Air conditioning system fault diagnosis method and device based on parameter correlation Bayesian network

Abstract

The invention provides an air conditioning system fault diagnosis method and device based on a parameter correlation Bayesian network, and the method comprises the steps: rapidly determining the real-time numerical values of key performance parameters of parts through the correlation between the material parameters of the parts, comparing the calibration numerical values of the parts when the parts leave a factory, and determining the real-time fault rates of various faults of the parts; and in combination with a Bayesian network model of the air conditioning system, calculating the real-time fault rate and the real-time posterior probability of each stage of sub-node in sequence, and judging whether the sub-node is a fault node or not by comparing the difference between the statistical posterior probability and the real-time posterior probability. According to the method, in real-time data measurement and processing, measurement is convenient, and operability is high; according to the method for judging the faults through the Bayesian network and the posterior probability, the diagnosis process is more flexible, fault positioning is more accurate, the method can be suitable for fault diagnosis of various types of air conditioning systems through slight adjustment, and high practical generalization is achieved.

Description

Technical Field

[0001] The invention relates to an air conditioning system fault diagnosis method and device based on a parameter-related Bayesian network, belonging to the technical field of reliability analysis. Background Art

[0002] Due to the harsh working conditions of construction machinery during operation, the external stresses on the air conditioning system are complex and changeable, and there are multiple failure modes. However, due to the complex structure and numerous equipment of the air conditioning system, it is difficult to locate the fault in the fault diagnosis when the system fails, which delays the maintenance progress, affects the construction period, and even causes safety problems. Therefore, it is of great significance to carry out research on fault detection and diagnosis methods of air conditioning systems.

[0003] Different types of construction machinery vehicles have different working conditions, which results in different external stresses on the air-conditioning system, leading to various failure modes. In addition, the fault diagnosis model has poor portability between air-conditioning systems of different models, making it difficult to promote and apply. On the other hand, due to the correlation between component performance parameters and the randomness of component performance degradation, when a system failure occurs, it is often due to the joint influence of multiple components, making it difficult to accurately locate the weak component. At present, there is a lack of a fault diagnosis model that takes into account the random degradation process and parameter correlation, can be quickly and portablely applied to different air-conditioning systems, and obtains accurate fault diagnosis results.

[0004] The main existing technical solutions are as follows:

[0005] 1. Troubleshooting method:

[0006] Troubleshooting is to gradually eliminate factors that may cause system failures, gradually narrow the scope of the failure, and finally find and repair the failure. Its basic process is to confirm the fault problem and collect information related to the fault, then perform data analysis and problem troubleshooting, and then verify the problem through experiments.

[0007] However, the troubleshooting method requires a high level of experience from the troubleshooter, and the process from information collection and analysis to troubleshooting and experimental verification is relatively lengthy, resulting in low troubleshooting efficiency and high maintenance costs. It is less feasible for troubleshooting and locating more complex systems.

[0008] 2. Artificial Intelligence Technology:

[0009] Artificial intelligence technology is applied to system fault diagnosis and location, and commonly used technologies are machine learning and data mining. It learns and identifies fault modes from a large amount of fault data, analyzes and models system fault data, and realizes intelligent diagnosis of potential faults and prediction of equipment life.

[0010] However, when artificial intelligence technology is applied to fault diagnosis, the diagnostic effect is affected by the amount of training sample data, so the data volume requirement is relatively high. It is difficult to obtain effective fault diagnosis results for sample data with small amount of data and strong randomness.

[0011] Real-time monitoring technology

[0012] By monitoring the system operating status in real time and based on the system's preset rules or thresholds, timely alarms are issued when abnormal data in equipment operation is found. Combined with machine learning algorithms and models, the real-time collected data is analyzed and diagnosed to identify possible causes of equipment failures and help technicians take timely countermeasures.

[0013] However, real-time monitoring technology is suitable for equipment status monitoring and fault diagnosis in fixed scenarios. When applied to outdoor equipment with changeable working conditions and harsh environments, the reliability of the monitoring system is difficult to guarantee, the application cost is high, and it may also lead to false status reports. Summary of the invention

[0014] The present invention belongs to the technical field of reliability analysis. Specifically, in view of the problem that in the fault diagnosis and analysis of engineering machinery air-conditioning systems, the system structure is complex and changeable, and the components have random degradation characteristics, which leads to complex system fault reasoning and analysis processes and difficulty in fault location, a method and device for air-conditioning system fault diagnosis based on parameter-dependent Bayesian networks are proposed to obtain good fault diagnosis results, and the method and device can be quickly transplanted and applied in air-conditioning systems with different structures.

[0015] To achieve the above object, the present invention is implemented by adopting the following technical solutions:

[0016] In a first aspect, the present invention provides a Bayesian network fault diagnosis method based on parameter correlation, comprising the following steps:

[0017] Step 1: Determine the fault type of the air conditioning system of the construction machinery. The same fault type corresponds to the same fault event in the fault tree. The fault event is used as the bottom event of the fault tree model to build a fault tree model of the air conditioning system.

[0018] Step 2: Mapping the fault tree model into a Bayesian network;

[0019] Step 3: Combined with historical fault data, calculate the probability of occurrence of the bottom event of the fault tree model, and use it as the prior probability of the leaf node of the Bayesian network corresponding to the bottom event of the fault tree model, and obtain the prior probability of each leaf node of the Bayesian network;

[0020] Step 4: According to the Bayesian network mapped by the fault tree model and the prior probability of each leaf node of the Bayesian network, the posterior probability of each level of sub-nodes in the Bayesian network is calculated;

[0021] Step 5: Obtain a regression model between component performance related parameters; the input of the regression model between component performance related parameters is the component performance related parameters that are easy to measure, and the output is the real-time key performance parameters of the component;

[0022] Step 6: Measure the real-time data of the performance-related parameters of the component, and combine the regression model between the performance-related parameters of the component to obtain the real-time data of the real-time key performance parameters of the component;

[0023] Step 7: Obtain a real-time failure rate assessment model for various faults of a component, wherein the input variables of the real-time failure rate assessment model are the factory calibration values ​​of the key performance parameters of the component, the real-time predicted values ​​of the key performance parameters of the component output by the relevant parameter regression model, and the market statistical failure rate of the component, and the output variable is the real-time failure rate of the component under the corresponding failure mode;

[0024] Step 8: Obtain the factory calibration values ​​of the key performance parameters of the components and the market statistical failure rate of the components, and combine the real-time data of the real-time key performance parameters of the components output by the regression model to calculate the real-time failure rate of the sub-nodes at each level of the Bayesian network, and use the real-time failure rate of the sub-nodes at each level as the real-time prior probability to calculate the real-time posterior probability of the sub-nodes at each level in the Bayesian network;

[0025] Step 9: Compare the real-time posterior probability and the posterior probability of the sub-nodes at each level in the Bayesian network, diagnose the fault status of the sub-nodes at each level, and obtain the fault diagnosis result;

[0026] Step 10: Output the fault diagnosis results.

