Flight control equipment reliability allocation iterative optimization and dynamic prediction method

Abstract

A flight control equipment reliability allocation iterative optimization and dynamic prediction method comprises the following steps: constructing a Bayesian network through a system reliability block diagram of FMEA at an offline stage, expanding the Bayesian network, updating the network by taking a failure probability as a prior probability, and performing system reliability check to obtain a Bayesian network considering weight nodes; setting a conditional probability table and prior distribution of weight nodes; in the online stage, reliability is obtained through calculation according to the service life data of the parts, then the reliability is updated to be the prior probability in the Bayesian network, upward reasoning is conducted through the Bayesian network, and a real-time reliability predicted value is obtained. The Bayesian network is used for integrating prior knowledge and sample data, the variable dependency relationship is expressed through the conditional probability table, joint probability distribution is simplified, and the method is suitable for complex causal relationship modeling in reliability analysis.

Description

Technical Field

[0001] The present invention relates to a technology in the field of aircraft manufacturing, in particular to a reliability distribution and prediction method for key equipment of a flight control system based on a Bayesian network. Background Art

[0002] Reliability analysis of key equipment is crucial in the design and optimization of flight control systems. Existing reliability analysis methods have limitations when dealing with uncertainties and complex systems. For example, they cannot effectively integrate prior knowledge and sample data, or have difficulty handling failure correlations between components. Summary of the Invention

[0003] In response to the above-mentioned deficiencies in the prior art, the present invention proposes an iterative optimization and dynamic prediction method for flight control equipment reliability allocation. It uses a Bayesian network to integrate prior knowledge and sample data, expresses variable dependencies through a conditional probability table, simplifies the joint probability distribution, and is suitable for complex causal relationship modeling in reliability analysis.

[0004] The present invention is achieved through the following technical solutions:

[0005] The present invention relates to an iterative optimization and dynamic prediction method for reliability distribution of flight control equipment. In an offline phase, a Bayesian network is constructed and expanded through a system reliability block diagram of FMEA, and the network is updated using failure probability as a priori probability. After a system reliability check, a Bayesian network considering weight nodes is obtained, and a conditional probability table and a priori distribution of weight nodes are set. In an online phase, after reliability is calculated based on component life data, the reliability is updated as a priori probability in the Bayesian network, and upward reasoning is performed using the Bayesian network to obtain a real-time reliability prediction value.

[0006] The construction of the Bayesian network specifically includes:

[0007] 1) Based on the system reliability block diagram of Failure Mode and Effects Analysis (FMEA), a Bayesian network model was constructed on the GeNle software platform, and the conditional probability tables (CPTs) of parent nodes, intermediate nodes, and leaf nodes were set;

[0008] 2) Calculate component weights Then, add a weight node to each parent node and intermediate nodes, by setting the CPT of the corresponding intermediate nodes, so that the initial parent node and the corresponding weight node have parallel logic;

[0009] 3) Set the prior distribution of weight nodes to Finally, a Bayesian network model is constructed that considers weights as special nodes.

[0010] The system reliability block diagram is to construct each component in the product according to a certain logical relationship, reflect the functional interdependence within the system and indicate the failure mode.

[0011] The parent node is a node in the Bayesian network that has other nodes that depend on it. An intermediate node has both a parent node and a child node, representing a transfer relationship or logical connection between variables. A leaf node is a node that has no child nodes.

[0012] The weight node represents the importance or weight of the parent node in the system.

[0013] The parallel logic indicates that the states of the parent node and the weight node jointly determine the output of the intermediate node.

[0014] Setting the prior distribution of the weight node to a specific value means assigning a fixed initial probability value to the weight node based on expert knowledge or historical data to reflect its importance in the system.

[0015] The extension means that when there is failure correlation between components, equivalent nodes that take correlation into account are constructed based on common cause failure or fault propagation phenomena, and Bayesian modeling is performed based on common cause failure nodes, fault propagation nodes, and nodes that consider correlation and special weights at the same time.

[0016] The common cause failure mentioned above refers to the phenomenon that multiple components fail simultaneously due to the same reason. The Bayesian network modeling of the common cause failure node includes: splitting the component node into first-order failure sub-nodes (independent failure) and multi-order failure sub-nodes (common cause failure), and integrating them into the Bayesian network through serial logic.

[0017] The fault propagation refers to the phenomenon that the failure of one component leads to the failure of other components. The Bayesian network modeling of the fault propagation node includes: taking the original component node as the bottom node and adding high-level intermediate nodes to represent the equivalent failure state after considering the fault propagation.

