Engine system assembly fault diagnosis method based on genetic programming

Abstract

The invention provides an engine system assembly fault diagnosis method based on genetic programming. The engine system assembly fault diagnosis method comprises the steps that 1, automobile engine assembly parameters and vibration data are collected; 2, data enhancement based on a neighborhood oversampling regression technology; 3, carrying out assembly parameter-vibration performance symbol regression modeling based on genetic programming; and step 4, system assembly fault diagnosis and key process parameter analysis are carried out. According to the method, a genetic programming symbol regression method is adopted, an explicit tree structure representing internal correlation between the assembly parameters and the vibration performance is automatically evolved, high-precision and highly interpretable vibration state classification is achieved, and fault diagnosis can be effectively supported. Through deep analysis of the optimal expression tree, key process parameters having significant influence on vibration performance are accurately identified, and a nonlinear coupling mechanism and a sensitive direction of the key process parameters are disclosed, so that a reliable theoretical basis and engineering guidance are provided for accurate optimization of an engine assembly process and effective traceability of vibration faults.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of engine assembly quality and fault diagnosis, and specifically provides an engine system assembly fault diagnosis method based on genetic programming. First, aiming at the small sample, sparse or unbalanced data distribution problems existing in the actual assembly process, the neighborhood oversampling regression technology is used to perform data enhancement on the original data set. Second, aiming at the problem that the complex nonlinear and strong coupling relationship between the assembly indexes is difficult to effectively depict by the traditional model, the genetic programming is used to automatically evolve an explicit mathematical expression tree structure describing the internal law between the assembly indexes and the engine performance, so as to realize high-precision and strong-explainability engine fault diagnosis and reliability evaluation. Finally, based on the obtained optimal tree structure, the key process parameters that have a significant impact on the engine performance are identified, which provides a direct basis for assembly process optimization and vibration fault tracing. The present application is suitable for fault diagnosis and process improvement of automobile engines and other complex mechatronic products in the assembly and manufacturing process. BACKGROUND

[0002] The automobile engine as the core power component of the vehicle, its reliability directly affects the safety, durability and use performance of the vehicle. In the engine manufacturing and assembly process, a large number of process parameters (such as bearing clearance, shaft runout, bolt tightening torque, fit size, etc.) are involved, which will have a significant impact on the vibration performance of the engine under the complex coupling effect. The vibration value is one of the key indicators to measure the smoothness and reliability of the engine, and its abnormality often indicates potential assembly defects or early faults.

[0003] At present, the fault diagnosis method for automobile engines mostly depends on physical models, empirical formulas or traditional statistical models. However, due to the complex structure of the engine system, the large number of assembly parameters and the strong nonlinear interaction, the traditional method is often difficult to accurately depict the internal law between “assembly parameters – vibration performance”. Especially in the actual industrial scene of small sample and high-dimensional features, the conventional regression or machine learning model is easy to overfit, or the prediction bias is caused due to the mismatch of prior assumptions.

[0004] In addition, the data collected in the engine manufacturing process usually has the problems of limited sample size and uneven feature distribution, which further limits the generalization ability and engineering applicability of the model. Although existing researches have tried to introduce neural networks, support vector machines and other black box models, their explainability is poor, and it is difficult to provide physical insight into the key process parameters, which is not conducive to process optimization and fault tracing.

[0005] Based on this, the application provides an engine system assembly fault diagnosis method based on genetic programming, which automatically evolves an interpretable mathematical expression from data by introducing symbolic regression technology, reveals the internal relationship between assembly parameters and vibration performance, and improves the modeling robustness under small samples by combining neighborhood oversampling regression data enhancement technology, thereby providing a scientific basis for fault diagnosis and process improvement of engine manufacturing process. SUMMARY

[0006] The purpose of the application is to provide an engine system assembly fault diagnosis method based on genetic programming, which solves the problems of numerous assembly process parameters, complex nonlinear coupling relationship between parameters and vibration performance, and small sample data in actual production, which leads to difficult modeling, insufficient prediction accuracy and poor model interpretability.

