Transmission lubricating oil system intelligent fault diagnosis method considering measurement uncertainty

Abstract

The invention discloses a transmission lubricating oil system intelligent fault diagnosis method considering measurement uncertainty, and the method comprises the steps: employing a layered injection strategy for a multi-source fault type of a target lubricating oil system; and then establishing a mapping relation between the system state and the test point response by constructing a fault-test correlation matrix (matrix D). And thirdly, under the imperfect test assumption, aiming at the dynamic uncertainty characteristics of the sensor measurement error, constructing a hybrid quantitative model based on a Bayesian probability framework to carry out testability model optimization on the target lubricating oil system. And finally, based on the optimized testability model, proposing a dynamic fault diagnosis strategy, constructing a multi-dimensional misdiagnosis cost model including false alarm, leak detection, system damage and the like, and realizing optimal distribution of test resources in combination with an information entropy algorithm and a cross-domain optimization technology. Through a real-time data-driven adaptive diagnosis mechanism and a multi-level fault identification framework, the fault detection accuracy under a complex working condition is significantly improved.

Description

Technical Field:

[0001] The present invention relates to an intelligent fault diagnosis method for a transmission lubricating oil system considering measurement uncertainty, which belongs to the field of system fault diagnosis. Background Art:

[0002] As an important lubricating and cooling system in mechanical equipment, the transmission lubricating oil system is widely used in aviation, navigation, and industrial transmission equipment. Its main functions are to reduce friction, lower temperature, and maintain the normal operation of the system. However, due to the complex working environment and the high-load operation of the system, the transmission lubricating oil system is vulnerable to various faults, such as lubricating oil leakage, pollutant mixing, insufficient lubrication, etc. These faults may lead to a decline in equipment performance and even major safety accidents, posing great challenges to industrial production and safe operation. Therefore, higher requirements are put forward for the fault diagnosis technology of the transmission lubricating oil system. Traditional fault diagnosis methods are mostly based on physical models or empirical rules. Although they are effective in single or clear fault modes, it is often difficult to ensure the diagnostic accuracy and efficiency when facing complex fault mechanisms and the coupling relationship of multiple variables. In addition, with the wide application of sensor technology, a large amount of sensor data provides a rich information source for the fault diagnosis of the lubricating oil system. However, random errors and systematic errors in the measurement process are inevitable, which makes traditional fault diagnosis methods have certain limitations in data processing and uncertainty handling.

[0003] In recent years, with the development of artificial intelligence technology, fault diagnosis methods based on machine learning and data-driven have received extensive attention. These methods can achieve more efficient and accurate diagnosis by mining the fault patterns hidden in big data. However, the data collected in actual engineering often contains noise and uncertainty, which poses new challenges to the performance of intelligent diagnosis methods. On the one hand, measurement uncertainty will affect the training and inference processes of the model, resulting in the instability of diagnostic results; on the other hand, the complex system dynamics and multi-factor coupling increase the difficulty of accurately extracting fault features. Existing research usually assumes that the measurement data is reliable and rarely considers the impact of measurement uncertainty on diagnostic results, resulting in insufficient reliability of diagnostic models in practical applications. Therefore, how to construct an intelligent and robust fault diagnosis method considering measurement uncertainty has become an important research direction in the current field of fault diagnosis of transmission lubricating oil systems. Summary of the Invention:

[0004] The present invention provides an intelligent fault diagnosis method for a transmission lubricating oil system considering measurement uncertainty to solve the above problems existing in the prior art.

[0005] The present invention adopts the following technical solutions: An intelligent fault diagnosis method for a transmission lubricating oil system considering measurement uncertainty, the steps are as follows:

[0006] Step 1: Lubricating oil system fault injection and test point selection and sensitivity analysis. A layered injection strategy is used for the multi-source fault types of the target lubricating oil system. For fault modes with physical reproducibility, physical fault injection experiments are conducted on the lubricating oil system bench test platform. For irreversible faults with complex coupling or significant nonlinear characteristics, a virtual model of the lubricating oil system is constructed based on multi-domain joint simulation technology to perform fault numerical simulation. Available test points are identified in the lubricating oil system bench test platform and the multi-domain joint simulation model, and threshold parameter calibration is completed for each preferred test point.

