IRUN-BRB diesel engine composite fault diagnosis method based on multiple labels

Abstract

The invention discloses an IRUN-BRB diesel engine composite fault diagnosis method based on multiple labels, and the method comprises the steps: determining a plurality of single fault types of a diesel engine, and obtaining a corresponding single fault data sample; a plurality of BRB models are correspondingly established, each BRB model corresponds to one single fault type, and the consequent attributes of the BRB model comprise a health state, the single fault type and other single fault types; determining a front part feature combination of each BRB model; based on the single-fault data sample, determining a antecedent feature combination of all BRB models, inputting the antecedent feature combination into the corresponding BRB model, calculating confidence distribution of consequent attributes of each BRB model, and taking the consequent attribute with the maximum confidence as output; all BRB models are optimized based on an improved Runge-Kutta algorithm; and obtaining a composite fault sample, and determining a composite fault based on the output combination of all the optimized BRB models. According to the method, the multi-label classification method is established and combined with the IRUN-BRB model, so that composite fault diagnosis of the diesel engine can be realized under the condition of zero samples.

Description

Technical Field

[0001] The present invention belongs to the technical field of fault detection, and particularly relates to a multi-label-based IRUN-BRB diesel engine compound fault diagnosis method for identifying compound faults of a diesel engine in a zero-sample situation. Background Technique

[0002] Emergency diesel generators are mainly used to automatically start diesel generator sets when off-site power is lost in nuclear power plants, provide necessary power supply for them, and should have reliability matching the requirements of safety systems that need their power support to prevent radioactive material leakage caused by core meltdown. Although the risk of a complete power outage in a nuclear power plant is very small, nuclear leakage accidents have still occurred due to the loss of in-plant power. Therefore, the health status monitoring and fault diagnosis of diesel generator sets are of great significance. As the key power system of an emergency diesel generator set, monitoring the status and diagnosing faults of the diesel engine, promptly grasping the operating status of the diesel engine, determining a reasonable maintenance plan, and avoiding sudden faults are also of great importance for ensuring the safety of relevant personnel.

[0003] During the normal operation of a diesel engine, due to the interaction between different components, faults generated by the coupling of different single faults are compound faults. The economic losses and personal injuries caused by the hazards of compound faults are much greater than those caused by single faults. Since the types of coupled faults are different, when diagnosing compound faults, the types of single faults cannot be determined before diagnosis. Therefore, the adaptability of the diagnosis method to various compound faults needs to be considered, and the diagnostic output should include as many compound fault combination forms as possible. However, due to the occasional and concurrent nature of faults in the diesel engine system, the number of compound faults grows exponentially compared to the number of single faults. It is difficult to label appropriate tags for each compound fault sample in the actual application of a diesel engine. Therefore, it is particularly important to propose an effective diesel engine compound fault diagnosis method.

[0004] According to different diesel engine fault state detection and analysis methods, fault diagnosis methods can be divided into oil detection methods, instantaneous speed detection methods, thermal parameter detection methods, and acoustic detection methods. The thermal parameter analysis method mainly measures parameters such as cylinder pressure, power, fuel consumption rate, exhaust temperature, and pressure during the operation of the diesel engine. They comprehensively reflect the working cycle state of the diesel engine, directly affect the process of converting the thermal energy of the diesel engine into mechanical energy, and are concentratedly reflected in the thermal parameters, having the advantage of accurately and directly reflecting the working state of the diesel engine. Summary of the Invention

[0005] The object of the present invention is to provide a multi-label-based IRUN-BRB diesel engine compound fault diagnosis method, which can accurately identify the compound faults of the diesel engine by establishing a multi-label classification method combined with the IRUN-BRB model in the case of lack of diesel engine compound fault sample data.

[0006] Next, in the first aspect of the present invention, a multi-label-based IRUN-BRB diesel engine compound fault diagnosis method is provided, and the method includes:

[0007] Determine various single fault types of the diesel engine and obtain corresponding single fault data samples;

[0008] For various single fault types of the diesel engine, establish multiple BRB models correspondingly, each BRB model corresponds to a single fault type, and the consequent attributes of the BRB model include the health state, this single fault type, and other single fault types;

[0009] Determine the antecedent feature combinations of each BRB model, and the antecedent feature combination is a combination of fault feature parameters corresponding to the single fault type of the BRB model;

[0010] Based on the single fault data samples, determine the antecedent feature combinations of all BRB models and input them into the corresponding BRB models. Calculate the confidence distribution of the consequent attributes of each BRB model through confidence conversion, rule activation, and evidence reasoning, and the consequent attribute with the maximum confidence is used as the output of the BRB model;

[0011] Optimize all BRB models based on the improved Runge-Kutta algorithm IRUN to make the outputs of all BRB models conform to the single fault type of the single fault data sample;

[0012] Obtain compound fault samples, determine the corresponding antecedent feature combinations and input them into all optimized BRB models, and determine the compound faults based on the output combinations of all optimized BRB models.

[0013] In some embodiments, the various single fault types of the diesel engine include the intercooler efficiency degradation fault F1, the crankcase gas leakage fault F2, the fuel injection timing fault F3, and the fuel supply reduction fault F4;

[0014] Among them, the antecedent feature combination of the intercooler efficiency degradation fault F1 is the intercooler outlet temperature Tco and the maximum combustion pressure Pf. The intercooler outlet temperature Tco is the main antecedent feature, and the maximum combustion pressure Pf is the auxiliary antecedent feature;

[0015] The antecedent feature combination of the crankcase gas leakage fault F2 is the fuel consumption rate Ge, the average exhaust valve temperature Tc, and the maximum pressure rise rate Rf. The fuel consumption rate Ge and the average exhaust valve temperature Tc are the main antecedent features, and the maximum pressure rise rate Rf is the auxiliary antecedent feature;

[0016] The antecedent feature combination of the fuel injection timing fault F3 is the maximum pressure rise rate Rf, the maximum combustion pressure Pf, and the mean effective pressure Pm. The maximum pressure rise rate Rf and the maximum combustion pressure Pf are the main antecedent features, and the mean effective pressure Pm is the auxiliary antecedent feature;

[0017] The antecedent feature combination of the fuel supply reduction fault F4 is the temperature Ti before the turbine, the temperature To after the turbine, and the rated power Pw. The temperature Ti before the turbine and the temperature To after the turbine are the main antecedent features, and the rated power Pw is the auxiliary antecedent feature;

[0018] The main antecedent features are used to identify the health state and this single fault type, and there are three reference levels L, M, and H; the auxiliary antecedent features are used to identify the health state and other single fault types, and there are two reference levels L and H.

