defect elimination knowledge graph is obtained by performing entity recognition on the defect elimination detection text data including the detection text data of various types of materials. The defect elimination knowledge graph includes the detection data of different material types and their relationships. Based on the target material type, the detection project is searched from the defect elimination knowledge graph to obtain the target detection project; based on the target detection project, the target material is numerically simulated to give the initial damage of the target material; based on the initial damage, the initial damage of the target material is updated to give the cumulative damage of the target material to complete the quality inspection of the target material. By analyzing the previous detection data, a defect elimination knowledge graph is constructed to understand the relationship between different material detection data, and based on this, the target material is analyzed to obtain the cumulative damage of the target material, and the quality inspection of the target material is completed, which reduces the manpower, material resources and time required in the quality inspection process to a certain extent, and improves the work efficiency of material quality inspection.

[0028] like Figure 1 As shown, the embodiment of the present invention provides a material quality detection method integrating knowledge graph and numerical simulation, and the specific steps are as follows: S101: Acquire defect detection text data.

[0029] Specifically, the fault elimination detection text data includes the detection text data of various types of materials. The fault elimination detection text data is text data obtained based on previous material detection. It should be understood that various types of materials include but are not limited to power transformers, distribution transformers, oil-immersed reactors, dry-type reactors, current transformers, voltage transformers, shunt capacitors, AC lightning arresters, arc suppression coils, combination electrical appliances, high-voltage circuit breakers, disconnectors, grounding switches, distribution transformers, pole switches, switch cabinets, ring network cabinets, cable branch boxes, power cables, cable accessories, overhead insulated wires, power fittings, steel pipe towers, angle steel towers, concrete poles, line insulators, control cables, etc. The detection text data includes the detection process and methods, detection standards and specifications, detection results and processing of various types of materials. Among them, the detection results include whether the materials are qualified or not, for example, whether they have passed the insulation resistance test, the withstand voltage test, etc. The detection results also include specific values, such as the voltage, current, insulation resistance and other parameters of the materials. There are also performance indicators, defect descriptions, etc., such as the short-circuit resistance of the transformer and the opening and closing performance of the switchgear. For another example, if the materials are unqualified, the defects or problems will be described in detail, such as damaged insulation layer, poor contact, etc.

[0030] S102: Based on the entity recognition model of fusion language preprocessing, entity recognition is performed on the defect detection text data and a defect elimination knowledge graph is constructed.

[0031] Reference Figure 2 Specifically, an entity recognition model fused with language preprocessing is used to perform entity recognition on the defect detection text data to obtain multiple entities and the attributes of each entity. A spectral clustering algorithm is used to divide multiple entities according to the attributes of each entity to obtain multiple entity subsets. Based on the attributes of each entity in each entity subset, the subset attributes of each entity subset are analyzed to give subset attribute values. According to the subset attribute values ​​of each entity subset, each entity subset is fused to obtain a defect elimination knowledge graph.

[0032] It can be understood that the defect elimination detection text contains rich defect information. Through information extraction, important information such as fault location, frequency, probability, etc. can be obtained, which can provide a multi-angle reference basis for the detection plan and detection process of the same type of materials. Therefore, by integrating the knowledge of defect elimination detection texts, detection forms, etc. of various types of materials, a comprehensive analysis of the material location, defect frequency, detection items, and index parameters can be used to give a more reasonable basis for the formulation of detection plans and optimization suggestions for the detection process. In a specific example, when the material to be detected is a power cable, for unstructured texts such as power cable detection records and defect elimination cases, information extraction and knowledge mining through named entity recognition can fully utilize the value contained in the data. It is mainly through the extraction, transformation, and fusion of knowledge on power cable parameters, detection forms, and power cable defect elimination texts to form a knowledge graph.

[0033] Reference Figure 3 , further, the defect detection text data that has completed text cleaning is segmented to give a first entity set. Based on the stop word set, the first entity set is screened, and the stop words in the first entity set are eliminated to obtain a second entity set. Each text sequence in the second entity set is filled or intercepted to obtain a sequence vector set, wherein the length of each sequence vector in the sequence vector set is the same. The attention mechanism layer is used to process each sequence vector in the sequence vector set, analyze the correlation between each sequence vector, and obtain multiple entities and the attributes of each entity.

[0034] In one implementation provided by an embodiment of the present invention, the fusion language preprocessing entity recognition model is a BiLSTM-CRF model. The BiLSTM-CRF model is a deep learning model that combines a bidirectional long short-term memory network (BiLSTM) and a conditional random field (CRF), and is applied to sequence labeling tasks such as named entity recognition (NER).

