Intelligent fault diagnosis method of fault tree based on genetic algorithm and gain factor dual drive

Abstract

The invention discloses an intelligent fault diagnosis method of a fault tree based on dual drive of a genetic algorithm and a gain factor, and the method comprises the following steps: 1, building a bidirectional mapping fault tree model according to the phenomenon and reason of a fault event; 2, in the fault tree model, a dynamic sub-tree lightening and extinguishing mechanism is introduced, and the reason probability is recalculated in real time according to a fault field observation result; 3, constructing a shared bottom event gain factor model, optimizing adjustable parameters, and adjusting the cause probability calculated in the step 2 by using the optimized model; and step 4, taking the adjusted cause probability as an evaluation index of fault diagnosis, and outputting a maintenance priority list. According to the method, the interpretability of the fault tree is maintained, meanwhile, the diagnosis precision and efficiency under the multi-concurrent fault scene are remarkably improved, and technical support can be provided for equipment maintenance, fault prediction and task decision making.

Description

Technical Field

[0001] This invention relates to a fault diagnosis method, and more particularly to an intelligent fault diagnosis method based on a fault tree driven by both genetic algorithm and gain factor. Background Technology

[0002] This section provides only background information relevant to this disclosure and is not necessarily prior art.

[0003] In the full lifecycle operation and maintenance of complex equipment and industrial systems, fault diagnosis is always a core element in ensuring mission continuity and reducing downtime losses. Since its inception in the 1960s, traditional Fault Tree Analysis (FTA), with its clear Boolean logic, intuitive graphical representation, and solid foundation in probability theory, has been widely applied in high-safety fields such as nuclear power, aerospace, and rail transportation. However, with the exponential expansion of system scale and the dramatic increase in functional density, the limitations of traditional FTA in terms of dynamism, concurrency, and data adaptability have become increasingly apparent. On the one hand, the continuous generation of condition monitoring data, maintenance records, and environmental disturbance information during equipment operation makes it difficult for static prior probabilities to reflect the true health status in real time. On the other hand, the prevalent "one cause, multiple effects," "multiple causes, one effect," and forced linkage phenomena in modern systems cause the probability contribution of shared base events to be repeatedly calculated or completely submerged across multiple fault subtrees, often leading to diagnostic conclusions that deviate from reality. Even more challenging is that in time-sensitive scenarios such as joint operations and emergency rescue, maintenance windows are compressed to minutes or even seconds. Traditional parameter tuning methods that rely on expert experience can no longer meet the decision-making needs of "fast, accurate, and explainable".

[0004] To address these challenges, academia and industry have successively introduced probabilistic graphical models such as dynamic fault trees, Bayesian networks, and Markov decision processes, aiming to capture the dynamic evolution of systems through state transitions or online learning mechanisms. However, while these methods achieve higher accuracy, they also bring new problems such as node explosion, difficulty in obtaining parameters, and severe black-box nature. Especially in battlefield or disaster environments, limited computing resources and scarce prior data make the deployment of highly complex models extremely difficult. In recent years, evolutionary computing techniques, represented by genetic algorithms and particle swarm optimization, have been attempted for FTA parameter tuning, trying to alleviate the lack of human experience through data-driven approaches. However, single-population evolutionary algorithms are prone to getting trapped in local optima when dealing with high-dimensional, multi-peak, and complexly constrained fault spaces, and the serial computing mode cannot fully utilize the parallel advantages of modern heterogeneous hardware.

[0005] It should be noted that the information disclosed in the background section above is only used to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] Purpose of the invention: The technical problem to be solved by the present invention is to provide an intelligent fault diagnosis method based on a fault tree driven by both genetic algorithm and gain factor, which addresses the shortcomings of the existing technology.

[0007] To address the aforementioned technical problems, this invention discloses an intelligent fault diagnosis method based on a fault tree driven by both genetic algorithms and gain factors, comprising the following steps:

[0008] Step 1: Based on the phenomena and causes of the fault events, establish a bidirectional mapping fault tree model;

[0009] Step 2: In the fault tree model, a dynamic subtree lighting and extinguishing mechanism is introduced to recalculate the cause probability in real time based on the fault site observation results.

[0010] Step 3: Construct a shared bottom event gain factor model and optimize the adjustable parameters. Use the optimized model to adjust the cause probabilities calculated in Step 2.

[0011] Step 4: Using the adjusted cause probability as the evaluation index for fault diagnosis, output a maintenance priority list.

