Industrial fault multi-scheme generation method and system fusing knowledge graph

Abstract

The invention provides an industrial fault multi-scheme generation method and system fused with a knowledge graph, and the method comprises the steps: obtaining a fault data set of target industrial equipment, carrying out the feature extraction processing, generating a fault feature set, calling a pre-constructed industrial knowledge graph, carrying out the entity association analysis processing, and generating a fault association feature set; fault propagation path reasoning processing is carried out based on the fault correlation feature set, a potential fault propagation path set is generated, then solution matching and feasibility evaluation processing are carried out on the potential fault propagation path set, and finally an optimized solution set is generated, so that professional knowledge in the industrial field is fused; fault propagation paths can be accurately analyzed, diversified and optimized solutions can be generated, and the accuracy of industrial fault diagnosis and the effectiveness of the solutions are improved.

Description

Technical Field

[0001] This invention relates to the field of smart industrial technology, and more specifically, to a method and system for generating multiple solutions for industrial faults that integrates knowledge graphs. Background Technology

[0002] In industrial production, the stable operation of industrial equipment is a key factor in ensuring production efficiency and product quality. However, due to various reasons such as equipment aging, improper operation, and environmental factors, industrial equipment frequently malfunctions. Traditional methods for industrial fault diagnosis and solution generation mainly rely on the experience and expertise of technicians, manually analyzing fault data to determine the cause of the fault and formulate solutions. This method is not only inefficient but also easily affected by human factors, resulting in a lack of assurance regarding the accuracy of fault diagnosis and the effectiveness of solutions.

[0003] With the rapid development of information technology, although some automated fault diagnosis systems have been introduced into industrial production, most of these systems rely solely on simple rule matching or machine learning algorithms for fault diagnosis, lacking a deep understanding of the complex operating environment and fault propagation mechanisms of industrial equipment. Furthermore, when generating solutions, these systems often only offer a single or limited set of options, failing to fully consider the diversity and complexity of faults and thus struggling to meet the needs of actual production. Summary of the Invention

[0004] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide a method for generating multiple solutions to industrial faults by integrating knowledge graphs, the method comprising:

[0005] Obtain a set of fault data for the target industrial equipment, perform feature extraction processing on the set of fault data, and generate a set of fault features corresponding to the set of fault data.

[0006] The pre-built industrial knowledge graph is invoked to perform entity association analysis on the fault feature set, generating a fault association feature set of the fault feature set in the industrial knowledge graph;

[0007] Based on the fault association feature set, fault propagation path reasoning is performed to generate a set of potential fault propagation paths for the target industrial equipment in the industrial knowledge graph.

[0008] The set of potential fault propagation paths is subjected to solution matching processing to generate a set of candidate solutions corresponding to each potential fault propagation path;

[0009] The candidate solution set is subjected to a feasibility assessment to generate an optimized solution set corresponding to the target industrial equipment.

[0010] In another aspect, embodiments of the present invention also provide an industrial fault multi-solution generation system that integrates knowledge graphs, including a processor and a machine-readable storage medium connected to the processor. The machine-readable storage medium is used to store programs, instructions, or code, and the processor is used to execute the programs, instructions, or code in the machine-readable storage medium to implement the above-described method.

[0011] Based on the above, this invention, by integrating knowledge graph technology, achieves in-depth mining and analysis of industrial fault data. It can accurately extract fault features and associate them with relevant entities in the industrial knowledge graph, thereby revealing the potential connections and propagation paths between faults. Through reasoning processing of fault propagation paths, it can generate a set of potential fault propagation paths for the target industrial equipment in the industrial knowledge graph. Furthermore, by performing solution matching and feasibility assessment on the potential fault propagation paths, it can generate a set of optimized solutions corresponding to each potential fault propagation path. This not only improves the accuracy of fault diagnosis and the effectiveness of solutions, but also greatly enhances the diversity and adaptability of solutions, thereby significantly improving the intelligence level of industrial fault diagnosis and solution generation, which helps to stabilize industrial production. Attached Figure Description

[0012] Figure 1 This is a schematic diagram of the execution flow of the industrial fault multi-solution generation method fused with knowledge graph provided in the embodiments of the present invention.

[0013] Figure 2 This is a schematic diagram of exemplary hardware and software components of the industrial fault multi-solution generation system that integrates knowledge graphs provided in an embodiment of the present invention. Detailed Implementation

[0014] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 This is a flowchart illustrating a method for generating multiple solutions to industrial faults by fusing knowledge graphs, provided in one embodiment of the present invention. The following is a detailed description of this method for generating multiple solutions to industrial faults by fusing knowledge graphs.

[0015] Step S110: Obtain the fault data set of the target industrial equipment, perform feature extraction processing on the fault data set, and generate a fault feature set corresponding to the fault data set.

[0016] In industrial settings, target industrial equipment can be various types of manufacturing, transportation, or energy equipment. This equipment is typically equipped with various types of sensors to monitor its operational status in real time. For example, temperature sensors monitor the temperature of critical components, pressure sensors detect pressure inside the equipment or in pipes, and vibration sensors sense vibrations during operation. These sensors collect data at regular time intervals and transmit it to a data storage system. Simultaneously, the target industrial equipment's operation log records all operations performed by the operator, including the time, content, and status of the equipment during operation. Furthermore, environmental factors can also affect the operation of the target industrial equipment; therefore, environmental monitoring equipment collects environmental parameters such as temperature, humidity, and air pressure. By integrating data from sensors, operation logs, and environmental monitoring equipment, a fault data set is obtained.

[0017] The next step is to perform feature extraction on the fault data set, which includes the following sub-steps:

[0018] Step S111: Perform time series analysis on the equipment operating parameters in the fault data set to generate the operating status fluctuation characteristics of the target industrial equipment in the time dimension.

[0019] Let the changes in equipment operating parameters over time form a sequence, denoted by R, where R = {r1, r2, ..., rn}, and ri represents the equipment operating parameter value at the i-th time point, and n is the total number of time points. To analyze the fluctuations in the equipment's operating state over time, we first calculate the differences in operating parameters between adjacent time points. For the i-th time point, the parameter difference between it and the next time point is denoted as Δri, calculated by subtracting the parameter value ri at the i-th time point from the parameter value ri+1 at the (i+1)-th time point, i.e., Δri = ri+1 - ri (i = 1, 2, ..., n-1). This yields a difference sequence ΔR = {Δr1, Δr2, ..., Δrn-1}.

[0020] To more comprehensively describe the fluctuation characteristics of the operating status, the statistical characteristics of the difference sequence can be further calculated. For example, the mean of the difference sequence, denoted as MΔR, can be calculated by adding all the differences in the sequence and then dividing by the number of differences (n-1), i.e., MΔR = (Δr1 + Δr2 + ... + Δrn-1) / (n-1). This mean reflects the average fluctuation of the equipment's operating parameters over time. Simultaneously, the variance of the difference sequence, denoted as VΔR, can be calculated, which reflects the dispersion of the fluctuations in the equipment's operating parameters. To calculate the variance, the square of the difference between each difference and the mean is calculated first, then these squares are added together and divided by the number of differences (n-1), i.e., VΔR = [(Δr1 - MΔR)² + (Δr2 - MΔR)² + ... + (Δrn-1 - MΔR)²] / (n-1). Through statistical characteristics such as the mean and variance, the fluctuation characteristics of the equipment's operating status over time can be described relatively accurately.

[0021] Step S112: Perform frequency domain transformation processing on the equipment abnormal signals in the fault data set to generate the energy distribution characteristics of the target industrial equipment in the frequency domain dimension.

[0022] Equipment anomaly signals are typically manifested in the time domain, such as vibration signals and current signals. Let the time domain representation of the equipment anomaly signal be S = {s1, s2, ..., sn}, where si represents the anomaly signal value at the i-th time point, and n is the total number of time points. To convert the time-domain signal to the frequency domain, frequency domain transformation methods, such as the Fourier transform, can be used. After the Fourier transform, the time-domain signal S will be converted into a frequency-domain signal F = {f1, f2, ..., fm}, where fi represents the signal value of the i-th frequency component, and m is the number of frequency components.

[0023] In the frequency domain, signals with different frequency components correspond to different energies. To calculate the energy of each frequency component, we can square each frequency component of the frequency domain signal, i.e., Ei = fi. 2 (i = 1, 2, ..., m), where Ei represents the energy of the i-th frequency component. Combining the energies of all frequency components yields the energy distribution characteristic of the device in the frequency domain: E = {E1, E2, ..., Em}. Analyzing this energy distribution characteristic reveals the energy concentration of abnormal signals at different frequencies, thus identifying potential faults in the device. For example, a concentrated energy distribution within a specific frequency range may indicate a mechanical or electrical fault related to that frequency.

[0024] Step S113: Perform semantic parsing processing on the equipment operation logs in the fault data set to generate behavioral pattern features of the target industrial equipment in the operation dimension.

[0025] Equipment operation logs are text-based records of various operational information provided by operators to the equipment. To perform semantic parsing of the operation logs, the text first needs to be preprocessed. This includes removing useless characters such as punctuation and spaces, and then performing word segmentation, which breaks down continuous text into meaningful words. Let the word sequence obtained after preprocessing and word segmentation be W = {w1, w2, ..., wk}, where wi represents the i-th word and k is the total number of words.

[0026] Next, each word needs to be assigned corresponding semantic information. A semantic dictionary can be built to associate different words with specific operational semantics. For example, words such as "start," "stop," "accelerate," and "decelerate" correspond to different operational behaviors of the device. For each word wi in the word sequence W, its corresponding semantic information is looked up according to the semantic dictionary, resulting in a semantic sequence S = {s1, s2, ..., sk}, where si represents the semantic information of the i-th word.

