An industrial equipment fault diagnosis system and method based on low-code modeling

Abstract

The application discloses an industrial equipment fault diagnosis system and method based on low-code modeling, relates to the technical field of industrial diagnosis, and comprises a low-code diagnosis platform, a data acquisition module, a data preprocessing module, a feature configuration and weight calculation module, a fault causal correlation knowledge graph module, a multi-fault decoupling module, and a fault judgment and primary and secondary differentiation module; through the multi-fault decoupling module, the ICA decoupling technology is adopted to perform blind decoupling processing on the coupled abnormal signals generated in the operation process of the industrial equipment, the working condition disturbance noise can be effectively stripped, the feature components of various independent faults are restored, the mutual interference and feature covering among different faults are avoided, and an accurate feature basis is provided for subsequent fault judgment; meanwhile, through the feature alignment, similarity calculation and threshold comparison process of the fault judgment and primary and secondary differentiation module, accurate identification of the coupled fault type is realized, the probability of fault misjudgment and omission is greatly reduced, and the reliability of fault diagnosis is improved.

Description

Technical Field

[0001] This invention relates to the field of industrial diagnostic technology, specifically to an industrial equipment fault diagnosis system and method based on low-code modeling. Background Technology

[0002] Industrial equipment, such as steam turbines and generators in gas-fired power plants, is prone to multiple fault coupling during long-term high-load operation. This means that two or more independent faults occur simultaneously in the same equipment, and these faults influence each other's parameter representations and mask each other's features, making it difficult to pinpoint the root cause. Existing low-code modeling-based fault diagnosis methods mostly use single-parameter threshold judgments and single-rule independent triggering, which cannot effectively decouple multiple fault coupling scenarios. They also suffer from drawbacks such as static and fixed rules, poor adaptability, reliance on a large number of labeled samples, and inability to quantify the contribution of faults. As a result, the accuracy of fault diagnosis and the precision of root cause localization are low, making it difficult to meet the actual operation and maintenance needs of industrial sites.

[0003] Therefore, there is an urgent need for an industrial equipment fault diagnosis method that can effectively solve the problem of multi-fault coupling and decoupling, adapt to the ease of use of low-code platforms, and realize the quantification of fault contribution and adaptive optimization of the model, so as to overcome the shortcomings of existing technologies. Summary of the Invention

[0004] The purpose of this invention is to provide an industrial equipment fault diagnosis system and method based on low-code modeling to solve the problems raised in the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for fault diagnosis of industrial equipment based on low-code modeling, the method comprising: Step S1: Initialize the low-code diagnostic environment, load the preset fault feature primitive components, decoupling components, knowledge graph components and contribution calculation components, bind the basic information of industrial equipment and equipment monitoring points, and establish a basic diagnostic carrier; Step S2: Collect real-time monitoring time-series data and historical fault data of industrial equipment. Preprocess the real-time monitoring time-series data to obtain a standardized, noise-free, and missing time-series data matrix. Based on the preprocessed time-series data and historical fault data, combined with the correlation analysis between faults and parameters and parameter sensitivity analysis, configure single fault feature primitives and the correlation weights between each monitoring parameter and the corresponding single fault to obtain a fault-parameter correlation weight table and a standard feature vector for each type of single fault. Step S3: Based on the fault-parameter association weight table, equipment component structural relationships, and standard characteristic curves of a single fault, build a visualized fault causal association knowledge graph through a low-code platform; Step S4: Input the standardized time series data matrix into the decoupling component built into the low-code platform. The decoupling component is a low-code component based on the independent component analysis algorithm. It removes the operating condition disturbance noise through four steps: data dataification, whitening processing, iterative optimization to solve the separation matrix, and signal separation. It performs blind decoupling processing on the mixed abnormal signals generated by multiple fault coupling to obtain multiple independent fault feature components. Step S5: Input the decoupled fault feature components into the fault causal association knowledge graph, calculate the similarity between each fault feature component and the standard features of various single faults through the feature matching module, and determine the type of coupled fault based on the similarity. Step S6: Based on the fault-parameter association weight table, the decoupled fault feature components, and standardized time series data, calculate the normalized contribution of various coupled faults, distinguish between primary and secondary faults according to the normalized contribution, and generate maintenance priority suggestions.

[0006] Furthermore, in step S2, the single fault feature primitives and the association weights between each monitoring parameter and the corresponding single fault are configured to obtain the fault-parameter association weight table and the standard feature vector for each type of single fault, including the following specific steps: Step S21: Filter out the time series data corresponding to a single fault from the historical fault data, remove data with multiple fault coupling and abnormal operating conditions, and perform a preprocessing process on the filtered single fault time series data that is completely consistent with the preprocessing of real-time monitoring time series data. Step S22: Extract the time-series data of single faults after filtering in step S21, and extract the exclusive parameter change pattern corresponding to each type of single fault. The parameter change pattern includes the parameter change trend, change rate and change threshold range. Encapsulate the parameter change pattern into a low-code drag-and-drop component to obtain the feature primitive of a single fault. This component supports subsequent modification of core parameters such as parameter change trend, rate and threshold based on changes in equipment operating status and maintenance feedback results, and can be directly called into the low-code diagnostic environment. Step S23: For each type of single fault, based on the single fault time series data preprocessed in step S31, calculate the Pearson correlation coefficient between each monitoring parameter and the corresponding fault when the fault occurs. The correlation coefficient quantifies the degree of linear correlation between each monitoring parameter and the single fault. The larger the absolute value of the correlation coefficient, the higher the correlation between the monitoring parameter and the single fault. Take the absolute value of the Pearson correlation coefficient as the quantified value of the correlation between the monitoring parameter and the corresponding single fault. Step S24: Using the controlled variable method, with the single fault time series data preprocessed in Step S31 as the benchmark, keep the values ​​of all other monitoring parameters unchanged, and only change the value of a certain target monitoring parameter. Observe the change range of the single fault characteristic. The larger the change range, the higher the sensitivity of the target monitoring parameter to the single fault, and the greater the corresponding correlation weight. Calculate the sensitivity quantification value S using the formula: S=ΔF / ΔP; where ΔF represents the absolute value of the difference in the fault characteristic quantification value of the single fault characteristic after the change of the target monitoring parameter, and ΔP represents the absolute value of the difference before and after the change of the target monitoring parameter. Calculate the sensitivity quantification value S for all monitoring parameters corresponding to each type of single fault using this method. Step S25: Combining the correlation quantification value |r| obtained in step S23 and the sensitivity quantification value S obtained in step S24, calculate the initial correlation weight w0 for each monitoring parameter to each type of single fault, w0 = 0.4 * |r| + 0.6 * S; normalize the calculated initial correlation weights of all monitoring parameters using the formula w kj =w 0(k,j) / ∑w 0(k,j) The normalized correlation weight w of the j-th monitoring parameter corresponding to the k-th type of single fault is obtained. kj ;w 0(k,j) This represents the initial association weight of the j-th monitoring parameter corresponding to the k-th type of single fault; ∑w 0(k,j) This indicates that the initial correlation weights from j=1 to j=m are summed; m represents the total number of monitoring parameters recorded under a single fault; for monitoring parameters that are not related to the single fault, their normalized correlation weights are set to 0. Step S26: Organize the name of each type of single fault, the corresponding name of each monitoring parameter, and the normalized association weight into a table to obtain the fault-parameter association weight table. The fixed arrangement order of all monitoring parameters is used as the vector dimension. The value of each dimension corresponds to the normalized association weight of the monitoring parameter to a single fault. The dimension value corresponding to irrelevant parameters is 0, and the dimension value corresponding to the associated parameters is the normalized weight value. Construct the standard feature vector of each type of single fault, with the dimension consistent with the number of monitoring parameters.