[0012] Furthermore, the establishment of the bidirectional mapping fault tree model described in step 1 includes:

[0013] Step 1-1 involves dividing various fault information obtained from equipment or industrial systems into a three-layer node structure: fault events, fault phenomena, and fault causes. Different layers are connected via AND or OR gates, forming parent-child node relationships.

[0014] Fault event S is a top-level node used to represent a system or device failure;

[0015] Fault phenomenon M is an intermediate node used to represent specific observable fault content; intermediate nodes are divided into first-level intermediate nodes and lower-level intermediate nodes according to their nesting relationship.

[0016] The cause of the failure, R, is the underlying node.

[0017] Steps 1-2: Based on the phenomena and causes of the fault events, the fault tree is divided into reusable subtree units; each subtree unit contains several intermediate nodes M and bottom-level nodes R.

[0018] Steps 1-3: Construct an independent base probability library for each subtree. For each non-first-level intermediate node in each subtree unit, assign a base probability as the initial cause probability of that node. If the same bottom-level node exists in multiple subtree units, assign a base probability to it independently in each subtree.

[0019] Furthermore, the dynamic subtree lighting and extinguishing mechanism described in step 2 includes:

[0020] When a fault M is observed on-site, its corresponding subtree is immediately lit up; otherwise, the unobserved subtree is turned off.

[0021] A subtree connected by an AND gate is an AND subtree, and all nodes connected by an AND gate together light up the upper-level node; a subtree connected by an OR gate is an OR subtree, and each lower-level node connected by an OR gate individually lights up the upper-level node.

[0022] When the OR subtree is lit, the cause probability of all its bottom-level nodes is forcibly set to 1; when the OR subtree is turned off, the cause probability of its bottom-level nodes is set to 0. If the bottom-level node exists under other OR subtrees, its base probability in other OR subtrees is set to 0, and the base probability of the bottom-level node is redistributed to the remaining lit nodes according to the initial proportion of the remaining lit nodes.

[0023] The priority rule is: OR gate off > OR gate on > AND gate off.

[0024] Furthermore, the shared bottom event gain factor model described in step 3 includes:

[0025] If the bottom node exist If a node is lit up repeatedly in a subtree, the cause probability of that node is adjusted according to the base probability. The specific calculation method is as follows:

[0026]

[0027] in, The adjusted causal probability, for the bottom-level nodes. The basic probability vector is represented as , For the bottom layer node In the The basic probability in a subtree. This indicates taking the maximum value; and This is an adjustable parameter.

[0028] Furthermore, the constraints in the shared bottom event gain factor model are as follows:

[0029] And the probability of cause With base probability There is a positive correlation;

[0030] And the probability of cause Number of child trees Monotonically increasing.

[0031] Furthermore, the optimization of the adjustable parameters described in step 3 refers to optimizing the adjustable parameters in the shared bottom event gain factor model. and Genetic algorithms are used for learning and optimization.

[0032] Furthermore, the optimization of the adjustable parameters described in step 3 specifically includes:

[0033] Using two-dimensional real-number encoding, each chromosome in the genetic algorithm is defined as a vector. ;

[0034] The population size in the genetic algorithm is set to... The initial chromosome is generated randomly.

[0035] Based on population size Half of the manually labeled examples serve as the baseline data;

[0036] Select similar sample pairs from the baseline data whose maximum probability value and total probability are close;

[0037] Based on the adjusted causal probability of each pair of similar samples , and Calculate the difference between predicted values and The details are as follows:

[0038]

[0039]

[0040] If not satisfied Otherwise, a penalty is imposed when calculating fitness;

[0041] Based on the calculated fitness, after evaluation, the population enters an evolutionary cycle integrating selection, crossover, and mutation until the number of iterations reaches a threshold or the fitness improvement rate is less than a threshold within a preset number of generations. This completes the genetic algorithm and yields optimized adjustable parameters. and .

[0042] Furthermore, the specific method for calculating fitness is as follows:

[0043] Fitness without penalty The calculation method is as follows:

[0044]

[0045] in, Penalties for violating the monotonicity constraint;

[0046] Fitness when punishment is applied The calculation method is as follows:

[0047]

[0048] in, , It is a constant;

[0049] The mean square error between the predicted and expected values ​​is calculated as follows:

[0050]

[0051] in, For the first The predicted value of the dataset, i.e., the adjusted causal probability calculated based on the shared bottom-event gain factor model. For the first The expected value of the set of data.