[0007] Step 2: Perform a testability analysis on the target lubricating oil system. Based on the output of the test points after the fault injection in Step 1, establish a "fault-test" correlation matrix to describe the correlation between the fault / normal state of the lubricating oil system and the test points.

[0008] Step 3: Under the assumption of imperfect testing, a hybrid quantitative model based on a Bayesian probabilistic framework is constructed to optimize the test model of the target lubricating oil system based on the dynamic uncertainty characteristics of the sensor measurement error used to collect lubricating oil system operating data.

[0009] In step 4, based on the testability model optimized in step 3, a dynamic fault diagnosis strategy suitable for the imperfect test assumption is further proposed.

[0010] Furthermore, in step 2, the "fault-test" correlation matrix, i.e., the testability mathematical model, describes the correlation between system faults / states and test points. It is a Boolean matrix containing only 0s and 1s, and its specific form is as follows:

[0011]

[0012] Where: ft ij All are Boolean values, ft ij =1 indicates measuring point T j Fault F can be detected i , ft ij =0 means no, the i-th row vector F i =[ft i1 ,ft i2 ,…,ft ij ] indicates fault F i The output of all available test points when it occurs is understood as fault F i The sign of the jth column vector T j =[ft 1j ,ft 2j ,…,ft ij ] T Indicates the available test point T jThe detectable faults are understood as the fault detection and isolation capabilities of measurement point T j

[0013] When performing testability analysis, the following assumptions are made:

[0014] (2.1) Hypothesis of only two states: The system only has normal or single fault states;

[0015] (2.2) Single fault hypothesis: At most one fault occurs at any moment;

[0016] (2.3) Test reuse hypothesis: Test points can be reused;

[0017] (2.4) Dynamic test hypothesis: The test results are affected by uncertainties and need to be dynamically corrected through a probability model.

[0018] Furthermore, step 3 specifically includes:

[0019] (3.1) Sources of uncertainties include: signal transmission distortion; sensor dynamic error; correlation variations caused by multi-fault coupling and component defects in the lubricating oil system; data reporting delay or loss;

[0020] (3.2) Let the sensor measurement error be ∈~N(0,σ 2 ), where σ 2 obeys the inverse gamma prior distribution σ 2 ~Inv-Gamma(α0,β0), representing historical accuracy data. Introduce the hidden state variable s t to describe the sensor drift mode caused by environmental interference and construct a Markov state transition network;

[0021] (3.3) Use the Markov chain Monte Carlo (MCMC) method to achieve real-time parameter update:

[0022] (3.4) Data flow trigger: When new test data y1, y2,..., y t is received, alternately update the drift state s t through Gibbs sampling, dynamically adjust the error variance σ 2 based on the acceptance rate of the Metropolis-Hastings algorithm, and update the posterior hyperparameters according to σ 2 |y,s~Inv-Gamma(α t ,β t );

[0023]

[0024] ​(3.5) Adaptive Convergence Diagnosis: Monitor the convergence of the MCMC chain using the PSRF (Potential Scale Reduction Factor), dynamically adjust the number of samplings, fuse the updated error distribution with the testability requirements, and extend the traditional 3σ criterion to a dynamic threshold Where is the expected value of the current posterior distribution, and construct a Fault Probability Ratio (FPR) metric:

[0025]

[0026] Dynamically adjust the diagnostic strategy through the FPR value to suppress false alarms;

[0027] (3.6) Online Verification: Use the Bayes Factor to compare the explanatory power of the new and old models for the observed data:

[0028]

[0029] When B 01 < 0.1, trigger model reconstruction, perform marginalization processing, and exponentially decay and weight the importance of historical data to balance model sensitivity and stability.