[0019] In some of these embodiments, the health state indicates that the diesel engine is in a fault-free state. This single fault type indicates that the fault state that occurs in the diesel engine at this time covers the single fault type corresponding to the corresponding BRB model. Other single fault types indicate that the diesel engine has a fault but the fault type cannot be determined.

[0020] In some of these embodiments, the k-th confidence rule R of the BRB model k is:

[0021]

[0022] with rule weight θ k and feature weight δ i (i = 1, 2, 3, …, M)

[0023] where, represents the j-th reference level of the input antecedent feature X M in the k-th rule of the BRB model, β n k represents the confidence degree of the n-th consequent attribute D n in the k-th rule, N represents the total number of consequent attributes, θ k represents the rule weight of the k-th rule, δ i represents the feature weight of the i-th antecedent feature, and M represents the number of antecedent features;

[0024] Each BRB model corresponds to output N confidence degrees β n , and each consequent attribute corresponds to a confidence degree β n . According to the magnitude of the confidence degree β n , predict the fault state of the diesel engine.

[0025] In some of these embodiments, the confidence distribution of the consequent attributes of each BRB model is calculated through confidence conversion, rule activation, and evidence reasoning, including:

[0026] (1) Calculate the confidence distribution of each input data x for a single fault i The confidence distribution corresponding to the reference level in the k-th rule is α ij+1 = 1 - α ij , (u(A i,j ) < x i < u(A i,j+1 ))), A ij represents the j-th reference level of the antecedent feature X i , α ij is the confidence of the j-th reference level of the single-fault data x i , u(A i,j ) represents the quantization value of the reference level corresponding to A ij ;

[0027] (2) Calculate the activation weight in the k-th rule where M represents the number of antecedent features, L represents the number of rules, represents the relative feature weight of the antecedent feature, represents the feature weight of the i-th antecedent feature in the k-th rule, θ k represents the rule weight of the k-th rule;

[0028] (3) Use the evidence reasoning ER algorithm for rule fusion to obtain the confidence distribution of the consequent attributes in different states of the diesel engine where, β n is the confidence of the n-th consequent attribute D n in the k-th rule;

[0029] Obtain the consequent attribute D n corresponding to the maximum confidence β n as the output of the BRB model.

[0030] In some of these embodiments, all BRB models are optimized based on the improved Runge-Kutta algorithm IRUN, including:

[0031] In the k-th rule, if the diesel engine is in a healthy state, the corresponding confidence is β n k , then β n k > β j k (j ≠ n, j = 1, 2,..., N) to serve as the equality constraint of the consequent attribute; where, β jk (j ≠ n, j = 1, 2, …, N) is the remaining confidence distribution in this rule;

[0032] Initialize the parameters of the Runge - Kutta algorithm RUN and perform sampling operations. Then, detect the confidence distribution of the consequent attributes of the k - th rule:

[0033] if β n k ≤ β j k (j ≠ n, j = 1, 2, …, N 1 ) where 1 ≤ N 1 ≤ N - 1, then add the expert knowledge constraint of compound fault coupling Then enter the normalization operation Satisfy the equality constraint of the consequent attributes; where, β jI k is the confidence distribution of this rule after adding the knowledge embedding operation;

[0034] if β n k > β j k (j ≠ n, j = 1, 2, …, N), then directly enter the normalization operation Finally, perform the screening operation of the Runge - Kutta algorithm RUN to iteratively calculate the optimal value of the objective function, and select the optimal model parameter Ω from the population min .

[0035] In some of the embodiments, the objective function is as follows:

[0036]

[0037] Among them, S represents the number of samples, is the actual output of the s - th sample, β n (s) is the predicted output of the s - th sample, and the model parameter Ω to be optimized = [δ i , θ k , β j k , and the parameter range is 0 ≤ δ i ≤ 1(i = 1, 2,..., M), 0 ≤ θ k ≤ 1(k = 1, 2,..., L),

[0038] In some of the embodiments, for the compound fault test samples that exceed the single - fault data sample range, use the method of confidence conversion of compound fault data, and the confidence distribution is:

[0039] After the conversion of the confidence of the compound fault, the confidence distribution β of the consequent attributes of all the optimized BRB models is used for output. n According to the fault state with the highest confidence, the compound fault label is determined.

[0040] According to a second aspect of the present invention, there is provided an electronic device, including: a processor and a memory, where the memory stores a program or instructions that can run on the processor, and when the program or instructions are executed by the processor, the steps of the multi-label-based IRUN-BRB diesel engine compound fault diagnosis method described in any one of the first aspects are implemented.

[0041] According to a third aspect of the present invention, there is provided a readable storage medium, on which a program or instructions are stored, and when the program or instructions are executed by the processor, the steps of the multi-label-based IRUN-BRB diesel engine compound fault diagnosis method described in any one of the first aspects are implemented.

[0042] Generally speaking, compared with the prior art by the above technical solution conceived by the present invention, the following beneficial effects can be achieved:

[0043] By establishing a multi-label classification method combined with the IRUN-BRB model, the present invention can utilize the thermal parameter data to realize the compound fault diagnosis of the diesel engine in the case of zero compound fault samples. Using the multi-label classification method, the compound fault is converted into a combination form of single faults, reducing the types of compound faults and the number of consequent attributes of the BRB model, and reducing the complexity of the model; for the inadaptability between the single fault data samples and the compound fault data samples, by establishing a compound fault confidence conversion method, the model is adapted to the compound fault data samples; by embedding the RUN algorithm into the expert knowledge-guided algorithm to generate the optimal model parameter solution, the accurate diagnosis of the compound fault of the diesel engine is realized in the case of only single fault data samples.