[0052] Furthermore, the step of selecting similar sample pairs from the benchmark data whose maximum probability value and total probability are close includes:

[0053] If the benchmark data contains data ,satisfy:

[0054]

[0055] The difference is <0.05

[0056] The difference is <0.1

[0057] in, , and They are the first , and The number of times the node is repeatedly lit in the sample group; , and These are the original probability vectors of the three groups of samples;

[0058] Then the data These are similar sample pairs.

[0059] Furthermore, the output maintenance priority list includes:

[0060] Calculate the adjusted cause probability for each node, sort them in descending order, and use this as the maintenance priority list for the bottom-level nodes.

[0061] Beneficial effects:

[0062] This invention uses the classic fault tree as its logical framework and superimposes three core mechanisms on it: dynamic probability recalculation, shared event gain amplification, and genetic parameter optimization. This allows for the simultaneous improvement of diagnostic accuracy and real-time performance while retaining the intuitive interpretability of Fault Tree Analysis (FTA). Attached Figure Description

[0063] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments, and the advantages of the present invention in the above and / or other aspects will become clearer.

[0064] Figure 1 This is an example diagram of a fault tree.

[0065] Figure 2 This is a schematic diagram of the bottom node that is lit up multiple times.

[0066] Figure 3 This is a schematic diagram of the overall process of the present invention.

[0067] Figure 4 This is a schematic diagram of a fault tree for a field survey vehicle in one embodiment. Detailed Implementation

[0068] This invention comprehensively considers the pain points of complex equipment and industrial systems in balancing multiple concurrent faults, real-time diagnosis, and interpretability, and proposes a fault tree intelligent diagnosis method driven by both rules and data.

[0069] This invention is achieved through the following technical solution: using a classic fault tree as the logical framework, and superimposing three core mechanisms on it: dynamic probability recalculation, shared event gain amplification, and genetic parameter optimization. This allows for the simultaneous improvement of diagnostic accuracy and real-time performance while retaining the intuitive interpretability of Fault Tree Analysis (FTA). The overall process of this method is as follows: Figure 3 As shown, the details are as follows:

[0070] 1. Establish a bidirectional mapping fault tree model

[0071] This invention, based on the classic fault tree, reclassifies and redefines the different levels of the fault tree, categorizing various fault information into a three-layer structure: fault event, fault phenomenon, and fault cause. Different levels are connected via AND / OR gates, forming parent-child node relationships, such as... Figure 1 As shown:

[0072] S: Fault event, which is a top-level node, usually representing a failure of a large system or device, such as a radio station, a vehicle, or a computer.

[0073] M: Fault phenomenon, which is an intermediate node and represents a specific, observable fault, such as communication failure, insufficient energy supply, or data anomaly. Intermediate nodes can be divided into first-level intermediate nodes and lower-level intermediate nodes based on their nesting relationship. In the figure, M1~M4 are first-level intermediate nodes, and M5 is a lower-level intermediate node.

[0074] R: Cause of failure, which is a low-level node, a detailed failure that is generally difficult to observe directly, such as a component failure, a circuit problem, or environmental interference.

[0075] During the system requirements decomposition phase, the fault tree is first divided into reusable subtree units based on the "phenomenon-cause" bidirectional mapping principle. Each subtree unit contains several intermediate node events (module or subsystem failures) and bottom-level nodes (basic failure causes). A base probability needs to be assigned to each non-first-level intermediate node. The same bottom-level node may exist in multiple subtree units (e.g., ...). Figure 1 In the case of R3 (which exists simultaneously in subtrees under M1 and M2), a base probability needs to be assigned independently to each subtree. This means establishing an independent base probability library for each subtree. Since the events at the bottom nodes of each subtree may occur simultaneously and are probabilistically compatible, the sum of the probabilities of the bottom nodes in each subtree is allowed to be greater than 1. This library not only integrates historical statistical data but also reserves a real-time monitoring interface, allowing the prior probabilities to be continuously updated based on sensor data during operation.

[0076] 2. Dynamic subtree lighting / extinguishing mechanism

[0077] When a fault is observed on-site, its corresponding subtree is immediately "lit up"; conversely, unobserved subtrees are "extinguished". Subtrees connected by AND gates are called AND subtrees ( Figure 1 The subtree containing the plus sign (+) and all nodes connected by the gate together light up the parent node; the subtree connected by the OR gate is the OR gate subtree (...). Figure 1 In the subtree containing the symbol (the one with the symbol), each lower-level node connected by a gate can individually light up the upper-level node. An example of a fault tree is shown below. Figure 1 As shown.