[0030] Furthermore, the dynamic fault diagnosis strategy applicable under the imperfect test hypothesis in step 4 is as follows:

[0031] (4.1) First, construct a dynamic error diagnosis cost model;

[0032] (4.2) During the dynamic diagnosis process, the system makes adaptive adjustments through real-time data feedback;

[0033] (4.3) Each time new data is obtained, adjust the weight coefficient in the diagnostic strategy based on the updated posterior distribution;

[0034] (4.4) Use the information entropy algorithm to optimize the selection of test points to ensure that the selection of each test point can maximize the information gain of the system;

[0035] (4.5) Combine the adaptive optimization strategy and select the optimal test point sequence by solving the objective function that minimizes the expected misdiagnosis cost.

[0036] Furthermore, in step 4.1, the error diagnosis cost model includes the following aspects:

[0037] Misdiagnosis cost: The maintenance cost generated when the system is misdiagnosed as faulty;

[0038] Missed diagnosis cost: The subsequent losses caused when the system is in a faulty state but fails to be diagnosed;

[0039] False alarm cost: When the system is in a normal state, a false diagnosis of a fault causes waste of resources and additional maintenance costs.

[0040] The mathematical expression of this model is as follows:

[0041] C error = α·C miss + β·C false alarm + γ·C damage + δ·C recovery (5)

[0042] Where: C error is the total cost of false diagnosis; C miss is the cost of missed diagnosis, representing the loss incurred when a fault is not detected; C false alarm is the false alarm cost, representing the cost incurred when the system is wrongly diagnosed as faulty when it is normal; C damage is the system damage cost, representing the repair cost of the system damage caused by misdiagnosis; C recovery is the recovery cost, representing the cost required for recovery measures due to missed diagnosis or misdiagnosis; α, β, γ, δ are weight coefficients, which are adjusted according to expert experience or historical data.

[0043] Furthermore, in step 4.2, the specific method is: by monitoring the change trend of the sensor output data and combining historical data, dynamically adjust the weight coefficients and thresholds in the diagnostic strategy to optimize the diagnostic results, and use the Bayesian update method to adjust the misdiagnosis cost model in real time. Assume that the initial diagnostic cost model is θ0. As data is collected and analyzed, continuously update the posterior distribution of the model through the Bayesian method:

[0044]

[0045] Where: P(θ|D) is the posterior probability of the model parameter θ under the data D; P(D|θ) is the likelihood function, representing the probability of the data occurring under the given model parameter; P(θ) is the prior distribution, representing the initial assumption of the model parameter; P(D) is the marginal probability of the data.

[0046] Furthermore, in step 4.3, during the implementation process, a multi-level fault identification framework is adopted. Specifically:

[0047] The first layer: The rapid screening stage, based on real-time measurement data for preliminary judgment to quickly determine whether there is a fault;

[0048] The second layer: The precise diagnosis stage, deeply analyze the type of suspected fault to determine the specific fault category;

[0049] The third layer: the optimization and repair stage, which selects the optimal repair plan by combining the current state of the system and the repair cost.

[0050] Further, in step 4.4, information entropy is used to measure the additional information provided by each test point in the system. The specific formula is:

[0051]

[0052] where: H(X) is the information entropy, representing the average uncertainty of the system state; p(x i ) is the probability distribution of the system state x i .

[0053] Further, in step 4.5, combining the adaptive optimization strategy of cross-domain optimization, the optimal test point selection problem is solved by maximizing the information gain and minimizing the misdiagnosis loss:

[0054]

[0055] where E[C error (t)] is the expected cost of false diagnosis after selecting the test point t.

[0056] The present invention has the following beneficial effects: By introducing the assumption of imperfect testing, the present invention systematically considers the errors caused by measurement uncertainty in the transmission lubricating oil system, including the situations of missed detection and false alarms. Aiming at the traditional method that only relies on the test results under ideal conditions, a fault diagnosis model integrating measurement uncertainty factors is established, enabling it to more accurately reflect the fault detection ability in actual working conditions. Description of the drawings:

[0057] Figure 1 is the flow chart of the intelligent fault diagnosis method for the transmission lubricating oil system considering measurement uncertainty of the present invention. Specific implementation manners:

[0058] The present invention will be further described below with reference to the drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention and cannot be used to limit the protection scope of the present invention.