[0027] To generate behavioral pattern characteristics of the device in the operational dimension, further analysis of the semantic sequences is required. This involves statistically analyzing the frequency of different semantic information and the relationships between them. For example, the frequency of a particular operational semantic can be calculated as the proportion of its occurrences to the total number of semantics. Simultaneously, the sequence and co-occurrence of different operational semantics can be analyzed to identify behavioral patterns in device operation. For instance, if the "start" operation is frequently followed by the "adjust parameters" operation, this constitutes an operational behavioral pattern. Combining these frequency information and relationships generates the behavioral pattern characteristics of the device in the operational dimension.

[0028] Step S114: Perform trend fitting processing on the equipment environmental parameters in the fault data set to generate the external interference characteristics of the target industrial equipment in the environmental dimension.

[0029] Equipment environmental parameters include data on the changes in temperature, humidity, and air pressure of the environment in which the equipment is located over time. Let the time series representation of a certain parameter (such as temperature) be T = {t1, t2, ..., tn}, where ti represents the environmental parameter value at the i-th time point, and n is the total number of time points.

[0030] To perform trend fitting, a polynomial fitting method can be used. Assuming a first-order polynomial fitting is used, the fitting function is y = ax + b, where x represents time, y represents the environmental parameter value, and a and b are the coefficients to be determined. The values ​​of coefficients a and b are determined using the least squares method. The goal of the least squares method is to minimize the sum of squared errors between the predicted values ​​of the fitted function and the actual environmental parameter values. Specifically, the predicted value yi = axi + b is calculated for each time point, and then the sum of squared errors E = (t1 - y1)² + (t2 - y2)² + ... + (tn - yn)² is calculated. By taking the partial derivatives of E with respect to a and b, and setting the partial derivatives to zero, the values ​​of coefficients a and b can be obtained by solving the system of equations.

[0031] Once the fitted function is obtained, the slope 'a' of the fitted function can be used to determine the changing trend of environmental parameters. If 'a' is greater than zero, it indicates that the environmental parameters are trending upwards; if 'a' is less than zero, it indicates that the environmental parameters are trending downwards; if 'a' is close to zero, it indicates that the environmental parameters are relatively stable. Combining the trend information of different environmental parameters generates the external interference characteristics of the equipment in the environmental dimension. For example, if the ambient temperature shows an upward trend while the humidity shows a downward trend when the equipment malfunctions, then this trend information may be the external interference factors causing the equipment malfunction.

[0032] Step S115: The operating state fluctuation characteristics, energy distribution characteristics, behavior pattern characteristics and external disturbance characteristics are standardized and transformed respectively, and then merged into the fault feature set.

[0033] The purpose of standardization is to unify the dimensions of different features, making them comparable in subsequent analyses. For the fluctuation characteristics of the operating state, let it be represented as A = {a1, a2, ..., ap}, where ai represents the i-th fluctuation characteristic value and p is the number of fluctuation characteristics. Standardization can be performed using the Z-score standardization method, calculating the Z-score value for each fluctuation characteristic value, i.e., Zi = (ai - MA) / SA, where MA is the mean of the fluctuation characteristic and SA is the standard deviation of the fluctuation characteristic. The mean MA is calculated by adding all fluctuation characteristic values ​​and then dividing by the number of characteristics p, i.e., MA = (a1 + a2 + ... + ap) / p; the standard deviation SA is calculated by first calculating the square of the difference between each fluctuation characteristic value and the mean, then adding these squares and dividing by the number of characteristics p, and finally taking the square root, i.e., SA = √[(a1 - MA)² + (a2 - MA)² + ... + (ap - MA)²] / p. After standardization, the standardized operating state fluctuation characteristics Z_A = {Z1, Z2, ..., Zp} are obtained.

[0034] For the energy distribution characteristics, let them be represented as B = {b1, b2, ..., bq}, where bi represents the i-th energy distribution characteristic value and q is the number of energy distribution characteristics. Using the Z-score standardization method, the Z-score value of each energy distribution characteristic is calculated, i.e., Zj = (bj - MB) / SB, where MB is the mean of the energy distribution characteristic and SB is the standard deviation of the energy distribution characteristic. After standardization, the standardized energy distribution characteristics Z_B = {Z1, Z2, ..., Zq} are obtained.

[0035] The behavioral pattern features and external interference features were also processed using a similar Z-score normalization method to obtain the normalized behavioral pattern features Z_C and the normalized external interference features Z_D.

[0036] Finally, the standardized operating state fluctuation characteristics Z_A, standardized energy distribution characteristics Z_B, standardized behavior pattern characteristics Z_C, and standardized external disturbance characteristics Z_D are merged to obtain the fault feature set. The merging method can be to sequentially concatenate these features to form a set containing all feature information.

[0037] Step S120: Call the pre-built industrial knowledge graph to perform entity association analysis on the fault feature set, and generate the fault association feature set of the fault feature set in the industrial knowledge graph.

[0038] A pre-built industrial knowledge graph is a knowledge base that contains various entities and their relationships in the industrial field. Entities can be fault phenomena, fault causes, solutions, etc., and relationships can be causal relationships, association relationships, etc.

[0039] Step S121: Perform similarity matching processing between the fault feature set and the fault phenomenon entities in the industrial knowledge graph to generate the fault phenomenon matching result corresponding to the fault feature set.

[0040] Let the set of fault features be F = {f1, f2, ..., fn}, where fi represents the i-th fault feature value and n is the number of fault features. The set of fault phenomenon entities in the industrial knowledge graph is P = {p1, p2, ..., pm}, where pj represents the j-th fault phenomenon entity and m is the number of fault phenomenon entities.

[0041] To perform similarity matching, the cosine similarity calculation method can be used. For the fault feature set F and each fault phenomenon entity pj, they are represented as vectors. Assume the fault feature vector is VF = (f1, f2, ..., fn), and the fault phenomenon entity vector is VPj = (pj1, pj2, ..., pjn), where pjk represents the value of the j-th fault phenomenon entity in the k-th feature dimension.

[0042] The cosine similarity between the fault feature vector VF and the fault phenomenon entity vector VPj is calculated using the formula Cos_Sim(VF, VPj)=(VF·VPj) / (||VF||×||VPj||), where VF·VPj represents the dot product of the vectors, and ||VF|| and ||VPj|| represent the magnitudes of the vectors, respectively. The dot product is calculated by multiplying the values ​​of the corresponding dimensions of the two vectors and then adding them together, i.e., VF·VPj=f1×pj1+f2×pj2+…+fn×pjn; the magnitude of the vectors is calculated by adding the squares of the values ​​of each dimension of the vectors and taking the square root, i.e., ||VF||=√(f1²+f2²+…+fn²), ||VPj||=√(pj1²+pj2²+…+pjn²).

[0043] By calculating the cosine similarity between the fault feature set and each fault phenomenon entity, a similarity score sequence S = {S1, S2, ..., Sm} is obtained, where Sj represents the similarity score between the fault feature set and the j-th fault phenomenon entity. Based on the similarity score, the matching result of the fault phenomenon corresponding to the fault feature set can be determined. For example, the top few fault phenomenon entities with the highest similarity scores can be selected as the matching results.

[0044] Step S122: Based on the fault phenomenon matching results, locate the set of fault cause entities associated with the fault feature set in the industrial knowledge graph.

[0045] Based on the fault phenomenon matching results obtained in the previous step, the fault cause entities that are associated with these fault phenomenon entities are searched in the industrial knowledge graph. The industrial knowledge graph records the causal or association relationships between fault phenomena and fault causes. For each matched fault phenomenon entity, the associated fault cause entities are found through the relationship chains in the knowledge graph.

[0046] Let the set of matched fault phenomenon entities be P_match = {p_match1, p_match2, ..., p_matchs}, where p_matchi represents the i-th matched fault phenomenon entity, and s is the number of matched fault phenomenon entities. For each p_matchi, search the knowledge graph for the set of fault cause entities Ci = {ci1, ci2, ..., cik} associated with it, where cij represents the j-th fault cause entity associated with p_matchi, and k is the number of fault cause entities associated with p_matchi.

[0047] Merge the sets of fault cause entities corresponding to all matched fault phenomenon entities, remove duplicate entities, and obtain the set of fault cause entities C = {c1, c2, ..., ct} associated with the fault feature set, where ct represents the t-th fault cause entity and t is the total number of fault cause entities.

[0048] Step S123: Perform causal chain analysis on the set of fault-causing entities to generate causal dependency features of the set of fault-causing entities in the industrial knowledge graph.

[0049] Step S1231: Extract the causal chain template set corresponding to the fault cause entity set from the industrial knowledge graph, and remove causal paths containing circular dependencies through a loop detection algorithm. The causal chain template set retains acyclic bidirectional causal paths.

[0050] In an industrial knowledge graph, each entity causing a failure may have causal relationships with other entities, and these causal relationships constitute causal chains. For a set of failure-causing entities C = {c1, c2, ..., ct}, all causal chains related to these entities are extracted from the knowledge graph. Let the set of causal chain templates be T = {T1, T2, ..., Tu}, where Ti represents the i-th causal chain template and u is the number of causal chain templates.

[0051] To ensure the validity of causal chains, loop detection algorithms are needed to eliminate causal paths containing circular dependencies. A circular dependency refers to a closed loop in the causal chain, where an entity returns to itself through a series of causal relationships. Loop detection algorithms can be implemented using depth-first search or breadth-first search methods. For each causal chain template Ti, a loop detection algorithm is used. If a circular dependency is found, the causal chain template is removed from the causal chain template set T. The final causal chain template set T only retains acyclic bidirectional causal paths.

[0052] Step S1232: Traverse each causal path in the causal chain template set and extract the causal transmission direction features between entities in the industrial knowledge graph of the fault cause entity set.

[0053] For each causal chain template Ti in the causal chain template set T, analyze the causal relationships within it to determine the direction of causal transmission between entities. In a causal chain, each causal relationship has a cause entity and a result entity, and the direction from the cause entity to the result entity is the causal transmission direction.