[0035] The BiLSTM-CRF model includes a text cleaning layer, a word segmentation layer, a stop word removal layer, a sequence filling layer, an embedding layer, and an attention mechanism layer. The sequence filling layer is used to fill or truncate text sequences of different lengths to a fixed length, and the embedding layer is used to convert the text sequence after word segmentation into a numerical vector and use a pre-trained word embedding method to map the numerical vector to a vector in a high-dimensional space.

[0036] The defect detection text data is input into the BiLSTM-CRF model, and first enters the text cleaning layer to clean the defect detection text data and remove noise in the text, such as special characters, HTML tags, etc. After the text cleaning is completed, it enters the word segmentation layer to segment the defect detection text data and divide the text into units at the word or character level to obtain the first entity set. The first entity set enters the stop word removal layer, and the stop words in the first entity set are removed in combination with the stop word set to obtain the second entity set, wherein the stop word set includes various stop words. When the first entity set contains stop words in the stop word set, the corresponding stop words in the first entity set are removed until the first entity set does not contain stop words in the stop word set to obtain the second entity set. The second entity set is input into the sequence filling layer, and each text sequence in the second entity set is filled or intercepted to obtain text sequences of the same length, and then the text sequences of the same length are input into the embedding layer to obtain a sequence vector set. The attention mechanism layer is used to process each sequence vector in the sequence vector set, analyze the correlation between each sequence vector, and obtain multiple entities and the attributes of each entity.

[0037] Among them, the specific process of sequence filling layer and embedding layer is as follows: In a specific example, the defect detection text data is T, and after the word segmentation layer, the first entity set is obtained. , the maximum sequence length is L, and the set after truncation and filling is , the length of each sequence after truncation and padding is L. The word embedding matrix in the embedding layer is E, the position embedding matrix is ​​P, and the fragment embedding matrix is ​​S. The final output vector X is expressed as:

[0038] in, Words The word embedding vector of is the position embedding vector for position i, is the segment embedding vector of segment j, where j=0 represents the first sentence.

[0039] After the sequence vector is obtained through the embedding layer, it is processed through the attention mechanism layer. Each sequence vector in the sequence vector set is input into the attention mechanism layer to obtain the context-weighted representation of each sequence vector. These representations not only contain the semantic information of the sequence vector itself, but also incorporate the relevant information of other words in the sentence. In other embodiments, since the association between words in the missing detection text data is relatively clear, the dependency between words can be captured by adopting local attention, unnecessary calculations can be reduced to improve efficiency, and weighted representation of words can be achieved, which is not limited to this.

[0040] The specific process of the attention mechanism layer is as follows: First, get the expansion size k and mask matrix M, and calculate the attention score of the i-th word in the j-th segment. Normalize the obtained attention score and multiply it with the value vector V to get the final output.

[0041] In the BiLSTM-CRF model, the BiLSTM network is used to merge the outputs of each word sequence through the forward and backward LSTM at the same time. The main function of the CRF model is to obtain the optimal prediction sequence through the relationship between adjacent labels. The CRF model considers the constraints of the corresponding label of a word and the adjacent labels.

[0042] Among them, the element in the i-th row and j-th column in the mask matrix is ​​specifically expressed as:

[0043] The attention score is specifically expressed as:

[0044] in, is the dimension of the key vector, is the query matrix corresponding to the i-th word, indicating the information that the i-th word wants to pay attention to. is the key matrix corresponding to the jth fragment, indicating the information that other words can be paid attention to. is the element value in the i-th row and j-th column of the mask matrix.

[0045] Furthermore, based on the attributes of each entity in each entity subset, the embedding vector of each entity is obtained. According to the embedding vector of each entity, the norm of the embedding vector is given. For each entity subset, the mean of the norms of the corresponding embedding vectors of each entity in the entity subset is used as the subset attribute value of the corresponding entity subset.

[0046] It is understandable that in the process of constructing the missing knowledge graph, it is necessary to determine the attributes of each entity in each entity subset, and based on this, complete the storage of the missing knowledge graph. The specific process is: The spectral clustering algorithm is used to create nodes representing each entity subset and edges representing the relationship between each entity subset according to each entity subset and the relationship between each entity subset. The attributes of each entity subset are analyzed. In a specific example, the average node importance of the entity subset is used as the attribute value of the entity subset. The specific steps for determining the average node importance are as follows: Each entity has an importance value, which is calculated by the norm of the embedding vector corresponding to each entity. The norm of the embedding vector is specifically expressed as:

[0047] in, is the embedding vector of the vth entity, d is the dimension of the embedding vector, is the value of the embedding vector of the vth entity in the uth dimension.