[0044] This method is for the compound fault diagnosis of the diesel engine in the zero-sample case, and does not require the data samples and labels of each compound fault, only the data samples of each single fault.

[0045] The reasoning process of this method is transparent and highly interpretable, can effectively combine the expert knowledge in the design and manufacture of the diesel engine and its operation, and can accurately identify the compound fault combination forms with different single fault weights. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 It is a schematic flowchart of a compound fault diagnosis method for a diesel engine based on a multi-label IRUN-BRB model provided by an embodiment of the present application;

[0047] Figure 2 It is an improved structural diagram of a RUN algorithm provided by an embodiment of the present application;

[0048] Figure 3 This is an experimental result diagram of a compound fault diagnosis situation of a diesel engine provided by an embodiment of the present application;

[0049] Figure 4 This is a schematic diagram of the hardware structure of a computer device provided by an embodiment of the present application. Detailed implementation manners

[0050] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention. Based on the embodiments provided in the present application, all other embodiments obtained by those of ordinary skill in the art without making creative efforts fall within the protection scope of the present invention.

[0051] Obviously, the accompanying drawings in the following description are only some examples or embodiments of the present application. For those of ordinary skill in the art, without making creative efforts, the present application can also be applied to other similar scenarios based on these drawings. In addition, it can also be understood that although the efforts made in this development process may be complex and lengthy, for those of ordinary skill in the art related to the content disclosed in the present application, some design, manufacturing or production changes made on the basis of the technical content disclosed in the present application are only conventional technical means and should not be understood that the content disclosed in the present application is insufficient.

[0052] Referring to "embodiment" in the present application means that the specific features, structures or characteristics described in combination with the embodiment can be included in at least one embodiment of the present application. The phrase appears in various positions in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment mutually exclusive with other embodiments. It is explicitly and implicitly understood by those of ordinary skill in the art that the embodiments described in the present application can be combined with other embodiments without conflict.

[0053] Unless otherwise defined, the technical terms or scientific terms involved in this application should be understood by people with ordinary skills in the technical field to which this application belongs. The words "one", "a", "a", "the" and the like involved in this application do not indicate a quantitative limitation, and may represent the singular or plural. The terms "include", "comprise", "have" and any of their variations involved in this application are intended to cover non-exclusive inclusions; for example, a process, method, system, product or device that includes a series of steps or modules (units) is not limited to the listed steps or units, but may also include steps or units that are not listed, or may also include other steps or units inherent to these processes, methods, products or devices. The words "connect", "connected", "coupled" and the like involved in this application are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. The "multiple" involved in this application refers to two or more. "And / or" describes the association relationship of associated objects, indicating that there may be three relationships, for example, "A and / or B" can represent: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the objects before and after are in an "or" relationship. The terms "first", "second", "third", etc. involved in this application are only used to distinguish similar objects and do not represent a specific ordering of the objects.

[0054] Since its proposal, the confidence rule base reasoning (BRB) method has been widely used in the field of fault diagnosis and has achieved remarkable results, especially in the expression and analysis of uncertain information, incomplete information, and qualitative and quantitative mixed information, showing its superior performance and advantages. In addition, the confidence rule base optimization method proposed later is centered on the rule-based data analysis and mining process. This process is different from the traditional data-driven parameter learning method. With the support of the rule expert system, it appropriately weakens the description of the complex internal relationship of the research object and adds an interpretable description to the research object, which reduces the requirements for the initial model construction, making it more suitable for dealing with high-complexity diesel engine composite fault diagnosis problems. Based on this, this application proposes a composite fault diagnosis method based on the improved Runge-Kutta algorithm optimized confidence rule base reasoning model (IRUN-BRB). The BRB method can not only mine the inherent connection of fault data, but also combine the expert knowledge of multi-fault coupling to dominate the optimization direction of BRB model parameters, so as to achieve accurate diagnosis of the combination form of diesel engine composite faults in the case of zero samples.

[0055] like Figure 1As shown in the figure, this application proposes a compound fault diagnosis method for diesel engines based on IRUN - BRB. First, for the single - fault data of diesel engines, a BRB model is established for each fault type between the fault thermotechnical parameters and the fault type. Each BRB model outputs three fault states, namely: healthy state, specific fault, and unknown fault. Select the fault features with larger weights as the antecedent features of the BRB model. Then, according to the differences in the ranges of single - fault data and compound - fault data, use the method of converting the confidence of compound faults to convert the compound - fault data outside the range into a confidence suitable for the BRB model. Use the evidential reasoning algorithm (ER) to output the posterior confidence of each state. After that, use the IRUN algorithm combined with compound - fault expert knowledge to mine the internal relationship of the single - fault data of diesel engines. Finally, decouple the fault combination form of compound faults through each sub - model. The steps are as follows:

[0056] Step 1: When a diesel engine fails, the process of converting thermal energy into mechanical energy of the diesel engine will be affected, which will be directly reflected in the changes of the thermotechnical parameters of the diesel engine. Common faults of diesel engines include: the fault of decreased inter - cooler efficiency (F1), the fault of crankcase gas leakage (F2), the fault of injection timing (F3), and the fault of reduced fuel supply (F4). Establish a BRB sub - model for each fault. The consequent attributes of each sub - model include three states: healthy state, specific fault, and unknown fault.

[0057] Step 2: Let X i (i = 1, 2, 3, …, M) be the fault feature parameters that can reflect the fault state of the diesel engine, which are called antecedent features. They are derived from the thermotechnical parameter set T of the diesel engine = {t q |q = 1, 2, 3, …, Q}. The elements in the set T include the power data, fuel consumption rate data, cylinder temperature and pressure data, and intake and exhaust temperature and pressure data of the diesel engine, with a total of Q thermotechnical parameters. Select the relevant thermotechnical parameters for each fault as the antecedent features, and establish a separate sub - model in combination with the multi - label classification problem. The multi - label coding method is different from the general classification coding method. It regards compound faults as a combination of multiple faults.