[0007] The application aims to comprehensively use neighborhood oversampling regression data enhancement technology and genetic programming symbolic regression method. Firstly, the data enhancement technology is used to solve the problem of insufficient generalization ability under small sample condition, and then the genetic programming is used to automatically explore the explicit mathematical expression describing the internal law of "assembly parameter-vibration performance" from the enhanced data. This method can effectively process small sample and high-dimensional feature data, avoid prior assumptions about model form, accurately quantify the influence of key assembly indicators (such as gap, runout, tightening torque, etc.) on vibration value, and finally realize accurate evaluation of engine system assembly reliability, effective identification of potential faults, and direct and interpretable guidance for assembly process optimization.

[0008] Based on the above theory and idea, the application provides an engine system assembly fault diagnosis method based on genetic programming, and the specific steps are as follows: Step 1: Collecting automobile engine assembly parameters and vibration data Suppose the automobile engine assembly process contains key procedures, the th procedure contains quality characteristic parameters, arrange all the parameters of the procedures in order to form a dimensional assembly parameter vector (wherein, ): (1) Wherein, the assembly parameter vector of the engine is comprehensively represented, represents a specific assembly parameter, represents Real number vector space. The assembly parameters cover various types such as size tolerance, geometric tolerance, fit clearance, bolt tightening torque, etc., including but not limited to: shaft runout, bearing fit interference / clearance, single-sided clearance of transmission shaft, gear meshing clearance, tightening torque and angle of each part of the bolt, etc.

[0009] The vibration test data of the complete machine after assembly, the vibration value as the key output index to measure the assembly quality and reliability of the automobile engine, accordingly, define the vibration state label of binary classification : (2) wherein, is an indicator function, is the effective value of vibration velocity, is the vibration limit value set according to expert experience or design specification.

[0010] Based on the data recorded during the assembly process of the engine, the original data set : (3) wherein, is the assembly parameter vector of the engine, is the vibration state label of the engine.

[0011] Step two: data enhancement based on neighborhood oversampling regression technology Engine assembly parameter data is usually collected by manual measurement, which makes it difficult to collect large samples. In addition, most of the post-assembly vibration values are within the normal range, resulting in unbalanced data. In order to solve this problem, the neighborhood oversampling regression technology is used to enhance the original data set.

[0012] Neighborhood oversampling regression is an adapted version of the KNNOR method (KNNOR) in regression tasks. The KNNOR method solves the class imbalance problem in classification by generating synthetic data points, while neighborhood oversampling regression applies this idea to regression tasks, aiming to generate synthetic data points that preserve the continuous target values of the original data set.

[0013] This method selects a reference sample point from the sparse area of the data, denoted as the assembly parameter vector of the sample point , from the nearest neighbor sample point of the reference sample point, denoted as the assembly parameter vector of the adjacent sample point . The goal is to generate a synthetic data point based on the selected point and the adjacent sample point a new synthetic data point is created at a position on the line between (4) where is a random number uniformly distributed in the interval (0, 1) and is used to control the position of the synthetic data point .

[0014] The synthetic feature vector is generated after the corresponding continuous vibration value is calculated by a distance-weighted average of the vibration values of the nearest neighbor samples: (5) where is the target value of the th nearest neighbor, is the Euclidean distance between the th nearest neighbor sample point and the th nearest neighbor sample point, is a parameter that controls the impact of distance, which also takes a value in the interval (0, 1).

[0015] Finally, according to the vibration qualification criterion, the calculated continuous vibration value is converted into a binary classification label : (6) This process is iterated for the sparse area sample points in the original data set, and a new synthetic data point is generated each time, effectively increasing the sample size and diversity of the training set.

[0016] Step three: symbolic regression modeling of "assembled parameters-vibration performance" based on genetic programming The basic idea of genetic programming (GP) is to randomly generate an initial population suitable for a given environment, i.e., the search space of the research problem. Similar to genetic algorithms, each individual in the population has a fitness value, and genetic operators are used to process individuals with high fitness to generate the next generation of population, and so on. The solution or approximate solution to the given problem will appear in a certain generation. In symbolic regression problems, genetic programming algorithms simulate the natural evolution process through selection operators, crossover operators and mutation operators, iteratively optimize mathematical expressions to fit the data set In this specification, refers to the assembled parameter vector of the engine,​ Refers to the label. The algorithm flow is as follows: Figure 1 As shown.