[0027] Furthermore, the engineering machinery is any one of a mining car, a loader, an excavator, and a crane.

[0028] Further, step 1: determine the fault type of the air conditioning system of the construction machinery, the same fault type corresponds to the same fault event in the fault tree, the fault event is used as the bottom event of the fault tree model, and the fault tree model of the air conditioning system is constructed, including:

[0029] In the fault tree model, the failure of the air-conditioning system is taken as the top event, the failures corresponding to each subsystem are taken as the first-level intermediate events, the failures corresponding to the components are taken as the second-level intermediate events, and the failures corresponding to the parts are taken as the third-level intermediate events. All failure events constitute the bottom event of the fault tree model.

[0030] Further, step 2: mapping the fault tree model into a Bayesian network, including:

[0031] When the fault tree model is converted into a Bayesian network, the structure of the fault tree model corresponds to the Bayesian network, and the fault events at all levels in the fault tree model correspond to the nodes at all levels in the Bayesian network: the bottom event of the fault tree model corresponds to the leaf node of the Bayesian network, and the first-level intermediate event, the second-level intermediate event, and the third-level intermediate event in the fault tree model correspond to the first-level sub-node, the second-level sub-node, and the third-level sub-node of the Bayesian network respectively;

[0032] The connections between fault events in the fault tree model correspond to the edges of the Bayesian network. The logic gates that connect events together in the fault tree model are replaced by the conditional probability table of the Bayesian network. The prior probability of the leaf nodes in the Bayesian network corresponds to the probability of occurrence of the bottom event in the fault tree model.

[0033] Further, step 3: Combined with the historical fault data, the probability of occurrence of the bottom event of the fault tree model is calculated, and it is used as the prior probability of the leaf node of the Bayesian network corresponding to the bottom event of the fault tree model, and the prior probability of each leaf node of the Bayesian network is obtained, including:

[0034] Combined with historical fault data, the probability of occurrence of the bottom event of the fault tree model is calculated and used as the prior probability of the corresponding Bayesian network leaf node.

[0035] Among them, the historical fault data includes all fault types in the air conditioning system and the frequency of occurrence of each fault type;

[0036] The probability of occurrence of the bottom event of the fault tree model is the probability of occurrence of various types of faults obtained after statistical analysis of the fault data, and it is used as the prior probability of the corresponding Bayesian network leaf node.

[0037] Further, step 4: according to the Bayesian network mapped by the fault tree model and the prior probability of each leaf node of the Bayesian network, the posterior probability of each level of sub-nodes in the Bayesian network is calculated, including:

[0038] In Bayesian networks, conditional probabilities are used to express the logical relationships between nodes. The logic gates in the fault tree model correspond to the conditional probability tables in the Bayesian network.

[0039] If the node Connect n child nodes , by express The probability of a node failing is expressed as:

[0040] (1)

[0041] Then the node When this happens, the i-th child node The formula for calculating the posterior probability of occurrence is expressed as:

[0042] (2)

[0043] in, express The i-th child node connected to the node The prior probability of , n is the total number of child nodes of node B; Represents a child node Node in case of The conditional probability of occurrence, Indicates that node B occurs when node The conditional probability of occurrence, that is, the i-th child node The posterior probability of occurrence;

[0044] n child nodes and The nodes correspond to leaf nodes and third-level child nodes respectively; , n is the total number of child nodes of node B; the posterior probability of all leaf nodes and the probability of occurrence of third-level child nodes are calculated;

[0045] The third-level child node As a higher level subnode The bottom event, if the secondary child node Connect m third-level child nodes ,by Representation Node The probability of failure, then the above steps can be used to obtain the node When this happens, the jth child node The posterior probability of occurrence, where m child nodes and nodes Corresponding to the third-level sub-nodes and the second-level sub-nodes respectively. Calculate the posterior probability of all third-level sub-nodes and the probability of the occurrence of the second-level sub-nodes;

[0046] The second-level child node C is used as the bottom event of the higher first-level child node, and the posterior probabilities of all leaf nodes, third-level child nodes, second-level child nodes and first-level child nodes of the compressor system are obtained in sequence through the above steps.

[0047] Further, step 5: obtaining a regression model between parameters related to component performance, including:

[0048] Obtain test data on component performance parameters and establish a performance parameter database;

[0049] Based on the performance parameter database, the parameter pairs with correlation are analyzed to obtain the corresponding relationship between the easily measurable component performance-related parameters and the component material key performance parameters, and obtain the correlated parameter pairs consisting of the easily measurable component performance-related parameters and the component material key performance parameters;

[0050] Based on the test data of the relevant parameter pairs consisting of the easily measurable component performance related parameters and the key performance parameters of the component materials, the regression model between the relevant parameters is established by using the support vector machine regression analysis:

[0051]

[0052] Among them, the variable The parameters related to component performance that are easy to measure are used as model input variables. Key performance parameters of component materials corresponding to easily measurable parameters related to component performance are used as model output variables.

[0053] Furthermore, the relevant parameter pairs consisting of easily measurable component performance related parameters and component material key performance parameters include:

[0054] There are usually multiple related parameter pairs for the same material. According to the actual problem requirements, the degradation data of the related parameter pairs in the performance parameter database are called to establish the corresponding regression model. The regression models established by different related parameter pairs of the same material or different materials are independent of each other.

[0055] The performance-related parameter pairs of commonly used materials for parts are shown in Table 4 below:

[0056] Table 4: Summary of parameters related to compressor parts materials

[0057] Material Related parameters Metals (steel, copper, aluminum) Hardness-tensile strength, toughness-shear modulus, temperature-conductivity Plastic, rubber Shear modulus-toughness Oil Viscosity-Specific Heat, Density-Thermal Conductivity

[0058] Further, step 6: measuring the real-time data of the performance-related parameters of the component, combining the regression model between the performance-related parameters of the component, and obtaining the real-time data of the real-time key performance parameters of the component, including:

[0059] For the key performance parameters of components that are difficult to measure, the corresponding regression model between the relevant parameters is obtained through the correspondence between the key performance parameters of the component performance that are easy to measure and the key performance parameters of the component materials;

[0060] By real-time monitoring of the easily measurable material parameters of the components, real-time data of parameters related to component performance are obtained;

[0061] Based on the real-time data of component performance-related parameters, the relevant parameter regression model is used to calculate and obtain the real-time data of the key performance parameters of the components.