[0048] For example, when When , the norm of the embedding vector .

[0049] For each entity subset, obtain the importance values ​​of all entities in the entity subset and determine the average node importance of the entity subset. The specific steps are as follows:

[0050] in, is the average node importance of the j-th entity subset, is the j-th entity subset, is the i-th entity in the j-th entity subset The importance value of .

[0051] Based on the average node importance of each entity subset and the importance value of each entity in each entity subset, the storage of the missing knowledge graph is completed.

[0052] S103: Based on the target material type, the inspection items are searched from the defect elimination knowledge graph to obtain the target inspection items.

[0053] Reference Figure 4 ,Specifically, based on the target material type, a search is performed from the deficiency elimination knowledge graph to obtain an entity subset corresponding to the target material type; Based on the entity subset corresponding to the target material type, the historical occurrence frequency of each entity in the entity subset is analyzed, and the priority of each entity in the entity subset is given based on the mapping relationship between the defective parts of the material and the inspection items; Determine the target detection items based on the priority of each entity.

[0054] In a specific implementation, each entity subset in the defect elimination knowledge graph represents different types of material data, and each entity in the entity subset represents different test items and / or test results in different types of material data and the material status corresponding to different test results. In the defect elimination knowledge graph, the material data corresponding to the target material type is queried, and the defective parts of the target material type corresponding to the material defects and the historical occurrence frequency of each defective part are found from these material data. Combined with the mapping relationship between the defective parts of the material and the test items, the priority of each entity corresponding to the target material type, that is, the priority of the test items, is given. Among them, the mapping relationship between the defective parts of the material and the test items is pre-established based on existing experience, and is pre-stored in the graph database and can be used directly. According to the priority of each entity, the target detection item is determined. The higher the priority of the entity, the more important the impact of the test item corresponding to the entity on the material of the target material type is, and the target detection item is determined among the entities with higher priority.

[0055] Reference Figure 5 ,Further, based on the entity subset corresponding to the target material type, the historical occurrence frequency of each entity in the entity subset is obtained; Based on the historical occurrence frequency of each entity in the entity subset, a state transition probability matrix is ​​constructed; Based on the initial state analysis and combined with the state transition probability matrix, the future state distribution is obtained; Prioritize each entity in the entity subset based on the probability of occurrence and / or severity of the future state distribution.

[0056] Furthermore, based on the historical occurrence frequency of each entity in the entity subset, the state occurrence frequency of different entities corresponding to different states is given; Based on the state occurrence frequencies of different entities corresponding to different states, a state transition frequency matrix is ​​constructed, where the elements in the state transition frequency matrix are the number of occurrences of the target material from one state to another; Analyze the state transition frequency matrix and give the state transition probability matrix.

[0057] In a specific implementation, the possible states of the material are first determined and a state transition probability matrix is ​​given based on the state transition probability matrix. For example, when the material is a power cable, the possible states of the power cable are defined as slight defects, medium defects, and severe defects. Based on these three states, the state occurrence frequencies of different entities corresponding to different states are counted, and a state transition frequency matrix is ​​constructed. Then, the state transition frequency matrix is ​​analyzed and adjusted to give a state transition probability matrix P, which is specifically expressed as:

[0058] If minor defects are defined as state 1, medium defects as state 2, and severe defects as state 3, P 12 is the probability of transitioning from state 1 to state 2. Similarly, the meanings of other elements in the state transition probability matrix can be known.

[0059] Based on the initial state analysis and combined with the state transition probability matrix, the future state distribution is obtained. In a specific example, the initial state distribution is , then the state distribution after m steps is the future state distribution .

[0060] The priority of each entity in the entity subset is determined according to the probability of occurrence and / or severity of the future state distribution. For example, the state with the highest probability of occurrence is taken as the final state of each entity, and then the priority of each entity is determined according to the severity of the final state of each entity, and the priority of each entity is determined in descending order of severity. In other embodiments, the priority of each entity can be determined by selecting the probability of occurrence or severity according to the actual situation, and this is not limited.

[0061] S104: Based on the target detection project, numerical simulation is performed on the target material to provide the initial damage condition of the target material.