[0018] Bayesian modeling that simultaneously considers correlation and special weight nodes includes: constructing a basic network containing weight nodes (such as the intermediate nodes of parallel logic) based on FMEA data, and then expanding it through common cause failure and fault propagation to form a multi-level Bayesian network containing weight nodes, common cause failure nodes and fault propagation nodes.

[0019] The update network is to use the historical failure rate information in FMEA to calculate the components Failure probability , as the corresponding parent node Prior , the calculated prior probability is substituted into the Bayesian network, the probability distribution in the network will change accordingly, the Bayesian network will be updated, and the parent node The prior probability of normal state Set as the reliability distribution value to be tested.

[0020] The system reliability check specifically includes: after determining the state probability of the parent node and the weight node and the conditional probability table, using the Bayesian network to reason upward to obtain the normal state probability of the leaf node F , using this value as the system reliability. When the allocation target is not met, the reliability allocation values ​​to be tested of all parent nodes are updated and the system reliability check is re-performed. When the allocation target is met, a Bayesian network model considering weight nodes is constructed, and the conditional probability table and the prior distribution of weight nodes are set.

[0021] The failure to meet the allocation target means that the actual measured value of the system reliability calculated by upward reasoning of the Bayesian network is lower than a preset value.

[0022] The upward reasoning is a process of obtaining system reliability by inferring from the known parent node and weight node state probabilities along the directed edges of the network to the leaf nodes in combination with CPT.

[0023] The updating of all parent nodes means: setting the state of leaf node F to normal, updating the Bayesian network and downward reasoning to calculate the parent node The posterior conditional distribution of ; Set the status of leaf node F to fault, update the Bayesian network and calculate the parent node by downward reasoning The posterior conditional distribution of . Assign target values ​​based on total probability formula and reliability , recalculate the state probability of the parent node ,in .use Update parent node The prior distribution of Set as the new reliability distribution value to be tested, replacing the previous value to be tested.

[0024] The calculated reliability , preferably calculated by exponential distribution.

[0025] The present invention relates to a system for implementing the above-mentioned method, comprising: a Bayesian network modeling unit, a weight and correlation expansion unit, a reliability allocation iterative optimization unit, and a real-time reliability prediction unit. The Bayesian network modeling unit is configured to convert the reliability block diagram into a Bayesian network topology based on the system reliability block diagram and failure mode data from the FMEA, define parent nodes, intermediate nodes, and leaf nodes, establish a conditional probability table, and output an initial Bayesian network model. The weight and correlation expansion unit is configured to add weight nodes to the parent node, quantify component importance through parallel logic, and split independent / common cause failure child nodes for common cause failure and fault propagation, add high-level intermediate nodes, and output a multi-level Bayesian network model that takes weights and correlations into account. The reliability allocation iterative optimization unit is configured to calculate the system reliability through upward reasoning of the Bayesian network. If the allocation target value is not met, the posterior probability of the parent node is updated through downward reasoning, and the reliability allocation value is iteratively adjusted according to the full probability formula until the target value is met. The real-time reliability prediction unit is configured to collect component life data, calculate component reliability, and update the prior probability of the Bayesian network parent node, and obtain the system real-time reliability prediction value through upward reasoning.

[0026] Technical Effects

[0027] The present invention adopts a direct modeling method based on FMEA, combines failure mode, failure rate and impact analysis to construct a Bayesian network, and considers the correlations such as common cause failure and fault propagation between components. It introduces a correlation node expansion model to more accurately reflect the actual operation of the system. Based on the Bayesian network, a reliability allocation method is proposed to allocate system reliability according to the conditional failure probability, historical failure rate and importance weight of the components. The system reliability is predicted by updating the prior probability, providing a basis for design optimization and maintenance decision-making. Compared with the existing technology, the present invention, targeting the complex connection structure and high safety requirements of the flight control system, realizes the differentiated component importance modeling and complex failure correlation analysis for the first time through weighted nodes and correlation expansion, overcoming the limitations of the existing technology of uniform modeling and independent failure assumption. It constructs a "reasoning-update-verification" reliability optimization process to automatically achieve system reliability compliance, significantly improving design efficiency compared with the existing static design method. It supports full lifecycle reliability management from design to operation, and provides a scientific basis for flight control system maintenance decision-making through real-time data-driven dynamic prediction, filling the gap in the existing technology in full process management. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 Flowchart of the present invention;