[0007] Furthermore, step S3 includes the following specific details: The knowledge graph includes equipment component nodes, monitoring parameter nodes, and fault nodes, and the connection labels between parameter nodes and fault nodes correspond to the associated weights. The knowledge graph is built through four steps: node creation, node association, weight binding, and feature benchmark import, realizing the visualization of causal relationships between faults, parameters, and equipment components. Node creation refers to creating equipment component nodes, monitoring parameter nodes, and fault nodes respectively through the knowledge graph configuration component of the low-code platform. Among them, equipment component nodes correspond to the core components of industrial equipment, monitoring parameter nodes correspond to the monitoring parameters in step S2, and fault nodes correspond to the single faults in step S2. Node association refers to establishing connections between nodes based on the installation correspondence between equipment components and monitoring parameters, and the causal relationship between monitoring parameters and faults. Among them, the relationship between equipment component nodes and monitoring parameter nodes is "monitoring-monitored", and the relationship between monitoring parameter nodes and fault nodes is "influence-influenced". Weight binding refers to binding the weight values ​​in the fault-parameter association weight table obtained in step S2 to the association line between the monitoring parameter node and the fault node one by one, so as to realize the visual annotation of the weight. Feature benchmark import refers to importing the standard feature curves of each type of single fault obtained in step S2 into the corresponding fault node as the benchmark for subsequent feature matching, thereby completing the construction of the fault causal association knowledge graph.

[0008] Furthermore, step S4 includes the following specific steps: Data centralization refers to the centralization of a standardized time-series data matrix, using the formula: x c (i,j)=x d (i,j)-x0(j), calculate the data x of the centralized i-th timestamp and j-th monitoring parameter. c (i,j), where x d (i,j) represents the standardized data of the i-th timestamp and the j-th monitoring parameter, and x0(j) represents the standardized mean of the j-th monitoring parameter; Whitening refers to the whitening process performed on the centered time-series data matrix. The process includes first calculating the covariance matrix of the centered data ∑=E[x] c x c T The covariance matrix is ​​decomposed into eigenvalues ​​to obtain the eigenvalue matrix A and the eigenvector matrix U. Then, the eigenvectors are obtained using the formula x. w =A -1 / 2 U T x c Obtain the whitened data x w ; Iterative optimization to solve the separation matrix refers to setting the decoupling dimension and the number of iterations, initializing the separation matrix W, and using a fixed-point iterative algorithm to optimize the separation matrix. The iteration objective is to make the separated signals independent of each other. During the iteration process, the independence of the signals is judged by the criterion of maximizing negative entropy, until the number of iterations reaches the preset value or the separation matrix converges. Signal separation refers to separating the whitened data xw Multiplying the separated component by the optimized separation matrix W yields the separated independent fault feature components, calculated as: y = Wx w , where y is the independent fault feature component matrix, and each column vector corresponds to an independent fault feature component.

[0009] Furthermore, step S5, determining the type of coupling fault based on similarity, includes the following steps: Step S51: Extract feature parameters from the decoupled fault feature components and the standard features of a single fault, respectively. The feature parameters include parameter change trend, change rate, peak value, valley value and stability index. Step S52: Align the feature parameters of the fault feature components with the feature parameters of the single fault standard feature to ensure that the feature dimensions and time dimensions of the two are consistent, and perform interpolation or pruning on inconsistent feature parameters. Step S53: Use the cosine similarity algorithm to calculate the similarity between the aligned feature parameters. The calculation formula is: sim=(a1·b1) / (|a1|×|b1|), where a1 is the feature parameter vector of the fault feature component, b1 is the feature parameter vector of the single fault standard feature, and sim is the similarity value, which takes the value range of [0,1]. Step S54: Set a similarity threshold. If the calculated similarity value is not lower than the threshold, the fault corresponding to the fault feature component is determined to be a single fault. If the similarity value is lower than the threshold, it is determined to be an unknown fault and manual confirmation is requested.

[0010] Furthermore, step S6 includes the following specific steps: Step S61: Calculate the abnormal amplitude of each monitoring parameter. The abnormal amplitude is the absolute value of the difference between the standardized actual value of the parameter and the standardized mean value of the parameter when the equipment is running normally. Calculate the sum of the characteristic amplitudes of each type of fault. The sum of the characteristic amplitudes is the sum of the products of the abnormal amplitudes of each monitoring parameter corresponding to the fault and the corresponding associated weights. Calculate the total characteristic amplitude of all coupled faults, that is, the sum of the characteristic amplitudes of all types of faults. Step S62: Calculate the normalized contribution of each type of fault, which is the ratio of the sum of the characteristic amplitudes of the fault type to the total characteristic amplitude, and then convert it into a percentage form; Step S63: Based on the normalized contribution, set the criteria for judging primary and secondary faults, complete the distinction between primary and secondary faults and the order of maintenance priorities. The specific criteria are as follows: If the normalized contribution of a certain type of fault is greater than or equal to 50%, then the fault is determined to be the main fault and the main maintenance target, and maintenance resources are given priority. If the normalized contribution of a certain type of fault is greater than or equal to 20% and less than 50%, then the fault is determined to be a secondary fault and is treated as a secondary maintenance target, which is carried out after the main fault maintenance is completed. If the normalized contribution of a certain type of fault is less than 20%, the fault is determined to be an auxiliary fault. It can be repaired at a later date according to the operation and maintenance plan after the main and secondary faults have been repaired, taking into account the equipment's operating status. Maintenance priorities are ranked according to the order of primary faults > secondary faults > auxiliary faults. Faults of the same level are ranked from high to low according to their normalized contribution. The higher the contribution, the higher the maintenance priority. If there are faults with the same normalized contribution, the fault with the higher correlation weight to the core operating parameters of the equipment is prioritized for maintenance.