[0059] The intelligent fault diagnosis method for the transmission lubricating oil system considering measurement uncertainty of the present invention, as Figure 1 shown, specifically includes the following steps:

[0060] Step 1: Fault injection, test point selection and sensitivity analysis for the lubricating oil system. For the multi-source fault types of the target lubricating oil system, a hierarchical injection strategy is adopted: for fault modes with physical reproducibility, physical fault injection experiments are carried out relying on the bench test platform of the lubricating oil system; for irreversible faults with significant complex coupling or nonlinear characteristics, a virtual model of the lubricating oil system is constructed based on multi-domain co-simulation technology for fault numerical simulation. Determine the available test points in the bench test platform of the lubricating oil system and the multi-domain co-simulation model, and calibrate the threshold parameters of each optimized measurement point according to the equipment operation specifications. Here, the equipment refers to the equipment related to the lubricating oil system, including:

[0061] 1. Bench test platform of the lubricating oil system: A physical platform for carrying out physical fault injection experiments.

[0062] 2. Multi-domain co-simulation model: A virtual model for simulating irreversible faults with significant complex coupling or nonlinear characteristics.

[0063] 3. Sensors and measuring equipment: Various sensors and measuring tools for collecting the operation data of the lubricating oil system.

[0064] 4. Control and monitoring equipment: Equipment for monitoring the status of the lubricating oil system and controlling the experimental conditions.

[0065] These equipments are jointly used for test point selection, sensitivity analysis and threshold parameter calibration.

[0066] In Step 1, the multi-source fault types of the target lubricating oil system include four categories: abnormal lubricating oil consumption, instability of the pressure system, imbalance of the temperature field, and medium pollution. The test points are selected according to the structural characteristics of the bench test platform of the lubricating oil system and the multi-domain co-simulation model, and five core test points of lubricating oil pressure, temperature, flow rate, liquid level, and rotational speed are determined. Among them, the lubricating oil pressure threshold is calibrated according to the standard pressure value of the lubricating oil system under normal operating conditions ±5.6 kPa, the temperature threshold is set as a dynamic range of ±2 °C according to the working conditions of the current system, the liquid level threshold is calibrated based on the geometric characteristics of the lubricating oil tank as ±3.5 mm, the flow rate threshold is calibrated according to the measurement accuracy level of the flow sensor as ±2%, and the rotational speed threshold is calibrated according to the accuracy of the Hall effect sensor group as a dynamic range of ±0.5%.

[0067] Step 2, conduct testability analysis on the target lubricating oil system. According to the output of the test points after fault injection in Step 1, establish a "fault - test" correlation matrix (i.e., D matrix) to describe the correlation relationship between the fault / normal state of the lubricating oil system and the test points.

[0068] The "fault - test" correlation matrix (i.e., the D matrix), which is the testability mathematical model, describes the correlation relationship between system faults / states and test points. It is a Boolean matrix containing only 0 and 1, and its specific form is as follows:

[0069]

[0070] where: ft ij are all Boolean values. ft ij = 1 means that the test point T j can detect the fault F i , and ft ij = 0 means it cannot. The i - th row vector F i = [ft i1 , ft i2 , …, ft ij represents the outputs of all available test points when the fault F i occurs, and can be understood as the symptoms of the fault F i ; the j - th column vector T j = [ft 1j , ft 2j , …, ft ij T represents the faults that the available test point T j can detect, and can be understood as the fault detection and isolation ability of the test point T j .

[0071] When conducting testability analysis, the following assumptions are made:

[0072] (2.1) Hypothesis of only two states: The system only has normal or single - fault states;

[0073] (2.2) Single - fault hypothesis: At most one fault occurs at any moment;

[0074] (2.3) Test reuse hypothesis: Test points can be reused;

[0075] (2.4) Dynamic test hypothesis: The test results are affected by uncertainties and need to be dynamically corrected through a probability model.