[0062] Further, step 7: obtaining a real-time failure rate assessment model for various component failures, including:

[0063] The prior probability of determining various failure modes of components is the statistical result of historical failure data, covering various operating conditions of all products throughout their life cycle, which is the expected value of the failure rate.

[0064] Using the real-time data of key performance parameters of components and combining them with their factory calibration values, a real-time failure rate evaluation model is established to calculate the real-time failure rates of various component failures.

[0065] If the factory calibration value of the key performance parameter corresponding to a certain failure mode of a component and the real-time prediction value under the implementation condition are and The market statistical failure rate of this failure mode is , then the real-time failure rate The evaluation model is expressed as:

[0066] (3)

[0067] The input variables of the real-time failure rate assessment model are the factory calibration values ​​of the key performance parameters of the components, the real-time prediction values ​​of the key performance parameters of the components output by the relevant parameter regression model, and the market statistical failure rate of the components. The output variable is the real-time failure rate of the components under the corresponding failure mode.

[0068] Further, step 8: obtaining the factory calibration values ​​of the key performance parameters of the components and the market statistical failure rate of the components, combining the real-time data of the real-time key performance parameters of the components output by the regression model, calculating the real-time failure rate of the sub-nodes at each level of the Bayesian network, and using the real-time failure rate of the sub-nodes at each level as the real-time prior probability, calculating the real-time posterior probability of the sub-nodes at each level in the Bayesian network, including:

[0069] Obtain the factory calibration values ​​of key performance parameters of components and the market statistical failure rates of components;

[0070] According to the factory calibration values ​​of the key performance parameters of the components and the market statistical failure rates of the components, combined with the real-time data of the real-time key performance parameters of the components output by the regression model, based on the real-time failure rate evaluation model, the real-time failure rates of various faults of the components are calculated, that is, the real-time failure rates of the leaf nodes in the Bayesian network.

[0071] Real-time failure rate of each component failure mode , according to the logical relationship between each node of the Bayesian network, determine the real-time failure rate of components :

[0072] by Indicates the real-time failure rate of components , Represents the real-time failure rate of the kth failure mode ,but:

[0073] (4)

[0074] in Indicates that when the kth failure mode occurs in real time, the component ( )The probability of failure occurring in real time.

[0075] Combining the multiple faults of components and the real-time failure rate of components, the real-time posterior probability of multiple failure modes of components is calculated using the posterior probability calculation formula, that is, the real-time posterior probability of the leaf node is obtained. Among the l failure modes, the real-time posterior probability of the kth failure mode is The calculation formula is:.

[0076] (5)

[0077] in, Indicates parts ( ) Under the premise of real-time failure, the probability of the kth failure mode occurring in real time among l failure modes;

[0078] Based on the real-time failure rate of components The logical relationship between each node of the Bayesian network determines the real-time failure rate of the component , and the real-time posterior probability of the parts, which corresponds to the real-time posterior probability of the third-level child nodes.

[0079] By analogy, the real-time failure rate and real-time a posteriori probability of the second-level sub-nodes and the real-time failure rate and real-time a posteriori probability of the first-level sub-nodes are calculated to obtain the real-time failure rate and real-time a posteriori probability of sub-nodes at all levels for application in system fault diagnosis and location.

[0080] Further, step 9: compare the real-time posterior probability and the posterior probability of the sub-nodes at each level in the Bayesian network, diagnose the fault status of the sub-nodes at each level, and obtain the fault diagnosis results, including:

[0081] According to the obtained real-time posterior probability of failure of each sub-node , by comparing the real-time posterior probability level by level and the posterior probabilities of the child nodes at each level in the Bayesian network The relative value of , calculates the abnormal degree of failure rate δ:

[0082] (6)

[0083] According to the abnormal degree of failure rate δ and abnormal failure rate threshold of nodes at each level To determine the node where the fault occurs, Indicates the abnormal failure rate threshold of the j-th fault of the i-th level child node.

[0084] Furthermore, according to the abnormal degree of failure rate δ and abnormal failure rate threshold of nodes at each level To determine the node where the fault occurred, including:

[0085] When the system fails When an event occurs, start troubleshooting from the next most serious event, and then from the system component failure with the highest statistical failure rate. Start troubleshooting by comparing the abnormality of the relative difference between the real-time a posteriori failure rate of the system component failure and the statistical a posteriori failure rate. If the abnormal degree of the failure rate of the sub-node is greater than the abnormal failure rate threshold, the sub-node is determined to be a faulty node.

[0086] If this system component If it is a non-faulty node, check the low-fault nodes of the same level in order of real-time fault rate from high to low;

[0087] If this system component is a faulty node, then its child nodes and the next-level child nodes of its child nodes are checked in sequence according to the node level of the Bayesian network.

[0088] Further, step 10: outputting the fault diagnosis result, including:

[0089] Obtain the fault judgment analysis results of the sub-nodes at all levels of the system. The output results are presented in the form of statistical tables or logical structure diagrams, including the historical failure rate, real-time failure rate, failure rate abnormal threshold reference value, and fault judgment result information of the nodes at all levels of the system, and form an output report and output it.

[0090] In a second aspect, the present invention provides a Bayesian network fault diagnosis device based on parameter correlation, comprising:

[0091] Memory, for storing computer programs / instructions;

[0092] A processor is used to execute the computer program / instructions to implement the steps of the method described in the first aspect.

[0093] In a third aspect, the present invention provides a computer-readable storage medium having a computer program / instruction stored thereon, which, when executed by a processor, implements the steps of the method described in the first aspect.

[0094] Compared with the prior art, the present invention has the following beneficial effects:

[0095] (1) Compared with the troubleshooting method, the present invention uses Bayesian network for troubleshooting, which is highly operable and efficient. By using the statistical analysis results of historical fault data as the prior probability of each sub-node of the system, it is easy to obtain preliminary information on relatively weak components and nodes, which does not require high experience of maintenance personnel and has strong operability; combined with the troubleshooting rules, through the parameter-related regression model and the component status evaluation model, combined with the Bayesian posterior probability formula, the fault location can be quickly determined with high efficiency.