[0011] An industrial equipment fault diagnosis system based on low-code modeling, the system includes a low-code diagnosis platform, a data acquisition module, a data preprocessing module, a feature configuration and weight calculation module, a fault causal association knowledge graph module, a multi-fault decoupling module, and a fault determination and primary / secondary distinction module. The low-code diagnostic platform is loaded with preset fault feature primitive components, ICA decoupling components, knowledge graph configuration components, and contribution calculation components; The data acquisition module is used to collect historical fault data and real-time monitoring time series data of industrial equipment, and to remove abnormal data; The data preprocessing module is used to perform noise filtering, missing value imputation, and standardization on the collected data, transforming the data into a standardized data matrix within a preset range; The feature configuration and weight calculation module is used to filter preprocessed historical data of single faults, encapsulate fault feature primitives, calculate the correlation degree and parameter sensitivity between faults and monitoring parameters using Pearson correlation coefficient and control variable method respectively, calculate and normalize the correlation weights by combining the preset proportions of correlation degree and sensitivity, and generate a fault-parameter correlation weight table and a standard feature vector of single faults. The fault causal association knowledge graph module is used to create three types of nodes—equipment components, monitoring parameters, and faults—through the knowledge graph configuration component of the low-code diagnostic platform. It establishes "monitoring-monitored" and "impact-impacted" associations between nodes, binds association weights, and imports standard feature curves of a single fault to complete the construction of the fault causal association knowledge graph. The multi-fault decoupling module is used to input the standardized data matrix into the ICA decoupling component, and sequentially perform data centering, covariance matrix eigenvalue decomposition and whitening processes. The separation matrix is ​​optimized through a fixed-point iterative algorithm to separate independent fault feature components. The fault determination and primary / secondary fault differentiation module is used to extract the core parameters of the decoupled fault feature components and the standard features of a single fault. After feature alignment, cosine similarity calculation and threshold comparison, the coupled fault type of industrial equipment is determined. Then, the normalized contribution of each type of fault is calculated by combining the correlation weight and the abnormal amplitude of the parameters, thus completing the differentiation of primary and secondary faults and the ranking of maintenance priorities.

[0012] Furthermore, the correlation degree and parameter sensitivity are included in the correlation weight calculation according to a preset ratio; the standardization process adopts a standardization method adapted to industrial time series data to transform the data into a unified dimension interval; the fixed-point iterative algorithm selects a nonlinear activation function adapted to the nonlinear characteristics of industrial signals, and the number of iterations and convergence threshold can be configured according to the type of industrial equipment and diagnostic needs.

[0013] Compared with the prior art, the beneficial effects of the present invention are: This invention employs a multi-fault decoupling module and ICA decoupling technology to blindly decouple coupled abnormal signals generated during the operation of industrial equipment. This effectively removes operating condition disturbance noise, restores the characteristic components of various independent faults, and avoids mutual interference and feature masking between different faults, providing an accurate feature basis for subsequent fault determination. At the same time, through feature alignment, similarity calculation, and threshold comparison processes of the fault determination and primary / secondary differentiation modules, it achieves accurate identification of coupled fault types, significantly reducing the probability of false or missed faults and improving the reliability of fault diagnosis.

[0014] This invention quantifies the correlation and sensitivity between faults and monitoring parameters by using feature configuration and weight calculation, employing Pearson correlation coefficient and control variable methods. Then, it calculates and normalizes the correlation weights by combining preset ratios. The entire process is based on pure technical logic analysis and does not rely on the subjective experience of equipment experts to assign values, ensuring the objectivity, standardization, and repeatability of the correlation weight configuration. At the same time, the generated fault-parameter correlation weight table and single fault standard feature vector provide a reliable basis for fault matching and contribution calculation.

[0015] By using a fault causal association knowledge graph, three types of nodes—equipment components, monitoring parameters, and faults—are structurally associated, clearly presenting the causal relationships between them. Corresponding association weights are bound to the association lines, and a single fault standard characteristic curve is imported. Maintenance personnel can intuitively view the association logic between faults and parameters and equipment components, quickly achieving root cause tracing and solving the problems of ambiguous causal relationships and difficulty in tracing faults in existing technologies. The normalized contribution of various coupled faults is calculated, enabling precise differentiation between primary and secondary faults. This clarifies maintenance priorities, guiding maintenance personnel to prioritize primary faults that have a greater impact on equipment operation, avoiding ineffective maintenance and wasted resources, optimizing maintenance processes, reducing maintenance costs, and ensuring the stable and efficient operation of industrial equipment. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the structure of an industrial equipment fault diagnosis method based on low-code modeling according to the present invention. Detailed Implementation

[0017] Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Example: Figure 1 As shown, this invention provides a method for fault diagnosis of industrial equipment based on low-code modeling, the method comprising: Step S1: Initialize the low-code diagnostic environment, load the preset fault feature primitive components, decoupling components, knowledge graph components, and contribution calculation components, bind the basic information of industrial equipment and equipment monitoring points, and establish a basic diagnostic carrier; for example, bind the basic information of a 300MW steam turbine: model, rated power, years of operation, core component composition, and 5 monitoring points, each corresponding to 5 monitoring parameters. Step S2: Collect real-time monitoring time-series data and historical fault data of industrial equipment. Preprocess the real-time monitoring time-series data to obtain a standardized, noise-free, and missing data matrix. Preprocessing includes noise filtering, missing value imputation, and standardization. Standardization is used to convert all monitoring parameters into the same dimension range to eliminate the influence of dimensions. Based on the preprocessed time-series data and historical fault data, combined with the correlation analysis between faults and parameters and parameter sensitivity analysis, configure single fault feature primitives and the correlation weights between each monitoring parameter and the corresponding single fault to obtain a fault-parameter correlation weight table and a standard feature vector for each type of single fault. The standard feature vector is a vector composed of the correlation weights of each monitoring parameter corresponding to each type of single fault. Its dimension is consistent with the number of monitoring parameters. The weights corresponding to irrelevant parameters are 0, and the sum of the weights corresponding to all correlated parameters is 1. Step S3: Based on the fault-parameter association weight table, equipment component structural relationships, and standard characteristic curves of a single fault, build a visualized fault causal association knowledge graph through a low-code platform; Step S4: Input the standardized time series data matrix into the decoupling component built into the low-code platform. The decoupling component is a low-code component based on the independent component analysis algorithm. It removes the operating condition disturbance noise through four steps: data dataification, whitening processing, iterative optimization to solve the separation matrix, and signal separation. It performs blind decoupling processing on the mixed abnormal signals generated by multiple fault coupling to obtain multiple independent fault feature components. Step S5: Input the decoupled fault feature components into the fault causal association knowledge graph, calculate the similarity between each fault feature component and the standard features of various single faults through the feature matching module, and determine the type of coupled fault based on the similarity. Step S6: Based on the fault-parameter association weight table, the decoupled fault feature components, and standardized time series data, calculate the normalized contribution of various coupled faults, distinguish between primary and secondary faults according to the normalized contribution, and generate maintenance priority suggestions.