[0058] Step 3: Build a BRB model. The antecedent features of the single - fault data of the diesel engine are used as the input of the BRB model, and the consequent output includes three states of the diesel engine. The set of reference values is D = {D n |n = 1, 2, 3}, where D 1 is the healthy state, indicating that the diesel engine is in a fault - free state; D 2 is the specific fault, indicating that the fault state of the diesel engine at this time covers this known fault; D 3 is the unknown fault state, indicating that the diesel engine has a fault but the fault type cannot be determined.

[0059] The k-th belief rule of the BRB model is as follows:

[0060]

[0061] with rule weight θ k and attribute weights δ i (i = 1, 2, 3, …, M)

[0062] where represents the j-th reference level of the input antecedent feature X M in the k-th rule of the model, β n k represents the confidence of the n-th consequent attribute in the k-th rule, θ k represents the rule weight of the k-th rule, δ i represents the feature weight of the i-th antecedent feature, and M represents the number of antecedent features.

[0063] Step 4: For the constructed belief rule base model, when fault data is input, the BRB model calculates the confidence distribution of each consequent attribute through steps such as confidence conversion, rule activation, and evidence reasoning. The specific process is as follows:

[0064] (41) Calculate the confidence distribution of each input data x i corresponding to the reference level in the k-th rule as α ij+1 = 1 - α ij , (u(A i,j ) < x i < u(A i,j+1 ))), α ij is the confidence of the j-th reference level of the single-fault input data x i , and u(A i,j ) represents the quantization value of the corresponding reference level of A ij . Since the data sample range of single faults cannot fully cover the data sample range of compound faults, a method for establishing the confidence conversion of compound fault data is established for data beyond the reference level of the single-fault data sample, and the confidence distribution is:

[0065] (42) BRB model rule activation. After the fault data is converted into the confidence of the reference level, the corresponding rules need to be activated. The activation weight in the k-th rule is where L represents the number of rules, represents the relative feature weight of the antecedent feature.

[0066] (43) Use the Evidential Reasoning (ER) algorithm for rule reasoning to obtain the posterior confidence distribution of different states of the diesel engine. Among them, β n is the confidence of the posterior attribute of the health state D in the k-th rule. Obtain the β with the maximum confidence. n The corresponding posterior attribute D n is used as the system output of the BRB model. n

[0067] Step 5: Establish an optimization model for the parameters of the BRB model. The optimized BRB can accurately reflect the reasoning process of the system. The optimization goal is to make the mean square error (MSE) between the estimated output and the actual output as small as possible, which can be described as: Among them, S represents the number of optimization samples. is the actual output of the s-th sample data, and β n (s) is the predicted output of the s-th sample data. The parameter to be optimized is Ω = [δ i , θ k , β j k . The parameter range is 0 ≤ δ i ≤ 1 (i = 1, 2,..., M), 0 ≤ θ k ≤ 1 (k = 1, 2,..., L).

[0068] Step 6: Knowledge embedding improvement operation. Use the improved Runge-Kutta algorithm (IRUN) to optimize the model parameters. The improvement method is to combine expert experience knowledge to guide the direction of generating the optimal solution during the optimization process of the RUN algorithm. According to the expert experience knowledge of diesel engine compound faults, in the k-th rule, the most likely health state of the diesel engine is denoted as . At this time, The corresponding confidence is β n k . Then β n k > β j k (j ≠ n, j = 1, 2,…, N). The model parameters optimized by the RUN algorithm cannot meet the expert knowledge of compound faults. It is necessary to add its knowledge embedding operation to increase the adaptability of the model parameters to compound fault data. After initializing the parameters of the RUN algorithm and performing the sampling operation, detect the confidence distribution of the posterior attribute state of the k-th rule. If β n k ≤ β j k (j ≠ n, j = 1, 2,…, N 1 ) where 1 ≤ N 1If ≤ N - 1, then add the expert knowledge constraint of compound fault coupling Then enter the normalization operation Satisfy the equality constraint of the consequent attribute; Detect if β n k > β j k (j ≠ n, j = 1, 2, …, N), then directly enter the normalization operation Finally, perform the screening operation of the RUN algorithm to iteratively calculate the minimum value of the mean square error, and select the optimal model parameter Ω from the population min .

[0069] Figure 2 The improved structure diagram of the RUN algorithm is shown. The RUN algorithm cannot effectively solve the adaptability of single - fault data samples to compound - fault test samples. Therefore, it is necessary to add the operation of embedding compound - fault expert knowledge to the RUN algorithm. The improved RUN algorithm is denoted as the IRUN algorithm. The expert - knowledge embedding operation is to detect the confidence - degree distribution of the consequent attribute of each generated rule. The confidence - degree distribution that does not conflict with the compound - fault expert knowledge directly enters the normalization operation to satisfy the equality constraint. For the confidence - degree distribution that does not conform to the expert knowledge, the following knowledge - embedding operation is performed:

[0070]

[0071] In the formula The confidence degree corresponding to the most likely health state of the diesel engine in the k - th rule in the compound - fault expert knowledge of the diesel engine, β j k is the remaining confidence - degree distribution in this rule is the confidence - degree distribution of this rule after adding the knowledge - embedding operation

[0072] Step 7: Pass the compound - fault test samples through the compound - fault confidence conversion and use the optimal model parameter Ω min Output the confidence - degree distribution β of the consequent attribute of the diesel - engine state of each sub - model n , and determine the compound - fault label according to the fault state with the maximum confidence