[0096] (2) Compared with artificial intelligence technology, the present invention uses historical fault data as the benchmark value, real-time measurement data of key performance parameters of components as the application value of fault diagnosis, introduces the concept of real-time failure rate, and combines the key importance of components to make fault judgments, which has good flexibility. It has good generalization in dealing with system fault diagnosis problems with strong randomness in the degradation process and harsh and changeable working conditions.

[0097] (3) Compared with real-time monitoring technology, the present invention quickly predicts and applies key performance parameters of components based on parameter correlation, achieves good equipment status monitoring effect at a lower monitoring cost, and has good operability in harsh outdoor environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0098] Figure 1 It is a technical solution flow chart;

[0099] Figure 2 This is the Bayesian network topology diagram of the air conditioning system;

[0100] Figure 3 This is the fault tree model diagram of the air conditioning compressor system. DETAILED DESCRIPTION

[0101] The present invention will be further described below in conjunction with the accompanying 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.

[0102] Embodiment 1:

[0103] This embodiment proposes a method for air conditioning system fault diagnosis based on parameter-dependent Bayesian network. Taking the fault diagnosis of the air conditioning system of a mining machine as an example, the following technical solution steps are adopted to implement it. The technical solution is as follows: Figure 1 As shown:

[0104] (1) Using the correlation between component material parameters, quickly determine the real-time values ​​of the key performance parameters of the components, and compare them with the calibration values ​​of the components when they leave the factory. Combined with the real-time performance status evaluation model, determine the real-time failure rate of various component failures;

[0105] (2) Combining the real-time failure rates of various component failures with the Bayesian network model of the air conditioning system, the real-time failure rates of sub-nodes at each level are calculated in turn;

[0106] (3) According to the real-time failure rate of each sub-node, the real-time posterior probability is calculated, and by comparing the difference between the statistical posterior probability and the real-time posterior probability, the abnormal threshold is used to determine whether it is a faulty node.

[0107] Specifically, the method of this embodiment includes the following steps:

[0108] Step 1: Determine the fault type of the mine car air conditioning system and build a fault tree model;

[0109] Specifically, the failure of the air-conditioning system is taken as the top event, the failures corresponding to each subsystem are taken as the first-level intermediate events, the failures corresponding to the components are taken as the second-level intermediate events, and the failures corresponding to the parts are taken as the third-level intermediate events. All the above failure events constitute the bottom events of the model.

[0110] The corresponding faults of the subsystems are cracking, damage, leakage, etc. of the fluorine tube assembly; damage, breakage, blockage, jamming, etc. of the indoor unit assembly; damage, jamming, breakage, etc. of the compressor.

[0111] The corresponding faults of the components are: the hot water pipe component falls off or breaks; the condenser fan is damaged; the thermostat component burns out or breaks, etc.

[0112] The component failures include bearing wear and breakage, loose tensioner, clutch damage, broken fan blades, core wear, loose and falling fan blades, broken pipes, loose joints, and aging of seals.

[0113] Step 2: Map the fault tree model into a Bayesian network;

[0114] Specifically, when the fault tree model is converted into a Bayesian network, the structure of the fault tree model corresponds to the Bayesian network, and each level of fault events in the fault tree model corresponds to each level of nodes in the Bayesian network: the bottom event of the fault tree model corresponds to the leaf node of the Bayesian network, and the first-level intermediate event, second-level intermediate event, and third-level intermediate event in the fault tree model correspond to the first-level child node, second-level child node, and third-level child node of the Bayesian network, respectively. The connection between the fault events in the fault tree model corresponds to the edge of the Bayesian network, and the logic gate that connects the events together in the fault tree model is replaced by the conditional probability table of the Bayesian network. The prior probability of the leaf node in the Bayesian network corresponds to the probability of occurrence of the bottom event in the fault tree model.

[0115] Figure 2 The Bayesian network topology diagram for the air conditioning system;

[0116] Step 3: Combine historical fault data to calculate the probability of occurrence of the bottom event of the fault tree model and use it as the prior probability of the corresponding Bayesian network leaf node;

[0117] Specifically, the probability of occurrence of the bottom event of the fault tree model is calculated in combination with the historical fault data, and is used as the prior probability of the corresponding Bayesian network leaf node. The probability of occurrence of the bottom event of the fault tree model is the probability of occurrence of various types of faults obtained after statistical analysis of the historical monitoring data of the air-conditioning system. Since the present invention has many subsystems, the probability of occurrence of the bottom event of the air-conditioning compressor system is taken as an example, as shown in Table 1, and the conditional probabilities of other subsystems are no longer listed.

[0118] Table 1: Statistical table of the probability of occurrence of bottom events in air conditioning compressor system

[0119] Code Bottom Event Occurrence probability (times / 1000h) Code Bottom Event Occurrence probability (times / 1000h) A1 Bearing wear 0.117 A8 Compressor internal leakage 0.014 A2 Bearing fracture 0.019 A9 Controller burnt out 0.017 A3 Seal aging 0.012 A10 Loose wiring harness 0.043 A4 Fasteners fall off 0.009 A11 Cover falling off 0.008 A5 Clutch sticking 0.007 A12 Belt breakage 0.005 A6 Tensioner wheel stuck 0.031 A13 Bracket cracking 0.005 A7 Tensioner wheel displacement 0.023

[0120] The bottom event fault codes of the compressor system are A1-A13, the third-level intermediate node fault codes are B1-B7, and so on. The other second-level intermediate node codes of the air-conditioning compressor system are C1-C4, and the first-level intermediate node compressor system code is D1. The faults of the sub-nodes of the compressor system at all levels are shown in Table 2, and the fault tree model diagram is shown in Figure 3 .

[0121] Table 2: Summary of faults of sub-nodes at all levels of the compressor system

[0122]

[0123] Step 4: According to step 2 and step 3, calculate the posterior probability of each level of child nodes;

[0124] Specifically, the posterior probability of each level of child nodes is calculated according to step 2 and step 3. In the Bayesian network, conditional probability is used to express the logical relationship between each node, and the logic gate in the fault tree model corresponds to the conditional probability table in the Bayesian network.