[0076] Step 3, under the imperfect test hypothesis, aiming at the dynamic uncertainty characteristics of sensor measurement errors, a hybrid quantization model based on the Bayesian probability framework is constructed to optimize the testability model of the target lubricating oil system. This method combines the static threshold determination of the 3 - sigma criterion with the dynamic error distribution estimation: First, a normal distribution N(0, σ 2 ​) is the measurement error probability model of the initial prior, and the posterior distribution parameters are updated iteratively online through the Markov Chain Monte Carlo (MCMC) method. Each time new test data is obtained, the Metropolis-Hastings algorithm is used for adaptive sampling to correct the mean offset μ caused by sensor drift in real time. t and variance to synchronously generate a dynamic 3σ confidence interval.

[0077] In step 3, under the imperfect test hypothesis, for the dynamic uncertainty of the sensor measurement error, a hybrid quantization model based on the Bayesian probability framework is constructed to optimize the testability model of the target lubricating oil system, specifically including:

[0078] (3.1) Sources of uncertainty: signal transmission distortion (temperature, flow rate, packet loss); sensor dynamic error (self-accuracy drift, environmental interference); correlation variation caused by multi-fault coupling and system component defects; data reporting delay or loss. Here, the system components refer to the components in the lubricating oil system, such as pumps, valves, lubricating oil filters, radiators, etc.

[0079] (3.2) Assume that the sensor measurement error is ∈~N(0,σ 2 ), where σ 2 obeys the inverse gamma prior distribution σ 2 ~Inv-Gamma(α0,β0), which characterizes the historical accuracy data. Introduce the hidden state variable s t to describe the sensor drift mode caused by environmental interference (such as temperature, vibration level), and construct a Markov state transition network.

[0080] (3.3) Use the Markov Chain Monte Carlo (MCMC) method to achieve real-time parameter update:

[0081] (3.4) Data stream trigger: When new test data y1, y2,…, y t is received, the drift state s t is alternately updated through Gibbs sampling, and the error variance σ 2 is dynamically adjusted based on the acceptance rate of the Metropolis-Hastings algorithm. According to σ 2 |y,s~Inv-Gamma(α t ,β t ), the posterior hyperparameters are updated:

[0082]

[0083] (3.5) Adaptive Convergence Diagnosis: Monitor the convergence of the MCMC chain using PSRF (Potential Scale Reduction Factor) and dynamically adjust the number of samplings. Integrate the updated error distribution with the testability requirements and extend the traditional 3σ criterion to a dynamic threshold. Where is the expected value of the current posterior distribution, and construct a Fault Probability Ratio (FPR) metric:

[0084]

[0085] Dynamically adjust the diagnostic strategy through the FPR value to suppress false alarms.

[0086] (3.6) Online Verification: Use the Bayes Factor to compare the explanatory power of the new and old models for the observed data:

[0087]

[0088] When B 01 < 0.1, trigger model reconstruction, perform marginalization processing, exponentially decay and weight the importance of historical data, and balance the model sensitivity and stability.

[0089] Step 4: Based on the optimized testability model described above, further propose a dynamic fault diagnosis strategy applicable to the imperfect test hypothesis. In this strategy, various complex factors affecting the diagnostic process are considered, including environmental fluctuations, sensor errors, and their interference with test results. By constructing a dynamic misdiagnosis cost model and introducing an adaptive diagnosis strategy based on real-time data feedback, combined with a multi-level fault identification framework, the detection ability and diagnostic accuracy of system faults are effectively improved. In the misdiagnosis cost calculation, multi-dimensional factors such as false alarms, missed detections, system damage, and recovery costs are comprehensively considered, and the allocation of test resources is optimized through risk assessment. Combining the information entropy algorithm, using an adaptive optimization strategy and cross-domain optimization technology, the accurate selection of the optimal test points is achieved.

[0090] In Step 4, under the imperfect test hypothesis, the test results are affected by various factors, which may lead to misdiagnosis and missed detection. To effectively measure these errors, first, a dynamic misdiagnosis cost model needs to be constructed. This model includes the following aspects:

[0091] Misdiagnosis Cost: The maintenance cost generated when the system is misdiagnosed as faulty.

[0092] Missed Detection Cost: The subsequent losses caused when the system is in a faulty state but fails to be diagnosed.

