Fault diagnosis method based on belief rule base and average causal effect

Abstract

The invention discloses a fault diagnosis method based on a belief rule base and an average causal effect, and the method is characterized in that the method comprises the following steps: constructing the belief rule base through data driving, carrying out the intervention of a precondition attribute, calculating the average causal effect, and obtaining a reference value, a precondition attribute weight, confidence distribution and a rule weight of the belief rule base; a causal interpretability constraint is introduced in the belief rule base optimization process, and the causal interpretability of belief rule base parameters is protected; and based on the confidence rule base, reasoning the input observation data to obtain a fault diagnosis reasoning result. According to the method, the causal theory is fused into data driving to construct the belief rule base, the model parameters are determined through the causal mechanism, the belief rule base is converted from statistical association to causal mechanism modeling, and the accuracy, interpretability and physical significance of fault diagnosis are improved.

Description

Technical Field

[0001] This invention relates to the field of fault diagnosis technology for complex equipment, and in particular to a fault diagnosis method based on a confidence rule base and average causal effect. Background Technology

[0002] The Belief Rule Base (BRB) model is an intelligent expert system approach based on evidence theory, fuzzy set theory, traditional IF-THEN rules, and machine learning methods. The rule structure of a Belief Rule Base consists of premise attributes, reference values, and confidence distributions. The reasoning process uses the Evidence-Based Reasoning (ER) algorithm to fuse the rules, ultimately producing transparent and interpretable decision results. Its core objective is to combine qualitative expert knowledge with quantitative observation data through an interpretable rule format, thereby solving all types of uncertainty problems in complex systems.

[0003] However, the practical application of fault diagnosis in complex mechanical systems is limited by the large number of features extracted from multi-source sensors. This leads to an exponential increase in the number of rules due to the growing scale of prerequisite attributes or reference values. As the complexity of the rules increases, it becomes difficult to set initial parameters solely based on expert knowledge, and model performance may be reduced due to the limited completeness and accuracy of expert experience.

[0004] In a BRB system, rules describe the causal relationships between presupposition attributes and their associated outcomes. Their interpretability depends on the rule parameters, which reflect the true physical causal relationships within the actual system. Using a data-driven approach to determine the initial parameters of the BRB model alleviates the problem of relying on experts for initialization. However, such methods capture statistical correlations, not the causal relationships described by the BRB rules. This can preserve spurious correlations and fail to establish causal connections with reality. Therefore, the parameters reflect spurious associations rather than genuine causal paths. Summary of the Invention

[0005] To address the above shortcomings, this invention provides a fault diagnosis method based on a confidence rule base and average causal effect. This solves the problem that existing confidence rule bases can only capture statistical correlations rather than true causal relationships, leading to difficulties in obtaining accurate inference results from the model. The specific technical solution is as follows: A fault diagnosis method based on a confidence rule base and average causal effect includes the following steps: By constructing a confidence rule base through data-driven approaches, the average causal effect is calculated by intervening in the premise attributes, resulting in the reference value, premise attribute weight, confidence distribution, and rule weight of the confidence rule base. Causal interpretability constraints are introduced during the optimization process of the confidence rule base to protect the causal interpretability of the confidence rule base parameters; Based on the confidence rule base, the fault diagnosis reasoning results are obtained by reasoning on the input observation data.

[0006] Preferably, the step of constructing a confidence rule base through data-driven methods, intervening in the premise attributes to calculate the average causal effect, and obtaining the reference value of the confidence rule base, the weight of the premise attributes, the confidence distribution, and the rule weights includes the following steps: The average causal effect is calculated for each reference interval of the intervention premise attribute to determine the reference value of the confidence rule base and quantify the impact of the premise attribute reference interval on the result. By intervening in the values ​​of prerequisite attributes, the average causal effect is calculated, the weights and confidence distributions of prerequisite attributes in the confidence rule base are determined, and the contribution of the reference values ​​of prerequisite attributes to the results is quantified. By calculating the average causal effect using the values ​​of the joint intervention premise attributes, the rule weights of the confidence rule base are determined, thus quantifying the impact of rule antecedents on the outcome.