[0054] Let the set of entities contained in the causal chain template Ti be Ei = {ei1, ei2, ..., eiv}, where eij represents the j-th entity in the i-th causal chain template, and v is the number of entities in the causal chain template. For each pair of adjacent entities eij and eij+1 in the causal chain template Ti, if eij is the cause entity and eij+1 is the result entity, then the causal transmission direction is recorded as from eij to eij+1; otherwise, the causal transmission direction is recorded as from eij+1 to eij.

[0055] By summarizing the causal transmission direction information between entities in all causal chain templates, we obtain the causal transmission direction feature D = {d1, d2, ..., dw} of the fault cause entity set in the industrial knowledge graph, where dw represents the w-th causal transmission direction information and w is the total number of causal transmission direction information.

[0056] Step S1233: Calculate the causal strength parameter between each entity in the fault cause entity set based on the causal transmission direction characteristics between the entities. The causal strength parameter reflects the direct or indirect causal force between the two entities.

[0057] To calculate the causal strength parameter between entities, multiple factors can be considered, such as the frequency of occurrence of causal relationships and the confidence level of causal relationships. Let the set of entities causing failures be C = {c1, c2, ..., ct}. For entity pairs (ci, cj), calculate the causal strength parameter Iij between them.

[0058] First, count the number of times (Nij1) entity ci appears as the cause entity and entity cj appears as the result entity in all causal chain templates; simultaneously, count the number of times (Nij2) entity cj appears as the cause entity and entity ci appears as the result entity. Then, consider the confidence level of the causal relationship. The confidence level can be determined based on factors such as the reliability of the source of the causal relationship in the knowledge graph and the number of verifications. Let the confidence level of the causal relationship between entity pair (ci, cj) be Cij.

[0059] The causal strength parameter Iij can be calculated by comprehensively considering the frequency of occurrence and the confidence level. For example, a weighted average method can be used: Iij = (α × Nij1 + β × Nij2) × Cij, where α and β are weighting coefficients, and α + β = 1. Through this calculation, the causal strength parameter matrix I between entities in the fault cause entity set is obtained, where the element Iij in the matrix represents the causal strength parameter between entity ci and entity cj.

[0060] Step S1234: Generate a multidimensional causal association matrix of the fault cause entity set in the industrial knowledge graph based on the causal transmission direction feature and the causal strength parameter.

[0061] We have already obtained the causal transmission direction characteristics and causal strength parameters among the entities in the fault-causing entity set. Now, we need to integrate this information to construct a multidimensional causal association matrix. Assume the fault-causing entity set is represented by letters, denoted as C, which contains multiple fault-causing entities, sequentially labeled c1, c2, and so on up to ct. The multidimensional causal association matrix is ​​denoted as M.

[0062] Each element Mij in matrix M represents the causal relationship between entities ci and cj. When determining the value of Mij, both the causal transmission direction characteristics and the causal strength parameter must be considered.

[0063] From the perspective of causal transmission direction characteristics, if the causal transmission direction characteristics indicate that entity ci is the cause of entity cj, then Mij will show a causal direction relationship from ci to cj; if entity cj is the cause of entity ci, then the causal direction of Mij will show a causal direction relationship from cj to ci; if there is no causal transmission relationship between the two, then the causal direction information in this part of the matrix will show no obvious direction or a zero value (representing no causal relationship).

[0064] Furthermore, considering the causal strength parameter, let's assume the causal strength parameter between entities ci and cj is denoted as Iij. If the causal transmission direction is from ci to cj, then the value of Mij will be determined based on the causal strength parameter Iij, and will include an identifier representing the direction from ci to cj. For example, we can define that when the causal direction is from ci to cj, Mij equals Iij multiplied by a coefficient representing the positive causal direction (this coefficient can be a constant with a specific meaning used to clarify the direction information); if the causal direction is from cj to ci, Mij equals Iij multiplied by a coefficient representing the negative causal direction.

[0065] In constructing matrix M, this process is performed on each pair of entities (ci, cj) in the fault cause entity set C. That is, all combinations of i from 1 to t and j from 1 to t are considered sequentially. For each pair (ci, cj), the corresponding position in matrix M is filled using the method described above for determining the Mij value based on the causal transmission direction characteristics and causal strength parameters. After all (ci, cj) combinations have been processed, a complete multidimensional causal association matrix M is obtained. This multidimensional causal association matrix comprehensively reflects the causal relationships of the fault cause entity set in the industrial knowledge graph.

[0066] Step S1235: Perform matrix decomposition on the multidimensional causal correlation matrix to extract the global causal dependency pattern and local causal dependency pattern of the set of fault cause entities.

[0067] The purpose of matrix decomposition on the multidimensional causal relationship matrix M is to split the matrix M into several submatrices with specific meanings, thereby extracting the global causal dependency pattern and local causal dependency pattern of the set of fault-causing entities.

[0068] Matrix decomposition can employ several common methods, such as singular value decomposition (SVD) or eigenvalue decomposition (EVD). Taking SVD as an example, it decomposes matrix M into the product of three matrices, denoted as U, Σ, and V (transposes of M). The column vectors of matrix U are the eigenvectors of the matrix obtained by multiplying M by its transpose; the column vectors of matrix V are the eigenvectors of the matrix obtained by multiplying the transpose of M by M; and matrix Σ is a diagonal matrix whose diagonal elements are the singular values ​​of matrix M.

[0069] Global and local causal dependency patterns are extracted from the matrices obtained from the decomposition. Global causal dependency patterns reflect the overall causal dependencies of the set of fault-causing entities, and can typically be extracted from the eigenvectors corresponding to large singular values ​​in matrices U and V. This is because large singular values ​​indicate that this part plays a significant role in the overall causal structure, and its corresponding eigenvectors contain information that has a substantial impact on the causal relationships of the entire set of fault-causing entities.

[0070] Local causal dependency patterns focus on the causal dependencies between a subset of entities within the set of entities causing the failure. These patterns can be extracted by analyzing the eigenvectors corresponding to smaller singular values ​​in matrices U and V. The eigenvectors corresponding to these smaller singular values ​​may only be related to a subset of entities; the causal relationships formed between these subsets constitute the local causal dependency pattern.

[0071] In the specific extraction process, for the extraction of global causal dependency patterns, we can first identify the larger singular values ​​in matrix Σ, determine the positions of these singular values ​​in the matrix, and then find the corresponding eigenvectors in matrices U and V based on these positions. Integrating and analyzing the information represented by these eigenvectors yields the global causal dependency pattern. For example, we can observe the values ​​of each element in the eigenvectors; entities with larger element values ​​may play a more crucial role in the global causal relationship. Analyzing the associations between these entities allows us to summarize the global causal dependency pattern.

[0072] To extract local causal dependency patterns, we similarly identify smaller singular values ​​in matrix Σ and find their corresponding eigenvectors in matrices U and V based on their positions. We analyze these eigenvectors to see which entities have non-zero element values ​​or obvious numerical characteristics; these entities may have local causal dependencies. We then organize and summarize these local causal dependencies to obtain the local causal dependency patterns.

[0073] Step S1236: Merge the global causal dependency pattern and the local causal dependency pattern into the causal dependency feature.

[0074] After obtaining the global causal dependency pattern and the local causal dependency pattern, they need to be merged to form the causal dependency feature. The merging process must ensure that the information from both patterns is fully preserved and that the relationship between them is clearly reflected.

[0075] First, the global and local causal dependency patterns are encoded. An encoding rule can be designed for each pattern, allowing the causal relationships to be represented in an ordered manner. For example, for the global causal dependency pattern, the causal relationships between entities can be encoded using specific symbols and orders according to their order in set C; similarly, for the local causal dependency pattern, the encoding follows the order of the relevant entities.

[0076] Then, during the merging process, a concatenation method can be used, which involves sequentially connecting the encodings of the global causal dependency pattern and the local causal dependency pattern. During this connection process, separators can be added to differentiate between the encodings of the global and local causal dependency patterns. This facilitates the distinction and identification of the two patterns when using causal dependency features later.

[0077] After merging, a causal dependency feature is obtained. This feature integrates global and local causal dependency information of the set of fault-causing entities, and can more comprehensively reflect the causal relationship between fault-causing entities.

[0078] Step S124: Perform weight allocation processing based on the causal dependency features and the fault feature set to generate the causal influence distribution features corresponding to the fault cause entity set.

[0079] Step S1241: Extract the influence weight benchmark value of each fault feature dimension from the fault feature set.

[0080] The fault feature set contains multiple fault feature dimensions, each of which has a certain influence on the occurrence and development of faults. In order to determine the baseline value of the influence weight of each fault feature dimension, an in-depth analysis of the fault feature set is required.

[0081] Let F be the set of fault features, which contains multiple fault feature dimensions, denoted as f1, f2, and so on, up to fn. The influence weight baseline value of each fault feature dimension can be determined through statistical analysis of historical fault data. For example, this involves statistically analyzing the frequency of each fault feature dimension in past fault cases and its correlation with the severity of the fault.

[0082] For each fault feature dimension *fi*, the frequency of its occurrence in all historical fault cases is recorded as *Ni*, and the sum of the fault severity corresponding to each occurrence of this feature dimension is recorded as *Si*. By calculating the ratio of *Si* to *Ni*, a preliminary measure of the impact of this feature dimension on the fault can be obtained. Then, this measure for all fault feature dimensions is normalized so that the sum of all impact weight baseline values ​​is 1. The normalization process can be achieved by dividing the measure of each feature dimension by the sum of the measures of all feature dimensions. After this processing, the impact weight baseline value for each fault feature dimension is obtained, denoted as *wi*, where *i* ranges from 1 to *n*.

[0083] Step S1242: Use an attention mechanism to perform dynamic dimension alignment mapping on the global causal dependency pattern in the causal dependency feature and the fault feature dimension to generate a causal mapping relationship between the fault cause entity set and the fault feature set.