[0078] When an AND gate subtree is lit, the probabilities of all its underlying events are forcibly set to 1 to ensure the instantaneous satisfaction of "multiple cause-effect" logic. When an OR gate subtree is extinguished, the probability of its underlying node is set to 0. If the underlying node exists in another OR gate subtree, its base probability in the other OR gate subtree is set to 0, and the base probability of the underlying node is redistributed to the remaining lit nodes according to the initial proportion of the remaining lit nodes, thereby avoiding deviations caused by static assumptions. The corresponding priority rule is OR gate extinguished > OR gate lit > AND gate extinguished, ensuring logical consistency.

[0079] Through this mechanism, the system can complete dynamic probability recalculation in milliseconds, solving the problem that traditional FTAs ​​are unable to adapt to real-time changes in operating conditions due to static probability lag.

[0080] 3 Shared bottom event gain factor model

[0081] like Figure 2 As shown, if the subtrees under M1 and M2 are both lit up, then R3 is lit up twice in the two subtrees. R3 is its shared base event, and its final probability should be calculated based on the base probabilities under the two subtrees.

[0082] For scenarios where multiple concurrent failures share the same underlying event, this invention proposes a gain factor model: if the underlying event... exist The original probability vector of a subtree that is repeatedly lit up is: Then the probability after processing by the gain factor model is:

[0083] (1)

[0084] in , This is a globally adjustable parameter. The function satisfies:

[0085] and With all Positive correlation ,and Follow Monotonically increasing, and in The changes are significant when the size is small.

[0086] During the inference phase, logical consistency is first checked, such as disallowing illogical situations like a parent subtree being off while a child subtree is on within a nested OR gate subtree. Then, using the post-gain probability as a fault indicator, all basic events are sorted in reverse order, directly outputting a maintenance priority list. This "rule-data" hybrid engine retains the "white-box" advantages of FTAs ​​while incorporating the adaptability of data-driven approaches, enabling rapid implementation in equipment maintenance, battle damage assessment, and industrial process safety monitoring.

[0087] 4. Optimization based on genetic algorithm

[0088] Step 1: Encoding and Initialization

[0089] Two-dimensional high-precision real number encoding is used, and each chromosome in the genetic algorithm is defined as a vector. To avoid gain saturation or failure, let... In order to ensure When smaller The amplification effect, let .

[0090] Set the population size to N (initially 100), and generate the initial chromosomes randomly:

[0091] ,

[0092] in The above-mentioned area can be ensured through pre-testing. It neither saturates quickly nor fails to meet the trend of monotonicity.

[0093] Step 2: Fitness Assessment

[0094] Using half of N, i.e., 50 manually annotated examples, as a benchmark, calculate the mean squared error between the predicted and expected values:

[0095] (2)

[0096] in For the first Predicted values ​​for the set of data (calculated according to Formula 1). For the first The expected value of the set of data.

[0097] Formula 2 inherently satisfies the gain factor constraint, therefore no additional penalty function is needed. However, X will inevitably increase as n increases; and when n is small, the effect of increasing n on X will be more pronounced. The penalty function needs to quantify the degree of violation of this constraint.

[0098] From 50 sets of data, select "controllable comparison sample pairs": Define "similar sample pairs" as pairs of samples with similar maximum probability values ​​and similar sum of probabilities.

[0099] data satisfy ,in, , and They are the first , , The number of times this node is repeatedly "lit up" in the sample group. (Difference < 0.05) (Difference < 0.1), where , , These are the original probability vectors for the three groups of samples, ensuring that n is the only principal variable. For each pair of similar samples, the difference in predicted values ​​is calculated. and :

[0100]

[0101]

[0102] The effect is more pronounced when n is small. (Requirement:) Otherwise, a penalty will be imposed.

[0103] Set the fitness function (used to quantify the adaptability of each individual in the genetic algorithm to the environment, i.e., to judge the quality of individuals with specific values, and to serve as the basis for terminating the genetic algorithm) as follows:

[0104] (3)

[0105] in , Set the value to 500 and set a constant term. Avoid denominators of 0 to ensure numerical stability.