[0017] First, define the function set. With terminal set V The function set contains basic mathematical operators and functions, for example, This is used to construct non-linear relationships within the terminal set. Terminal set It consists of all assembly parameter variables and random constants, i.e. .

[0018] Secondly, design the fitness function. The fitness function is a measure of an individual's quality and a primary basis for natural selection in genetic programming. Treating automobile engine vibration fault diagnosis as a binary classification problem, fitness is defined as the accuracy of this expression in classifying all samples. Higher fitness means a greater probability of the individual being retained and propagating.

[0019] Next, an initial population is created, containing several symbolic expression trees. A symbolic expression tree is a binary tree where internal nodes are mathematical operators and terminal nodes are variables or constants. For example, Figure 2 The mathematical expression represented by the symbolic expression tree shown is: There is a one-to-one correspondence between symbolic expression trees and mathematical expressions. In genetic programming, the population is a set of mathematical expressions. The expression trees in the initial population can be generated using the following three strategies: 1. Growth strategy: Randomly select mathematical operators, variables, or constants at each node. Variables and constants become leaf nodes, while operators continue to grow. This typically generates an asymmetric and shallow expression tree; 2. Full strategy: Given a maximum depth, all non-leaf nodes are randomly selected mathematical operators, while leaf nodes are randomly selected variables or constants; 3. Half and Half Strategy: Combining the grow strategy and the full strategy, each strategy generates half of the expression tree.

[0020] After the initial population is created, the genetic programming algorithm enters an iterative "evaluation-evolution" loop. It evaluates the population using a fitness function. If the expression error in the population is less than a preset threshold or the population reaches its generation limit, the algorithm terminates and outputs the optimal expression. If the algorithm does not terminate, it proceeds to the evolutionary stage. The evolutionary stage includes four parts: selection, crossover, mutation, and replication.

[0021] 1. Selection: In the evolutionary cycle of genetic programming, the selection operator plays a crucial role in weeding out the weak based on individual fitness. The core objective is to pass on superior genes to the next generation with a high probability, resulting in a highly fitness-oriented subtree structure, while maintaining sufficient population diversity to avoid premature convergence. This scheme employs a tournament selection mechanism, with a tournament size of [missing information]. ,generally , For population size, Indicates the first For each generation of the population, the execution of each round of the tournament involves three steps: (1) Random sampling: from the current population Uniformly and independently selected individual ( ), forming a temporary tournament set ; (2) Competition evaluation: Calculate the set Each individual within fitness value According to Quantitatively evaluate the error performance or prediction performance under the objective function; (3) Determine the winner: Select the individual with the highest fitness. As the winner of this round of the tournament, .

[0022] After multiple rounds of tournaments, the winning individuals will enter the crossover, mutation, or replication phases with a given probability, and will be used to generate the next generation of the population.

[0023] 2. Crossover: The crossover operator is an important mechanism in genetic programming for achieving structural recombination. The basic idea is to generate new individuals with characteristics of both parents by exchanging gene segments between them, thus enabling a broader search in the solution space. Figure 3 As shown.

[0024] Figure 3 In the middle, the top shows the expression trees of two parent individuals, and the bottom shows the expression trees of two child individuals. Each circle represents an operand or terminal node of the expression. , , , , , These are computation nodes, inside the circle. 3. 7 is the terminal node. The structures of the two parent trees are respectively and After a crossover operation, the gene segments selected by the dashed lines in the two parent trees are exchanged, resulting in the following structures for the child trees: and .

[0025] Let the two parent individuals be represented as expression trees. and Based on the set crossover probability Randomly select two parent expression trees from the selected individuals. and Subsequently, in... and Randomly select a subtree node, denoted as . , The crossover operation generates a new descendant structure by swapping the two subtrees: (7) Among them, symbols Indicates in Subtrees in China Replace the original subtree .

[0026] 3. Mutation: Mutation operators are an important mechanism in genetic programming for maintaining population diversity and exploring potential new solution spaces. The core idea is to randomly perturb the local structure of individual individuals to escape local optima and enhance the algorithm's global search capability, such as... Figure 4 As shown.