[0084] The attention mechanism allows this embodiment to better focus on the correlation between the global causal dependency pattern and the fault feature dimension in the causal dependency features. During dynamic dimension alignment mapping, the global causal dependency pattern and the fault feature dimension must first be represented as features.

[0085] The global causal dependency pattern has already been encoded and represented as a series of symbols and sequence information. Similarly, the fault feature dimension can be represented as a vector, where each element represents an attribute or feature value of that dimension.

[0086] When employing an attention mechanism, an attention score is calculated between the global causal dependency pattern and each fault feature dimension. The attention score reflects the degree of attention the global causal dependency pattern pays to each fault feature dimension. This attention score can be calculated using an attention function that takes the feature representation of the global causal dependency pattern and the vector representation of the fault feature dimension as input, and outputs an attention score.

[0087] After calculating the attention scores for all fault feature dimensions, these attention scores are normalized so that their sum is 1. The normalized attention scores can be understood as the weight allocation of the global causal dependency pattern for each fault feature dimension.

[0088] Based on these weight assignments, a dynamic dimensional alignment mapping is performed between the global causal dependency pattern and the fault feature dimensions. That is, based on the attention score, the causal relationship information in the global causal dependency pattern is mapped to the corresponding fault feature dimension. For example, if a fault feature dimension has a high attention score, then the causal relationship information related to it in the global causal dependency pattern will be more closely associated with that fault feature dimension.

[0089] After such dynamic dimension alignment mapping, a causal mapping relationship is generated from the set of fault cause entities to the set of fault features. This causal mapping relationship shows how the causal relationships in the set of fault cause entities are related to each feature dimension in the set of fault features.

[0090] Step S1243: Calculate the causal matching score of each fault cause entity in the fault cause entity set on the fault feature dimension based on the causal mapping relationship.

[0091] After obtaining the causal mapping relationship from the set of fault-causing entities to the set of fault features, we can calculate the causal matching score of each fault-causing entity in the set of fault-causing entities on the fault feature dimension.

[0092] For each fault-causing entity ci in the fault-causing entity set C, determine the degree of association between it and each fault feature dimension fj in the fault feature set F, based on the causal mapping relationship. The degree of association can be obtained from the causal mapping relationship, for example, by examining the connection strength or association identifier between ci and fj in the causal mapping relationship.

[0093] Then, for each fault feature dimension fj, the causal matching score of the fault-causing entity ci on that fault feature dimension is calculated by combining the previously obtained influence weight benchmark value wj. The correlation degree can be multiplied by the influence weight benchmark value to obtain the causal matching score of ci on fj.

[0094] This calculation is performed for all fault feature dimensions (j from 1 to n) to obtain the causal matching score sequence of the fault causal entity ci on all fault feature dimensions. The same calculation is performed for each fault causal entity in the set of fault causal entities C, finally resulting in a matrix where the rows of the matrix represent fault causal entities, the columns represent fault feature dimensions, and the elements in the matrix are the causal matching score of each fault causal entity on each fault feature dimension.

[0095] Step S1244: Perform linear superposition processing based on the causal matching score and the influence weight benchmark value to generate the preliminary weight distribution of each fault cause entity in the fault cause entity set.

[0096] After obtaining the causal matching degree score matrix of each fault cause entity in the fault cause entity set on the fault feature dimension, the matrix is ​​linearly superimposed with the influence weight benchmark value of each fault feature dimension.

[0097] For each fault-causing entity ci in the set of fault-causing entities C, its causal matching score on each fault feature dimension is multiplied by the corresponding influence weight benchmark value for that fault feature dimension, and then these products are summed. For example, if the causal matching score of ci on fault feature dimension fj is mij, and the influence weight benchmark value for fault feature dimension fj is wj, then the initial weight calculation method for ci is to multiply mij by wj, and then sum over all j from 1 to n.

[0098] This calculation is performed on each fault-causing entity in the fault-causing entity set C to obtain a preliminary weight for each entity. These preliminary weights are then combined to generate a preliminary weight distribution for each fault-causing entity in the set. This preliminary weight distribution reflects the initial influence of each entity on the occurrence of the fault, taking into account both the influence weights of fault feature dimensions and the causal matching degree.

[0099] Step S1245: Normalize the preliminary weight distribution and generate the cause influence distribution feature corresponding to the fault cause entity set based on the normalization result.

[0100] The initial weight values ​​in the initial weight distribution may have a large range. In order to better conduct subsequent analysis and comparison, they need to be normalized. The purpose of normalization is to map the weight values ​​in the initial weight distribution to a specific range, usually the interval [0, 1], and to make the sum of the normalized weight values ​​of all fault-causing entities equal to 1.

[0101] Normalization can be achieved using common methods, such as dividing the initial weight of each fault-causing entity by the sum of the initial weights of all fault-causing entities. Let pi be the initial weight of fault-causing entity ci in the set of fault-causing entities C, and let P be the sum of the initial weights of all fault-causing entities (P equals p1 + p2 + ... + pt). Then, the normalized weight qi of ci is equal to pi divided by P.

[0102] For each fault-causing entity in the fault-causing entity set C, such normalization calculation is performed to obtain a normalized weight value sequence. This normalized weight value sequence is the distribution characteristic of the causal influence degree corresponding to the fault-causing entity set, which shows the relative influence degree of each fault-causing entity in the overall fault-causing system.

[0103] Step S125: Based on the distribution characteristics of the cause influence degree, perform dynamic aggregation processing on the set of fault cause entities to generate the set of fault association features.

[0104] The distribution characteristics of causal influence reflect the relative influence of each fault-causing entity within the overall fault-causing system. Dynamic aggregation of the fault-causing entity set based on these characteristics aims to rationally combine related entities to generate a more representative and comprehensive set of fault-related features.

[0105] First, the set of fault-causing entities is sorted according to the distribution characteristics of their causal influence. Arranging these entities in descending order of their causal influence clearly shows which entities have a more significant impact on the fault.

[0106] Then, dynamic aggregation is performed based on the ranking results. An aggregation threshold can be set, prioritizing the aggregation of fault-causing entities with high causal influence. Aggregation can be achieved by concatenating or merging the feature information of these entities. For example, if the feature information of the fault-causing entities can be represented by vectors, these vectors can be concatenated to form a higher-dimensional vector, which represents the features of the aggregated fault-causing entity group.

[0107] For fault-causing entities with low causal impact, the decision to add them to an existing entity group can be based on their correlation with other aggregated entity groups. The correlation can be determined by analyzing factors such as causal relationships and semantic connections within the industrial knowledge graph. If the correlation is high, they are added to the corresponding entity group; if the correlation is low, they can be grouped separately or left unaggregated for the time being.

[0108] During the aggregation process, the aggregated feature information must be continuously updated. Each time a new fault-causing entity is added to the aggregation group, the feature representation of the aggregation group must be recalculated to ensure that the features of the aggregation group accurately reflect the comprehensive information of all entities within the group.

[0109] After all the fault-causing entities have been processed according to the above rules, multiple aggregated entity groups are obtained. Combining the feature information of these entity groups generates a fault association feature set. This fault association feature set integrates the information of the fault-causing entities and, through dynamic aggregation processing, highlights important fault-causing relationships.

[0110] Step S130: Perform fault propagation path reasoning processing based on the fault association feature set to generate a set of potential fault propagation paths for the target industrial equipment in the industrial knowledge graph.

[0111] Step S131: Determine the initial set of fault nodes in the industrial knowledge graph based on the set of fault association features.

[0112] The fault association feature set contains comprehensive information related to the fault, which can be used to determine the initial set of fault nodes in the industrial knowledge graph. The industrial knowledge graph is a knowledge base containing various industrial entities and their relationships. Nodes represent various entities, such as fault phenomena, fault causes, and equipment components, while edges represent relationships between entities, such as causal relationships and correlational relationships.

[0113] In an industrial knowledge graph, nodes that match the feature information in the fault-related feature set are searched. Matching methods can be based on semantic matching, feature similarity matching, etc. For example, if the fault-related feature set contains feature descriptions of certain fault causes, nodes with similar feature descriptions in the industrial knowledge graph are likely to be fault-related nodes.

[0114] For the found matching nodes, based on the importance of each feature in the fault association feature set (such as the previously obtained causal influence distribution features), nodes highly correlated with important features are selected as initial fault nodes. These initial fault nodes are the starting points of fault propagation, representing potential root causes or early manifestations of faults in the industrial knowledge graph. These selected nodes are then combined to generate the initial fault node set.

[0115] Step S132: Perform neighborhood expansion processing on the initial set of faulty nodes to generate an expanded set of nodes within a preset inference depth.

[0116] Once the initial set of faulty nodes is determined, its neighborhood needs to be expanded to identify possible fault propagation paths. The preset inference depth is a pre-defined parameter that specifies the maximum number of steps to expand from the initial faulty node.

[0117] Starting with each node in the initial set of faulty nodes, find its directly connected neighbor nodes in the industrial knowledge graph. These neighbor nodes are connected to the initial faulty node through edges in the knowledge graph, representing relationships between nodes such as causal relationships and association relationships. Add these neighbor nodes to a temporary extended node set.

[0118] Then, a new round of neighborhood lookup is performed on the nodes in the temporary expanded node set. That is, the neighboring nodes of these nodes are found, and the newly found neighboring nodes are added to the temporary expanded node set. This process is repeated until the preset inference depth is reached.

[0119] During the expansion process, care should be taken to avoid adding duplicate nodes and to ensure that all nodes in the expanded node set are unique. A marking mechanism can be used to mark nodes that have already been processed, so that they will not be searched again in subsequent expansion processes.

[0120] Once the preset inference depth is reached, the resulting temporary extended node set becomes the extended node set of the initial fault node set within the preset inference depth. This extended node set contains all nodes that can be reached from the initial fault node within a preset number of steps, and these nodes constitute the potential propagation range of the fault. The nodes in the extended node set may represent different fault-related entities, such as different fault causes, fault phenomena, or equipment components. By analyzing these nodes, the propagation path of the fault in the industrial knowledge graph can be further explored.