[0125] Taking the air conditioning compressor system as an example, the corresponding leaf nodes in its Bayesian network are A1-A13, and its upper nodes B1-B7 are the third-level child nodes. The third-level child node compressor bearing B1 is connected to two root nodes bearing wear and bearing fracture , then the probability of failure of the corresponding third-level child node B1 can be expressed as:

[0126]

[0127] The leaf node and The posterior probability of occurrence is expressed as:

[0128]

[0129] Similarly, the third-level child nodes As a higher level subnode The bottom event, if the secondary child node Connect m third-level child nodes ,by Representation Node The probability of failure, then the above steps can be used to obtain the node When this happens, the jth child node The posterior probability of occurrence, where m child nodes and nodes They correspond to the third-level child nodes and the second-level child nodes in step 2 respectively.

[0130] Specifically, the five third-level child nodes B1-B5 are used as higher-level second-level child nodes. The bottom event, Representation Node The probability of failure, then the above steps can be used to obtain the node When it happens, the posterior probability of each of the five child nodes occurring.

[0131] By analogy, the posterior probabilities of all leaf nodes, third-level child nodes, second-level child nodes, and first-level child nodes of the compressor system can be obtained in sequence.

[0132] After calculation, the results are shown in Table 3 below:

[0133] Table 3: Posterior probability table of all sub-nodes of the compressor system

[0134]

[0135] Step 5: Using the database related to component performance parameters, establish a regression model between the parameters related to component performance;

[0136] Specifically, a regression model between the performance parameters of components is established using a database related to the performance parameters of components. The physical and chemical parameters of component materials are the intrinsic properties of the materials. There is a strong correlation between some parameters, and the determination of some parameters is relatively simple and convenient. Therefore, this correlation can be used to calculate the key performance parameters of components under different working conditions through regression analysis methods, and evaluate the real-time failure rate of components.

[0137] In the process of maintaining existing products on the market, we use the test data of parts performance parameters to establish a performance parameter database, analyze the parameter pairs with correlation, and use support vector machine regression analysis to establish a regression model between related parameters. , where the variable To facilitate the measurement of component material performance parameters, It is the key performance parameter of component materials.

[0138] Taking the compressor system as an example, the performance-related parameter pairs of commonly used materials for components are shown in Table 4 below:

[0139] Table 4: Summary of parameters related to compressor parts materials

[0140] Material Related parameters Metals (steel, copper, aluminum) Hardness-tensile strength, toughness-shear modulus, temperature-conductivity Plastic, rubber Shear modulus-toughness Oil Viscosity-Specific Heat, Density-Thermal Conductivity

[0141] The obtained results are obtained by calculation based on the correlation between the performance parameters of component materials, using the measured data of the parameters that are easy to measure and the relevant regression model. The regression model between the relevant parameters has the input of the measured parameter data that is easy to measure, and the output is the key performance parameters of the components that are difficult to measure or destructive during measurement. The types of models include but are not limited to support vector machine regression model, polynomial regression model, etc.

[0142] Step 6: Measure the real-time data of component performance parameters, combine the regression model between related parameters, and analyze the real-time key performance parameters of the components;

[0143] Specifically, the real-time data of component performance parameters is measured, and the real-time data of the key performance parameters of the components is obtained by using the regression model between the related parameters. For the key performance parameters of components that are difficult to measure, the real-time data of the key performance parameters of the components can be obtained by real-time monitoring of the easily measured component material parameters and using the regression model of the related parameters to quickly calculate and obtain the real-time data of the key performance parameters of the components.

[0144] Taking the compressor bearing of the compressor system as an example, its tensile strength and shear modulus are and The historical failure rates of wear and fracture are recorded as and In actual working conditions, the hardness and toughness are measured to be and , using regression model , the predicted real-time tensile strength and shear modulus are and ,Right now:

[0145]

[0146] Step 7: Establish a real-time failure rate assessment model for various component failures and analyze the real-time functional status of the components;

[0147] Specifically, based on the calibrated values ​​of multiple performance parameters of components when they leave the factory, a real-time failure rate assessment model for various failures of components is established to analyze the real-time functional status of components.

[0148] The prior probability of a component is the statistical result of historical failure data, covering various working conditions of all products throughout their life cycle, and is the expected value of the failure rate. However, for the components of a single product, they are affected by factors such as the working environment and performance parameter degradation, and they exhibit the characteristics of random degradation. Therefore, the historical failure rate is difficult to describe the real-time failure rate of a single product component in any period. Therefore, it is necessary to use the real-time data of the key performance parameters of the components, combined with their factory calibration values, to establish a real-time failure rate evaluation model to calculate the real-time failure rate of various failures of the components.

[0149] If the calibrated value of a key performance parameter of a component at the factory and the value measured under the implementation conditions are and , the market statistical failure rate is , then the real-time failure rate The evaluation model is:

[0150]

[0151] Taking the compressor bearings mentioned above as an example, the real-time failure rate of bearing wear and fracture is and , which can be expressed as:

[0152]

[0153] The mentioned component real-time performance status assessment model uses the calibrated values ​​of the component's key performance parameters when leaving the factory and the real-time predicted values ​​as input variables, including but not limited to exponential models, logarithmic models, linear models, etc., and the output result is the real-time failure rate of various component failures.

[0154] Step 8: Combine steps 3 and 7 to calculate the real-time failure rate of each sub-node. The real-time failure rate of various failures of components is calculated and obtained by combining the Bayesian network model of the air-conditioning system to describe the current failure rate of each sub-node in the air-conditioning system.

[0155] Specifically, in step 7, the real-time failure rate of multiple faults of components is the real-time failure rate of leaf nodes in the Bayesian network. Combined with step 4, the real-time failure rate of components is determined using the Bayesian network conditional probability, and the real-time posterior probability of components is calculated:

[0156] Real-time failure rate of each component failure mode , according to the logical relationship between each node of the Bayesian network, determine the real-time failure rate of components .by Indicates the real-time failure rate of components , Represents the real-time failure rate of l failure modes respectively ,but:

[0157]

[0158] in Indicates that when the kth failure mode occurs in real time, the component ( )The probability of failure occurring in real time.

[0159] Combining the multiple faults of components and the real-time failure rate of components, the real-time posterior probability of multiple failure modes of components can be calculated using the posterior probability calculation formula in step 4. For example, among the l failure modes, the real-time posterior probability of the kth failure mode is The calculation formula is:.

[0160]

[0161] Right now Indicates parts ( ) Under the premise of real-time failure, the probability of the kth failure mode occurring in real time among l failure modes, where , , The meaning is the same as above and will not be repeated here.