[0093] False alarm cost: When the system is in a normal state, a fault is misdiagnosed, resulting in wasted resources and additional maintenance costs.

[0094] The mathematical expression of this model is as follows:

[0095] C error = α·C miss + β·C false alarm + γ·C damage + δ·C recovery (5)

[0096] Where: C error is the total cost of misdiagnosis; C miss is the cost of missed diagnosis, representing the loss incurred when a fault is not detected; C false alarm is the false alarm cost, representing the cost incurred when the system is misdiagnosed as faulty when it is normal; C damage is the system damage cost, representing the repair cost of system damage caused by misdiagnosis; C recovery is the recovery cost, representing the cost required for recovery measures due to missed diagnosis or misdiagnosis; α, β, γ, δ are weight coefficients, which are adjusted based on expert experience or historical data.

[0097] In the dynamic diagnosis process, the system makes adaptive adjustments through real-time data feedback. The specific method is: by monitoring the change trend of the sensor output data and combining historical data, dynamically adjust the weight coefficients and thresholds in the diagnosis strategy to optimize the diagnosis results. For this purpose, the Bayesian update method is used to adjust the misdiagnosis cost model in real time.

[0098] Assume that the initial diagnosis cost model is θ0. As data is collected and analyzed, the posterior distribution of the model is continuously updated through the Bayesian method:

[0099]

[0100] Where: P(θ|D) is the posterior probability of the model parameter θ under the data D; P(D|θ) is the likelihood function, representing the occurrence probability of the data under the given model parameters; P(θ) is the prior distribution, representing the initial assumption of the model parameters;

[0101] P(D) is the marginal probability of the data.

[0102] Each time new data is obtained, the weight coefficients in the diagnosis strategy are adjusted based on the updated posterior distribution, thereby improving the robustness of the system to misdiagnosis and missed diagnosis.

[0103] In the implementation process, a multi-level fault identification framework is adopted to enhance the diagnostic ability of the system. Specifically:

[0104] The first layer: The rapid screening stage, which makes a preliminary judgment based on real-time measurement data and quickly determines whether there is a fault.

[0105] The second layer: The precise diagnosis stage, which deeply analyzes the type of suspected fault and determines the specific fault category.

[0106] The third layer: The optimization and repair stage, which selects the optimal maintenance plan by combining the current state of the system and the repair cost.

[0107] When optimizing the selection of test points, the information entropy algorithm is adopted to ensure that the selection of each test point can maximize the information gain of the system. Information entropy is used to measure the additional information provided by each test point in the system, and the specific formula is:

[0108]

[0109] where: H(X) is the information entropy, representing the average uncertainty of the system state; p(x i ) is the probability distribution of the system state x i .

[0110] Then, combined with the adaptive optimization strategy, by solving the objective function that minimizes the expected misdiagnosis cost, the optimal test point sequence is selected. The selection of the optimal test point aims to reduce the overall diagnosis cost and improve the fault detection ability of the system.

[0111] In practical applications, in addition to relying on the real-time data of the current device, the system should also combine cross-domain optimization techniques to obtain information from different fields (such as sensor technology, data acquisition, fault models, etc.) to improve the accuracy of test point selection. Through cross-domain learning, the data and models of each field are fused to optimize the test strategy.

[0112] Cross-domain optimization can solve the problem of optimal test point selection by maximizing the information gain and minimizing the misdiagnosis loss:

[0113]

[0114] where, E[C error (t)] is the expected wrong diagnosis cost after selecting the test point t.

[0115] The above is only the preferred embodiment of the present invention. It should be noted that for those of ordinary skill in the art, several improvements can be made without departing from the principle of the present invention, and these improvements should also be regarded as the protection scope of the present invention.