[0007] Preferably, the step of determining the reference value of the confidence rule base by calculating the average causal effect of each reference interval of the intervention premise attribute includes the following steps: The parameters are initialized, and the baseline reference interval is determined. The average causal effect of each reference interval is calculated based on the baseline reference interval, and the causal effect threshold is calculated. The causal effect threshold is used as a threshold standard to judge whether the causal effect is significant. The reference interval is adjusted based on the causal effect threshold to determine the reference value.

[0008] Preferably, the method of adjusting the reference interval based on the causal effect threshold to determine the reference value includes the following steps: The causal effect interval is determined based on the causal effect threshold; If a reference interval contains a low causal effect interval, then the reference intervals are merged; if no low causal effect intervals are found in the reference interval, then the reference interval is the final reference interval. After merging the intervals, determine whether the conditions for determining the final reference interval are met. If they are met, output the reference value; otherwise, recalculate the average causal effect of the merged intervals and determine whether there are intervals with causal effects based on the new causal effect threshold.

[0009] Preferably, the step of calculating the average causal effect by intervening in the values ​​of the premise attributes, and determining the premise attribute weights and confidence distribution of the confidence rule base, includes the following steps: Determine the baseline reference value for each prerequisite attribute; By combining the baseline reference values ​​with the values ​​of the intervention prerequisite attributes, the average causal effect of the reference values ​​is calculated to form an average causal effect matrix, which represents the average causal effect of the prerequisite attributes under different intervention settings. The weights of the premise attributes in the confidence rule base are calculated based on the mean of the average causal effect matrix of each premise attribute. Preferably, the step of calculating the average causal effect by intervening in the values ​​of the premise attributes, and determining the premise attribute weights and confidence distribution of the confidence rule base, includes the following steps: The average causal effect matrix is ​​normalized and transformed into a causal confidence matrix; Calculate the confidence distribution of the result for each rule based on the causal confidence matrix; Based on the causal confidence matrix, the confidence distribution of the confidence rule base is determined by ER fusion.

[0010] Preferably, the step of determining the rule weights of the confidence rule base by calculating the average causal effect through the values ​​of the joint intervention premise attributes includes the following steps: Discretize the prerequisite attributes. Determine the baseline reference value combination and the reference value combination for each rule; The joint intervention calculates the average causal effect of each rule, and the average causal effect value is standardized and normalized to be mapped to rule weights.

[0011] Preferably, the process of introducing causal interpretability constraints during the optimization of the confidence rule base to protect the causal interpretability of the confidence rule base parameters includes the following steps: The following three causal interpretability constraints are introduced during the optimization process: The first approach is to add a Euclidean distance term to the objective function to achieve multi-objective optimization. The second approach is to apply the principle of non-optimization to inactive rules, and retain the initial values ​​of the inactive rule parameters determined by the average causal effect. The third approach is to introduce average causal effect boundary constraints on confidence level and rule weights.

[0012] This invention also provides a fault diagnosis system based on a confidence rule base and average causal effect, comprising: The rule base construction module builds a confidence rule base through data-driven construction and, in conjunction with the average causal effect, intervenes in the reference values, premise attribute weights, confidence distributions, and rule weights in the confidence rule base in a targeted manner. The rule base causal interpretability optimization module is used to introduce causal interpretability constraints during the optimization process of the confidence rule base to protect the causal interpretability of the confidence rule base parameters. The fault diagnosis reasoning module is used to receive input observation data based on the optimized confidence rule base, perform reasoning operations through rule matching and confidence reasoning algorithms, and output the corresponding fault diagnosis reasoning results.

[0013] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention fully considers the causal relationship between premise attributes and results in the selection of initial model parameters. It utilizes the average causal effect to intervene in the reference values ​​of the confidence rule base, quantifying the different causal effects of premise reference intervals on the results. By reducing reference intervals with low average causal effect values ​​through a causal significance threshold, it more accurately determines more reasonable and concise reference values ​​for premise attributes, effectively reducing the number of rules and mitigating the combinatorial explosion problem.