[0162] Similarly, if a subsystem consists of The real-time failure rate is The real-time failure rate of the subsystem can be determined based on the logical relationship between the nodes of the Bayesian network. ,as well as By analogy, the real-time failure rate and real-time posterior probability of each component can be obtained for the fault diagnosis and location of the system.

[0163] Taking the compressor system as an example, bearing wear and fracture failure and The real-time failure rate is and , then the bearing Real-time failure rate for:

[0164]

[0165] The faulty node and , that is, the real-time posterior probability formulas of bearing wear and fracture are:

[0166]

[0167] Similarly, compressor fault C1 is composed of five component faults B1-B5 in Table 2, and the real-time failure rates are , then the real-time failure rate of the compressor can be determined according to the above Bayesian network conditional probability formula , and the real-time posterior probabilities of the five components.

[0168] By analogy, the real-time failure rate and real-time posterior probability of sub-nodes at each level of the compressor system can be obtained.

[0169] Step 9: Compare the posterior probabilities of the sub-nodes at each level and diagnose the fault status of the sub-nodes at each level;

[0170] Specifically, the real-time posterior probability of failure of each sub-node is obtained by step 8: , by comparing the real-time posterior probability level by level and the statistical posterior probability , calculate the abnormal degree of failure rate, according to the abnormal threshold of failure rate of nodes at all levels To determine the node where the fault occurs. Determined by the importance of components and historical experience, for example, the threshold of key and important components is low, and the abnormal threshold of redundant design components can be higher. The order principle of checking nodes at each level step by step is "high level first, then low level; high fault first, then low fault".

[0171] According to the obtained real-time posterior probability of failure of each sub-node , by comparing the real-time posterior probability level by level and the posterior probabilities of the child nodes at each level in the Bayesian network The relative value of , calculates the abnormal degree of failure rate δ:

[0172]

[0173] According to the abnormal degree of failure rate δ and abnormal failure rate threshold of nodes at each level To determine the node where the fault occurs, if the abnormal degree of failure rate δ of each node is greater than the corresponding abnormal failure rate threshold , then this node is judged to be a faulty node, where Indicates the abnormal failure rate threshold of the j-th fault of the i-th level child node.

[0174] Taking the compressor system as an example, when the system fails When an event occurs, start troubleshooting from the next most serious event, and start with the compressor component failure with the highest statistical failure rate. Start troubleshooting. By comparing the real-time post-test failure rate of compressor component failures and the statistical posterior failure rate The relative difference of the node determines whether it is a faulty node, and the abnormal threshold of the relative difference , determined by experience, that is, the basis for determining a faulty node is:

[0175]

[0176] If the compressor component is a non-faulty node, Real-time post-hoc failure rate of nodes Compared to the statistical posterior failure rate Does not exceed the threshold , then follow the principle of "high fault first, low fault later" to check the low fault nodes of the same level in turn, such as tensioner fault is the second highest failure rate, the real-time posterior probability and abnormal threshold are and , further investigate the fault node of the tensioner;

[0177] If the compressor component is a faulty node, Real-time post-hoc failure rate of nodes Compared to the statistical posterior failure rate Threshold exceeded , then check the three-level sub-nodes in sequence according to the principle of "high-level first, then low-level", such as bearing failure As the third-level child node, the real-time posterior probability and anomaly threshold are and .

[0178] The obtained real-time failure rate of each sub-node is used as the prior probability and is calculated layer by layer using the Bayesian algorithm; the statistical posterior probability of each sub-node of the system is obtained by the statistical results of historical fault data and the Bayesian network calculation and analysis; the abnormal threshold of the posterior probability is the upper limit of the relative difference allowed between the real-time posterior probability and the statistical posterior probability when there is no fault in the sub-nodes at all levels of the system, which is used to determine whether the current sub-node is a faulty node. The abnormal threshold includes but is not limited to relative value, absolute difference, dynamic difference according to the situation, etc.

[0179] Step 10: Output the fault diagnosis results.

[0180] Through step 9, the fault judgment analysis results of the sub-nodes at all levels of the system are obtained, and the output results are presented in the form of statistical tables or logical structure diagrams, including the historical failure rate, real-time failure rate, failure rate abnormal threshold reference value, fault judgment results and other information of the nodes at all levels of the system, and an output report is formed and handed over to technical operation and maintenance personnel for processing.

[0181] The method of this embodiment is convenient to measure and has strong operability in real-time data measurement and processing; the method of fault judgment using Bayesian network and posterior probability makes the diagnosis process more flexible and the fault location more accurate. It can be applied to fault diagnosis of various models of air-conditioning systems with slight adjustments, and has strong practical generalization.

[0182] In addition, it should be noted that:

[0183] In step 1, the air conditioning system uses a four-level fault tree model to construct the fault mode, which can be replaced by a three-level, five-level or even other level fault tree model to achieve the same purpose;

[0184] In step 5, the regression analysis model between the parameters related to the performance of the parts can be established by using methods such as logistic regression, linear regression, and nonlinear polynomial regression. The method can be selected according to the actual calculation accuracy requirements, data volume, etc., and the same purpose can be achieved.

[0185] In step 7, the real-time fault assessment model established in combination with the key performance parameters is not limited to the exponential model selected in the present invention. Other models, such as linear, exponential, logarithmic, power functions and corresponding variants, used to describe the relationship between the statistical failure rate and the real-time failure rate can also achieve the same purpose.

[0186] In step 9, the abnormal threshold of the posterior probability of each sub-node is determined by the statistical posterior probability and the real-time posterior probability, and is not limited to the relative value used in the present invention, and can be replaced by the absolute difference or the dynamic difference according to the situation, without affecting the purpose of the present invention.

[0187] Embodiment 2:

[0188] This embodiment provides a Bayesian network fault diagnosis device based on parameter correlation, including:

[0189] Memory, for storing computer programs / instructions;

[0190] A processor is used to execute the computer program / instructions to implement the steps of the method described in embodiment 1.

[0191] Embodiment three:

[0192] This embodiment provides a computer-readable storage medium on which a computer program / instruction is stored. When the computer program / instruction is executed by a processor, the steps of the method described in the first embodiment are implemented.