[0121] Step S133: Perform transition probability calculation on the adjacent nodes in the extended node set to generate the path transition probability for each pair of adjacent nodes in the extended node set.

[0122] To determine the likelihood of a fault propagating within the expanded node set, it is necessary to calculate the path transition probability between each pair of adjacent nodes. This can be achieved by considering multiple factors, and the specific steps are as follows:

[0123] Step S1331: Obtain the historical co-occurrence frequency of each pair of adjacent nodes in the industrial knowledge graph in the extended node set, and generate co-occurrence frequency parameters between adjacent nodes.

[0124] In the historical data of the industrial knowledge graph, the frequency of co-occurrence of each pair of adjacent nodes is counted. Let the extended node set contain node A and its adjacent node B. The number of times nodes A and B co-occur in past records is denoted as N_AB. Simultaneously, the total number of times node A appears is denoted as N_A. The co-occurrence frequency parameter P_AB_cooccur between adjacent nodes can be determined by the ratio of N_AB to N_A, i.e., by dividing the number of times nodes A and B co-occur by the total number of times node A appears, thus obtaining the co-occurrence frequency parameter reflecting their frequency of co-occurrence in historical data. This statistical calculation is performed on all adjacent node pairs in the extended node set to obtain the co-occurrence frequency parameter between each pair of adjacent nodes.

[0125] Step S1332: Extract the causal strength parameters of the adjacent nodes in the industrial knowledge graph, wherein the causal strength parameters reflect the transmission strength of the causal chain between nodes.

[0126] In industrial knowledge graphs, causal relationships between nodes have varying strengths. We previously calculated the causal strength parameters between entities when performing causal chain analysis on the set of entities causing failures. For adjacent nodes in the extended node set, we directly extract their causal strength parameters from the industrial knowledge graph. Let the causal strength parameter between adjacent nodes A and B be P_AB_causal, which reflects the transmission strength of the causal chain from node A to node B, i.e., the probability that node A causes node B to occur.

[0127] Step S1333: Calculate the topological distance parameter of the adjacent nodes in the industrial knowledge graph, wherein the topological distance parameter is the reciprocal of the hop count of the shortest path between nodes.

[0128] In the topology of an industrial knowledge graph, the shortest path hop count between adjacent nodes is calculated. Let H_AB be the shortest path hop count between node A and node B, and let P_AB_to pology be the reciprocal of H_AB. The topological distance parameter reflects the closeness of nodes in the knowledge graph structure; the fewer the hop count, the closer the nodes are in the graph structure, and the larger the topological distance parameter.

[0129] Step S1334: After standardizing and transforming the co-occurrence frequency parameter, the causal strength parameter and the topological distance parameter, the initial transition probability of each pair of adjacent nodes is generated by weighted fusion processing through dynamic weight allocation rules.

[0130] Since the co-occurrence frequency parameter, causality strength parameter, and topological distance parameter may have different ranges and dimensions, they need to be standardized first. Standardization can be achieved using common methods, such as subtracting the mean from each parameter and then dividing by its standard deviation, to ensure that all parameters have the same scale. Let the standardized co-occurrence frequency parameter be P_AB_cooccur_std, the standardized causality strength parameter be P_AB_causal_std, and the standardized topological distance parameter be P_AB_topology_std.

[0131] Then, the three standardized parameters are weighted and fused using a dynamic weight allocation rule. This dynamic weight allocation rule can be adjusted according to the actual situation. For example, different weights can be assigned to the co-occurrence frequency parameter, causality strength parameter, and topological distance parameter based on the characteristics of different types of faults. Let the weights be w_cooccur, w_causal, and w_topology, respectively, and w_cooccur + w_causal + w_topology = 1. The initial transition probability P_AB_initial for each pair of adjacent nodes can be calculated as follows: multiply the standardized co-occurrence frequency parameter by its corresponding weight, add the standardized causality strength parameter multiplied by its corresponding weight, and add the standardized topological distance parameter multiplied by its corresponding weight, i.e., P_AB_initial = w_cooccur * P_AB_cooccur_std + w_causal * P_AB_causal_std + w_topology * P_AB_topology_std.

[0132] Step S1335: Perform neighborhood normalization on the initial transition probability to generate the path transition probability of each pair of adjacent nodes in the extended node set.

[0133] For each node in the expanded node set, the sum of its initial transition probabilities from all its neighboring nodes may not be 1. To ensure that these probabilities have a reasonable distribution, neighborhood normalization is required for the initial transition probabilities.

[0134] Suppose node A has neighboring nodes B1, B2, ..., Bk, with initial transition probabilities to node A of B1, B2, ..., P_A_Bk_initial, respectively. The sum of these initial transition probabilities is denoted as S_A = P_A_B1_initial + P_A_B2_initial + ... + P_A_Bk_initial.

[0135] The path transition probability P_A_Bi from node Bi to node A can be obtained by dividing its initial transition probability by the sum S_A, i.e., P_A_Bi = P_A_Bi_initial / S_A. This normalization process is performed on all adjacent node pairs of all nodes in the extended node set to obtain the path transition probability for each pair of adjacent nodes.

[0136] Step S134: Perform path search processing on the extended node set according to the path transition probability to generate a set of candidate propagation paths for the target industrial equipment in the industrial knowledge graph.

[0137] After obtaining the path transition probability of each pair of adjacent nodes in the extended node set, path search can be performed based on these probabilities to find possible fault propagation paths.

[0138] Graph search algorithms, such as breadth-first search or improved versions of depth-first search, can be used. Taking breadth-first search as an example, the path starts from a node in the initial set of faulty nodes.

[0139] For each starting node, based on the path transition probability, select the neighboring node with the higher transition probability as the next path node. Combine the starting node and the selected neighboring nodes to form a preliminary path.

[0140] Then, using the newly added neighboring node as the current node, the next neighboring node is selected again based on the path transition probability and added to the path to form a longer path. This process is repeated until the preset path length is reached or no more suitable neighboring nodes can be found.

[0141] During the search process, all generated paths are recorded. These paths may have different lengths and node combinations. By summarizing them, a set of candidate propagation paths for the target industrial equipment in the industrial knowledge graph is obtained.

[0142] Step S135: Perform path validity verification processing on the candidate propagation path set to generate the potential fault propagation path set.

[0143] The candidate propagation path set may contain some paths that are not realistic or unreasonable, and their effectiveness needs to be verified.

[0144] First, check whether the nodes in the path conform to the causal logic in the industrial knowledge graph. For example, in the knowledge graph, some nodes do not have a causal relationship, but such a connection appears in the candidate path. Such a path is invalid and needs to be removed from the candidate propagation path set.

[0145] Secondly, consider the sum of the transition probabilities of the paths. For each candidate path, multiply the path transition probabilities of the adjacent nodes in the path to obtain the sum of the transition probabilities of the path. If the sum of the transition probabilities is lower than a preset threshold, it means that the probability of this path occurring is very small, and it is also regarded as an invalid path and eliminated.

[0146] After such validity verification, the remaining candidate paths constitute the set of potential fault propagation paths. The paths in this set of potential fault propagation paths are selected based on a comprehensive consideration of path transition probability and causal logic of knowledge graph, and are more likely to be actual fault propagation paths.

[0147] Step S140: Perform solution matching processing on the set of potential fault propagation paths to generate a set of candidate solutions corresponding to each potential fault propagation path.

[0148] Step S141: Perform node parsing processing on each potential fault propagation path in the set of potential fault propagation paths to generate a fault node sequence corresponding to each potential fault propagation path.

[0149] The set of potential fault propagation paths contains multiple possible fault propagation paths, each consisting of a series of nodes. Node parsing for each potential fault propagation path involves extracting the nodes from the path in sequence to form a fault node sequence.

[0150] For example, if a potential fault propagation path contains nodes A, B, and C, then after performing node parsing on this path, the resulting sequence of faulty nodes is {A, B, C}. This node parsing process is performed on all paths in the set of potential fault propagation paths to obtain the sequence of faulty nodes for each potential fault propagation path.

[0151] Step S142: Based on the fault node sequence, retrieve historical solution records in the industrial knowledge graph to generate a historical solution reference set corresponding to each potential fault propagation path.

[0152] In addition to fault-related nodes and relationships, the industrial knowledge graph also records historical solutions for different faults. Based on the fault node sequence corresponding to each potential fault propagation path, the historical solution records related to these nodes are searched in the industrial knowledge graph.

[0153] For each node in the fault node sequence, find the associated solution node in the knowledge graph. These solution nodes represent solutions previously implemented for the fault situation represented by that node. By aggregating all solution nodes related to all nodes in the fault node sequence, a historical solution reference set corresponding to each potential fault propagation path is obtained.

[0154] Step S143: Perform semantic similarity calculation on the historical solution reference set to generate a current semantic matching solution set corresponding to each potential fault propagation path.

[0155] The historical solution reference set contains multiple solutions, but not all of these solutions may be suitable for the current failure situation. To select more suitable solutions, semantic similarity calculations need to be performed on the historical solution reference set.

[0156] Each solution in the historical solution reference set is represented in text form, for example, by converting the solution description into a vector format. Simultaneously, relevant information about the current fault is also represented in text form and converted into a vector.

[0157] Calculate the semantic similarity between each solution vector in the historical solution reference set and the current fault information vector. This can be done using methods such as cosine similarity, which measures the similarity between two vectors by calculating the cosine of the angle between them.

[0158] Based on the semantic similarity calculation results, solutions with higher similarity are selected to form the current semantic matching solution set. The solutions in this current semantic matching solution set are semantically closer to the current fault situation and are more likely to be effective solutions.

[0159] Step S144: Perform constraint filtering on the current semantic matching solution set to generate a subset of feasible solutions for each potential fault propagation path.