[0062] Reference Figure 6 ,Specifically, based on the target material type, a finite element model is constructed; Combined with the target detection project, the load is added to the finite element model to obtain the structural time domain response of the target material; The response results are extracted and analyzed from the structural time-domain response of the target material to give the initial damage status of the target material.

[0063] The finite element model is a model established when the finite element analysis method is used. It is a group of units that are connected only at the nodes, transmit forces only at the nodes, and are constrained only at the nodes. The finite element model is the result of the discretization of the mechanical model and is a digital model for numerical calculation. The finite element model is a global behavior model of the entire system that is constructed by discretizing a complex physical structure or system into a group of small units (finite elements) and using numerical methods to simulate and analyze the physical behavior of these units. Each unit is connected through a node, and the behavior of each unit is described by physical laws (such as mechanics, heat conduction, or fluid mechanics). Using the finite element model, the target material in each state is numerically simulated to obtain the structural time domain response of the target object in different states.

[0064] Further, based on the target detection items, a load vector of the finite element model is determined; Based on the load vector, the finite element model is load adjusted to simulate different states of the target material and obtain the material dynamics equation; Combined with the finite element model, the material dynamics equation is solved, the data of different nodes of the target material are analyzed, and the structural time domain response of the target material is given.

[0065] The data of different nodes of the target material are based on the differences in the materials and the load vectors. They can be the data of different positions of the target material or the data of the target material at different time nodes. They are determined according to actual needs and are not limited to this.

[0066] In a specific implementation, the target material is a cable. First, a finite element model of the cable is established, and then loads are continuously applied to the finite element model to simulate various states of the cable and obtain the structural time domain response of the cable. When simulating the structural time domain response of the cable, a physical model needs to be established to describe the dynamic behavior of the cable under different loads. The finite element can be used to solve the dynamic equation of the cable, that is, the material dynamic equation, which is specifically expressed as:

[0067] Among them, M is the total mass matrix, C is the damping matrix, K is the total stiffness matrix, U is the node displacement vector, and R(t) is the time-dependent load vector. The above total mass matrix, damping matrix, and total stiffness matrix are related to the physical properties and geometric shape of the cable. The total mass matrix, damping matrix, and total stiffness matrix can be considered as constant matrices.

[0068] Apply a load vector to solve the dynamic equation of the cable. The specific steps include: applying a pulse load at one end of the cable Where A is the pulse amplitude, is the pulse generation time. The kinetic equations are iteratively solved using the explicit time integration method with a time step of . Extract the displacement and acceleration response of each node of the cable and analyze the dynamic behavior of the cable. When iteratively solving the dynamic equation, the displacement vector of each node on the cable at each time step is obtained. , velocity vector and the acceleration vector .

[0069] In a specific example, the applied pulse load of the cable is , the displacement vector of the cable midpoint response is obtained by solving , according to the displacement vector The time domain waveform shows that its amplitude is 5 cm and the period is 0.1 second. It can be found that the displacement response amplitude of the midpoint of the cable at a load frequency of 10 Hz is 5 cm, indicating that the cable has a large dynamic response at this frequency, giving the initial damage situation.

[0070] It can be understood that when the target materials are different, the corresponding target detection items are different, and the given loads are also different. Taking cables as an example, when the target detection item is to detect the load-bearing capacity of the cable, the applied load is the static pressure or weight that the cable bears during use; when the target detection item is to detect the vibration bearing capacity of the cable, the applied load is the tension, vibration, etc. that the cable may be subjected to during the laying process. For another example, when the target detection item is the bearing capacity of the cable under different weather conditions, the applied load is the load caused by environmental factors.

[0071] S105: Based on the initial damage condition, the initial damage condition of the target material is updated, and the cumulative damage condition of the target material is given to complete the quality inspection of the target material.

[0072] Reference Figure 7 ,Specifically, obtain the prior probability of the target material in different states; Combined with the initial damage situation, the likelihood probability of the target material in different states is given; Based on the prior probability and likelihood probability of the target material, the posterior probability is given; Determine the relationship between the prior probability and the posterior probability, and update the initial damage status of the target material; The posterior probability is used as the new prior probability, the posterior probability is updated, and combined with the updated initial damage situation, the cumulative damage situation of the target material is given to complete the quality inspection of the target material.