[0073] Specifically, for the diesel engine compound fault diagnosis method based on the multi-label IRUN-BRB model in the embodiments of the present application, the single fault data samples of the diesel engine are converted into a set of thermal parameters of the healthy state and fault state of the diesel engine. According to the 4 types of single faults of the diesel engine, corresponding antecedent feature combinations are selected for each fault type in the set of thermal parameters to identify such specific faults, and the corresponding BRB sub-models are established. Each BRB model needs to be activated separately. By converting the single fault data into the confidence of the corresponding fault reference level and activating the rule weight coefficients composed of the confidence reference levels, the ER inference algorithm is used to fuse the activated rules and output the confidence distribution of the consequent attributes of the diesel engine state. The antecedent feature weights, rule weights, and the confidence distribution of the consequent attributes under each rule are used as the parameter vector to be optimized of the BRB model. The mean square error between the diesel engine state predicted by the model and the actual operating state of the diesel engine is used as the objective function, and parameter feasible region constraints, expert knowledge constraints on compound fault coupling, and normalization constraints on the consequent attributes are added to the parameter vector of the objective function. The optimal parameter vector Ω of the BRB model is iteratively obtained by using the improved RUN algorithm with the embedding of the single fault data and expert knowledge of the diesel engine. min For the compound fault test samples beyond the range of single fault data, a method for converting the compound fault confidence is established, which is converted into the confidence of the corresponding fault reference level, and the optimal model parameter Ω min is used to output the confidence distribution β n of the consequent attributes of the diesel engine state D n of each sub-model, and the compound fault label is determined according to the fault state with the highest confidence.

[0074] Figure 3 This is the fault diagnosis accuracy result of the diesel engine compound fault diagnosis method based on the multi-label IRUN-BRB model in the embodiments of the present application. Through the optimal parameters of the model optimized by the IRUN algorithm, after the calculation of the compound fault test samples with the optimal parameter Ω min , the average recognition accuracy of the diesel engine fault state is 93.2%. Among them, the average accuracy of the F1 fault BRB model is 95.9%, the average accuracy of the F2 fault BRB model is 95.2%, the average accuracy of the F3 fault BRB model is 89.4%, and the average accuracy of the F4 fault BRB model is 92.2%. However, in the F3 fault BRB model, the recognition rate of a specific fault is 81.3%, and the recognition rate of the healthy state in the F3 fault BRB model is 76.5%. This proves that in the case of lacking compound fault data samples, by using the single fault data of the diesel engine and the compound fault expert knowledge, the single fault combination form in the compound fault can be accurately identified through the diesel engine compound fault diagnosis method based on the multi-label IRUN-BRB model.

[0075] The compound fault diagnosis steps using the above fault diagnosis method are as follows:

[0076] Step 1: When a fault occurs in the diesel engine, the process of converting thermal energy into mechanical energy in the diesel engine will be affected, which will be directly reflected in the changes of the diesel engine's thermal parameters. The collected thermal parameters of the diesel engine are: rated power (Pw), fuel consumption rate (Ge), maximum combustion pressure (Pf), maximum pressure rise rate (Rf), mean effective pressure (Pm), intercooler outlet temperature (Tco), average temperature after the exhaust valve (Tc), temperature before the turbine (Ti), and temperature after the turbine (To). Common faults of the diesel engine are: intercooler efficiency decline fault (F1), crankcase blow-by fault (F2), fuel injection timing fault (F3), and fuel supply reduction fault (F4). A BRB sub-model is established for each fault, and the consequent attributes of each sub-model include three states: healthy state, specific fault, and unknown fault.

[0077] Step 2: Let X i (i = 1, 2, 3, …, M) be the fault characteristic parameters that can reflect the fault state of the diesel engine, which are called the antecedent characteristics. They are derived from the diesel engine's thermal parameter set. For fault F1, Tco is selected as the main antecedent characteristic, and Pf is selected as the auxiliary antecedent characteristic; for fault F2, Ge and Tc are selected as the main antecedent characteristics, and Rf is selected as the auxiliary antecedent characteristic; for fault F3, Rf and Pf are selected as the main antecedent characteristics, and Pm is selected as the auxiliary antecedent characteristic; for fault F4, Ti and To are selected as the main antecedent characteristics, and Pw is selected as the auxiliary antecedent characteristic. For example, the reference levels L, M, H of the fuel consumption rate Ge are 190.18 g·kW -1 h -1 , 203.26 g·kW -1 h -1 , 242.64 g·kW -1 h -1 .

[0078] The main characteristics are used to identify the healthy state and specific fault states with three reference levels L, M, H, and the auxiliary characteristics are used to identify the healthy state and other fault states with two reference levels L, H. The quantization values of the reference levels L, H are determined by each fault characteristic of the single-fault data, and the quantization value of the reference level M is determined by the thresholds of the normal state and specific fault state of the fault characteristic. Finally, the multi-label classification model is shown in Table 1 below.

[0079] Table 1 Multi-label classification model table

[0080]

[0081]

[0082] For example, if all of Tag 1, Tag 2, and Tag 3 are 1, it indicates that the diesel engine has a combined fault of F1, F2, and F3 at this time. Among them, each tag corresponds to the output of a BRB model.

[0083] Step 3: Construct a BRB model. The antecedent features of the single-fault data of the diesel engine are used as the input of the BRB model, and the consequent output includes three states of the diesel engine. The set of its reference values is D = {D n |n = 1, 2, 3}, where D 1 is the healthy state, indicating that the diesel engine is in a fault-free state; D 2 is a specific fault, indicating that the fault state occurring at this time of the diesel engine covers this known fault; D 3 is the unknown fault state, indicating that the diesel engine has a fault but the type of the fault cannot be determined. The k-th confidence rule is:

[0084]

[0085] with rule weight θ k and feature weight δ i (i = 1, 2, 3, …, M)

[0086] where represents the j-th reference level of the input antecedent feature X M in the k-th rule of the model, β n k represents the confidence degree of the n-th consequent attribute in the k-th rule, θ k represents the rule weight of the k-th rule, δ i represents the feature weight of the i-th antecedent feature, and M represents the number of antecedent features. Each consequent attribute output by the BRB model has three, namely: D 1 , D 2 , and D 3 , and each consequent attribute corresponds to a confidence degree β n , and the fault state of the diesel engine is predicted according to its magnitude.