[0027] Figure 4 In the image, the top section shows the parent individual expression tree before the mutation operation, and the bottom section shows the child expression tree after the mutation operation. Each circle represents an operand or terminal node of the expression. , , , These are computation nodes, inside the circle. 3. 7 is the terminal node. The parent tree structure is as follows: After mutation, the selected local structure is randomly replaced, resulting in the following child tree structures: .

[0028] Let the current individual be represented as an expression tree. Mutation operations are based on probability. Trigger, for The mutation operator replaces or reorganizes parts of the nodes or subtree structure. Mutation operators include three forms: point mutation, lifting mutation, and subtree mutation.

[0029] The expression for point mutation can be represented as: ,in, Individual One of the nodes, It is a newly generated random node.

[0030] Boosting mutation aims to alter the structural depth and logical relationships of an expression by reorganizing its subtree levels. The expression can be represented as follows: ,in, Individual A subtree in which is denoted as ; yes A subtree in which is denoted as .

[0031] Subtree mutation is the type of mutation that has the greatest impact on the individual structure, and can be expressed as: ,in, Individual A subtree in which is denoted as ; It is a newly generated random subtree.

[0032] 4. Replication: The replication operator is a crucial link in the genetic programming evolutionary mechanism for maintaining the genetic continuity of superior individuals. Its main function is to directly preserve individuals exhibiting high fitness to the next generation without any structural modifications, thus ensuring that the information of superior genes is not destroyed during the population search process. Unlike crossover and mutation, the replication operation does not change the topological structure of the expression tree; it only achieves lossless transmission from parent to offspring.

[0033] Let the first generation population , It is the population size. ( An individual is an part of a population. The fitness function is During the replication phase, an elite preservation strategy is adopted, defining the top-fitness individuals in the current generation. Individuals constitute an elite group Formal representation: (8) in, Refers to a collection of elites The number of individuals included, and Reserve a preset number of elites. The replication operator sets the elites together. Individuals are directly incorporated into the next generation of the population. : (9) in, It is the set of the remaining offspring individuals generated through crossover and mutation operations.

[0034] The introduction of the copying operator enables the entire "selection-crossover-mutation-copying" evolutionary loop to form a balanced search framework that takes into account both the preservation of excellence and exploration, thus ensuring the final acquisition of the globally optimal expression.

[0035] Ultimately, after multiple generations of evolution, the optimal expression tree structure for system assembly fault diagnosis was generated.

[0036] Step 4: System Assembly Fault Diagnosis and Key Process Parameter Analysis The optimal expression tree is applied to independent test sets or the original dataset to calculate metrics such as classification accuracy, thus validating the model's generalization ability. Genetic programming can be used to quickly predict new combinations of engine assembly parameters and determine whether the engine's vibration values ​​are acceptable, thereby achieving reliable assembly quality assessment and fault identification.

[0037] Furthermore, a deep analysis of the optimal expression tree is performed to identify key process parameters. Since the expression tree is essentially a mathematical function structure, the importance of parameters can be assessed by analyzing its topological characteristics and variable distribution. Specifically, based on the frequency of each parameter's occurrence in the expression, its node depth, and the influence of its coefficients or substructures on the overall output, corresponding structural importance indices can be calculated. This quantifies the contribution of each parameter to the vibration response of the automotive engine, providing a clear basis for subsequent process optimization.

[0038] It is important to note that before implementing the overall genetic programming algorithm, the population size needs to be reasonably determined based on factors such as the complexity of the specific problem, available computing resources, and expert experience. Maximum number of algebras to run G The value of . To prevent structural degradation and tree depth explosion, it is also necessary to design the maximum tree depth for generating individuals. These parameters typically do not have a unified analytical expression and often need to be determined through experimental optimization or empirical criteria, and are dynamically monitored and adjusted during algorithm operation.

[0039] The advantages and beneficial effects of this invention are as follows: 1. The symbolic regression model generated by genetic programming in this invention has a completely transparent mathematical expression structure. This model can clearly demonstrate the quantitative mathematical relationship between various assembly parameters and vibration response, not only providing prediction results but also revealing the underlying mechanisms. This white-box characteristic greatly enhances engineers' understanding and trust in the model's decision-making process, providing a direct and reliable theoretical basis for the precise improvement of assembly processes.