Claims

1. An intelligent fault diagnosis method for a transmission lubricating oil system considering measurement uncertainty, characterized in that: The steps are as follows: Step 1: Fault injection and test point selection and sensitivity analysis for the lubricating oil system. For the multi-source fault types of the target lubricating oil system, a hierarchical injection strategy is adopted: for the fault modes with physical reproducibility, physical fault injection experiments are carried out relying on the lubricating oil system bench test platform; For irreversible faults with complex coupling or significant nonlinear characteristics, a virtual model of the lubricating oil system is constructed based on multi-domain co-simulation technology for fault numerical simulation. Available test points are determined in the lubricating oil system bench test platform and the multi-domain co-simulation model, and the threshold parameters of each selected measurement point are calibrated; Step 2: Conduct testability analysis on the target lubricating oil system. According to the outputs of the test points after fault injection in Step 1, establish a "fault - test" correlation matrix to describe the correlation relationship between the fault / normal state of the lubricating oil system and the test points; Step 3: Under the imperfect test hypothesis, aiming at the dynamic uncertainty characteristics of the sensor measurement errors for collecting the operating data of the lubricating oil system, construct a hybrid quantization model based on the Bayesian probability framework to optimize the testability model of the target lubricating oil system; Step 4: Based on the optimized testability model in Step 3, further propose a dynamic fault diagnosis strategy applicable to the imperfect test hypothesis.

2. The intelligent fault diagnosis method for a transmission lubricating oil system considering measurement uncertainty as described in claim 1, characterized in that: In Step 2, the "fault - test" correlation matrix, that is, the testability mathematical model, describes the correlation relationship between the system fault / state and the test points, and is a Boolean matrix containing only 0 and 1, and the specific form is as follows: where: ft ij are all Boolean values, and ft ij = 1 indicates that the measurement point T j can detect the fault F i , and ft ij = 0 indicates that it cannot. The i-th row vector F i = [ft i1 , ft i2 , …, ft ij represents the outputs of all available measurement points when the fault F i occurs, and is understood as the symptom of the fault F i . The j-th column vector T<00000\alpha>= [ft 1j , ft 2j , …, ft ij T represents the faults that the available measurement point T j can detect, and is understood as the fault detection and isolation ability of the measurement point T j .It should be noted that there seems to be a typo in the original text where "<00000\alpha>" should probably be " j ". This has been left as is in the translation to match the original as closely as possible. When conducting testability analysis, the following assumptions are made: (2.1) State uniqueness assumption: The system only has normal or single fault states; (2.2) Single fault assumption: At most one fault occurs at any moment; (2.3) Test reuse assumption: The test points can be reused; (2.4) Dynamic test assumption: The test results are affected by uncertainty and need to be dynamically corrected through a probability model.

3. The intelligent fault diagnosis method for a transmission lubricating oil system considering measurement uncertainty according to claim 2, characterized in that: Step 3 specifically includes: (3.1) The sources of uncertainty include: signal transmission distortion; sensor dynamic error; correlation variation caused by multi-fault coupling and component defects in the lubricating oil system; data reporting delay or loss; (3.2) Let the sensor measurement error be ∈ ~ N(0, σ 2 ), where σ 2 follows an inverse gamma prior distribution σ 2 ~ Inv-Gamma(α0, β0), representing historical accuracy data. Introduce the hidden state variable s t to describe the sensor drift mode caused by environmental interference and construct a Markov state transition network; (3.3) Use the Markov chain Monte Carlo (MCMC) method to achieve real-time parameter update: (3.4) Data stream trigger: When new test data y1, y2, …, y t is received, the drift state s is alternately updated through Gibbs sampling t , and the error variance σ is dynamically adjusted based on the acceptance rate of the Metropolis-Hastings algorithm 2 . According to σ 2 |y, s ∼ Inv-Gamma(α t , β t ), the posterior hyperparameters are updated: (3.5) Adaptive Convergence Diagnosis: Monitor the convergence of the MCMC chain using the PSRF (Potential Scale Reduction Factor), dynamically adjust the number of samplings, fuse the updated error distribution with the testability requirements, and extend the traditional 3σ criterion to a dynamic threshold where is the expected value of the current posterior distribution, and construct the Fault Probability Ratio (FPR) metric: Dynamically adjust the diagnostic strategy through the FPR value to suppress false alarms; (3.6) Online verification: Use the Bayes Factor to compare the ability of the new and old models to explain the observed data: When B 01 When <0.1, trigger model reconstruction, perform marginalization processing, and exponentially decay-weight the importance of historical data to balance model sensitivity and stability.