[0073] In a specific implementation, based on historical experience, the prior probability P(H) of the dynamic behavior of the cable under different loads is defined. For example, the probability of the cable resonating under a high-frequency load is 0.3. Based on the parameters obtained during the operation or experiment of the cable, evidence E related to the dynamic behavior of the cable is collected. For example, the obtained parameters show that the displacement response amplitude of the cable at a certain frequency increases significantly. The probability that the cable resonates under a high-frequency load and the displacement response amplitude increases significantly at this frequency is calculated to be 0.8, that is, the likelihood probability P(E|H)=0.8. Then the posterior probability P(H|E) is calculated, which is specifically expressed as:

[0074] in, is the probability that hypothesis H is not true, and is the probability that evidence E appears when hypothesis H is not true.

[0075] Determine the relationship between prior probability and posterior probability. Update the initial damage of the target material. If the posterior probability is higher than the prior probability, it means that the evidence E corresponding to the likelihood probability supports hypothesis H, that is, the dynamic behavior of the cable under this load tends to resonate.

[0076] The posterior probability As a new prior probability, it is used for the next data collection and analysis. Repeat this process. As more data is obtained, the posterior probability will be continuously updated to obtain the cumulative damage of the target material. For example, suppose at multiple time points Gathered evidence , then the probability of the cumulative behavior is , by gradually accumulating evidence, the assessment of the cable dynamic structure is updated and the cumulative damage situation is given.

[0077] Reference Figure 8 , an embodiment of the present invention provides a material quality detection device integrating knowledge graph and numerical simulation, including: The data acquisition module 201 is used to acquire defect elimination and detection text data, wherein the defect elimination and detection text data includes detection text data of various types of materials; A graph construction module 202, for performing entity recognition on the defect elimination detection text data and constructing a defect elimination knowledge graph based on an entity recognition model of fusion language preprocessing; The project determination module 203 is used to search for inspection projects from the defect elimination knowledge graph based on the target material type to obtain the target inspection project; The damage determination module 204 is used to perform numerical simulation on the target material based on the target detection item and provide the initial damage condition of the target material; The damage update module 205 is used to update the initial damage status of the target material based on the initial damage status, and provide the cumulative damage status of the target material to complete the quality inspection of the target material.

[0078] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the described module can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0079] Although preferred embodiments of the present invention have been described, additional changes and modifications may be made to these embodiments by those skilled in the art once the basic inventive concepts are known. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention. Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A material quality detection method integrating knowledge graph and numerical simulation, characterized in that: include: Acquire defect elimination and detection text data, wherein the defect elimination and detection text data includes detection text data of various types of materials; Based on the entity recognition model that integrates language preprocessing, entity recognition is performed on the defect detection text data and a defect elimination knowledge graph is constructed; Based on the target material type, the inspection items are searched from the defect elimination knowledge graph to obtain the target inspection items; Based on the target detection project, numerical simulation is performed on the target material to give the initial damage of the target material; Based on the initial damage situation, the initial damage situation of the target material is updated, and the cumulative damage situation of the target material is given to complete the quality inspection of the target material.

2. The material quality detection method integrating knowledge graph and numerical simulation as claimed in claim 1 is characterized in that: Based on the entity recognition model that integrates language preprocessing, entity recognition is performed on the defect detection text data and a defect elimination knowledge graph is constructed, which includes: The entity recognition model of fusion language preprocessing is used to perform entity recognition on the defect detection text data to obtain multiple entities and the attributes of each entity; Using the spectral clustering algorithm, multiple entities are divided according to the attributes of each entity to obtain multiple entity subsets; Based on the attributes of each entity in each entity subset, analyze the subset attributes of each entity subset and give the subset attribute value; According to the subset attribute values ​​of each entity subset, the entity subsets are fused to obtain the missing knowledge graph.

3. The material quality detection method integrating knowledge graph and numerical simulation as claimed in claim 2 is characterized in that: The entity recognition model of fusion language preprocessing is used to perform entity recognition on the defect detection text data to obtain multiple entities and the attributes of each entity, including: Segment the defect detection text data after text cleaning to obtain a first entity set; Based on the stop word set, the first entity set is screened, the stop words in the first entity set are removed, and a second entity set is obtained; Filling or truncating each text sequence in the second entity set to obtain a sequence vector set, wherein the lengths of each sequence vector in the sequence vector set are the same; The attention mechanism layer is used to process each sequence vector in the sequence vector set, analyze the correlation between each sequence vector, and obtain multiple entities and the attributes of each entity.