[0014] 2. This invention utilizes a data-driven approach combined with average causal effects to determine key model parameters such as rule confidence distribution, premise attribute weights, and rule weights, replacing the traditional method that relies on expert knowledge. This approach not only enhances objectivity but also significantly improves the efficiency and accuracy of subsequent high-dimensional parameter optimization. By providing an initial model with high inference accuracy, it establishes a robust starting point for optimization, accelerates convergence, and improves the performance of the final model.

[0015] 3. This invention utilizes causal theory to explain the logic of rule generation, introduces causal interpretability constraints into the model, and enhances the interpretability of model parameters through causal logic by improving the causal interpretability of initial parameters during the optimization process. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.

[0017] Figure 1 A flowchart illustrating the system fault diagnosis method based on confidence rule base and average causal effect provided by the present invention; Figure 2 The application flowchart of the system fault diagnosis method based on confidence rule base and average causality effect provided by the present invention; Figure 3 A schematic diagram illustrating the principle of the method for determining the confidence distribution of the rule results in this invention; Figure 4 A schematic diagram illustrating the principle of the method for determining the confidence distribution of the rule results in this invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0019] In the description of this invention, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0020] In the description of this invention, "several" means one or more, "multiple" means two or more, "greater than," "less than," and "exceeding" are understood to exclude the stated number, while "above," "below," and "within" are understood to include the stated number. Where the terms "first," "second," and "third" are used for descriptive purposes and to distinguish technical features, they should not be construed as indicating or implying relative importance, or implicitly indicating the number of indicated technical features, or implicitly indicating the sequential relationship of the indicated technical features.

[0021] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances. Furthermore, the technical features involved in the different embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0022] See Figure 1 A fault diagnosis method based on confidence rule base and average causal effect includes the following steps: constructing a confidence rule base through data-driven approach, intervening in the premise attributes to calculate the average causal effect, and obtaining the reference value of the confidence rule base, the weight of the premise attributes, the confidence distribution, and the rule weights; Causal interpretability constraints are introduced during the optimization process of the confidence rule base to protect the causal interpretability of the confidence rule base parameters; Based on the confidence rule base, the fault diagnosis reasoning results are obtained by reasoning on the input observation data.

[0023] The method for obtaining fault diagnosis inference results based on input observation data using a confidence rule base includes the following steps: constructing an initial confidence rule base according to determined initial parameters; inputting the observation data into an optimization algorithm with added causal interpretability constraints to train and obtain the final parameters of the confidence rule base; calculating the matching degree between the premise attributes and reference values ​​in the observation data based on the optimized confidence rule base; calculating the activation weight of each rule in the confidence rule base; and using the ER parsing algorithm to calculate the final confidence level of the evaluation results to obtain the final fault diagnosis result.

[0024] Compared with existing technologies, this invention proposes to integrate average causal effects into the confidence rule base to construct a novel system fault diagnosis method. It uses average causal effects to quantify the complex nonlinear causal relationship between premise attributes and outcomes, determining the reference values, premise attribute weights, rule weights, and outcome confidence distributions of the confidence rule base, thus aligning the parameters of the rule base with the actual physical mechanisms of the system. This invention shifts the confidence rule base from statistical correlation to causal mechanism modeling, improving the accuracy, interpretability, and physical meaning of fault diagnosis.

[0025] The aforementioned data-driven confidence rule base is constructed to intervene in the premise attributes and calculate the average causal effect. By calculating the ACE curve of a certain premise attribute with respect to the result, the inflection point or threshold of its significant change is directly set as a reference value. For example, if the ACE analysis shows that the set temperature is the critical point at which the causal effect of the corresponding equipment failure increases sharply, the set temperature can be set as a higher-level cutoff point.

[0026] For the weighting of prerequisite attributes, traditional optimization methods often fail to reflect the true importance of the attributes. The method described above constructs an initial confidence rule base based on determined initial parameters; the observed data is input into an optimization algorithm with added causal interpretability constraints to train and obtain the final parameters of the confidence rule base. Weights are assigned to each attribute based on its ACE value; a larger ACE value indicates a stronger causal driving force of the attribute on the result, and the weight should be increased accordingly. For example, if the causal effect of vibration on a fault is twice that of temperature, its weight should also be approximately twice that of temperature, making the fault diagnosis results more accurate.