[0040] 2. The method proposed in this invention does not rely on manually preset model forms, but automatically discovers the inherent laws between assembly parameters and vibration performance from data through evolutionary calculations. This method can effectively identify nonlinear coupling effects, threshold effects, and interactions between parameters, breaking through the limitations of model form assumptions in traditional regression analysis. This allows for a more accurate depiction of the potential complex correlations in the actual assembly process, providing a new technical approach for a deeper understanding of the formation mechanism of automotive engine vibration characteristics.

[0041] 3. By analyzing the optimal expression, the method proposed in this invention can accurately identify the key process parameters that are most sensitive to the vibration performance of automobile engines, thus achieving a leap from "fault diagnosis" to "parameter optimization". Attached Figure Description

[0042] Figure 1 This is a flowchart illustrating the overall process framework of the engine system assembly fault diagnosis method based on genetic programming proposed in this invention.

[0043] Figure 2 This is an example of the symbolic expression tree structure used in genetic programming in this invention.

[0044] Figure 3 This is a schematic diagram of the crossover operator operation in genetic programming.

[0045] Figure 4 This is a schematic diagram of the mutation operator operation in genetic programming.

[0046] Figure 5 It is the optimal expression tree structure obtained from a specific four-cylinder car engine implementation case. Detailed Implementation

[0047] The following example uses a certain type of car engine, combined with... Figure 1 The present invention provides a more detailed description of the method for diagnosing assembly faults in automotive engine systems based on genetic programming.

[0048] Step 1: Collect engine assembly parameters and vibration data This type of automotive engine is widely used in passenger vehicles. The assembly process involves 12 steps and a total of 151 key quality characteristic parameters. The dataset covers multiple dimensions, including bearing clearance, shaft runout, and bolt tightening torque. Vibration performance, a core indicator of engine assembly quality, has a pass / fail threshold set at 18 mm / s. Based on this threshold, the measured vibration value of each sample is converted into a binary classification label. Due to the high cost of collecting actual production data, only 65 valid samples were ultimately obtained, forming the original dataset. ,in, Indicates the first A 151-dimensional assembly parameter vector for each sample. For the first The dataset contains binary classification labels for a given number of samples. This dataset represents a typical small sample size scenario, posing a challenge for subsequent modeling.

[0049] Step 2: Data Augmentation Based on Neighborhood Oversampling Regression Technique The original assembly system dataset contained 65 samples and 151 input features, which were small in size and significantly unevenly distributed in the feature space. Therefore, neighborhood oversampling regression data augmentation was employed. After this augmentation process, the total number of samples was expanded to 842, while the feature dimensions remained unchanged. The expanded dataset is defined as follows: ,in, Indicates the first The assembly parameter vector of each sample For the first Each sample corresponds to a binary classification label. This significantly improves the density and diversity of the data distribution, providing a sufficient and statistically significant training foundation for subsequent genetic programming modeling.

[0050] Step 3: Symbolic Regression Modeling of Assembly Parameters-Vibration Performance Based on Genetic Programming First, based on the practical background of engine system assembly reliability, the function set and terminal set used in genetic programming are defined. The function set includes operators such as addition, subtraction, multiplication, division, maximum value, minimum value, and trigonometric functions, aiming to capture the complex nonlinear relationship that may exist between assembly parameters and vibration response. The terminal set consists of system assembly parameter variables. (Such as perpendicularity, bearing size, clearance, etc.) and random constants constitute.

[0051] Secondly, for the genetic programming classification task in this study, a fitness function is defined to measure the ability of an individual's expression to discriminate the performance category of the assembly system. The fitness function is defined as follows: To predict the consistency rate between the predicted category and the true category: (10) in, The total number of samples, This represents a logical judgment function. For the first The predicted category for each sample, For the first The true category of each sample.

[0052] Next, an initial population is generated using a half-and-half strategy. After creation, an iterative cycle of "evaluation-evolution" is entered, with specific parameter settings shown in Table 1.