[0019] Step S2 configures the single fault feature primitives and the association weights between each monitoring parameter and the corresponding single fault, resulting in a fault-parameter association weight table and a standard feature vector for each type of single fault. This includes the following specific steps: Step S21: Filter out the time series data corresponding to a single fault from the historical fault data, and remove data with multiple fault coupling and abnormal operating conditions, such as equipment start-up and shutdown phases and load change phases. Perform a preprocessing process on the filtered single fault time series data that is completely consistent with the preprocessing of real-time monitoring time series data. Step S22: Extract the time-series data of single faults filtered in Step S21, and extract the exclusive parameter change pattern corresponding to each type of single fault. The parameter change pattern includes the parameter change trend, change rate, and change threshold range. The parameter change trend includes rising, falling, and stable. The change rate refers to the amount of parameter change per unit time. The change threshold range refers to the critical value at which the parameter deviates from the normal range when the fault occurs. The parameter change pattern is obtained by mining the time-series change pattern of parameters corresponding to single faults in historical fault data. The parameter change pattern is encapsulated into a low-code draggable component to obtain the feature primitive of a single fault. This component supports subsequent modification of core parameters such as parameter change trend, rate, and threshold based on changes in equipment operating status and maintenance feedback results, and can be directly called into the low-code diagnostic environment. Step S23: For each type of single fault, based on the single fault time series data preprocessed in step S31, calculate the Pearson correlation coefficient between each monitoring parameter and the corresponding fault when the fault occurs. The correlation coefficient quantifies the degree of linear correlation between each monitoring parameter and the single fault. The larger the absolute value of the correlation coefficient, the higher the correlation between the monitoring parameter and the single fault. Take the absolute value of the Pearson correlation coefficient as the quantified value of the correlation between the monitoring parameter and the corresponding single fault. The specific calculation is as follows: ; Where xi represents the time series data of a certain monitoring parameter after preprocessing when the fault occurs, and yi represents the quantified value of the severity of a single fault at the corresponding time. It is set according to the actual impact of the fault and is quantified as a value of 0-1. The more severe the fault, the closer the quantified value is to 1. This represents the mean of the time-series data for the corresponding monitoring parameter. The mean value of the quantified severity of a single fault is represented by n, where n is the number of timestamps in the time series; the calculated correlation coefficient r ranges from [-1, 1]. Step S24: Using the controlled variable method, with the single fault time series data preprocessed in Step S31 as the benchmark, keep the values ​​of all other monitoring parameters unchanged, and only change the value of a certain target monitoring parameter. The change range is controlled within ±10% to ±30% of the normal range of this parameter, which is close to the actual operating fluctuation range of industrial equipment. Observe the change range of the single fault characteristic. The larger the change range, the higher the sensitivity of the target monitoring parameter to the single fault, and the greater the corresponding correlation weight. Calculate the sensitivity quantification value S using the formula: S=ΔF / ΔP; where ΔF represents the absolute value of the difference in the fault characteristic quantification value of the single fault characteristic after the change of the target monitoring parameter, and ΔP represents the absolute value of the difference before and after the change of the target monitoring parameter. Calculate the sensitivity quantification value S for all monitoring parameters corresponding to each type of single fault using this method. Step S25: Combining the correlation quantification value |r| obtained in step S23 and the sensitivity quantification value S obtained in step S24, calculate the initial correlation weight w0 for each monitoring parameter to each type of single fault, w0 = 0.4 * |r| + 0.6 * S; normalize the calculated initial correlation weights of all monitoring parameters using the formula w kj =w 0(k,j) / ∑w 0(k,j) The normalized correlation weight w of the j-th monitoring parameter corresponding to the k-th type of single fault is obtained. kj ;w 0(k,j) This represents the initial association weight of the j-th monitoring parameter corresponding to the k-th type of single fault; ∑w 0(k,j) This indicates that the initial correlation weights from j=1 to j=m are summed; m represents the total number of monitoring parameters recorded under a single fault; for monitoring parameters unrelated to the single fault, their normalized correlation weights are set to 0; unrelatedness can be quantified as a correlation quantification value |r| < 0.1 and a sensitivity quantification value S < 0.1. Step S26: Organize the name of each type of single fault, the corresponding name of each monitoring parameter, and the normalized association weight into a table to obtain the fault-parameter association weight table. The fixed arrangement order of all monitoring parameters is used as the vector dimension. The value of each dimension corresponds to the normalized association weight of the monitoring parameter to a single fault. The dimension value corresponding to irrelevant parameters is 0, and the dimension value corresponding to the associated parameters is the normalized weight value. Construct the standard feature vector of each type of single fault, with the dimension consistent with the number of monitoring parameters.