[0027] See Figure 2In a preferred embodiment, the step of constructing a confidence rule base through data-driven methods, intervening in the premise attributes to calculate the average causal effect, and obtaining the reference value, premise attribute weight, confidence distribution, and rule weight of the confidence rule base includes the following steps: determining the reference value of the confidence rule base by calculating the average causal effect of each reference interval of the premise attribute, and quantifying the impact of the premise attribute reference interval on the result; determining the premise attribute weight and confidence distribution of the confidence rule base by calculating the average causal effect of the values ​​of the premise attributes, and quantifying the contribution of the premise attribute reference value to the result; and determining the rule weight of the confidence rule base by jointly intervening in the values ​​of the premise attributes to calculate the average causal effect, and quantifying the impact of the rule antecedent on the result.

[0028] In BRB systems, rules are used to describe the causal relationships between presupposition attributes and their associated outcomes. Their interpretability depends on the rule parameters, which reflect the true physical causal relationships of the actual system. Setting initial parameters through expert experience or data-driven methods both have inherent limitations. Compared to existing technologies, this invention integrates causal theory into a data-driven approach, transforming the BRB model from a statistical association to an expression of causal mechanisms, offering the following advantages: (1) Through the causal mechanism, the parameters can be aligned with the real physical causal relationship, thereby ensuring the physical meaning of the parameters and improving the traceability of the decision.

[0029] (2) Causal theory can identify the causal influence of premise attributes in a rule on the result. Using a significance threshold can reduce weakly correlated parameters and retain only rules with strong causal relationships, thereby simplifying the rule base.

[0030] (3) Causal influence can quantify the contribution of prior attributes to the result, thus enabling them to be converted into initial parameters. Compared with random initialization or pure data fitting, causal prior parameters reduce the parameter search space, accelerate convergence efficiency, and improve global optimality.

[0031] See Figure 2 In a preferred embodiment, the step of determining the reference value of the confidence rule base by calculating the average causal effect of each reference interval of the intervention premise attribute includes the following steps: initializing the parameters and determining the baseline reference interval; calculating the average causal effect of each reference interval based on the baseline reference interval and calculating the causal effect threshold, wherein the causal effect threshold is used as a threshold standard for judging whether the causal effect is significant; adjusting the reference interval based on the causal effect threshold, thereby determining the reference value.

[0032] In a specific implementation plan, when determining the reference values ​​for the confidence rule base: The parameter initialization process and baseline reference interval determination include: First, statistically determining the global boundaries of the data. The upper and lower limits of the sample space are directly set as the boundaries of the reference intervals based on historical data. Then, initial reference values ​​are generated through equal-width binning. For each antecedent attribute, its data range is divided into several equal-width intervals, each serving as a potential reference interval. After constructing the initial reference interval set, the baseline reference intervals used for calculating the average causal effect (ACE) are determined by statistically identifying the extreme values ​​of each antecedent attribute under normal conditions.

[0033] The average causal effect of each reference interval is calculated based on the baseline reference interval using the following formula: Formula I In Formula I This represents the average causal effect of reference interval A on outcome variable D. A represents the reference interval of the premise attribute being evaluated, while B represents the baseline reference interval of that attribute under normal conditions. This indicates that the prerequisite attribute value is set to belong to an intervention within a specific interval A. For each reference interval A, the average causal effect is the difference between the mean of the outcome variable within that interval and the mean of the outcome variable under the baseline reference interval conditions. This difference constitutes the causal effect of that interval on the outcome variable.

[0034] The causal effect threshold is determined by calculating the causal effect threshold using the following formula: Formula II Formula III Formula IV Based on the ACE values ​​calculated in the previous step, a quantile-based method is used to determine the causal effect threshold for each premise attribute. This is a sequence of ordered absolute ACE values ​​for a given reference interval of the premise attributes. and specified quantile level Calculate the ACE threshold Q for each prerequisite attribute.