[0053] Table 1. Genetic programming algorithm parameter settings in this case.

[0054] To prevent overfitting in the complex function search space of genetic programming, the following constraints and control strategies are introduced during the evolutionary process: (1) Set the maximum tree depth to 8 to limit the model complexity; (2) A slight penalty term is introduced for the complexity of candidate individuals, and expressions with simple structure but high prediction performance are retained first; (3) Use the elite retention mechanism to directly retain the best individual in each generation to prevent the excellent solution from being destroyed by random operations.

[0055] After multiple generations of evolution, the algorithm successfully generated the optimal expression tree for system assembly reliability, with the tree structure as follows: Figure 5 As shown. Circles in the tree represent operands or terminal nodes of the expression: operands include... (division), (Subtraction) (addition), (multiplication), , , , These nodes correspond to the operations in the expression; terminal nodes are... The initial identifier, such as , These represent various key assembly parameters in the automotive engine system.

[0056] Step 4: System Assembly Fault Diagnosis and Key Process Parameter Analysis Based on the obtained optimal expression tree, the accuracy reached 0.9917 on the data-augmented automotive engine dataset and 0.9231 on the original dataset. The confusion matrix on the original dataset is shown in Table 2, where TP is the true negative class, FN is the false negative class, FP is the false positive class, and TN is the true negative class.

[0057] Table 2. Confusion matrix results on the original dataset

[0058] Based on the confusion matrix, we can further calculate various performance metrics of the model on the original automotive engine dataset, including precision, recall, specificity, and F1 score: (11) (12) (13) (14) The results show that the genetic programming model maintains high classification performance on the original dataset, effectively identifying the "faulty" and "fault-free" states of a car engine. Although there are a few cases of misclassification, the overall classification trend is highly consistent with the real situation.

[0059] To further verify the superiority of the proposed method, its performance was compared with that of several typical existing methods in engine fault diagnosis. The results are shown in Table 3.

[0060] Table 3 Performance Comparison Results of Fault Diagnosis Models

[0061] As shown in Table 3, the genetic programming method proposed in this invention exhibits comprehensive and significant advantages in engine fault diagnosis. The accuracy of the proposed method reaches 92.31%, a 1.7% improvement compared to the best-performing BPNN, and a 33.3% improvement compared to logistic regression. Crucially, the proposed method achieves the highest F1 score (82.76%), a comprehensive measure of model performance, representing a 7.59% improvement over the best-performing BPNN. This indicates that the proposed method achieves a better balance between precision and recall, effectively reducing false positives while minimizing false negatives. Furthermore, the recall rate of the proposed method is significantly higher than that of the best-performing BPNN, demonstrating the stronger ability of the proposed genetic programming method to detect real-world problems, which is essential for engine diagnosis with extremely high reliability requirements.

[0062] Furthermore, based on the results of the optimal expression analysis, the key process parameters that have the most significant impact on engine vibration were identified, including coaxiality. , fit clearance Bearing clearance bounce and median value . , The increasing trend of will significantly exacerbate the vibration response, while , , and The combined changes produce periodic dynamic changes in the vibration value through nonlinear coupling.

[0063] Therefore, in production control, the deviation range of the above-mentioned key indicators should be closely monitored and strictly limited, especially when and The difference is too large, or , When vibration risks exceed design tolerances, the risk increases dramatically. These parameters should be identified as critical process control points during system assembly, providing clear guidance for improving engine assembly stability and operational reliability.

[0064] In summary, taking a certain type of automobile engine as an example, this invention effectively expands the training sample size based on small-sample assembly data collected in actual production, using neighborhood oversampling regression data augmentation technology, thus solving the problems of data sparsity and uneven distribution. Building upon this, a genetic programming symbolic regression method is employed to automatically evolve an explicit tree structure representing the intrinsic relationship between assembly parameters and vibration performance, achieving high-precision and highly interpretable vibration state classification, which can effectively support fault diagnosis. Finally, through deep analysis of the optimal expression tree, key process parameters with significant impact on vibration performance are accurately identified, revealing their nonlinear coupling mechanism and sensitive directions. This provides a reliable theoretical basis and engineering guidance for the precise optimization of engine assembly processes and the effective tracing of vibration faults.