[0106] Step 3: Integrated Evolution of Selection, Crossover, and Mutation

[0107] After fitness assessment, the population enters a unified evolutionary cycle: first, tournament selection is used (three individuals are randomly selected each round, and the one with the highest fitness is retained), repeated N times until a new population of the same size as the parents is formed; then, linear crossover is performed on the selected individuals with a crossover probability of 0.85, and the parent vectors are... , Generate offspring ( If the offspring exceed the boundary, the loop is mapped to the feasible region. Finally, Gaussian mutation is triggered on the crossover individuals with a probability of 0.02. Random perturbations of N(0,0.01²) and N(0,0.1²) are added to the k and a distributions. If the offspring exceed the boundary, the loop is truncated to the boundary value. Thus, selection, crossover and mutation operations are completed simultaneously in one iteration, achieving efficient parameter space exploration and local refinement.

[0108] Step 4: Termination Conditions

[0109] The algorithm's fitness improvement rate is less than 1×10⁻ after 200 iterations or 25 consecutive iterations. 5 Stop when the time comes.

[0110] Example:

[0111] The following detailed implementation of the technical solution of this invention, in conjunction with the accompanying drawings and a fault diagnosis scenario of the electrical system of a field survey vehicle, ensures that the technical solution is reproducible and verifiable, while avoiding duplication with the framework description of the "Invention Content" and focusing on "operational details + data verification".

[0112] 1. Implementation Scenarios and Fault Tree Construction

[0113] Taking the electrical system of a certain type of field survey vehicle as the application object, a bidirectional mapping fault tree of "phenomenon-cause" was constructed (e.g., Figure 4 As shown in the figure, the hierarchy and logical relationship of this fault tree correspond to the model definition in Invention Content 1, and the specific structure is as follows:

[0114] (1) Fault event (top node S): Electrical system failure of field survey vehicle (manifested as concurrent abnormal function of multiple subsystems);

[0115] (2) Fault phenomenon (intermediate node M): The first-level intermediate node includes "radar failure (M1)", "engine cannot start (M2)", "driving power interruption (M3)", "navigation signal instability (M4)" and "radio signal interruption (M5)". Among them, "radio signal interruption (M5)" is a nested lower-level intermediate node, which is associated with two bottom events: "satellite signal obstruction (R6)" and "positioning module failure (R7)".

[0116] (3) Causes of failure (bottom node R): including “short circuit in radar power supply line (R1)”, “radar signal receiver failure (R2)”, “battery depletion (R3)”, “fuel line blockage (R4)”, “excessive carbon brush wear (R5)”, etc. Among them, “battery depletion (R3)” exists in both “engine cannot start (M2)” and “driving power interruption (M3)”, and is a shared bottom event;

[0117] (4) Construction of basic probability library: Each subtree is configured with a basic probability library, which integrates the maintenance records of this type of field survey vehicle for the past 5 years (a total of 1200 fault data), and reserves the vehicle CAN bus interface, which can access sensor data such as voltage, current, and signal strength in real time (sampling frequency 1Hz), and dynamically update the prior probability of the basic event (e.g., the initial probability of "battery depletion (R3)" is 0.3 in the M2 subtree and 0.2 in the M3 subtree).

[0118] 2. Execution flow of dynamic subtree lighting / extinguishing mechanism

[0119] (1) Scenario 1: Radar failure (M1) occurs alone. When "no signal output from radar" is observed, the M1 subtree is lit up; since the bottom nodes R1 and R2 of the M1 subtree are connected by an AND gate (both must fail simultaneously to cause M1), the probabilities of R1 and R2 are forcibly set to 1 according to the rules. At this time, the failure probability calculation result of the M1 subtree is 1 (satisfying the "multiple causes and one effect" logic), and the diagnosis takes 0.3ms.

[0120] (2)Scenario 2: The engine fails to start (M2) occurs, but the driving power is interrupted (M3) does not occur.

[0121] a. Light up the M2 subtree and turn off the M3 subtree; the bottom nodes of both the M2 and M3 subtrees contain the shared event R3 (battery depletion), and both subtrees are OR gate logic (a single bottom event failure can cause the upper layer phenomenon).

[0122] b. According to the OR gate extinguishing rule, set the probability of R3 in the M3 subtree to 0, and at the same time, redistribute its initial probability (0.2) in the M3 subtree according to the initial probability ratio (0.3:0.4:0.3) of the remaining lit nodes (R3, R4, R5) in the M2 subtree. After redistribution, the probability of R3 becomes 0.3+0.2×(0.3 / (0.3+0.4+0.3))=0.36, the probability of R4 becomes 0.4+0.2×(0.4 / 1)=0.48, and the probability of R5 becomes 0.3+0.2×(0.3 / 1)=0.36.