[0035] Formula II is used to calculate the position index of a quantile in an ordered sequence. pos This represents the calculated location index value, where n represents the ordered sequence of absolute ACE values ​​of the reference interval for the given attribute. The number of elements in the sequence, n, must be greater than or equal to 2, because if the sequence has only one element, there is no point in calculating the threshold based on quantiles.

[0036] Formula III is used to calculate the position index value obtained in Formula II. pos It is decomposed into integer and fractional parts for subsequent linear interpolation to calculate the threshold.m Yes pos The integer part obtained by rounding down, r is pos The decimal part.

[0037] Formula IV is used based on the integer part m and decimal part r The final causal effect threshold Q is determined in the ordered sequence using linear interpolation. Q represents the ACE threshold of each premise attribute, which will be used to determine whether the causal effect of the premise attribute is significant. S m For ordered sequences The index is m The element value is the starting point for linear interpolation. S m+1 For an ordered sequence with index 1 m+1 The element value.

[0038] See Figure 2 In a more preferred embodiment, the step of adjusting the reference interval based on the causal effect threshold to determine the reference value includes the following steps: judging the causal effect interval of the reference interval according to the causal effect threshold; if the reference interval has a causal effect interval, then the reference interval is merged; if the reference interval does not have a causal effect interval, then the reference interval is the final reference interval; after merging the intervals, it is determined whether the conditions for determining the final reference interval are met; if they are met, the reference value is output; if they are not met, the average causal effect of the merged intervals is recalculated, and the causal effect interval is judged based on the new causal effect threshold to determine whether the merged interval has a causal effect interval.

[0039] The process involves adjusting reference intervals, controlled by a causal effect threshold. Intervals with absolute ACE values ​​below the causal effect threshold are merged, prioritizing the direction with the smaller absolute ACE value among adjacent intervals, and removing their corresponding boundaries. Boundary intervals must be retained; those below the causal effect threshold cannot be deleted but can be merged with adjacent intervals. Left boundary intervals can only be merged to the right, removing their right boundary. Right boundary intervals can only be merged to the left, removing their left boundary. Furthermore, to ensure sufficient discriminatory power among reference values, at least two reference intervals must be retained after merging. After each merge operation, the ACE value of the currently retained intervals is recalculated. This process repeatedly checks for the existence of low-ACE intervals. This cycle of merging, recalculating, and checking continues until all intervals meet the threshold requirements or reach the minimum interval limit.

[0040] See Figure 2In a preferred embodiment, determining the weights and confidence distributions of the premise attributes by intervening in the values ​​of the premise attributes through the average causal effect includes the following steps: Determine the baseline reference value for each prerequisite attribute; combine the baseline reference value with the values ​​of the prerequisite attributes under different intervention settings to calculate the average causal effect of the reference value, forming an average causal effect matrix to represent the average causal effect of the prerequisite attribute under different intervention settings; calculate the prerequisite attribute weight based on the mean of the average causal effect matrix of each prerequisite attribute; calculate the causal confidence matrix based on the average causal effect matrix; and determine the confidence distribution based on the causal confidence matrix and ER fusion.

[0041] In the specific implementation plan, the determination of the prerequisite attribute weights for the confidence rule base is carried out in the following manner: (1) Determine each prerequisite attribute X i Baseline reference value b i .

[0042] (2) By intervening in the values ​​of the prerequisite attributes, the average causal effect of each reference value on the outcome is calculated using the following formula: Formula V In formula V, Representative reference value The average causal effect on the outcome. This represents the j-th reference value of the i-th prerequisite attribute. This represents the level of the nth evaluation result. Indicates the prerequisite attributes for intervention The value is i=1,...,M, where M is the number of prerequisite attributes, and n=1,...,N, where N is the number of evaluation result levels.

[0043] (3) Generate an average causal effect matrix by calculating the average causal effect of each reference value on the result. The average causal effect matrix describes the average causal effect of the premise attribute under different intervention settings, and is given by the following formula: Formula VI In formula VI, J i Indicates the prerequisite attribute X i Number of reference values, average causal effect matrix It is a matrix that quantifies the causal influence of the precondition attribute range on the result.