[0160] Step S1441: Obtain the set of equipment maintenance resource constraints for the target industrial equipment. The set of equipment maintenance resource constraints includes maintenance tool availability constraints, maintenance personnel skill constraints, and maintenance time window constraints.

[0161] The maintenance of target industrial equipment is subject to various resource constraints. The maintenance tool availability constraint refers to whether the types and quantities of currently available maintenance tools can meet the needs of the solution. The maintenance personnel skill constraint indicates whether the skills and expertise of maintenance personnel are sufficient to implement the corresponding solution. The maintenance time window constraint refers to the time frame within which the equipment can be maintained.

[0162] These constraint information are obtained from data sources such as equipment maintenance management systems, personnel skill files, and equipment operation plans, and then aggregated to form a set of equipment maintenance resource constraints.

[0163] Step S1442: Perform tool matching verification processing on each solution in the current semantic matching solution set to generate a first set of filtering solutions that meets the maintenance tool availability constraints.

[0164] For each solution in the current set of semantic matching solutions, analyze its required maintenance tools. Compare the required tools with the tools available in the maintenance tool availability constraints.

[0165] If all the tools required for a solution are available, the solution passes the tool matching verification and is added to the first set of filtered solutions. If some required tools are not available, the solution does not meet the maintenance tool availability constraint and is removed from the current semantic matching solution set.

[0166] Step S1443: Perform skill matching verification on each solution in the first set of filtering solutions to generate a second set of filtering solutions that meets the skill constraints of the maintenance personnel.

[0167] For each solution in the first set of filtering solutions, analyze its skill requirements for maintenance personnel. Compare these requirements with the skills required of maintenance personnel as specified in the maintenance personnel skill constraints.

[0168] If all the skills required for a solution are possessed by maintenance personnel, then the solution passes the skill matching verification and is added to the second set of filtered solutions. If some required skills are not possessed by maintenance personnel, then the solution does not meet the maintenance personnel skill constraints and is removed from the first set of filtered solutions.

[0169] Step S1444: Perform time window verification processing on each solution in the second filtering solution set to generate a third filtering solution set that conforms to the maintenance time window constraint.

[0170] Analyze the time required for each solution in the second set of filtering solutions. Compare the required time with the time range within the maintenance time window constraint during which the equipment can be maintained.

[0171] If a solution's required time can be completed within the equipment's maintainable time window, then the solution passes the time window validation and is added to the third set of filtered solutions. If the required time exceeds the time window range, then the solution does not meet the maintenance time window constraint and is removed from the second set of filtered solutions.

[0172] Step S1445: Generate the subset of feasible solutions based on the verification results of the third set of filtering solutions.

[0173] The solutions in the third filtering solution set simultaneously satisfy the constraints of maintenance tool availability, maintenance personnel skills, and maintenance time window. This third filtering solution set is then considered as a subset of feasible solutions for each potential fault propagation path.

[0174] Step S145: Combine and optimize the subset of feasible solutions according to a preset dimension to generate the set of candidate solutions.

[0175] The feasible solution subset contains multiple feasible solutions. In order to obtain a better combination of solutions, it is necessary to perform combination optimization processing according to preset dimensions.

[0176] The preset dimensions can include aspects such as the cost, effectiveness, and implementation difficulty of the solution. For each solution in the subset of feasible solutions, its performance on the preset dimensions is evaluated.

[0177] For example, in terms of cost, assess the material and human resource costs required for each solution; in terms of effectiveness, assess the likelihood and extent to which each solution resolves the failure; and in terms of implementation difficulty, assess the technical difficulty and operational complexity required for the solution.

[0178] Based on these evaluation results, combinatorial optimization algorithms, such as genetic algorithms and simulated annealing algorithms, are used to combine solutions from a subset of feasible solutions. By continuously adjusting the combination of solutions, the optimal combination is found across preset dimensions.

[0179] By combining these optimized solutions, a candidate solution set is generated. The solutions in this candidate solution set, after comprehensively considering various factors, are more likely to be effective solutions for potential fault propagation paths.

[0180] Step S150: Perform a feasibility assessment on the candidate solution set to generate an optimized solution set corresponding to the target industrial equipment.

[0181] Step S151: Perform cost parameter calculation processing on each solution in the candidate solution set to generate an economic evaluation index corresponding to each solution.

[0182] For each solution in the candidate solution set, calculate its cost parameters. Cost parameters can include multiple aspects, such as material costs, labor costs, and time costs.

[0183] Material costs refer to the expenses of various materials required to implement the solution, such as the cost of replacing parts and the cost of raw materials consumed. Labor costs refer to the salaries and other expenses of personnel involved in implementing the solution. Time costs refer to the costs incurred due to the time spent implementing the solution, such as production losses caused by equipment downtime.

[0184] These cost parameters are summarized and quantified to obtain the total cost of each solution. This total cost is then used as the economic evaluation metric for each solution, reflecting the economic cost of implementing that solution.

[0185] Step S152: Perform risk parameter prediction processing on each solution in the candidate solution set to generate a risk assessment index corresponding to each solution.

[0186] For each solution, risk parameters are predicted, taking into account various risks that may be encountered during implementation. Risks may include technical risks, security risks, environmental risks, etc.

[0187] Technical risk refers to the possibility that a solution may be technically unfeasible or fail to achieve the expected results. Safety risk refers to the possibility that the implementation of a solution may pose a threat to personnel safety. Environmental risk refers to the possibility that a solution may have adverse effects on the surrounding environment.

[0188] By analyzing historical data and referencing expert experience, these risk parameters for each solution are predicted and quantified. For example, an assessment value for the probability of occurrence and the degree of impact can be set for each risk type. These assessment values ​​are then combined to obtain a risk score for each solution. This risk score serves as a risk assessment indicator for each solution, reflecting the magnitude of the risk faced in implementing the solution.

[0189] Step S153: Perform execution difficulty analysis on each solution in the candidate solution set to generate an operational evaluation index for each solution.

[0190] Analyze the execution difficulty of each solution. Execution difficulty can be measured from multiple aspects, such as technical difficulty, operational complexity, and requirements for equipment and personnel.

[0191] Technical difficulty refers to the level of technical knowledge and skills required to implement the solution. Operational complexity refers to the complexity of the operational steps involved in implementing the solution. Equipment and personnel requirements refer to the requirements for the condition of the equipment, the number of personnel, and their qualifications in relation to implementing the solution.

[0192] Based on these assessments, an execution difficulty score is assigned to each solution. This execution difficulty score serves as an operational assessment metric for each solution, reflecting the ease or difficulty of implementing the solution.

[0193] Step S154: The economic evaluation index, the risk evaluation index and the operational evaluation index are respectively scored and converted to generate standardized scores.

[0194] Since the value ranges and dimensions of economic assessment indicators, risk assessment indicators, and operational assessment indicators may differ, it is necessary to convert them into standardized scores to facilitate comprehensive evaluation.

[0195] Scoring transformations can be performed using methods such as linear transformations. For example, for each evaluation indicator, its maximum and minimum values ​​can be identified, and the actual value of each indicator can be mapped to a preset scoring range, such as [0, 100]. This transformation ensures that each evaluation indicator has the same scoring range, facilitating subsequent comprehensive calculations.

[0196] Step S155: Assign dynamic weight parameters according to preset business priority rules, and perform weighted processing on the standardized score to generate a comprehensive score result.

[0197] The pre-defined business priority rules are determined based on the company's actual needs and strategic goals. For example, if the company currently focuses more on cost control, then the weight of economic evaluation indicators will be relatively high; if the company focuses more on safety and reliability, then the weight of risk evaluation indicators will be relatively high.

[0198] Based on the business priority rules, dynamic weight parameters are assigned to the economic assessment indicators, risk assessment indicators, and operational assessment indicators, respectively. Let the weights be w_economy, w_risk, and w_operability, and w_economy + w_risk + w_operability = 1.

[0199] For each solution, the standardized score of its economic evaluation indicator is multiplied by w_econom y, the standardized score of its risk evaluation indicator is multiplied by w_risk, and the standardized score of its operability evaluation indicator is multiplied by w_operability. Then, these three products are added together to obtain the comprehensive score of the solution.

[0200] Step S156: Sort the candidate solution set in descending order according to the comprehensive scoring results to generate a sorted list of the candidate solution set.

[0201] Each solution in the candidate solution set is ranked from highest to lowest according to its overall score. A higher score indicates that the solution performs better in terms of overall economic efficiency, risk, and operability.

[0202] The sorting process yields a sorted list of candidate solutions, which shows the order of merit of each solution.

[0203] Step S157: Extract the top solution group from the sorted list according to the preset filtering threshold to generate the optimized solution set.

[0204] The preset screening threshold is a standard set based on actual business needs and objectives, used to select the most suitable solution from the ranking list. When determining the screening threshold, factors such as the company's resources, the severity of the fault, and expectations for the solution's effectiveness need to be considered. For example, if the company has limited resources, the screening threshold may be set high, selecting only solutions with very high overall scores; if the fault is urgent and needs to be resolved quickly, the screening threshold may be appropriately lowered to obtain more available solutions.

[0205] Starting from the head of the sorted list, solutions are selected sequentially in descending order of their overall scores until either the number of selected solutions or the overall score reaches the filtering threshold. These selected solutions constitute the head solution group.

[0206] It is important to note that when selecting the top solution group, it is crucial to ensure that the chosen solutions are technically feasible and well-suited to the specific characteristics of the target industrial equipment. For example, some solutions may theoretically have a high overall score, but due to factors such as the equipment's unique structure or operating environment, they may be practically impossible to implement; such solutions should not be selected for the optimized solution set.

[0207] Once the leading solution group is determined, an optimized solution set corresponding to the target industrial equipment is generated. The solutions in this optimized solution set have been comprehensively weighed in terms of economy, risk and operability, and after screening, they are the relatively optimal solution combinations, which can provide effective guidance for solving the faults of the target industrial equipment.

[0208] Step S210: Perform execution effect tracking processing on the optimized solution set to generate an actual execution effect record for each solution in the optimized solution set.