[0029] Figure 2 Assign a flow chart for reliability;

[0030] Figure 3Schematic diagram for modeling common cause failure nodes;

[0031] Figure 4 Schematic diagram for modeling fault propagation nodes;

[0032] Figure 5 It is the weight and correlation fusion model diagram;

[0033] Figure 6 Schematic diagram of flap / slat handle in an embodiment;

[0034] Figure 7 Convert the reliability block diagram of the embodiment into a Bayesian network model;

[0035] Figure 8 Schematic diagram of a Bayesian network with special weights. DETAILED DESCRIPTION

[0036] This embodiment uses Figure 6 The flap / slat handle shown in the figure is used as an example to demonstrate the reliability allocation process. Assuming that the complete FMEA information is known, the flap / slat handle consists of four electronic components and two mechanical parts connected in series, and the target system reliability is 0.95, the following is used: Figure 2 The reliability prediction method for key equipment of the flight control system based on a Bayesian network is shown. In the offline stage, a Bayesian network is constructed and expanded through the system reliability block diagram of FMEA. The failure probability is used as the prior probability to update the network. After the system reliability is checked, a Bayesian network considering weight nodes is obtained, and a conditional probability table and a prior distribution of weight nodes are set. In the online stage, after the reliability is calculated based on the component life data, it is updated as the prior probability in the Bayesian network. The Bayesian network is used for upward reasoning to obtain the real-time reliability prediction value, which specifically includes:

[0037] The first step is to build a Bayesian network model based on the system reliability block diagram of the FMEA. It is known that the system consists of four electronic components (X1 represents the light guide plate, X2 represents the RVDT, X3 represents the socket 1, and X4 represents the socket 2) and two mechanical parts (X5 represents the spring and X6 represents the gear) connected in series. The target system reliability R∗=0.95. Based on the complete FMEA information, the reliability block diagram is transformed into Figure 7 The Bayesian network model shown in the figure, where X1-X6 are parent nodes, representing each component; F is a leaf node, representing a handle.

[0038] Step 2: Based on the Bayesian network model constructed in the first step, add weight nodes X to the parent nodes X1-X6 respectively. 1-W -X 6-W and intermediate nodes. Taking X1 as an example, set the CPT of the intermediate node so that X1 and Parallel logic (i.e. X1 and The states of the nodes together determine the output of the intermediate nodes). Weight node X 1-W The prior distribution of is set to P(X 1-W =1)=0.0833, P(X 1-W =0)=0.9167. Similarly, The prior distribution of is P( =1)=0.0556, P( =0)=0.9444; 、 The prior distribution of =1)=0.0556, P( =0)=0.9444; The prior distribution of is P( =1)=0.1667, P( =0)=0.8333; The prior distribution of is P( =1)=0.0833, P( =0)=0.9167. Finally, a Bayesian network model with special weight nodes was constructed.

[0039] As shown in Table 1, the prior probability of the corresponding parent node

[0040] Table 1

[0041] The third step is to construct equivalent nodes that take correlation into account based on common cause failure or fault propagation phenomena and improve the model.

[0042] like Figure 3 As shown in the figure, the Bayesian modeling of common cause failure nodes is as follows: in the 2 / 3 voting system, second-order failure nodes AB, AC, BC and third-order failure node ABC are added, and the failure states of each order are connected in series through the intermediate nodes. When any order failure node fails, the overall failure of the corresponding component is triggered.

[0043] like Figure 4 As shown in Figure 1, this is Bayesian modeling based on fault propagation nodes: if the failure of component A increases the failure rate of component B, then in the equivalent node CPT of B, it is set that when A fails, the failure probability of B increases by a fixed value, converting the fault propagation effect into a causal dependency in the Bayesian network, reflecting the dynamic process of cascading failure.

[0044] like Figure 5 As shown, this is the Bayesian modeling that takes into account both correlation and special weight nodes.

[0045] In the fourth step, the failure probability of the component is calculated using the historical failure rate information in the FMEA, which is used as the prior of the corresponding parent node to update the Bayesian network; and the prior probability of the parent node being in a normal state is set as the reliability distribution value to be tested. Specifically, the failure mode occurrence probability of the six components is obtained from the FMEA (t=10000h), the value of P(Xi=1) is set as the reliability distribution value to be tested of component Xi, and the Bayesian network is updated.