[0087] Step 4: For the constructed confidence rule base model, when fault data is input, the BRB model calculates the confidence degree distribution of each consequent attribute through steps such as confidence degree conversion, rule activation, and evidence reasoning. The specific process is as follows:

[0088] (41) Calculate the confidence degree distribution of the single-fault each input data x i corresponding to the reference level in the k-th rule as α ij+1 = 1 - α ij , (u(A i,j ) < x i < u(Ai,j+1 )),α ij is the confidence of the j-th reference level of the single-fault input data x i , and u(A i,j ) represents the quantization value of A ij corresponding to the reference level.

[0089] (42) BRB model rule activation. After the fault data is converted into the confidence of the reference level, the corresponding rules need to be activated. The activation weight in the k-th rule where where L represents the number of rules, represents the relative feature weight of the antecedent feature.

[0090] (43) Use the evidential reasoning (ER) algorithm for rule fusion to obtain the consequent confidence distribution of different states of the diesel engine where, β n is the confidence of the consequent attribute of the health state D n in the k-th rule. Obtain the β n with the maximum confidence, and the corresponding consequent attribute D n is used as the system output of the BRB model.

[0091] It should be noted that if a certain BRB model outputs a specific fault, other BRB models should output unknown faults. Only when all BRB models output the healthy state, the diesel engine is in the healthy state.

[0092] Step 5: Establish an optimization model for the BRB model parameters. The optimized BRB can accurately reflect the inference process of the system. The optimization goal is to make the mean square error (MSE) between the estimated output and the actual output as small as possible, which can be described as: where S represents the number of optimization samples, is the actual output of the s-th sample data, and β n (s) is the predicted output of the s-th sample data. The parameter to be optimized Ω = [δ i , θ k , β j k , and the parameter range is 0 ≤ δ i ≤ 1 (i = 1, 2,..., M), 0 ≤ θ k ≤ 1 (k = 1, 2,..., L).

[0093] Step 6: Knowledge embedding improvement operation. Use the improved Runge-Kutta algorithm (IRUN) to optimize the model parameters. The improvement method is to combine expert experience knowledge to guide the direction of generating the optimal solution during the optimization process of the RUN algorithm. According to the expert experience knowledge of diesel engine compound faults, in the k-th rule, the most likely health state of the diesel engine is recorded as At this time The corresponding confidence level is β n k , then β n k >β j k (j≠n, j = 1, 2, …, N). The model parameters optimized by the RUN algorithm cannot meet the compound fault expert knowledge. It is necessary to increase its knowledge embedding operation to increase the adaptability of the model parameters to the compound fault data. After initializing the parameters of the RUN algorithm and performing the sampling operation, detect the confidence level distribution of the consequent attribute state of the k-th rule. If β n k ≤β j k (j≠n, j = 1, 2, …, N 1 ) where 1 ≤ N 1 ≤N - 1, then add the expert knowledge constraint of compound fault coupling Then enter the normalization operation Satisfy the equality constraint of the consequent attribute; detect if β n k >β j k (j≠n, j = 1, 2, …, N), then directly enter the normalization operation Finally, perform the screening operation of the RUN algorithm to iteratively calculate the minimum value of the mean square error, and select the optimal model parameter Ω min . The fault models of 4 BRBs are shown in Tables 1, 2, 3 and 4.

[0094] Table 1 Optimal parameter table of F1 fault model

[0095]

[0096]

[0097] Table 2 Optimal parameter table of F2 fault model

[0098] Number <![CDATA[θ k > Ge(0.7) Ti(1) Rf(0.7) <![CDATA[{NF(β 1 ),F2(β 2 ),UF(β 3 )}]]> 1 1 L L L {0,0,1} 2 1 L M L {0,0,1} 3 0.716 L H L {0,0.501,0.499} 4 1 M M L {0.001,0.002,0.997} 5 0.702 M H L {0.001,0.499,0.5} 6 0.939 H M L {0.064,0.83,0.106} 7 1 H H L {0,1,0} 8 1 L L H {0,0.02,0.998} 9 0.7 L M H {0.866,0.016,0.118} 10 0.706 L H H {0.414,0.416,0.17} 11 0.7 M M H {0.501,0.499,0} 12 1 M H H {0,1,0} 13 1 H M H {0,1,0} 14 1 H H H {0,1,0}

[0099] Table 3 Optimal parameter table of F3 fault model

[0100]

[0101]

[0102] Table 4 Optimal Parameter Table of F4 Fault Model

[0103] Number <![CDATA[θ k > Ti(1) To(0.7) Pw(0.7) <![CDATA[{NF(β 1 ),F4(β 2 ),UF(β 3 )}]]> 1 0.951 L L L {0,1,0} 2 0.841 L M L {0.009,0.623,0.368} 3 1 M L L {0.,0.834,0.166} 4 0.7 M M L {0.09,0.737,0.173} 5 0.731 M H L {0,0.091,0.909} 6 0.700 H M L {0,0.124,0.876} 7 1 H H L {0.116 0.002,0.882} 8 1 L L H {0.004,0.994,0.002} 9 0.7 L M H {0.028,0.760,0.212} 10 0.7 M L H {0.205,0.577,0.218} 11 0.7 M M H {0.857,0.001,0.142} 12 1 M H H {0,0.21,0.79} 13 0.775 H M H {0,0.219,0.781} 14 1 H H H {0,0,1}

[0104] Step 7: Input the compound fault test samples into the 4 fault sub-models. For the compound fault test samples that exceed the data sample range of single faults, use the method of converting the confidence of compound fault data, and the confidence distribution is as follows: After the confidence conversion of the compound fault and using the optimal model parameter Ω min Output the consequent attribute confidence distribution β of the diesel engine state of each sub-model n , and determine the compound fault label according to the fault state with the maximum confidence.