[0020] As shown in the example: From historical fault data, 50 sets of single fault data for "slight wear of bearing bush" and 50 sets of single fault data for "oil pump output reduction" were selected. Data with multiple fault coupling and abnormal operating conditions were removed. The selected data were subjected to the same preprocessing process as S2 to ensure that the units of measurement were consistent with the real-time data. Historical data analysis revealed that when this fault occurred, the bearing vibration amplitude V showed an upward trend with a trend coefficient of 0.8, a rate of change of 0.02 mm / s·min, a threshold range of 0.3-0.5 mm / s, and a standardized value of 0.5-0.75. The bearing temperature T showed a slow upward trend with a trend coefficient of 0.3, a rate of change of 0.1℃ / min, a threshold range of 65-75℃, and a standardized value of 0.625-0.875. Other parameters showed no significant changes. This change pattern was encapsulated as a low-code draggable component named "Bearing Slight Wear Feature Primitive Element," supporting subsequent adjustments to the trend coefficient, rate of change, and threshold. Historical data analysis revealed that when this fault occurred, the lubricating oil pressure P showed a decreasing trend with a trend coefficient of -0.7, a rate of change of 0.005 MPa / min, a threshold range of 0.15-0.25 MPa, and a standardized value of 0.25-0.5. The return oil flow rate Q also showed a decreasing trend with a trend coefficient of -0.6, a rate of change of 0.2 m³ / h·min, a threshold range of 10-15 m³ / h, and a standardized value of 0-0.25. Other parameters showed no significant changes. This change pattern was encapsulated as a low-code draggable component and named "Oil Pump Output Attenuation Characteristic Element".

[0021] Using Pearson correlation coefficients, taking "slight bearing wear" as an example, xi represents the standardized time-series data of each monitoring parameter, and yi represents the quantified value of the fault severity, ranging from 0.2 to 0.8, with the more severe the wear, the closer it is to 0.8. The calculated quantified values ​​of the correlation metrics for each parameter, |r|, are as follows: T=0.72, P=0.08, Q=0.06, V=0.85, G=0.12. Similarly, the quantified values ​​of the correlation metrics for "oil pump output reduction", |r|, are as follows: T=0.09, P=0.88, Q=0.76, V=0.11, G (oil particle size)=0.07. Using the controlled variable method, with other parameters fixed, and the target parameter changed by ±20%, the following parameters were observed for slight wear of the bearing: V change ±20%, fault characteristic change range 0.34, S=0.34 / 0.2=1.7; T change ±20%, fault characteristic change range 0.12, S=0.12 / 0.2=0.6; other parameter changes ±20%, fault characteristic change range <0.02, S<0.1. For reduced oil pump output: P change ±20%, fault characteristic change range 0.38, S=0.38 / 0.2=1.9; Q change ±20%, fault characteristic change range 0.28, S=0.28 / 0.2=1.4; other parameter changes ±20%, fault characteristic change range <0.02, S<0.1. The bearing bush has slight wear. The initial weights are w_initial = 0.4 × |r| + 0.6 × S, resulting in: T = 0.4 × 0.72 + 0.6 × 0.6 = 0.648; V = 0.4 × 0.85 + 0.6 × 1.7 = 1.36; P = 0.4 × 0.08 + 0.6 × 0.05 = 0.062; Q = 0.4 × 0.06 + 0.6 × 0.04 = 0.048; G = 0.4 × 0.12 + 0.6 × 0.06 = 0.084. After normalization, the sum is 1, and |r| and S for P, Q, and G are all < 0.1, so they are set to 0. The final weights are: T = 0.32, V = 0.68, P = 0, Q = 0, G = 0.

[0022] The oil pump output decreases. The initial weights are w_initial=0.4×|r|+0.6×S, resulting in: P=0.4×0.88+0.6×1.9=1.532; Q=0.4×0.76+0.6×1.4=1.184; T=0.4×0.09+0.6×0.05=0.066; V=0.4×0.11+0.6×0.06=0.08; G=0.4×0.07+0.6×0.04=0.052. After normalization, T, V, and G are set to 0, resulting in the final weights: P=0.56, Q=0.44, T=0, V=0, G=0.

[0023] The standard eigenvectors for slight wear of the bearing bush are: [0.32, 0, 0, 0.68, 0]; and the oil pump output decrease is: [0, 0.56, 0.44, 0, 0].

[0024] Step S3 includes the following specific contents: The knowledge graph includes equipment component nodes, monitoring parameter nodes, and fault nodes, and the connection labels between parameter nodes and fault nodes correspond to the associated weights. The knowledge graph is built through four steps: node creation, node association, weight binding, and feature benchmark import, realizing the visualization of causal relationships between faults, parameters, and equipment components. Node creation refers to creating equipment component nodes, monitoring parameter nodes, and fault nodes respectively through the knowledge graph configuration component of the low-code platform. Among them, equipment component nodes correspond to the core components of industrial equipment, monitoring parameter nodes correspond to the monitoring parameters in step S2, and fault nodes correspond to the single faults in step S2. Node association refers to establishing connections between nodes based on the installation correspondence between equipment components and monitoring parameters, and the causal relationship between monitoring parameters and faults. Among them, the relationship between equipment component nodes and monitoring parameter nodes is "monitoring-monitored", and the relationship between monitoring parameter nodes and fault nodes is "influence-influenced". Weight binding refers to binding the weight values ​​in the fault-parameter association weight table obtained in step S2 to the association line between the monitoring parameter node and the fault node one by one, so as to realize the visual annotation of the weight. Feature benchmark import refers to importing the standard feature curves of each type of single fault obtained in step S2 into the corresponding fault node as the benchmark for subsequent feature matching, thereby completing the construction of the fault causal association knowledge graph.

[0025] Step S4 includes the following specific steps: Data centralization refers to the centralization of a standardized time-series data matrix, using the formula: x c (i,j)=x d (i,j)-x0(j), calculate the data x of the centralized i-th timestamp and j-th monitoring parameter. c (i,j), where x d (i,j) represents the standardized data of the i-th timestamp and the j-th monitoring parameter, and x0(j) represents the standardized mean of the j-th monitoring parameter; Whitening refers to the whitening process performed on the centered time-series data matrix. The process includes first calculating the covariance matrix of the centered data ∑=E[x] c x c T The covariance matrix is ​​decomposed into eigenvalues ​​to obtain the eigenvalue matrix A and the eigenvector matrix U. Then, the eigenvectors are obtained using the formula x. w =A -1 / 2 U T x c Obtain the whitened data x w ; Iterative optimization to solve the separation matrix refers to setting the decoupling dimension and the number of iterations, initializing the separation matrix W, and using a fixed-point iterative algorithm to optimize the separation matrix. The iteration objective is to make the separated signals independent of each other. During the iteration process, the independence of the signals is judged by the criterion of maximizing negative entropy, until the number of iterations reaches the preset value or the separation matrix converges. Signal separation refers to separating the whitened data x wMultiplying the separated component by the optimized separation matrix W yields the separated independent fault feature components, calculated as: y = Wx w , where y is the independent fault feature component matrix, and each column vector corresponds to an independent fault feature component.