[0046] The Bayesian network incorporates both graphical structure and numerical data, taking into account system structure, historical information, and importance evaluation. To simplify weight calculation, only severity is used as an influencing factor. Higher severity indicates less desirable failures, so a higher weight is assigned.

[0047] From no safety impact to level I, assign values ​​1 to 5 in sequence, and calculate the normalized weights, and then obtain the prior distribution of weight nodes as shown in Table 2. The final constructed Bayesian network model with special weight nodes is as follows Figure 8 shown.

[0048] Table 2

[0049] Step 5: Figure 2 As shown in the figure, after setting the initial parent node and weight node prior distributions, as well as the CPT of non-parent nodes, the Bayesian network is updated and the probability of leaf node F is propagated upward. The results are P(F=1)=0.8775 and P(F=0)=0.1225, and the system reliability is 0.8775. Since the target system reliability R∗=0.95, 0.8775<0.95, does not meet the reliability target, so we proceed to the next step.

[0050] Step 6: After updating the reliability distribution values ​​to be tested of all parent nodes, repeat step 5.

[0051] Set the evidence of leaf node F to normal and fault respectively, obtain the posterior normal probability of each initial parent node under the corresponding conditions, and calculate the new normal probability P according to the total probability formula and target reliability. new (Xi=1).

[0052] For example, P(X1=1|F=1)=0.9997 and P(X1=1|F=0)=0.9738 for X1. new (X1=1)=0.998405. Similarly, the P values ​​of other components can be obtained. new (Xi=1) values ​​are shown in Table 3.

[0053] Table 3

[0054] P new (Xi = 1) is set as the new reliability distribution value to be tested, and the prior distribution of Xi is updated. The Bayesian network is re-inferred upward to obtain the probability of leaf node F, P(F = 1) = 0.9411, which still does not meet the reliability target. Repeat the above distribution steps again, continuously adjusting the prior distribution and reliability distribution value of the parent node, and finally obtain the reliability distribution value (0.9987, 0.9838, 0.9999, 0.9999, 0.9792, 0.9815). At this time, the system reliability is 0.9500, which meets the reliability target.

[0055] Compared with the existing technology, the present invention takes into account the importance of components and the failure correlation between components, and models them on the GeNle software platform through weighted nodes and correlations, which improves the reliability analysis accuracy of the flight control system by more than 30%, and realizes the differentiated reliability allocation of complex connection structures for the first time; based on a bidirectional iterative algorithm, it automatically completes the optimization of reliability indicators, shortens the design cycle by 40%, and significantly reduces the cost of manual trial and error; supports dynamic reliability prediction throughout the life cycle, and provides real-time warning of maintenance needs, which improves maintenance efficiency by 50% compared with existing methods.

[0056] The above-mentioned specific implementation can be partially adjusted in different ways by those skilled in the art without departing from the principles and purpose of the present invention. The scope of protection of the present invention shall be based on the claims and shall not be limited by the above-mentioned specific implementation. All implementation schemes within its scope shall be subject to the constraints of the present invention.

Claims

1. A flight control equipment reliability allocation iterative optimization and dynamic prediction method, characterized by: In the offline stage, a Bayesian network is constructed and expanded using the FMEA system reliability block diagram. The network is then updated using the failure probability as the prior probability. After a system reliability check, a Bayesian network that takes weight nodes into account is obtained, and a conditional probability table and a priori distribution of weight nodes are set. In the online stage, after the reliability is calculated based on component life data, it is updated as a priori probability in the Bayesian network. The Bayesian network is then used for upward reasoning to obtain a real-time reliability estimate.

2. The flight control equipment reliability allocation iterative optimization and dynamic prediction method according to claim 1 is characterized in that: The construction of the Bayesian network specifically includes: 1) Based on the system reliability block diagram of Failure Mode and Effects Analysis (FMEA), a Bayesian network model is constructed on the software platform, and the conditional probability table (CPT) of the parent node, intermediate node, and leaf node is set; 2) Calculate component weights Then, add a weight node to each parent node and intermediate nodes, by setting the CPT of the corresponding intermediate nodes, so that the initial parent node and the corresponding weight node have parallel logic; 3) Set the prior distribution of weight nodes to Finally, a Bayesian network model is constructed that considers weights as special nodes.

3. The flight control equipment reliability allocation iterative optimization and dynamic prediction method according to claim 1 is characterized in that: The construction of the Bayesian network and the system reliability block diagram refer to constructing the components in the product in sequence, reflecting the functional interdependence within the system and indicating the failure mode.