[0105] In addition, combined with Figure 1 The multi-label-based IRUN-BRB diesel engine compound fault diagnosis method described in the embodiments of the present application can all be implemented by a computer device. Figure 4 It is a schematic diagram of the hardware structure of the computer device according to the embodiment of the present application. As Figure 4 shown, the device may include a processor 201 and a memory 202 storing computer program instructions.

[0106] Specifically, the above-mentioned processor 201 may include a central processing unit (CPU), or an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present application.

[0107] Among them, the memory 202 may include a mass storage for data or instructions. By way of example and not limitation, the memory 202 may include a hard disk drive (HDD), a floppy disk drive, a solid state drive (SSD), a flash memory, an optical disc, a magneto-optical disc, a magnetic tape, or a universal serial bus (USB) drive, or a combination of two or more of these. In a suitable case, the memory 202 may include a removable or non-removable (or fixed) medium. In a suitable case, the memory 202 may be internal or external to the data processing device. In a specific embodiment, the memory 202 is a non-volatile memory. In a specific embodiment, the memory 202 includes a read-only memory (ROM) and a random access memory (RAM). In a suitable case, the ROM may be a mask-programmed ROM, a programmable ROM (PROM), an erasable PROM (EPROM), an electrically erasable PROM (EEPROM), an electrically alterable ROM (EAROM), or a flash memory (FLASH), or a combination of two or more of these. In a suitable case, the RAM may be a static random access memory (SRAM) or a dynamic random access memory (DRAM), where the DRAM may be a fast page mode dynamic random access memory (FPMDRAM), an extended data out dynamic random access memory (EDODRAM), a synchronous dynamic random access memory (SDRAM), etc.

[0108] The memory 202 can be used to store or cache various data files required for processing and / or communication, as well as possible computer program instructions executed by the processor 201.

[0109] By reading and executing the computer program instructions stored in the memory 202, the processor 201 implements any one of the above-described multi-label-based IRUN-BRB diesel engine compound fault diagnosis methods in the embodiments.

[0110] In some embodiments, the point cloud generation device may further include a communication interface 203 and a bus 200. Among them, as Figure 4 shown, the processor 201, the memory 202, and the communication interface 203 are connected through the bus 200 and complete communication with each other.

[0111] The communication interface 203 is used to implement communication between various modules, devices, units, and / or devices in the embodiments of the present application. The communication interface 203 can also implement data communication with other components such as external devices, image / data acquisition devices, databases, external storage, and image / data processing workstations.

[0112] The bus 200 includes hardware, software, or both, and couples the components of the point cloud generation device to each other. The bus 200 includes, but is not limited to, at least one of the following: Data Bus, Address Bus, Control Bus, Expansion Bus, Local Bus. By way of example and not limitation, the bus 200 may include an Accelerated Graphics Port (AGP) or other graphics bus, an Extended Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), a Hyper Transport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an InfiniBand interconnect, a Low Pin Count (LPC) bus, a memory bus, a MicroChannel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local Bus (VLB) bus, or other suitable buses, or a combination of two or more of these. In suitable cases, the bus 200 may include one or more buses. Although the embodiments of the present application describe and illustrate specific buses, the present application contemplates any suitable bus or interconnect.

[0113] The computer device can execute the multi-label based IRUN-BRB diesel engine compound fault diagnosis method in the embodiments of the present application, so as to implement the combination Figure 1 of the described multi-label based IRUN-BRB diesel engine compound fault diagnosis method.

[0114] In addition, in combination with the multi-label based IRUN-BRB diesel engine compound fault diagnosis method in the above embodiments, an embodiment of the present application can provide a computer-readable storage medium to implement. Computer program instructions are stored on the computer-readable storage medium; when the computer program instructions are executed by a processor, any one of the multi-label based IRUN-BRB diesel engine compound fault diagnosis methods in the above embodiments is implemented.

[0115] It should be noted that the technical features of the above-described embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification. In addition, according to the needs of implementation, each step / component described in the present application can be split into more steps / components, or two or more steps / components or partial operations of steps / components can be combined into new steps / components to achieve the purpose of the present invention.

[0116] Those skilled in the art can easily understand that the above-described embodiments only represent several implementation manners of the present application, and the description is relatively specific and detailed, but it should not be construed as a limitation on the scope of the invention patent. It should be pointed out that for those of ordinary skill in the art, without departing from the concept of the present application, several deformations and improvements can still be made, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application should be subject to the appended claims.

Claims

1. A multi-label based IRUN-BRB diesel engine composite fault diagnosis method, characterized in that: The method includes: Determine multiple single fault types of diesel engines and obtain corresponding single fault data samples; For various single fault types of diesel engines, multiple BRB models are established. Each BRB model corresponds to a single fault type. The consequent attributes of the BRB model include health status, the single fault type, and other single fault types. Determine a preceding feature combination of each BRB model, where the preceding feature combination is a combination of fault feature parameters corresponding to a single fault type of the BRB model; Based on the single fault data sample, the antecedent feature combination of all BRB models is determined and input into the corresponding BRB model. The confidence distribution of the consequent attribute of each BRB model is calculated through confidence conversion, rule activation and evidence reasoning. The consequent attribute with the largest confidence is taken as the output of the BRB model. Based on the improved Runge-Kutta algorithm IRUN, all BRB models are optimized so that the outputs of all BRB models conform to the single fault type of the single fault data sample; The composite fault samples are obtained, the corresponding antecedent feature combinations are determined and input into all optimized BRB models, and the composite faults are determined based on the output combinations of all optimized BRB models.