[0044] (4) Calculate each prerequisite attribute X iThe mean of the average causality matrix is ​​used as the mean, and the obtained value is mapped to the premise attribute weights by performing a minimum-maximum normalization. δ i This allows us to obtain the premise attribute weights of the confidence rule base. δ i .

[0045] The confidence distribution of the confidence rule base is determined in the following way: (1) First, calculate the average causal effect matrix. Normalization process to convert into causal confidence matrix Formula VII (2) Based on the causal confidence matrix CBM i The confidence distribution of the results for each rule is determined according to the ER fusion formula, specifically by calculating the confidence distribution of the results for each rule using the following formula: Formula VIII Formula IX In the above ER fusion mathematical formula, express The Middle i Under the premise attribute, the first j The interval for the first n The causal confidence of the result. μ This represents the normalization factor.

[0046] (3) Finally, based on the causal confidence matrix Fusion with Evidence-Based Reasoning (ER) to determine the confidence distribution like Figure 3 As shown, according to the ER fusion principle, if rule R k The antecedent is X1=A 1，k ∧X2=A 2，k ∧⋯∧X M =A M，k (M is the total number of prerequisite attributes, A) M，k It is the first M If the k-th interval of an attribute is selected, then multiple corresponding events will be triggered. CBM matrix.

[0047] For each triggered Extract the corresponding confidence vector from the matrix. , ... N is the total number of outcome categories, serving as evidence of the premise attribute's influence on the outcome. The evidence of all premise attributes is fused using an evidence reasoning algorithm to ultimately obtain the confidence distribution of the confidence rule.β 1,k , β 1,k ... β 1N,k ].

[0048] See Figure 1 and see Figure 2 In a preferred embodiment, determining the rule weights by assigning values ​​to the prerequisite attributes for joint intervention includes the following steps: Discretize the premise attributes; determine the baseline reference value combination and the reference value combination for each rule; jointly intervene to calculate the average causal effect of each rule, and standardize and normalize the average causal effect value to map it into rule weights.

[0049] In specific implementation plans, such as Figure 4 As shown, the average causal effect of the antecedent of each rule on the outcome is calculated using the following causal inference intervention model formula: Formula X The above formula, a method for determining rule weights through joint intervention, uses joint intervention to calculate the average causal effect of the antecedents of each rule on the outcome, where, This represents the combination of reference values ​​for the k-th rule. For the baseline reference value combination, k=1,...L, where L is the number of rules. Let represent the average causal effect of the combination of antecedents of the k-th rule on the outcome D.

[0050] The object of intervention for each rule k is determined by formula X. Calculate the average causal effect for each rule. Quantification rule antecedent combination c k The strength of the causal influence on the failure outcome D. The standardization and normalization processes are applied to map the weights to rule k.

[0051] See Figure 1 and see Figure 2 In a preferred embodiment, the step of introducing causal interpretability constraints based on rule weights and confidence distributions to optimize the causal interpretability of parameters includes the following steps: Based on rule weights and confidence distributions, the following three causal interpretability constraints are introduced during the optimization process: First, a Euclidean distance term is added to the objective function to achieve multi-objective optimization; second, the principle of non-optimization is applied to inactive rules, and the parameters of inactive rules retain their initial values ​​determined by the average causal effect; third, the average causal effect boundary constraint is introduced for confidence and rule weights.

[0052] When constructing confidence rules by determining the reference values, premise attribute weights, rule weights, and confidence distributions of the confidence rule base, the confidence rule base is optimized by introducing the aforementioned three causal interpretability constraints to maintain the causal interpretability of the parameters. In a specific implementation plan, for the first approach, a Euclidean distance term is added to the optimization objective function to achieve multi-objective optimization: [The specific implementation details the Euclidean distance term and its implications for the optimization objective function.] Formula XI In formula XI above, This is the Euclidean distance term, which measures the degree of deviation between the optimized parameters and the initial causal parameters. It is used to protect the causal interpretability of the initial parameters and should be controlled to prevent it from becoming too large.

[0053] For the second approach, the principle of non-optimization is applied to inactive rules. The parameters of inactive rules retain their initial values ​​determined by the average causal effect. Inactive rules are not optimized, and the initial parameters determined by the average causal effect are retained.