[0026] Step S5, determining the type of coupling fault based on similarity, includes the following steps: Step S51: Extract feature parameters from the decoupled fault feature components and the standard features of a single fault, respectively. The feature parameters include parameter change trend, change rate, peak value, valley value and stability index. Step S52: Align the feature parameters of the fault feature components with the feature parameters of the single fault standard feature to ensure that the feature dimensions and time dimensions of the two are consistent, and perform interpolation or pruning on inconsistent feature parameters. Step S53: Use the cosine similarity algorithm to calculate the similarity between the aligned feature parameters. The calculation formula is: sim=(a1·b1) / (|a1|×|b1|), where a1 is the feature parameter vector of the fault feature component, b1 is the feature parameter vector of the single fault standard feature, and sim is the similarity value, which takes the value range of [0,1]. Step S54: Set a similarity threshold. If the calculated similarity value is not lower than the threshold, the fault corresponding to the fault feature component is determined to be a single fault. If the similarity value is lower than the threshold, it is determined to be an unknown fault and manual confirmation is requested.

[0027] Step S6 includes the following specific steps: Step S61: Calculate the abnormal amplitude of each monitoring parameter. The abnormal amplitude is the absolute value of the difference between the standardized actual value of the parameter and the standardized mean of the parameter when the equipment is running normally. Calculate the sum of the characteristic amplitudes of each type of fault. The sum of the characteristic amplitudes is the sum of the products of the abnormal amplitudes of each monitoring parameter corresponding to the fault and the corresponding associated weights. Calculate the total characteristic amplitude of all coupled faults, that is, the sum of the characteristic amplitudes of all types of faults. The standardized mean of the parameters when the equipment is running normally is obtained by statistically analyzing the monitoring parameter data under normal operating conditions of industrial equipment, standardizing the data, and taking the mean value. It is pre-stored in the low-code platform and can be updated according to changes in the operating status of the equipment. Step S62: Calculate the normalized contribution of each type of fault, which is the ratio of the sum of the characteristic amplitudes of the fault type to the total characteristic amplitude, and then convert it into a percentage form; Step S63: Based on the normalized contribution, set the criteria for judging primary and secondary faults, complete the distinction between primary and secondary faults and the order of maintenance priorities. The specific criteria are as follows: If the normalized contribution of a certain type of fault is greater than or equal to 50%, then the fault is determined to be the main fault and the main maintenance target, and maintenance resources are given priority. If the normalized contribution of a certain type of fault is greater than or equal to 20% and less than 50%, then the fault is determined to be a secondary fault and is treated as a secondary maintenance target, which is carried out after the main fault maintenance is completed. If the normalized contribution of a certain type of fault is less than 20%, the fault is determined to be an auxiliary fault. It can be repaired at a later date according to the operation and maintenance plan after the main and secondary faults have been repaired, taking into account the equipment's operating status. Maintenance priorities are ranked as follows: primary faults > secondary faults > auxiliary faults. Faults of the same level are ranked from highest to lowest based on their normalized contribution; the higher the contribution, the higher the maintenance priority. If faults with the same normalized contribution exist, the fault with the higher correlation weight to the core operating parameters of the equipment is prioritized for maintenance. Refer to the Fault-Parameter Correlation Weight Table for correlation weights.

[0028] An industrial equipment fault diagnosis system based on low-code modeling, the system includes a low-code diagnosis platform, a data acquisition module, a data preprocessing module, a feature configuration and weight calculation module, a fault causal association knowledge graph module, a multi-fault decoupling module, and a fault determination and primary / secondary distinction module. The low-code diagnostic platform is loaded with preset fault feature primitive components, ICA decoupling components, knowledge graph configuration components, and contribution calculation components; The data acquisition module is used to collect historical fault data and real-time monitoring time series data of industrial equipment, and to remove abnormal data; The data preprocessing module is used to perform noise filtering, missing value imputation, and standardization on the collected data, transforming the data into a standardized data matrix within a preset range; The feature configuration and weight calculation module is used to filter preprocessed historical data of single faults, encapsulate fault feature primitives, calculate the correlation degree and parameter sensitivity between faults and monitoring parameters using Pearson correlation coefficient and control variable method respectively, calculate and normalize the correlation weights by combining the preset proportions of correlation degree and sensitivity, and generate a fault-parameter correlation weight table and a standard feature vector of single faults. The fault causal association knowledge graph module is used to create three types of nodes—equipment components, monitoring parameters, and faults—through the knowledge graph configuration component of the low-code diagnostic platform. It establishes "monitoring-monitored" and "impact-impacted" associations between nodes, binds association weights, and imports standard feature curves of a single fault to complete the construction of the fault causal association knowledge graph. The multi-fault decoupling module is used to input the standardized data matrix into the ICA decoupling component, and sequentially perform data centering, covariance matrix eigenvalue decomposition and whitening processes. The separation matrix is ​​optimized through a fixed-point iterative algorithm to separate independent fault feature components. The fault determination and primary / secondary fault differentiation module is used to extract the core parameters of the decoupled fault feature components and the standard features of a single fault. After feature alignment, cosine similarity calculation and threshold comparison, the coupled fault type of industrial equipment is determined. Then, the normalized contribution of each type of fault is calculated by combining the correlation weight and the abnormal amplitude of the parameters, thus completing the differentiation of primary and secondary faults and the ranking of maintenance priorities.

[0029] The correlation degree and parameter sensitivity are included in the correlation weight calculation according to a preset ratio; the standardization process adopts a standardization method adapted to industrial time series data to transform the data into a unified dimension interval; the fixed-point iterative algorithm selects a nonlinear activation function adapted to the nonlinear characteristics of industrial signals, and the number of iterations and convergence threshold can be configured according to the type of industrial equipment and diagnostic needs.