2. The IRUN-BRB diesel engine composite fault diagnosis method based on multiple labels according to claim 1 is characterized in that: The various single fault types of diesel engines include intercooler efficiency reduction fault F1, crankcase blowby fault F2, injection timing fault F3, and fuel supply reduction fault F4; Among them, the antecedent feature combination of the intercooler efficiency reduction fault F1 is the intercooler outlet temperature Tco and the maximum combustion pressure Pf, the intercooler outlet temperature Tco is the main antecedent feature, and the maximum combustion pressure Pf is the auxiliary antecedent feature; The antecedent feature combination of crankcase blowby fault F2 is fuel consumption rate Ge, average temperature after exhaust valve Tc and maximum pressure rise rate Rf. Fuel consumption rate Ge and average temperature after exhaust valve Tc are the main antecedent features, and maximum pressure rise rate Rf is the auxiliary antecedent feature. The antecedent feature combination of the injection timing fault F3 is the maximum pressure rise rate Rf, the maximum combustion pressure Pf and the mean effective pressure Pm. The maximum pressure rise rate Rf and the maximum combustion pressure Pf are the main antecedent features, and the mean effective pressure Pm is the auxiliary antecedent feature. The antecedent feature combination of the oil supply reduction fault F4 is the turbine inlet temperature Ti, the turbine outlet temperature To and the rated power Pw. The turbine inlet temperature Ti and the turbine outlet temperature To are the main antecedent features, and the rated power Pw is the auxiliary antecedent feature. The main antecedent feature is used to identify the healthy state and the single fault type, and has three reference levels L, M and H; the auxiliary antecedent feature is used to identify the healthy state and other single fault types, and has two reference levels L and H.

3. The IRUN-BRB diesel engine composite fault diagnosis method based on multiple labels according to claim 1 is characterized in that: The healthy state indicates that the diesel engine is in a fault-free state. This single fault type indicates that the fault state of the diesel engine at this time covers the single fault type corresponding to the corresponding BRB model. Other single fault types indicate that the diesel engine has a fault but the fault type cannot be determined.

4. The IRUN-BRB diesel engine composite fault diagnosis method based on multiple labels according to claim 1 is characterized in that: The kth confidence rule R of the BRB model k for: with rule weight θ k and feature weight δ i (i=1,2,3,…,M) in, Indicates that the antecedent feature X is input into the kth rule of the BRB model M The jth reference level of n k Indicates the nth consequent attribute D in the kth rule n , N represents the total number of consequent attributes, θ k represents the rule weight of the kth rule, δ i represents the feature weight of the i-th antecedent feature, and M represents the number of antecedent features; Each BRB model outputs N confidence scores β n , each consequent attribute corresponds to a confidence β n , according to the confidence level β n Size predicts the fault status of diesel engines.

5. The IRUN-BRB diesel engine composite fault diagnosis method based on multiple labels according to claim 1 is characterized in that: The confidence distribution of the consequent attributes of each BRB model is calculated through confidence conversion, rule activation and evidence reasoning, including: (1) Calculate the single fault for each input data x i The confidence distribution corresponding to the reference level in the kth rule is α ij+1 =1-α ij ,(u(A i,j ) <x i <u(A i,j+1 )), A ij Represents the antecedent feature X i The jth reference level of ij For single fault data x i The confidence level of the j-th reference level, u(A i,j ) means A ij Quantitative value corresponding to the reference level; (2) Calculate the activation weight in the kth rule in M represents the number of antecedent features, L represents the number of rules, represents the relative feature weight of the antecedent feature, represents the feature weight of the i-th antecedent feature in the k-th rule, θ k represents the rule weight of the kth rule; (3) Using the evidence reasoning ER algorithm to perform rule fusion, we can obtain the confidence distribution of the consequent attributes of different states of the diesel engine. in, β n is the nth consequent attribute D in the kth rule n Confidence level; Get the β with the highest confidence n The corresponding consequent attribute D n As the output of the BRB model.

6. The IRUN-BRB diesel engine composite fault diagnosis method based on multiple labels according to claim 5 is characterized in that: Based on the improved Runge-Kutta algorithm, IRUN optimizes all BRB models, including: In the kth rule, if the diesel engine is in a healthy state, the corresponding confidence is β n k , then β n k >β j k (j≠n,j=1,2,…,N) as the equality constraint of the consequent attribute; where β j k (j≠n,j=1,2,…,N) is the remaining confidence distribution in this rule; Initialize the parameters of the Runge-Kutta algorithm RUN and perform sampling operations, then detect the confidence distribution of the consequent attribute of the kth rule: if β n k ≤β j k (j≠n,j=1,2,…,N1), where 1≤N1≤N-1, then add the expert knowledge constraint β of compound fault coupling j k =β j k ×β n k (j≠n,j=1,2,…,N1), then enter the normalization operation Satisfy the equality constraints of the consequent attributes; where β jI k The confidence distribution of the rule after adding the knowledge embedding operation; if β n k >β j k (j≠n,j=1,2,…,N), then directly enter the normalization operation Finally, the Runge-Kutta algorithm RUN screening operation is performed to iteratively calculate the optimal value of the objective function and select the optimal model parameter Ω from the population. min .

7. The multi-label based IRUN-BRB diesel engine composite fault diagnosis method according to claim 6, characterized in that: The objective function is as follows: Among them, S represents the number of samples, is the actual output of the sth sample, β n (s) is the predicted output of the sth sample, and the model parameter to be optimized Ω=[δ i ,θ k , β j k ], the parameter range is 0≤δ i ≤1(i=1,2,...,M),0≤θ k ≤1(k=1,2,...,L), 8. The IRUN-BRB diesel engine composite fault diagnosis method based on multiple labels according to claim 5 is characterized in that: For compound fault test samples that exceed the range of single fault data samples, the confidence conversion method of compound fault data is used, and the confidence distribution is: After the composite fault confidence conversion, all optimized BRB models are used to output the consequent attribute confidence distribution β n , determine the composite fault label according to the fault state with the highest confidence.

9. An electronic device, characterized in that: include: A processor and a memory, the memory storing programs or instructions that can be run on the processor, and when the programs or instructions are executed by the processor, the steps of the IRUN-BRB diesel engine composite fault diagnosis method based on multiple tags as described in any one of claims 1 to 8 are implemented.

10. A readable storage medium, characterized in that: Programs or instructions are stored thereon, and when the programs or instructions are executed by the processor, the steps of the IRUN-BRB diesel engine composite fault diagnosis method based on multiple tags as described in any one of claims 1 to 8 are implemented.

Images