[0193] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0194] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0195] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0196] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0197] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A Bayesian network fault diagnosis method based on parameter correlation, characterized in that: The following steps are involved: Step 1: Determine the fault type of the air conditioning system of the construction machinery. The same fault type corresponds to the same fault event in the fault tree. The fault event is used as the bottom event of the fault tree model to build a fault tree model of the air conditioning system. Step 2: Mapping the fault tree model into a Bayesian network; Step 3: Combined with historical fault data, calculate the probability of occurrence of the bottom event of the fault tree model, and use it as the prior probability of the leaf node of the Bayesian network corresponding to the bottom event of the fault tree model, and obtain the prior probability of each leaf node of the Bayesian network; Step 4: According to the Bayesian network mapped by the fault tree model and the prior probability of each leaf node of the Bayesian network, the posterior probability of each level of sub-nodes in the Bayesian network is calculated; Step 5: Obtain a regression model between component performance related parameters; the input of the regression model between component performance related parameters is the component performance related parameters that are easy to measure, and the output is the real-time key performance parameters of the component; Step 6: Measure the real-time data of the performance-related parameters of the component, and combine the regression model between the performance-related parameters of the component to obtain the real-time data of the real-time key performance parameters of the component; Step 7: Obtain a real-time failure rate assessment model for various faults of a component, wherein the input variables of the real-time failure rate assessment model are the factory calibration values ​​of the key performance parameters of the component, the real-time predicted values ​​of the key performance parameters of the component output by the relevant parameter regression model, and the market statistical failure rate of the component, and the output variable is the real-time failure rate of the component under the corresponding failure mode; Step 8: Obtain the factory calibration values ​​of the key performance parameters of the components and the market statistical failure rate of the components, and combine the real-time data of the real-time key performance parameters of the components output by the regression model to calculate the real-time failure rate of the sub-nodes at each level of the Bayesian network, and use the real-time failure rate of the sub-nodes at each level as the real-time prior probability to calculate the real-time posterior probability of the sub-nodes at each level in the Bayesian network; Step 9: Compare the real-time posterior probability and the posterior probability of the sub-nodes at each level in the Bayesian network, diagnose the fault status of the sub-nodes at each level, and obtain the fault diagnosis result; Step 10: Output the fault diagnosis results.

2. The Bayesian network fault diagnosis method based on parameter correlation according to claim 1 is characterized in that: Step 1: Determine the fault type of the air conditioning system of the construction machinery. The same fault type corresponds to the same fault event in the fault tree. Take the fault event as the bottom event of the fault tree model and build a fault tree model of the air conditioning system, including: In the fault tree model, the failure of the air-conditioning system is taken as the top event, the failures corresponding to each subsystem are taken as the first-level intermediate events, the failures corresponding to the components are taken as the second-level intermediate events, and the failures corresponding to the parts are taken as the third-level intermediate events. All failure events constitute the bottom event of the fault tree model.

3. The Bayesian network fault diagnosis method based on parameter correlation according to claim 2 is characterized in that: Step 2: Mapping the fault tree model into a Bayesian network, including: When the fault tree model is converted into a Bayesian network, the structure of the fault tree model corresponds to the Bayesian network, and each level of fault events in the fault tree model corresponds to each level of nodes in the Bayesian network: the bottom event of the fault tree model corresponds to the leaf node of the Bayesian network, and the first-level intermediate event, second-level intermediate event, and third-level intermediate event in the fault tree model correspond to the first-level sub-node, second-level sub-node, and third-level sub-node of the Bayesian network respectively; The connections between fault events in the fault tree model correspond to the edges of the Bayesian network. The logic gates that connect events together in the fault tree model are replaced by the conditional probability table of the Bayesian network. The prior probability of the leaf nodes in the Bayesian network corresponds to the probability of occurrence of the bottom event in the fault tree model.

4. The Bayesian network fault diagnosis method based on parameter correlation according to claim 3 is characterized in that: Step 3: Combined with historical fault data, calculate the probability of occurrence of the bottom event of the fault tree model, and use it as the prior probability of the leaf node of the Bayesian network corresponding to the bottom event of the fault tree model, and obtain the prior probability of each leaf node of the Bayesian network, including: Combined with historical fault data, the probability of occurrence of the bottom event of the fault tree model is calculated and used as the prior probability of the corresponding Bayesian network leaf node. Among them, the historical fault data includes all fault types in the air conditioning system and the frequency of occurrence of each fault type; The probability of occurrence of the bottom event of the fault tree model is the probability of occurrence of various types of faults obtained after statistical analysis of the fault data, and it is used as the prior probability of the corresponding Bayesian network leaf node.

5. The Bayesian network fault diagnosis method based on parameter correlation according to claim 1 is characterized in that: Step 4: According to the Bayesian network mapped by the fault tree model and the prior probability of each leaf node of the Bayesian network, the posterior probability of each level of sub-nodes in the Bayesian network is calculated, including: In Bayesian networks, conditional probabilities are used to express the logical relationships between nodes. The logic gates in the fault tree model correspond to the conditional probability tables in the Bayesian network. If the node Connect n child nodes , by express The probability of a node failing is expressed as: （1） Then the node When this happens, the i-th child node The formula for calculating the posterior probability of occurrence is expressed as: （2） in, express The i-th child node connected to the node The prior probability of , n is the total number of child nodes of node B; Represents a child node Node in case of The conditional probability of occurrence, Indicates that node B occurs when node The conditional probability of occurrence, that is, the i-th child node The posterior probability of occurrence; n child nodes and The nodes correspond to leaf nodes and third-level child nodes respectively; , n is the total number of child nodes of node B; the posterior probability of all leaf nodes and the probability of occurrence of third-level child nodes are calculated; The third-level child node As a higher level child node The bottom event, if the secondary child node Connect m third-level child nodes ,by Representation Node The probability of failure, then the above steps can be used to obtain the node When this happens, the jth child node The posterior probability of occurrence, where m child nodes and nodes Corresponding to the third-level sub-nodes and the second-level sub-nodes respectively; calculating the posterior probability of all the third-level sub-nodes and the probability of the second-level sub-nodes occurring; The second-level child node C is used as the bottom event of the higher first-level child node, and the posterior probabilities of all leaf nodes, third-level child nodes, second-level child nodes and first-level child nodes of the compressor system are obtained in sequence through the above steps.

6. The Bayesian network fault diagnosis method based on parameter correlation according to claim 1 is characterized in that: Step 5: Obtain a regression model between parameters related to component performance, including: Obtain test data on component performance parameters and establish a performance parameter database; Based on the performance parameter database, the parameter pairs with correlation are analyzed to obtain the corresponding relationship between the easily measurable component performance-related parameters and the component material key performance parameters, and obtain the correlated parameter pairs consisting of the easily measurable component performance-related parameters and the component material key performance parameters; Based on the test data of the relevant parameter pairs consisting of the easily measurable component performance related parameters and the key performance parameters of the component materials, the regression model between the relevant parameters is established by using the support vector machine regression analysis: ； Among them, the variable The parameters related to component performance that are easy to measure are used as model input variables. Key performance parameters of component materials corresponding to easily measurable parameters related to component performance are used as model output variables.