[0123] c. The dynamic probability recalculation process takes 0.5ms, avoiding the problem of "underestimation of R3 failure probability" caused by the static probability of traditional FTA.

[0124] (3) Scenario 3: Navigation signal instability (M4) occurs, radio signal interruption (M5) does not occur. Light up the M4 subtree and turn off the M5 subtree; M5 is a lower-level intermediate node, which has a base probability (0.15). According to the rules, set the posterior probability of M5 to 0, and distribute this probability according to the initial probability ratio (0.6:0.4) of its lower-level nodes R6 and R7. After the distribution, the probability of R6 changes from 0.5 to 0.5+0.15×0.6=0.59, and the probability of R7 changes from 0.3 to 0.3+0.15×0.4=0.36, ensuring the consistency of the probability distribution logic of nested intermediate nodes.

[0125] 3. Application of the shared bottom event gain factor model

[0126] For the scenario 2 where "engine cannot start (M2) + driving power interruption (M3) occurs concurrently" (both subtrees are lit, and R3 is lit twice), the final probability of R3 is calculated according to the gain factor formula in invention content 3. The specific steps are as follows:

[0127] (1) Parameter substitution: The original probability of R3 in the M2 subtree is P1=0.3, and the original probability in the M1 subtree is P2=0.2. The number of times it is repeatedly lit is n=2. The global parameters after optimization by the genetic algorithm are K=1.2 (range [0.5, 2.0]) and a=1.8 (range [1.2, 3.0]). Substitute them into the formula:

[0128]

[0129] The calculation yields:

[0130]

[0131] (2) Probability adjustment and logic verification:

[0132] a. After gaining, the final probability of R3 is 0.54. Simultaneously, according to the "gain node priority," the probabilities of other bottom nodes in the M1 and M2 subtrees are reduced by their original proportions (e.g., the probability of R4 is reduced from 0.4 to...). The R5 probability decreased from 0.3 to Ensure that the sum of probabilities within the subtree is reasonable;

[0133] b. The logical consistency check of the M2 subtree was completed, and no abnormalities such as "the parent subtree is off while the child subtree is on" were found. The pass rate of the check was 100%.

[0134] 4. Implementation Process of Genetic Algorithm Parameter Optimization

[0135] Based on the optimization framework of Invention Content 4, adaptive optimization of parameters K and a was performed on 50 manually labeled samples (including single-fault, double-fault, and triple-fault scenarios, with a labeling accuracy of 99%) of the fault tree of the field survey vehicle. The specific steps and results are as follows:

[0136] (1) Step 1: Encoding and initialization adopt two-dimensional real number encoding, the chromosome is [K,a]; the population size N=100, and the initial chromosome is randomly generated (such as [1.1,1.5], [0.8,2.2], [1.5,1.3], etc.) to ensure that K∈[0.5,2.0] and a∈[1.2,3.0] to avoid gain saturation or failure.

[0137] (2) Step 2: Fitness assessment

[0138] a. Using the mean squared error (MSE) between the predicted and expected values ​​of 50 samples as a benchmark, select “similar sample pairs” (e.g., sample A: n=1, P=[0.4], expected X=0.48; sample B: n=2, P=[0.4,0.3], expected X=0.95) to verify the constraint that “X changes more significantly when n is smaller”.

[0139] b. Calculate the initial population's MSE = 0.05. For samples that violate the monotonicity constraint (e.g., for a chromosome [0.9, 1.5], the calculated X = 0.45 when n = 1 and X = 0.43 when n = 2), apply a penalty function P = 0.6 and substitute it into the fitness function:

[0140] F=1 / (MSE+P+10⁻ 6 )

[0141] The calculated fitness of this chromosome is F = 1 / (0.05 + 0.6 + 10⁻). 6 The mean squared error is approximately 1.538, while the mean squared error of the optimal initial chromosome [1.2, 1.8] is 0.01, with no penalty, and the mean squared error is approximately 90.91.

[0142] (3) Step 3: Selection, crossover and mutation

[0143] a. Selection: A tournament selection strategy is adopted, and three individuals are randomly selected each round (e.g., [1.2,1.8], [1.0,1.6], [0.9,2.0]). The individual with the highest fitness, [1.2,1.8], is retained. This process is repeated 100 times to form a new population.