[0209] When implementing various solutions within the optimization solution set, it is necessary to track their performance. For example, before implementing a solution, a series of evaluation indicators need to be clearly defined. These indicators may include the degree of fault repair, the time it takes for the equipment to return to normal operation, and the resource consumption during implementation. For instance, the degree of fault repair can be evaluated by comparing the various operating parameters of the equipment before and after the solution implementation; the time it takes for the equipment to return to normal operation can be recorded from the start of the solution implementation until the various indicators of the equipment return to normal range; and resource consumption can be calculated by tracking the costs of materials, manpower, and time used during implementation.

[0210] During solution implementation, relevant data is continuously collected. The operating status of the equipment can be monitored in real time through sensors on the equipment, and changes in various operating parameters can be recorded; at the same time, information such as the quantity of materials used and the working time of the personnel involved in the implementation process is also recorded.

[0211] After the solution is implemented, the effectiveness is evaluated based on the collected data and according to pre-defined evaluation indicators. The evaluation results are compiled into an actual implementation effectiveness record, which should include detailed basic information about each solution, key data during the implementation process, and the final evaluation results. For example, for a specific solution, the record may include the solution's name, implementation time, material costs consumed, the specific time it took for equipment to return to normal operation, and the specific extent of fault repair.

[0212] Step S220: Based on the actual execution effect record, perform entity relationship update processing on the industrial knowledge graph to generate an updated industrial knowledge graph.

[0213] Step S221: Perform feature parsing processing on the successful solutions in the actual execution effect record to generate feature enhancement parameters for the successful solutions.

[0214] A successful solution is one that achieves the expected results during actual implementation, effectively resolving faults or significantly improving equipment operation. Feature analysis is performed on these successful solutions to extract key features related to their success and quantify them as feature enhancement parameters.

[0215] First, conduct a detailed analysis of the actual implementation records of successful solutions to identify factors closely related to their success. These factors may include the technical means employed, the types of materials used, and the operational steps performed. For example, if a successful solution employed a novel maintenance technique during implementation, and this technique played a crucial role in fault repair, then this novel maintenance technique can be considered a key factor.

[0216] Next, these key factors are quantified. The importance of each key factor can be measured by setting quantitative indicators. For example, for the type of materials used, the frequency of that material's appearance in successful solutions can be statistically analyzed; for the operational steps performed, the contribution of each step to fault repair can be evaluated.

[0217] Combining these quantified key factors generates the feature enhancement parameters for a successful solution. These parameters will be used for subsequent enhancement processing of entity relationships in the industrial knowledge graph.

[0218] Step S222: Perform feature parsing processing on the failed solutions in the actual execution effect record to generate feature suppression parameters for the failed solutions.

[0219] Failed solutions refer to solutions that, in practice, fail to achieve the expected results, do not effectively resolve faults, or do not significantly improve equipment operation. Feature analysis is performed on these failed solutions to identify the key features that led to their failure and quantify them as feature suppression parameters.

[0220] Similarly, a thorough analysis of the actual implementation records of failed solutions is crucial to identify key factors related to the solution's failure. These factors may include limitations in technical means, material quality issues, and unreasonable operational procedures. For example, if a failed solution used substandard components during implementation, resulting in the failure to resolve the problem, then component quality issues could be considered a key factor.

[0221] These key factors are quantified, and corresponding quantitative indicators are set to measure the impact of each key factor on solution failure. For example, regarding the limitations of technical means, the probability that the technology cannot solve the problem in practical applications can be assessed; regarding unreasonable operational procedures, the negative impact of the procedure on the overall solution effect can be analyzed.

[0222] Combining these quantified key factors generates feature suppression parameters for failed solutions. These parameters will be used for subsequent weakening of entity relationships in the industrial knowledge graph.

[0223] Step S223: Perform weight enhancement processing on the corresponding entity relationships in the industrial knowledge graph according to the feature enhancement parameters to generate an enhanced entity relationship subgraph.

[0224] In industrial knowledge graphs, entity relationships are represented by edges, each with a corresponding weight that reflects the strength of the relationship between entities. Based on the feature enhancement parameters of successful solutions, the entity relationships in the industrial knowledge graph related to these features are weighted and strengthened.

[0225] For key factors involved in the feature enhancement parameters, the corresponding entities and entity relationships are found in the industrial knowledge graph. For example, if the feature enhancement parameters mention a new maintenance technology, then the entities related to this maintenance technology (such as the maintenance technology itself, solutions using the technology, related equipment components, etc.) and the relationships between them are found in the knowledge graph.

[0226] Based on the quantified value of each key factor in the feature enhancement parameters, the weights of these entity relationships are adjusted. Entity relationships can be strengthened by increasing their weights, and the magnitude of the increase can be determined according to the importance of the key factor. For example, for key factors that play a crucial role in the success of the solution, the corresponding entity relationship weight can be increased relatively more; for key factors with lower importance, the weight increase can be relatively smaller.

[0227] After weighting all relevant entity relationships, these strengthened entity relationships are extracted to form a strengthened entity relationship subgraph, which reflects the stronger associations between entities involved in a successful solution.

[0228] Step S224: Perform weight weakening processing on the corresponding entity relationships in the industrial knowledge graph according to the feature suppression parameters to generate a weakened entity relationship subgraph.

[0229] Similar to weight enhancement, the weights of entity relationships related to these features in the industrial knowledge graph are weakened based on the feature suppression parameters of the failed solutions.

[0230] In the industrial knowledge graph, identify the entities and entity relationships that correspond to the key factors in the feature suppression parameters. For example, if the feature suppression parameters mention component quality issues, then find the entities related to that component (such as the component itself, solutions using that component, related equipment parts, etc.) and the relationships between them.

[0231] Based on the quantified value of each key factor in the feature suppression parameters, the weights of these entity relationships are adjusted. Entity relationships are weakened by reducing their weights, with the reduction amount determined by the degree of influence of the key factor on the solution's failure. For example, the weight reduction can be relatively large for major key factors causing solution failure, and relatively small for key factors with less influence.

[0232] After weighting all relevant entity relationships, these weakened entity relationships are extracted to form a weakened entity relationship subgraph, which reflects the weaker associations between entities involved in the failed solution.

[0233] Step S225: Merge the enhanced entity relationship subgraph with the weakened entity relationship subgraph to generate the updated industrial knowledge graph.

[0234] When merging the enhanced and weakened entity relationship subgraphs, it is necessary to ensure the integrity and consistency of entity and entity relationship information during the merging process.

[0235] First, the entities in the two subgraphs are merged. For duplicate entities, only one is retained, and its related attribute information is merged. Then, the entity relationships in the two subgraphs are merged. For relationships between identical entity pairs, if they exist in both the strengthening and weakening subgraphs, they need to be updated based on the weighted results; if they exist only in one of the subgraphs, the relationship is directly retained.

[0236] During the merger process, it is also necessary to check whether the relationships between the merged entities are logical and consistent with reality. For example, check for unreasonable weight values ​​or contradictory relationships. If problems are found, appropriate adjustments and corrections need to be made.

[0237] After merging, an updated industrial knowledge graph is obtained, which reflects the latest information after adjusting entity relationships based on actual implementation results.

[0238] Step S230: Based on the updated industrial knowledge graph, perform correlation analysis on the subsequent fault data set to generate an optimized fault correlation feature set.

[0239] When a new set of subsequent fault data appears, the updated industrial knowledge graph is used to perform correlation analysis on it. For details, please refer to the description of the corresponding embodiment above, which will not be repeated here.

[0240] By using the updated industrial knowledge graph, the optimized fault association feature set can more accurately reflect the relationship between subsequent fault data sets and various entities in the industrial knowledge domain. By continuously updating the industrial knowledge graph and using the updated graph to analyze new fault data, the efficiency and accuracy of fault diagnosis and resolution can be continuously improved.

[0241] Figure 2 The diagram illustrates exemplary hardware and software components of an industrial fault multi-solution generation system 100 that integrates a knowledge graph to implement the inventive ideas, according to some embodiments of the present invention. For example, a processor 120 may be used in the industrial fault multi-solution generation system 100 integrating a knowledge graph and to perform the functions of the present invention.

[0242] The industrial fault multi-solution generation system 100 integrating knowledge graphs may include a network port 110 connected to a network, one or more processors 120 for executing program instructions, a communication bus 130, and various forms of storage media 140, such as a disk, ROM, or RAM, or any combination thereof. Exemplarily, the industrial fault multi-solution generation system 100 integrating knowledge graphs may also include program instructions stored in ROM, RAM, or other types of non-transitory storage media, or any combination thereof. The method of the present invention can be implemented according to these program instructions. The industrial fault multi-solution generation system 100 integrating knowledge graphs also includes an I / O interface 150 between the computer and other input / output devices.

[0243] Furthermore, this embodiment of the invention also provides a readable storage medium, wherein computer-executable instructions are preset in the readable storage medium, and when the processor executes the computer-executable instructions, the above-mentioned method for generating multiple solutions for industrial faults by integrating knowledge graphs is implemented.

[0244] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

Claims

1. A method for generating multiple solutions to industrial faults by integrating knowledge graphs, characterized in that, The method includes: Obtain a set of fault data for the target industrial equipment, perform feature extraction processing on the set of fault data, and generate a set of fault features corresponding to the set of fault data. The pre-built industrial knowledge graph is invoked to perform entity association analysis on the fault feature set, generating a fault association feature set of the fault feature set in the industrial knowledge graph; Based on the fault association feature set, fault propagation path reasoning is performed to generate a set of potential fault propagation paths for the target industrial equipment in the industrial knowledge graph. The set of potential fault propagation paths is subjected to solution matching processing to generate a set of candidate solutions corresponding to each potential fault propagation path; The candidate solution set is subjected to a feasibility assessment to generate an optimized solution set corresponding to the target industrial equipment.