4. The material quality detection method integrating knowledge graph and numerical simulation as claimed in claim 2 is characterized in that: Based on the attributes of each entity in each entity subset, analyze the subset attributes of each entity subset and give the subset attribute values, including: Based on the attributes of each entity in each entity subset, the embedding vector of each entity is obtained; According to the embedding vector of each entity, the norm of the embedding vector is given; For each entity subset, the mean of the norms of the embedding vectors corresponding to each entity in the entity subset is taken as the subset attribute value of the corresponding entity subset.

5. The material quality detection method integrating knowledge graph and numerical simulation as claimed in claim 1 is characterized in that: Based on the target material type, the inspection items are searched from the defect elimination knowledge graph to obtain the target inspection items, including: Based on the target material type, search from the deficiency elimination knowledge graph to obtain the entity subset corresponding to the target material type; Based on the entity subset corresponding to the target material type, the historical occurrence frequency of each entity in the entity subset is analyzed, and the priority of each entity in the entity subset is given based on the mapping relationship between the defective parts of the material and the inspection items; Determine the target detection items based on the priority of each entity.

6. The material quality detection method integrating knowledge graph and numerical simulation as claimed in claim 5 is characterized in that: Based on the entity subset corresponding to the target material type, the historical occurrence frequency of each entity in the entity subset is analyzed, and the priority of each entity in the entity subset is given in combination with the mapping relationship between the material defect parts and the inspection items, including: Based on the entity subset corresponding to the target material type, the historical occurrence frequency of each entity in the entity subset is obtained; Based on the historical occurrence frequency of each entity in the entity subset, the state occurrence frequency of different entities corresponding to different states is given; Based on the state occurrence frequencies of different entities corresponding to different states, a state transition frequency matrix is ​​constructed, where the elements in the state transition frequency matrix are the number of occurrences of the target material from one state to another; Analyze the state transition frequency matrix and give the state transition probability matrix; Based on the initial state analysis and combined with the state transition probability matrix, the future state distribution is obtained; Prioritize each entity in the entity subset based on the probability of occurrence and / or severity of the future state distribution.

7. The material quality detection method integrating knowledge graph and numerical simulation as claimed in claim 1 is characterized in that: Based on the target detection project, the target material is numerically simulated to give the initial damage of the target material, including: Construct a finite element model based on the target material type; Combined with the target detection project, the load is added to the finite element model to obtain the structural time domain response of the target material; The response results are extracted and analyzed from the structural time-domain response of the target material to give the initial damage status of the target material.

8. The material quality detection method integrating knowledge graph and numerical simulation as claimed in claim 7 is characterized in that: Combined with the target detection project, the load is added to the finite element model to obtain the structural time domain response of the target material, including: Determine the load vector of the finite element model based on the target detection items; Based on the load vector, the finite element model is load adjusted to simulate different states of the target material and obtain the material dynamics equation; Combined with the finite element model, the material dynamics equation is solved, the data of different nodes of the target material are analyzed, and the structural time domain response of the target material is given.

9. The material quality detection method integrating knowledge graph and numerical simulation as claimed in claim 1, characterized in that: Based on the initial damage situation, the initial damage situation of the target material is updated, and the cumulative damage situation of the target material is given to complete the quality inspection of the target material, including: Obtain the prior probability of the target material in different states; Combined with the initial damage situation, the likelihood probability of the target material in different states is given; Based on the prior probability and likelihood probability of the target material, the posterior probability is given; Determine the relationship between the prior probability and the posterior probability, and update the initial damage status of the target material; The posterior probability is used as the new prior probability, the posterior probability is updated, and combined with the updated initial damage situation, the cumulative damage situation of the target material is given to complete the quality inspection of the target material.

10. A material quality detection device integrating knowledge graph and numerical simulation, characterized in that: A material quality detection method integrating knowledge graph and numerical simulation as described in any one of claims 1 to 9 is adopted, comprising: A data acquisition module is used to acquire defect elimination and detection text data, wherein the defect elimination and detection text data includes detection text data of various types of materials; A graph construction module is used to perform entity recognition on the defect detection text data and construct a defect elimination knowledge graph based on the entity recognition model of the fusion language preprocessing; The project determination module is used to search for inspection projects from the defect elimination knowledge graph based on the target material type to obtain the target inspection project; The damage determination module is used to perform numerical simulation on the target material based on the target detection project and give the initial damage status of the target material; The damage update module is used to update the initial damage condition of the target material based on the initial damage condition, and provide the cumulative damage condition of the target material to complete the quality inspection of the target material.

Images