4. The intelligent fault diagnosis method for a transmission lubricating oil system considering measurement uncertainty according to claim 3, characterized in that: The dynamic fault diagnosis strategy applicable to the imperfect test hypothesis in Step 4 is specifically as follows: (4.1) First, construct a dynamic error diagnosis cost model; (4.2) During the dynamic diagnosis process, the system makes adaptive adjustments through real-time data feedback; (4.3) Each time new data is obtained, adjust the weight coefficient in the diagnostic strategy based on the updated posterior distribution; (4.4) Use the information entropy algorithm to optimize the selection of test points to ensure that the selection of each test point can maximize the information gain of the system. (4.5) Combine the adaptive optimization strategy and select the optimal test point sequence by solving the objective function that minimizes the expected misdiagnosis cost.

5. The intelligent fault diagnosis method for a transmission lubricating oil system considering measurement uncertainty according to claim 4, characterized in that: In step 4.1, the error diagnosis cost model includes the following aspects: Misdiagnosis cost: The maintenance cost generated when the system is misdiagnosed as faulty. Missed diagnosis cost: The subsequent losses caused when the system is in a faulty state but fails to be diagnosed. False alarm cost: The waste of resources and additional maintenance cost caused by misdiagnosing the system as faulty when it is in a normal state. The mathematical expression of this model is as follows: C error = α·C miss + β·C false alarm + γ·C damage + δ·C recovery (5) Where: C error is the total cost of error diagnosis; C miss is the cost of missed diagnosis, representing the loss incurred when a fault is not detected; C false alarm is the cost of false alarm, representing the cost incurred when the system is wrongly diagnosed as faulty when it is normal; C damage is the cost of system damage, representing the repair cost of system damage caused by misdiagnosis; C recovery is the recovery cost, representing the cost required for recovery measures caused by missed diagnosis or misdiagnosis; α, β, γ, δ are weight coefficients, which are adjusted according to expert experience or historical data.

6. The intelligent fault diagnosis method for a transmission lubricating oil system considering measurement uncertainty according to claim 5, characterized in that: In step 4.2, the specific method is: By monitoring the change trend of the sensor output data and combining historical data, dynamically adjust the weight coefficients and thresholds in the diagnosis strategy to optimize the diagnosis results, and use the Bayesian update method to adjust the misdiagnosis cost model in real time. Assume the initial diagnosis cost model is θ0. As data is collected and analyzed, continuously update the posterior distribution of the model through the Bayesian method: Where: P(θ|D) is the posterior probability of the model parameter θ under the data D; P(D|θ) is the likelihood function, representing the occurrence probability of the data given the model parameters; P(θ) is the prior distribution, representing the initial assumption about the model parameters; P(D) is the marginal probability of the data.

7. The intelligent fault diagnosis method for a transmission lubricating oil system considering measurement uncertainty according to claim 6, characterized in that: Step 4.3, in the implementation process, adopt a multi-level fault identification framework. Specifically: The first layer: The rapid screening stage, make a preliminary judgment based on the real-time measurement data to quickly determine whether there is a fault. The second layer: The precise diagnosis stage, conduct in-depth analysis on the suspected fault type to determine the specific fault category. The third layer: The optimization and repair stage, combine the current state of the system and the repair cost to select the optimal repair plan.

8. The intelligent fault diagnosis method for a transmission lubricating oil system considering measurement uncertainty according to claim 7, characterized in that: Step 4.4, Information entropy is used to measure the additional information provided by each test point in the system. The specific formula is: Among them: H(X) is the information entropy, representing the average uncertainty of the system state; p(x i ) is the probability distribution of the system state x i .

9. The intelligent fault diagnosis method for a transmission lubricating oil system considering measurement uncertainty according to claim 8, characterized in that: Step 4.5, Combine the adaptive optimization strategy of cross-domain optimization and solve the optimal test point selection problem by maximizing the information gain and minimizing the misdiagnosis loss: Among them, E[C error (t)] is the expected error diagnosis cost after selecting the test point t.

Images