[0144] b. Crossover: Perform linear crossover with a crossover probability of 0.85. For example, the parent generation [1.2,1.8] and [1.3,1.7] generate the offspring generation [1.25,1.75]. When the offspring generation goes out of bounds, it is cyclically mapped to the feasible region (e.g., the offspring generation [2.1,1.6] is adjusted to [0.5,1.6]).

[0145] c. Mutation: Gaussian mutation is triggered with a probability of 0.02. N(0,0.01²) perturbation is added to K (e.g., 1.2→1.203), and N(0,0.1²) perturbation is added to a (e.g., 1.8→1.812). When the value exceeds the limit, it is truncated to the boundary value (e.g., when a is mutated to 3.1, it is truncated to 3.0).

[0146] (4) Step 4: Termination Condition and Optimization Results When the algorithm iterates to the 180th generation, the fitness improvement rate (ΔF / F0) for 25 consecutive generations is 8 × 10⁻ 6 <1×10⁻ 5 The termination condition is met; the final output is the optimal parameters K=1.2 and a=1.8. At this time, the MSE of 50 samples drops to 0.008, the monotonicity constraint satisfaction rate is 100%, and the optimization time is 2.3 seconds (which meets the requirement of "second-level diagnosis under limited on-site resources").

[0147] 5. Diagnostic Result Output and Verification

[0148] The above example is just a simplified illustration; actual fault tree structures can be much more complex. Based on the above process, for each equipment failure, all basic events are sorted using "gain probability (X)" as the evaluation metric, and the maintenance priority ranking is output.

[0149] Verification results: When maintenance is performed according to this priority, the average maintenance time is reduced from 15 minutes in the traditional FTA to 2.8 minutes, which verifies the advantages of this invention in "improving diagnostic accuracy and efficiency in multi-fault concurrent scenarios".

[0150] In its specific implementation, this application provides a computer storage medium and a corresponding data processing unit. The computer storage medium is capable of storing a computer program, which, when executed by the data processing unit, can run the invention's content regarding an intelligent fault diagnosis method based on a fault tree driven by both genetic algorithms and gain factors, as well as some or all of the steps in various embodiments. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0151] Those skilled in the art will clearly understand that the technical solutions in the embodiments of the present invention can be implemented using computer programs and their corresponding general-purpose hardware platforms. Based on this understanding, the technical solutions in the embodiments of the present invention, or the parts that contribute to the prior art, can be embodied in the form of computer programs, i.e., software products. These computer program software products can be stored in a storage medium and include several instructions to cause a device containing a data processing unit (which may be a personal computer, server, microcontroller, MCU, or network device, etc.) to execute the methods described in various embodiments or certain parts of the embodiments of the present invention.

[0152] This invention provides an intelligent fault diagnosis method based on a fault tree driven by both genetic algorithms and gain factors. Many methods and approaches exist for implementing this technical solution; the above description is merely a preferred embodiment. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of this invention, and these improvements and modifications should also be considered within the scope of protection of this invention. All components not explicitly stated in this embodiment can be implemented using existing technologies.

Claims

1. An intelligent fault diagnosis method based on a fault tree driven by both genetic algorithm and gain factor, characterized in that, Includes the following steps: Step 1: Based on the phenomena and causes of the fault events, establish a bidirectional mapping fault tree model; Step 2: In the fault tree model, a dynamic subtree lighting and extinguishing mechanism is introduced to recalculate the cause probability in real time based on the fault site observation results. Step 3: Construct a shared bottom event gain factor model and optimize the adjustable parameters. Use the optimized model to adjust the cause probabilities calculated in Step 2. Step 4: Using the adjusted cause probability as the evaluation index for fault diagnosis, output a maintenance priority list.

2. The intelligent fault diagnosis method based on a fault tree driven by both genetic algorithm and gain factor as described in claim 1, characterized in that, The establishment of the bidirectional mapping fault tree model described in step 1 includes: Step 1-1 involves dividing various fault information obtained from equipment or industrial systems into a three-layer node structure: fault events, fault phenomena, and fault causes. Different layers are connected via AND or OR gates, forming parent-child node relationships. Fault event S is a top-level node used to represent a system or device failure; Fault phenomenon M is an intermediate node used to represent specific observable fault content; intermediate nodes are divided into first-level intermediate nodes and lower-level intermediate nodes according to their nesting relationship. The cause of the failure, R, is the underlying node; Steps 1-2: Based on the phenomena and causes of the fault events, the fault tree is divided into reusable subtree units; each subtree unit contains several intermediate nodes M and bottom-level nodes R. Steps 1-3: Construct an independent base probability library for each subtree. For each non-first-level intermediate node in each subtree unit, assign a base probability as the initial cause probability of that node. If the same bottom-level node exists in multiple subtree units, assign a base probability to it independently in each subtree.