[0054] For the third approach, an average causal effect boundary constraint is introduced for the confidence level and rule weights, using the following functional formula: Formula XII In the above formula XII, and These represent the initial confidence level and the initial rule weight, respectively, and ε is a constant representing the parameter adjustment range.

[0055] Boundary constraints on the confidence score of the confidence rule base model, and the optimized confidence score. β n,k It must fall within the initial confidence level Centered on ε×(1- ACE k ) represents the range of the interval length. Where (1- ACE k This reflects the constraint of causal effect on the adjustment range. The stronger the causal effect, the smaller the adjustment range, thus more strictly protecting the initial causal information.

[0056] Boundary constraints on rule weights of the confidence rule base model, and optimized rule weights. θ k It must fall under the initial rule weight Centered on ε×(1- ACE k The range is defined as the interval length. The principle is consistent with the confidence constraint, and the purpose is to protect the causal interpretability of the rule weights.

[0057] This invention also provides a fault diagnosis system based on a confidence rule base and average causal effect, comprising: The rule base construction module builds a confidence rule base through data-driven approaches, intervenes in the premise attributes, calculates the average causal effect, and determines the reference value, premise attribute weight, confidence distribution, and rule weight in the confidence rule base.

[0058] The rule base causal interpretability optimization module is used to introduce causal interpretability constraints during the optimization process of the confidence rule base to protect the causal interpretability of the confidence rule base parameters.

[0059] The fault diagnosis reasoning module is used to receive input observation data based on the optimized confidence rule base, perform reasoning operations through rule matching and confidence reasoning algorithms, and output the corresponding fault diagnosis reasoning results.

[0060] In summary, this invention integrates causal theory into data-driven construction of a confidence rule base, determines model parameters through causal mechanisms, and shifts the confidence rule base from statistical correlation to causal mechanism modeling, thereby improving the accuracy, interpretability, and physical meaning of fault diagnosis.

[0061] Those skilled in the art will recognize that the units of the various examples described in connection with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of the invention.

[0062] In the embodiments provided by the present invention, it should be understood that the division of units is only a logical functional division. In actual implementation, there may be other division methods, such as multiple units can be combined into one unit, one unit can be split into multiple units, or some features can be ignored.

[0063] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0064] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0065] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A fault diagnosis method based on a confidence rule base and average causal effect, characterized in that, Includes the following steps: By constructing a confidence rule base through data-driven approaches, the average causal effect is calculated by intervening in the premise attributes, resulting in the reference value, premise attribute weight, confidence distribution, and rule weight of the confidence rule base. Causal interpretability constraints are introduced during the optimization process of the confidence rule base to protect the causal interpretability of the confidence rule base parameters; Based on the confidence rule base, the fault diagnosis reasoning results are obtained by reasoning on the input observation data.

2. The fault diagnosis method based on confidence rule base and average causal effect according to claim 1, characterized in that, The process of constructing a confidence rule base through data-driven methods, intervening in the premise attributes to calculate the average causal effect, and obtaining the reference value of the confidence rule base, the weight of the premise attributes, the confidence distribution, and the rule weights includes the following steps: The average causal effect is calculated for each reference interval of the intervention premise attribute to determine the reference value of the confidence rule base and quantify the impact of the premise attribute reference interval on the result. By intervening in the values ​​of prerequisite attributes, the average causal effect is calculated, the weights and confidence distributions of prerequisite attributes in the confidence rule base are determined, and the contribution of the reference values ​​of prerequisite attributes to the results is quantified. By calculating the average causal effect using the values ​​of the joint intervention premise attributes, the rule weights of the confidence rule base are determined, thus quantifying the impact of rule antecedents on the outcome.

3. The fault diagnosis method based on confidence rule base and average causality effect according to claim 2, characterized in that, The method of determining reference values ​​for the confidence rule base by calculating the average causal effect across reference intervals of intervention premise attributes includes the following steps: The parameters are initialized, and the baseline reference interval is determined. The average causal effect of each reference interval is calculated based on the baseline reference interval, and the causal effect threshold is calculated. The causal effect threshold is used as a threshold standard to judge whether the causal effect is significant. The reference interval is adjusted based on the causal effect threshold to determine the reference value.