Claims

1. A method for diagnosing assembly faults in an engine system based on genetic programming, characterized in that: The steps are as follows: Step 1: Collect automotive engine assembly parameters and vibration data; Step 2: Data Augmentation Based on Neighborhood Oversampling Regression Technique; Use neighborhood oversampling regression technique to augment the original dataset; The goal is to generate synthetic data points that retain continuous target values ​​from the original dataset; Step 3: Symbolic regression modeling of assembly parameters and vibration performance based on genetic programming; Genetic operators are used to obtain highly fit individuals, generating the next generation of the population. A solution or approximate solution to a given problem will appear in a certain generation. In symbolic regression problems, genetic programming algorithms simulate natural evolution through selection, crossover, and mutation operators, iteratively optimizing mathematical expressions to fit the dataset. , Refers to the engine's assembly parameter vector. Refers to labels; Step 4: System assembly fault diagnosis and key process parameter analysis; The optimal expression tree is applied to an independent test set or the original dataset to calculate the classification accuracy of the optimal expression tree and verify the generalization ability of the model. Genetic programming is used to quickly predict new combinations of engine assembly parameters and determine whether the engine vibration value is qualified, so as to achieve reliable assembly quality assessment and fault identification.

2. The method for diagnosing engine system assembly faults based on genetic programming according to claim 1, characterized in that: In step one, let's assume the car engine assembly process includes... The first key process, the Each process includes Each quality characteristic parameter is arranged sequentially with all process parameters to form a... 3D assembly parameter vector ,in, ; (1) in, Refers to the engine's assembly parameter vector. This represents a specific assembly parameter. , express A dimensional real vector space.

3. A method for diagnosing engine system assembly faults based on genetic programming according to claim 1 or 2, characterized in that: In step one, the vibration test data of the assembled machine is used. The vibration value is a key output indicator for measuring the assembly quality and reliability of the automobile engine. Based on this, a two-category vibration state label is defined. : (2) in, For indicator functions, This is the effective value of the vibration velocity. These are vibration limit values ​​set based on expert experience or design specifications. based on Data recorded during the assembly process of the engine was used to construct the original dataset. : (3) in, For the first The assembly parameter vector of the engine. For the first Vibration status label for the engine.

4. The method for diagnosing engine system assembly faults based on genetic programming according to claim 1, characterized in that: In step two, a reference sample point is selected from the sparse data region, and the assembly parameter vector of the reference sample point is denoted as... From the benchmark sample points Randomly select one of the nearest neighbor sample points, and denote the assembly parameter vector of the neighbor sample point as... ; The goal is based on the selected point and neighboring sample points Use the positions on the connecting lines to create a new synthetic data point. Newly synthesized data points for: (4) in, It is a random number uniformly distributed within the interval (0, 1), used to control the synthetic data points. The location.

5. A method for diagnosing engine system assembly faults based on genetic programming according to claim 1 or 4, characterized in that: In step two, a synthetic feature vector is generated. after, Corresponding continuous vibration value Through synthesized feature vectors Calculated by distance-weighted average of vibration values ​​from the nearest neighbor samples: (5) in, It is the first The target value of the nearest neighbor, yes With the The Euclidean distance between the nearest neighbor sample points It is a parameter that controls the influence of distance, and its value range is (0, 1); Based on the vibration compliance criteria, the calculated continuous vibration values Convert to binary labels : (6) The process iterates through the sparse region sample points in the original dataset, generating a new synthetic data point in each iteration. This increases the sample size and diversity of the training set.

6. The method for diagnosing engine system assembly faults based on genetic programming according to claim 1, characterized in that: In step three, firstly, the function set is defined. With terminal set V The function set contains basic mathematical operators and functions. To construct a non-linear relationship between terminal sets; Terminal set It consists of all assembly parameter variables and random constants, i.e. ; Secondly, design the fitness function. The diagnosis of automotive engine vibration faults is considered as a binary classification problem. Fitness is defined as the accuracy of classification on all samples. The higher the fitness, the greater the probability that an individual will be retained and reproduced. Next, an initial population is created, containing several symbolic expression trees; A symbolic expression tree is a binary tree in which internal nodes are mathematical operators and terminal nodes are variables or constants. Ultimately, after multiple generations of evolution, the optimal expression tree structure for system assembly fault diagnosis was generated.