3. The intelligent fault diagnosis method based on a fault tree driven by both genetic algorithm and gain factor as described in claim 2, characterized in that, The dynamic subtree lighting and extinguishing mechanism described in step 2 includes: When a fault M is observed on-site, its corresponding subtree is immediately lit up; otherwise, the unobserved subtree is turned off. A subtree connected by an AND gate is an AND subtree, and all nodes connected by an AND gate together light up the upper-level node; a subtree connected by an OR gate is an OR subtree, and each lower-level node connected by an OR gate individually lights up the upper-level node. When the OR subtree is lit, the cause probability of all its bottom-level nodes is forcibly set to 1; when the OR subtree is turned off, the cause probability of its bottom-level nodes is set to 0. If the bottom-level node exists under other OR subtrees, its base probability in other OR subtrees is set to 0, and the base probability of the bottom-level node is redistributed to the remaining lit nodes according to the initial proportion of the remaining lit nodes. The priority rule is: OR gate off > OR gate on > AND gate off.

4. The intelligent fault diagnosis method based on a fault tree driven by both genetic algorithm and gain factor as described in claim 3, characterized in that, The shared bottom event gain factor model mentioned in step 3 includes: If the bottom node exist If a node is lit up repeatedly in a subtree, the cause probability of that node is adjusted according to the base probability. The specific calculation method is as follows: ； in, For the adjusted cause probability, the bottom-level node The basic probability vector is represented as , For the bottom layer node In the The basic probability in a subtree. This indicates taking the maximum value; and This is an adjustable parameter.

5. The intelligent fault diagnosis method based on a fault tree driven by both genetic algorithm and gain factor as described in claim 4, characterized in that, The constraints in the shared bottom event gain factor model are as follows: And the probability of cause With base probability There is a positive correlation; And the probability of cause Number of child trees Monotonically increasing.

6. The intelligent fault diagnosis method based on a fault tree driven by both genetic algorithm and gain factor as described in claim 5, characterized in that, The optimization of the adjustable parameters described in step 3 refers to optimizing the adjustable parameters in the shared bottom event gain factor model. and Genetic algorithms are used for learning and optimization.

7. The intelligent fault diagnosis method based on a fault tree driven by both genetic algorithm and gain factor as described in claim 6, characterized in that, Step 3, which involves optimizing the adjustable parameters, specifically includes: Using two-dimensional real-number encoding, each chromosome in the genetic algorithm is defined as a vector. ; The population size in the genetic algorithm is set to... The initial chromosome is generated randomly. Based on population size Half of the manually labeled examples serve as the baseline data; Select similar sample pairs from the baseline data whose maximum probability value and total probability are close; Based on the adjusted causal probability of each pair of similar samples , and Calculate the difference between predicted values and ; If not satisfied Otherwise, a penalty is imposed when calculating fitness; Based on the calculated fitness, after evaluation, the population enters an evolutionary cycle integrating selection, crossover, and mutation until the number of iterations reaches a threshold or the fitness improvement rate is less than a threshold within a preset number of generations. This completes the genetic algorithm and yields optimized adjustable parameters. and .

8. The intelligent fault diagnosis method based on a fault tree driven by both genetic algorithm and gain factor as described in claim 7, characterized in that, The specific method for calculating fitness is as follows: Fitness without penalty The calculation method is as follows: ； in, Penalties for violating the monotonicity constraint; Fitness when punishment is applied The calculation method is as follows: ； in, , It is a constant; This represents the mean square error between the predicted and expected values.

9. The intelligent fault diagnosis method based on a fault tree driven by both genetic algorithm and gain factor as described in claim 8, characterized in that, The method of selecting similar sample pairs from the benchmark data whose maximum probability value and total probability are close includes: If the benchmark data contains data ,satisfy: ； The difference is <0.05 The difference is <0.1 in, , and They are the first , and The number of times the node is repeatedly lit in the sample group; , and These are the original probability vectors of the three groups of samples; Then the data These are similar sample pairs.

10. The intelligent fault diagnosis method based on a fault tree driven by both genetic algorithm and gain factor as described in claim 9, characterized in that, The output maintenance priority list includes: Calculate the adjusted cause probability for each node, sort them in descending order, and use this as the maintenance priority list for the bottom-level nodes.

Images