4. The fault diagnosis method based on confidence rule base and average causal effect according to claim 3, characterized in that, The method of adjusting the reference interval based on the causal effect threshold to determine the reference value includes the following steps: The causal effect interval is determined based on the causal effect threshold; If a reference interval contains a low causal effect interval, then the reference intervals are merged; if no low causal effect intervals are found in the reference interval, then the reference interval is the final reference interval. After merging the intervals, determine whether the conditions for determining the final reference interval are met. If they are met, output the reference value; otherwise, recalculate the average causal effect of the merged intervals and determine whether there are intervals with low causal effect based on the new causal effect threshold.

5. The fault diagnosis method based on confidence rule base and average causality effect according to claim 2, characterized in that, The method of calculating the average causal effect by intervening in the values ​​of premise attributes, and determining the premise attribute weights and confidence distribution of the confidence rule base, includes the following steps: Determine the baseline reference value for each prerequisite attribute; By combining the baseline reference values ​​with the values ​​of the intervention prerequisite attributes, the average causal effect of the reference values ​​is calculated to form an average causal effect matrix, which represents the average causal effect of the prerequisite attributes under different intervention settings. The weights of the premise attributes in the confidence rule base are calculated based on the mean of the average causal effect matrix of each premise attribute.

6. The fault diagnosis method based on confidence rule base and average causality effect according to claim 5, characterized in that, It also includes the following steps: The average causal effect matrix is ​​normalized and transformed into a causal confidence matrix; Calculate the confidence distribution of the result for each rule based on the causal confidence matrix; Based on the causal confidence matrix, the confidence distribution of the confidence rule base is determined by ER fusion.

7. The fault diagnosis method based on confidence rule base and average causal effect according to claim 2, characterized in that, The method of determining the rule weights of the confidence rule base by calculating the average causal effect through the values ​​of joint intervention premise attributes includes the following steps: Discretize the prerequisite attributes. Determine the baseline reference value combination and the reference value combination for each rule; The joint intervention calculates the average causal effect of each rule, and the average causal effect value is standardized and normalized to be mapped to rule weights.

8. The fault diagnosis method based on confidence rule base and average causal effect according to claim 1, characterized in that, The method of introducing causal interpretability constraints during the confidence rule base optimization process to protect the causal interpretability of the confidence rule base parameters includes the following steps: The following three causal interpretability constraints are introduced during the optimization process: The first approach is to add a Euclidean distance term to the objective function to achieve multi-objective optimization. The second approach is to apply the principle of non-optimization to inactive rules, and retain the initial values ​​of the inactive rule parameters determined by the average causal effect. The third approach is to introduce average causal effect boundary constraints on confidence level and rule weights.

9. The fault diagnosis method based on confidence rule base and average causal effect according to claim 1, characterized in that, The method of obtaining fault diagnosis inference results by reasoning from the input observation data based on the confidence rule base includes the following steps: Construct an initial confidence rule base based on the determined initial parameters; The observed data is input into an optimization algorithm with added causal interpretability constraints, and the final parameters of the confidence rule base are obtained through training. Based on the optimized confidence rule base, the degree of matching between the premise attribute and the reference value in the observed data is calculated; Calculate the activation weight of each rule in the confidence rule base; The final confidence level of the evaluation results is calculated using the ER analytical algorithm to obtain the final fault diagnosis result.

10. A fault diagnosis system based on a confidence rule base and average causal effect, characterized in that, include: The rule base construction module builds a confidence rule base through data-driven construction, intervenes in the premise attributes, calculates the average causal effect, and determines the reference value, premise attribute weight, confidence distribution, and rule weight in the confidence rule base. The rule base causal interpretability optimization module is used to introduce causal interpretability constraints during the optimization process of the confidence rule base to protect the causal interpretability of the confidence rule base parameters. The fault diagnosis reasoning module is used to receive input observation data based on the optimized confidence rule base, perform reasoning operations through rule matching and confidence reasoning algorithms, and output the corresponding fault diagnosis reasoning results.

Images