7. A method for diagnosing engine system assembly faults based on genetic programming according to claim 1 or 6, characterized in that: In step three, the expression tree in the initial population is generated using the following three strategies: The grow strategy randomly selects mathematical operators, variables, or constants at each node; variables and constants become leaf nodes, while operators continue to grow; this generates an asymmetric and shallow expression tree. The full strategy: Given a maximum depth, all non-leaf nodes are randomly selected mathematical operators, while leaf nodes are randomly selected variables or constants. The half-and-half strategy combines the grow strategy and the full strategy, with each strategy generating half of the expression tree.

8. The method for diagnosing engine system assembly faults based on genetic programming according to claim 7, characterized in that: In step three, after the initial population is created, the genetic programming algorithm enters the evaluation-evolution iterative loop, using the fitness function to evaluate the population. If the expression error in the population is less than the preset threshold or the population generation reaches the upper limit, the algorithm terminates and outputs the best expression. If the algorithm does not terminate, it proceeds to the evolutionary step.

9. The method for diagnosing engine system assembly faults based on genetic programming according to claim 8, characterized in that: In step three, the evolutionary stage includes four parts: selection, crossover, mutation, and replication. Selection: In the evolutionary cycle of genetic programming, the selection operator plays a crucial role in weeding out the weak based on individual fitness. A tournament selection mechanism is adopted, with the tournament size set at... , , For population size, Indicates the first For each generation of the population, the execution of each round of the tournament involves three steps: (1) Random sampling: from the current population Uniformly and independently selected individual , This constitutes a temporary tournament set. ; (2) Competition evaluation: Calculate the set Each individual within fitness value According to Quantitatively evaluate the error performance or prediction performance under the objective function; (3) Determine the winner: Select the individual with the highest fitness. As the winner of this round of the tournament, ; After multiple rounds of tournaments, the winning individuals will enter the crossover, mutation, or replication phase with a given probability to generate the next generation of the population. Crossover: Let the two parent individuals be represented as expression trees. and Based on the set crossover probability Randomly select two parent expression trees from the selected individuals. and ; Subsequently, in and Randomly select a subtree node, denoted as . , The crossover operation generates a new descendant structure by swapping the two subtrees. (7) Among them, symbols Indicates in Subtrees in China Replace the original subtree ; Mutation: Let the current individual be represented as an expression tree. Mutation operations are based on probability. Trigger, for Some nodes or subtree structures are replaced or reorganized; mutation operators include three forms: point mutation, lifting mutation and subtree mutation. The expression for point mutation is as follows: ,in, Individual One of the nodes, These are newly generated random nodes; Boost mutation is represented as ,in, Individual A subtree in which is denoted as ; yes A subtree in which is denoted as ; Subtree mutation is the type of mutation that has the greatest impact on the individual structure, and is expressed as: ,in, Individual A subtree in which is denoted as ; It is a newly generated random subtree; Copying: The copying operation does not change the topology of the expression tree; it only achieves lossless transfer from parent to child. generation population , It is the population size. It is an individual within the population. ,individual The fitness function is During the replication phase, an elite preservation strategy is adopted, defining the top fitness rankings in the current generation. Individuals constitute an elite group Formal representation: (8) in, Refers to a collection of elites The number of individuals included, and Reserve a preset number of elites; the replication operator sets the elites together. Individuals are directly incorporated into the next generation of the population. : (9) in, It is the set of the remaining offspring individuals generated through crossover and mutation operations.

10. The method for diagnosing engine system assembly faults based on genetic programming according to claim 1, characterized in that: In step four, the optimal expression tree is deeply analyzed to identify key process parameters; the importance of parameters is evaluated by analyzing topological features and variable distribution; specifically, based on the frequency of each parameter in the expression, the depth of the node it is located in, and the degree of influence of the coefficient or substructure on the overall output, the corresponding structural importance index is calculated, thereby quantifying the contribution of each parameter to the vibration response of the automobile engine and providing a clear basis for subsequent process optimization.

Images