2. The method for generating multiple solutions to industrial faults by integrating knowledge graphs according to claim 1, characterized in that, The step of performing feature extraction processing on the fault data set to generate a fault feature set corresponding to the fault data set includes: Time series analysis is performed on the equipment operating parameters in the fault data set to generate the operating status fluctuation characteristics of the target industrial equipment in the time dimension. The abnormal equipment signals in the fault data set are subjected to frequency domain transformation to generate the energy distribution characteristics of the target industrial equipment in the frequency domain dimension. Semantic parsing is performed on the equipment operation logs in the fault data set to generate behavioral pattern features of the target industrial equipment in the operation dimension. The equipment environmental parameters in the fault data set are subjected to trend fitting processing to generate the external interference characteristics of the target industrial equipment in the environmental dimension. The operational status fluctuation characteristics, energy distribution characteristics, behavioral pattern characteristics, and external disturbance characteristics are standardized and transformed respectively, and then merged into the fault feature set.

3. The method for generating multiple solutions to industrial faults by integrating knowledge graphs according to claim 1, characterized in that, The step of calling a pre-built industrial knowledge graph to perform entity association analysis on the fault feature set, generating a fault association feature set of the fault feature set in the industrial knowledge graph, includes: The fault feature set is matched with the fault phenomenon entities in the industrial knowledge graph to generate the fault phenomenon matching result corresponding to the fault feature set. Based on the fault phenomenon matching results, locate the set of fault cause entities associated with the fault feature set in the industrial knowledge graph; A causal chain analysis is performed on the set of fault-causing entities to generate causal dependency features of the set of fault-causing entities in the industrial knowledge graph. Based on the causal dependency features and the fault feature set, a weight allocation process is performed to generate the causal influence distribution features corresponding to the fault cause entity set. Based on the distribution characteristics of the cause influence degree, the set of fault cause entities is dynamically aggregated to generate the set of fault association features. The step of performing causal chain analysis on the set of fault-causing entities to generate causal dependency features of the set of fault-causing entities in the industrial knowledge graph includes: The causal chain template set corresponding to the fault cause entity set is extracted from the industrial knowledge graph, and causal paths containing circular dependencies are removed by loop detection algorithm. The causal chain template set retains acyclic bidirectional causal paths. Traverse each causal path in the causal chain template set and extract the causal transmission direction features between entities in the industrial knowledge graph of the fault cause entity set; Based on the causal transmission direction characteristics between entities, the causal strength parameter between each entity in the fault cause entity set is calculated. The causal strength parameter reflects the direct or indirect causal force between two entities. Based on the causal transmission direction characteristics and the causal strength parameters, a multidimensional causal association matrix of the fault cause entity set in the industrial knowledge graph is generated. The multidimensional causal relationship matrix is ​​subjected to matrix decomposition to extract the global causal dependency pattern and local causal dependency pattern of the set of fault cause entities. The global causal dependency pattern and the local causal dependency pattern are merged into the causal dependency feature; The step of performing weight allocation processing based on the causal dependency features and the fault feature set to generate the causal influence distribution features corresponding to the fault cause entity set includes: Extract the influence weight baseline value for each dimension of the fault feature from the fault feature set; An attention mechanism is used to perform dynamic dimension alignment mapping between the global causal dependency pattern in the causal dependency feature and the fault feature dimension, thereby generating a causal mapping relationship between the fault cause entity set and the fault feature set. Based on the causal mapping relationship, calculate the causal matching score of each fault-causing entity in the fault-causing entity set on the fault feature dimension; The preliminary weight distribution of each fault-causing entity in the fault-causing entity set is generated by linearly superimposing the causal matching score and the influence weight benchmark value. The preliminary weight distribution is normalized, and the causal influence distribution feature corresponding to the fault cause entity set is generated based on the normalization result.

4. The method for generating multiple solutions to industrial faults by integrating knowledge graphs according to claim 1, characterized in that, The step of performing fault propagation path reasoning based on the fault association feature set to generate a set of potential fault propagation paths for the target industrial equipment in the industrial knowledge graph includes: Based on the set of fault association features, an initial set of fault nodes is determined in the industrial knowledge graph; The initial set of faulty nodes is subjected to neighborhood expansion processing to generate an expanded set of nodes within a preset inference depth. The transition probability is calculated for adjacent nodes in the extended node set to generate the path transition probability for each pair of adjacent nodes in the extended node set. Based on the path transition probability, a path search process is performed on the extended node set to generate a set of candidate propagation paths for the target industrial equipment in the industrial knowledge graph. The candidate propagation path set is subjected to path validity verification processing to generate the potential fault propagation path set; The step of calculating the transition probability of adjacent nodes in the expanded node set to generate the path transition probability of each pair of adjacent nodes in the expanded node set includes: Obtain the historical co-occurrence frequency of each pair of adjacent nodes in the industrial knowledge graph from the extended node set, and generate co-occurrence frequency parameters between adjacent nodes; The causal strength parameters of the adjacent nodes in the industrial knowledge graph are extracted, and the causal strength parameters reflect the transmission strength of the causal chain between nodes; Calculate the topological distance parameter of the adjacent nodes in the industrial knowledge graph, where the topological distance parameter is the reciprocal of the hop count of the shortest path between nodes; After standardization transformation based on the co-occurrence frequency parameter, the causal strength parameter, and the topological distance parameter, a weighted fusion process is performed using a dynamic weight allocation rule to generate the initial transition probability for each pair of adjacent nodes. The initial transition probabilities are normalized by neighborhood to generate the path transition probabilities for each pair of adjacent nodes in the expanded node set.

5. The method for generating multiple solutions to industrial faults by integrating knowledge graphs according to claim 1, characterized in that, The step of performing solution matching processing on the set of potential fault propagation paths to generate a set of candidate solutions corresponding to each potential fault propagation path includes: For each potential fault propagation path in the set of potential fault propagation paths, perform node parsing processing to generate a fault node sequence corresponding to each potential fault propagation path; Based on the fault node sequence, historical solution records are retrieved from the industrial knowledge graph to generate a historical solution reference set corresponding to each potential fault propagation path. The historical solution reference set is subjected to semantic similarity calculation to generate the current semantic matching solution set corresponding to each potential fault propagation path; The current set of semantic matching solutions is subjected to constraint filtering to generate a subset of feasible solutions for each potential fault propagation path; The subset of feasible solutions is combined and optimized according to a preset dimension to generate the set of candidate solutions.

6. The method for generating multiple solutions to industrial faults by integrating knowledge graphs according to claim 5, characterized in that, The step of performing constraint filtering on the current semantic matching solution set to generate a subset of feasible solutions for each potential fault propagation path includes: Obtain the set of equipment maintenance resource constraints for the target industrial equipment, which includes maintenance tool availability constraints, maintenance personnel skill constraints, and maintenance time window constraints. For each solution in the current semantic matching solution set, perform tool matching verification processing to generate a first set of filtering solutions that meets the maintenance tool availability constraints; Each solution in the first set of filtering solutions is subjected to skill matching verification to generate a second set of filtering solutions that meets the skill constraints of the maintenance personnel. Perform time window verification on each solution in the second set of filtering solutions to generate a third set of filtering solutions that conforms to the maintenance time window constraints; The feasible solution subset is generated based on the verification results of the third set of filtering solutions.

7. The method for generating multiple solutions to industrial faults by integrating knowledge graphs according to claim 1, characterized in that, The step of performing a feasibility assessment on the candidate solution set to generate an optimized solution set corresponding to the target industrial equipment includes: Cost parameters are calculated for each solution in the candidate solution set to generate an economic evaluation index for each solution. For each solution in the candidate solution set, risk parameter prediction processing is performed to generate a risk assessment index corresponding to each solution; For each solution in the candidate solution set, an execution difficulty analysis is performed to generate an operational evaluation index for each solution; The economic evaluation indicators, risk evaluation indicators, and operational evaluation indicators are scored and converted to generate standardized scores. Dynamic weight parameters are assigned according to preset business priority rules, and the standardized scores are weighted to generate a comprehensive score result. The candidate solution set is sorted in descending order based on the comprehensive scoring results to generate a sorted list of the candidate solution set; The optimized solution set is generated by extracting the top solution group from the sorted list based on a preset filtering threshold.

8. The method for generating multiple solutions to industrial faults by integrating knowledge graphs according to claim 1, characterized in that, The method further includes: The optimization solution set is subjected to execution effect tracking processing to generate an actual execution effect record for each solution in the optimization solution set; Based on the actual execution effect record, the industrial knowledge graph is updated with entity relationships to generate an updated industrial knowledge graph. Based on the updated industrial knowledge graph, the subsequent fault data set is subjected to correlation analysis and processing to generate an optimized fault correlation feature set.

9. The method for generating multiple solutions to industrial faults by integrating knowledge graphs according to claim 8, characterized in that, The step of updating the entity relationship of the industrial knowledge graph based on the actual execution effect record to generate an updated industrial knowledge graph includes: The successful solutions in the actual execution effect record are subjected to feature parsing processing to generate feature enhancement parameters for the successful solutions; The failure solutions in the actual execution effect record are subjected to feature parsing processing to generate feature suppression parameters for the failure solutions; Based on the feature enhancement parameters, the corresponding entity relationships in the industrial knowledge graph are weighted and enhanced to generate an enhanced entity relationship subgraph. The corresponding entity relationships in the industrial knowledge graph are weighted and weakened according to the feature suppression parameters to generate a weakened entity relationship subgraph. The enhanced entity relationship subgraph and the weakened entity relationship subgraph are merged to generate the updated industrial knowledge graph.

10. A multi-solution generation system for industrial faults integrating knowledge graphs, characterized in that, The method includes a processor and a memory, the memory and the processor being connected. The memory is used to store programs, instructions or code, and the processor is used to execute the programs, instructions or code in the memory to implement the industrial fault multi-solution generation method based on the fusion knowledge graph as described in any one of claims 1-9.

Images