[0030] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for fault diagnosis of industrial equipment based on low-code modeling, characterized in that: The method includes: Step S1: Initialize the low-code diagnostic environment, load the preset fault feature primitive components, decoupling components, knowledge graph components and contribution calculation components, bind the basic information of industrial equipment and equipment monitoring points, and establish a basic diagnostic carrier; Step S2: Collect real-time monitoring time-series data and historical fault data of industrial equipment. Preprocess the real-time monitoring time-series data to obtain a standardized, noise-free, and missing time-series data matrix. Based on the preprocessed time-series data and historical fault data, combined with the correlation analysis between faults and parameters and parameter sensitivity analysis, configure single fault feature primitives and the correlation weights between each monitoring parameter and the corresponding single fault to obtain a fault-parameter correlation weight table and a standard feature vector for each type of single fault. Step S3: Based on the fault-parameter association weight table, equipment component structural relationships, and standard characteristic curves of a single fault, a visualized fault causal association knowledge graph is built using a low-code platform; Step S4: Input the standardized time-series data matrix into the decoupling component built into the low-code platform. The decoupling component is a low-code component based on the independent component analysis algorithm. It removes the operating condition disturbance noise through four steps: data dataification, whitening processing, iterative optimization to solve the separation matrix, and signal separation. It performs blind decoupling processing on the mixed abnormal signals generated by multiple fault coupling to obtain multiple independent fault feature components. Step S5: Input the decoupled fault feature components into the fault causal association knowledge graph, calculate the similarity between each fault feature component and the standard features of various single faults through the feature matching module, and determine the type of coupled fault based on the similarity. Step S6: Based on the fault-parameter association weight table, the decoupled fault feature components, and standardized time series data, calculate the normalized contribution of various coupled faults, distinguish between primary and secondary faults according to the normalized contribution, and generate maintenance priority suggestions.

2. The industrial equipment fault diagnosis method based on low-code modeling according to claim 1, characterized in that: Step S2 involves configuring single fault feature primitives and the association weights between each monitoring parameter and the corresponding single fault to obtain a fault-parameter association weight table and a standard feature vector for each type of single fault. This includes the following specific steps: Step S21: Filter out the time series data corresponding to a single fault from the historical fault data, remove data with multiple fault coupling and abnormal operating conditions, and perform a preprocessing process on the filtered single fault time series data that is completely consistent with the preprocessing of real-time monitoring time series data. Step S22: Extract the time-series data of individual faults filtered in Step S21, and extract the specific parameter change pattern corresponding to each type of individual fault. The parameter change pattern includes the parameter change trend, change rate, and change threshold range. Step S23: For each type of single fault, based on the single fault time series data preprocessed in step S31, calculate the Pearson correlation coefficient between each monitoring parameter and the corresponding fault when the fault occurs, and take the absolute value of the Pearson correlation coefficient as the quantified value of the correlation between the monitoring parameter and the corresponding single fault. Step S24: Using the controlled variable method, with the single fault time series data preprocessed in step S31 as the benchmark, keep the values ​​of all other monitoring parameters unchanged, only change the value of a certain target monitoring parameter, observe the change range of the single fault characteristic, and use the formula: S=ΔF / ΔP to calculate the sensitivity quantification value S. Where ΔF represents the absolute value of the difference in the quantified values ​​of the fault characteristics of a single fault after the change of the target monitoring parameters, and ΔP represents the absolute value of the difference before and after the change of the target monitoring parameters; for all monitoring parameters corresponding to each type of single fault, their respective sensitivity quantified values ​​S are calculated using this method. Step S25: Combining the correlation quantification value |r| obtained in step S23 and the sensitivity quantification value S obtained in step S24, calculate the initial correlation weight w0 for each monitoring parameter to each type of single fault, w0 = 0.4 * |r| + 0.6 * S; normalize the calculated initial correlation weights of all monitoring parameters using the formula w kj =w 0(k,j) / ∑w 0(k,j) The normalized correlation weight w of the j-th monitoring parameter corresponding to the k-th type of single fault is obtained. kj ;w 0(k,j) This represents the initial association weight of the j-th monitoring parameter corresponding to the k-th type of single fault; ∑w 0(k,j) This indicates that the initial association weights from j=1 to j=m are summed. m represents the total number of monitoring parameters recorded under a single fault; For monitoring parameters that are not related to the single fault, their normalized correlation weight is set to 0; Step S26: Organize the name of each type of single fault, the corresponding name of each monitoring parameter, and the normalized association weight into a table to obtain the fault-parameter association weight table. The fixed arrangement order of all monitoring parameters is used as the vector dimension. The value of each dimension corresponds to the normalized association weight of the monitoring parameter to a single fault. The dimension value corresponding to irrelevant parameters is 0, and the dimension value corresponding to associated parameters is the normalized weight value. Construct a standard feature vector for each type of single fault. The dimension is consistent with the number of monitoring parameters.

3. The industrial equipment fault diagnosis method based on low-code modeling according to claim 1, characterized in that: Step S3 includes the following specific contents: The knowledge graph includes equipment component nodes, monitoring parameter nodes, and fault nodes, and the connection labels between parameter nodes and fault nodes correspond to the associated weights. The construction of the knowledge graph is completed through four steps: node creation, node association, weight binding, and feature benchmark import, so as to realize the visualization of the causal relationship between faults, parameters, and equipment components. The node creation refers to creating equipment component nodes, monitoring parameter nodes, and fault nodes respectively through the knowledge graph configuration component of the low-code platform. The equipment component nodes correspond to the core components of industrial equipment, the monitoring parameter nodes correspond to the monitoring parameters in step S2, and the fault nodes correspond to the single faults in step S2. The node association refers to establishing connections between nodes based on the installation correspondence between equipment components and monitoring parameters, and the causal relationship between monitoring parameters and faults. Among them, the relationship between equipment component nodes and monitoring parameter nodes is "monitoring-monitored", and the relationship between monitoring parameter nodes and fault nodes is "influence-influenced". The weight binding refers to binding the weight values ​​in the fault-parameter association weight table obtained in step S2 to the association line between the monitoring parameter node and the fault node one by one, so as to realize the visual annotation of the weight. The feature benchmark import refers to importing the standard feature curves of each type of single fault obtained in step S2 into the corresponding fault node as the benchmark for subsequent feature matching, thereby completing the construction of the fault causal association knowledge graph.