7. The Bayesian network fault diagnosis method based on parameter correlation according to claim 1 is characterized in that: Step 6: Measure the real-time data of the performance-related parameters of the component, and combine the regression model between the performance-related parameters of the component to obtain the real-time data of the real-time key performance parameters of the component, including: For the key performance parameters of components that are difficult to measure, the corresponding regression model between the relevant parameters is obtained through the correspondence between the key performance parameters of the component performance that are easy to measure and the key performance parameters of the component materials; By real-time monitoring of the easily measurable material parameters of the components, real-time data of parameters related to component performance are obtained; Based on the real-time data of component performance-related parameters, the relevant parameter regression model is used to calculate and obtain the real-time data of the key performance parameters of the components.

8. The Bayesian network fault diagnosis method based on parameter correlation according to claim 1 is characterized in that: Step 7: Obtain a real-time failure rate assessment model for various component failures, including: The prior probability of multiple failure modes of components is determined as the statistical result of historical failure data, covering various working conditions of all products in the whole life cycle, which is the expected value of failure rate; Using the real-time data of key performance parameters of components and combining them with their factory calibration values, a real-time failure rate evaluation model is established to calculate the real-time failure rates of various component failures. If the factory calibration value of the key performance parameter corresponding to a certain failure mode of a component and the real-time prediction value under the implementation condition are and The market statistical failure rate of this failure mode is , then the real-time failure rate The evaluation model is expressed as: （3） The input variables of the real-time failure rate assessment model are the factory calibration values ​​of the key performance parameters of the components, the real-time prediction values ​​of the key performance parameters of the components output by the relevant parameter regression model, and the market statistical failure rate of the components. The output variable is the real-time failure rate of the components under the corresponding failure mode.

9. The Bayesian network fault diagnosis method based on parameter correlation according to claim 1 is characterized in that: Step 8: Obtain the factory calibration values ​​of the key performance parameters of the components and the market statistical failure rate of the components, and combine the real-time data of the real-time key performance parameters of the components output by the regression model to calculate the real-time failure rate of the sub-nodes at each level of the Bayesian network, and use the real-time failure rate of the sub-nodes at each level as the real-time prior probability to calculate the real-time posterior probability of the sub-nodes at each level in the Bayesian network, including: Obtain the factory calibration values ​​of key performance parameters of components and the market statistical failure rates of components; According to the factory calibration values ​​of the key performance parameters of the components and the market statistical failure rates of the components, combined with the real-time data of the real-time key performance parameters of the components output by the regression model, based on the real-time failure rate evaluation model, the real-time failure rates of various faults of the components are calculated, that is, the real-time failure rates of the leaf nodes in the Bayesian network; Real-time failure rate of each component failure mode , according to the logical relationship between each node of the Bayesian network, determine the real-time failure rate of components : by Indicates the real-time failure rate of components , Represents the real-time failure rate of the kth failure mode ,but: （4） in Indicates that when the kth failure mode occurs in real time, the component ( ) The probability of failure occurring in real time; Combining the multiple faults of components and the real-time failure rate of components, the real-time posterior probability of multiple failure modes of components is calculated using the posterior probability calculation formula, that is, the real-time posterior probability of the leaf node is obtained. Among the l failure modes, the real-time posterior probability of the kth failure mode is The calculation formula is: （5） in, Indicates parts ( ) Under the premise of real-time failure, the probability of the kth failure mode occurring in real time among l failure modes; Based on the real-time failure rate of components The logical relationship between each node of the Bayesian network determines the real-time failure rate of the component , and the real-time posterior probability of the parts, the real-time posterior probability of the parts corresponds to the real-time posterior probability of the third-level child nodes; By analogy, the real-time failure rate and real-time a posteriori probability of the second-level sub-nodes and the real-time failure rate and real-time a posteriori probability of the first-level sub-nodes are calculated to obtain the real-time failure rate and real-time a posteriori probability of sub-nodes at all levels for application in system fault diagnosis and location.

10. The Bayesian network fault diagnosis method based on parameter correlation according to claim 1, characterized in that: Step 9: Compare the real-time posterior probability and the posterior probability of each sub-node in the Bayesian network, diagnose the fault status of each sub-node, and obtain the fault diagnosis results, including: According to the obtained real-time posterior probability of failure of each sub-node , by comparing the real-time posterior probability level by level and the posterior probabilities of the child nodes at each level in the Bayesian network The relative value of , calculates the abnormal degree of failure rate δ: （6） According to the abnormal degree of failure rate δ and abnormal failure rate threshold of nodes at each level To determine the node where the fault occurs, Indicates the abnormal failure rate threshold of the j-th fault of the i-th level child node.

11. The Bayesian network fault diagnosis method based on parameter correlation according to claim 10, characterized in that: According to the abnormal degree of failure rate δ and abnormal failure rate threshold of nodes at each level To determine the node where the fault occurred, including: When the system fails When an event occurs, start troubleshooting from the next most serious event, and then from the system component failure with the highest statistical failure rate. Start troubleshooting by comparing the abnormality of the relative difference between the real-time a posteriori failure rate of the system component failure and the statistical a posteriori failure rate. If the abnormal degree of the failure rate of the sub-node is greater than the abnormal failure rate threshold, the sub-node is determined to be a faulty node. If this system component If it is a non-faulty node, check the low-fault nodes of the same level in order of real-time fault rate from high to low; If this system component is a faulty node, then its child nodes and the next-level child nodes of its child nodes are checked in sequence according to the node level of the Bayesian network.

12. The Bayesian network fault diagnosis method based on parameter correlation according to claim 1, characterized in that: Step 10: Output the fault diagnosis results, including: Obtain the fault judgment analysis results of sub-nodes at all levels of the system; the output results are presented in the form of statistical tables or logical structure diagrams, including the historical failure rate, real-time failure rate, failure rate abnormal threshold reference value, and fault judgment result information of nodes at all levels of the system, and an output report is formed and output.

13. A Bayesian network fault diagnosis device based on parameter correlation, comprising: Memory, for storing computer programs / instructions; A processor, configured to execute the computer program / instructions to implement the steps of the method according to any one of claims 1 to 12.

Images