4. The industrial equipment fault diagnosis method based on low-code modeling according to claim 1, characterized in that: Step S4 includes the following specific steps: The data centralization refers to the centralization of a standardized time-series data matrix, using the formula: x c (i,j)=x d (i,j)-x0(j), calculate the data x of the centralized i-th timestamp and j-th monitoring parameter. c (i,j), where x d (i,j) represents the standardized data of the i-th timestamp and the j-th monitoring parameter, and x0(j) represents the standardized mean of the j-th monitoring parameter; The whitening process refers to whitening the centered time-series data matrix. The process includes first calculating the covariance matrix of the centered data ∑=E[x] c x c T The covariance matrix is ​​decomposed into eigenvalues ​​to obtain the eigenvalue matrix A and the eigenvector matrix U. Then, the eigenvectors are obtained using the formula x. w =A -1 / 2 U T x c Obtain the whitened data x w ; The iterative optimization solution of the separation matrix refers to setting the decoupling dimension and the number of iterations, initializing the separation matrix W, and using a fixed-point iterative algorithm to optimize the separation matrix. The iteration objective is to make the separated signals independent of each other. During the iteration process, the independence of the signals is judged by the negative entropy maximization criterion until the number of iterations reaches the preset value or the separation matrix converges. The signal separation refers to separating the whitened data x w Multiplying the separated component by the optimized separation matrix W yields the separated independent fault feature components, calculated as: y = Wx w , where y is the independent fault feature component matrix, and each column vector corresponds to an independent fault feature component.

5. The industrial equipment fault diagnosis method based on low-code modeling according to claim 1, characterized in that: The step S5, which determines the type of coupling fault based on similarity, includes the following steps: Step S51: Extract feature parameters from the decoupled fault feature components and the standard features of a single fault, respectively. The feature parameters include parameter change trend, change rate, peak value, valley value and stability index. Step S52: Align the feature parameters of the fault feature components with the feature parameters of the single fault standard feature to ensure that the feature dimensions and time dimensions of the two are consistent, and perform interpolation or pruning on inconsistent feature parameters. Step S53: Use the cosine similarity algorithm to calculate the similarity between the aligned feature parameters. The calculation formula is: sim=(a1·b1) / (|a1|×|b1|), where a1 is the feature parameter vector of the fault feature component, b1 is the feature parameter vector of the single fault standard feature, and sim is the similarity value, which takes the value range of [0,1]. Step S54: Set a similarity threshold. If the calculated similarity value is not lower than the threshold, the fault corresponding to the fault feature component is determined to be a single fault. If the similarity value is lower than the threshold, it is determined to be an unknown fault and manual confirmation is requested.

6. The industrial equipment fault diagnosis method based on low-code modeling according to claim 1, characterized in that: Step S6 includes the following specific steps: Step S61: Calculate the abnormal amplitude of each monitoring parameter, wherein the abnormal amplitude is the absolute value of the difference between the standardized actual value of the parameter and the standardized mean value of the parameter when the equipment is running normally; Calculate the sum of characteristic amplitudes for each type of fault, where the sum of characteristic amplitudes is the sum of the products of the abnormal amplitudes of each monitoring parameter corresponding to the fault and their corresponding associated weights; Calculate the total characteristic amplitude of all coupled faults, which is the sum of the characteristic amplitudes of all types of faults; Step S62: Calculate the normalized contribution of each type of fault, which is the ratio of the sum of the characteristic amplitudes of the fault type to the total characteristic amplitude, and then convert it into a percentage form; Step S63: Based on the normalized contribution, set the criteria for judging primary and secondary faults, complete the distinction between primary and secondary faults and the order of maintenance priorities. The specific criteria are as follows: If the normalized contribution of a certain type of fault is greater than or equal to 50%, then the fault is determined to be the main fault and the main maintenance target, and maintenance resources are given priority. If the normalized contribution of a certain type of fault is greater than or equal to 20% and less than 50%, then the fault is determined to be a secondary fault and is treated as a secondary maintenance target, which is carried out after the main fault maintenance is completed. If the normalized contribution of a certain type of fault is less than 20%, the fault is determined to be an auxiliary fault. It can be repaired at a later date according to the operation and maintenance plan after the main and secondary faults have been repaired, taking into account the equipment's operating status. The maintenance priority ranking is based on the priority of primary faults > secondary faults > auxiliary faults. Faults of the same level are ranked from high to low according to their normalized contribution. If there are faults with the same normalized contribution, the fault with the higher correlation weight to the core operating parameters of the equipment is prioritized for maintenance.

7. An industrial equipment fault diagnosis system based on low-code modeling, using the industrial equipment fault diagnosis method based on low-code modeling as described in any one of claims 1-6, characterized in that: The system includes a low-code diagnostic platform, a data acquisition module, a data preprocessing module, a feature configuration and weight calculation module, a fault causal association knowledge graph module, a multi-fault decoupling module, and a fault determination and primary / secondary distinction module. The low-code diagnostic platform is loaded with preset fault feature primitive components, ICA decoupling components, knowledge graph configuration components, and contribution calculation components. The data acquisition module is used to collect historical fault data and real-time monitoring time series data of industrial equipment, and to remove abnormal data; The data preprocessing module is used to perform noise filtering, missing value imputation and standardization on the collected data, and transform the data into a standardized data matrix within a preset range. The feature configuration and weight calculation module is used to filter preprocessed historical data of single faults, encapsulate fault feature primitives, calculate the correlation degree and parameter sensitivity between faults and monitoring parameters using Pearson correlation coefficient and control variable method respectively, calculate and normalize the correlation weights by combining the preset proportions of correlation degree and sensitivity, and generate a fault-parameter correlation weight table and a standard feature vector of a single fault. The fault causal association knowledge graph module is used to create three types of nodes—equipment components, monitoring parameters, and faults—through the knowledge graph configuration component of the low-code diagnostic platform, establish "monitoring-monitored" and "impact-impacted" associations between nodes, bind association weights, and import single fault standard feature curves to complete the construction of the fault causal association knowledge graph. The multi-fault decoupling module is used to input the standardized data matrix into the ICA decoupling component, and sequentially perform data centralization and covariance matrix eigenvalue decomposition and whitening processes. The separation matrix is ​​optimized by a fixed-point iterative algorithm to separate independent fault feature components. The fault determination and primary / secondary distinction module is used to extract the core parameters of the decoupled fault feature components and the standard features of a single fault. After feature alignment, cosine similarity calculation and threshold comparison, the coupled fault type of the industrial equipment is determined. Then, the normalized contribution of each type of fault is calculated by combining the correlation weight and the abnormal amplitude of the parameters, thus completing the distinction between primary and secondary faults and the ranking of maintenance priorities.

8. The industrial equipment fault diagnosis system based on low-code modeling according to claim 7, characterized in that: The correlation degree and parameter sensitivity are used in the correlation weight calculation according to a preset ratio; the standardization process adopts a standardization method adapted to industrial time series data to transform the data into a unified dimension interval; the fixed-point iterative algorithm selects a nonlinear activation function adapted to the nonlinear characteristics of industrial signals, and the number of iterations and convergence threshold can be configured according to the type of industrial equipment and diagnostic needs.

Images