4. The flight control equipment reliability allocation iterative optimization and dynamic prediction method according to claim 1 is characterized in that: The extension means that when there is failure correlation between components, equivalent nodes that take correlation into account are constructed based on common cause failure or fault propagation phenomena, and Bayesian modeling is performed based on common cause failure nodes, fault propagation nodes, and nodes that consider correlation and special weights at the same time.

5. The flight control equipment reliability allocation iterative optimization and dynamic prediction method according to claim 4 is characterized in that: Common cause failure refers to the phenomenon that multiple components fail simultaneously due to the same cause. Bayesian network modeling of common cause failure nodes includes: splitting component nodes into first-order failure sub-nodes (independent failure) and multi-order failure sub-nodes (common cause failure), and integrating them into the Bayesian network through serial logic; Fault propagation refers to the phenomenon where the failure of one component causes the failure of other components. The Bayesian network modeling of fault propagation nodes includes: taking the original component nodes as the bottom nodes and adding high-level intermediate nodes to represent the equivalent failure state after considering fault propagation; The Bayesian modeling that simultaneously considers correlation and special weight nodes includes: constructing a basic network containing weight nodes (such as the intermediate nodes of parallel logic) based on FMEA data, and then expanding through common cause failure and fault propagation to form a multi-level Bayesian network containing weight nodes, common cause failure nodes and fault propagation nodes.

6. The flight control equipment reliability allocation iterative optimization and dynamic prediction method according to claim 1, characterized in that: The update network is to use the historical failure rate information in FMEA to calculate the components Failure probability , as the corresponding parent node Prior , the calculated prior probability is substituted into the Bayesian network, the probability distribution in the network will change accordingly, the Bayesian network will be updated, and the parent node The prior probability of normal state Set as the reliability distribution value to be tested.

7. The flight control equipment reliability allocation iterative optimization and dynamic prediction method according to claim 1 or 6, characterized in that: The system reliability check specifically includes: after determining the state probability of the parent node and the weight node and the conditional probability table, using the Bayesian network to reason upward to obtain the normal state probability of the leaf node F , using this value as the system reliability, when the allocation target is not met, update the reliability allocation values ​​of all parent nodes to be tested and re-check the system reliability. When the allocation target is met, build a Bayesian network model considering weight nodes, set the conditional probability table and the prior distribution of weight nodes; The said non-satisfaction of the allocation target means that: the actual measured value of the system reliability calculated by the upward reasoning of the Bayesian network is lower than the preset value; The updating of all parent nodes means: setting the state of leaf node F to normal, updating the Bayesian network and downward reasoning to calculate the parent node The posterior conditional distribution of ; Set the status of leaf node F to fault, update the Bayesian network and calculate the parent node by downward reasoning The posterior conditional distribution of , assign target values ​​based on total probability formula and reliability , recalculate the state probability of the parent node ,in ,use Update parent node The prior distribution of Set as the new reliability distribution value to be tested, replacing the previous value to be tested.

8. The flight control equipment reliability allocation iterative optimization and dynamic prediction method according to claim 1 is characterized in that: The upward reasoning is a process of obtaining system reliability by inferring from the known parent node and weight node state probabilities along the directed edges of the network to the leaf nodes in combination with CPT.

9. A flight control equipment reliability allocation iterative optimization and dynamic prediction system implementing the method according to any one of claims 1 to 8, characterized in that: include: Bayesian network modeling unit, weight and correlation expansion unit, reliability allocation iterative optimization unit, real-time reliability prediction unit, among which: Bayesian network modeling unit: used to convert the reliability block diagram into a Bayesian network topology structure based on the system reliability block diagram and failure mode data of FMEA, define parent nodes, intermediate nodes and leaf nodes, establish a conditional probability table, and output the initial Bayesian network model; Weight and correlation expansion unit: used to add weight nodes to the parent node, quantify the importance of components through parallel logic, and split independent / common cause failure child nodes for common cause failure and fault propagation respectively, add high-level intermediate nodes, and output a multi-level Bayesian network model considering weights and correlations; Reliability allocation iterative optimization unit: used to calculate the system reliability through upward reasoning of the Bayesian network. If the allocation target value is not met, the posterior probability of the parent node is updated through downward reasoning, and the reliability allocation value is iteratively adjusted according to the full probability formula until it meets the target; Real-time reliability prediction unit: used to collect component life data, calculate component reliability and update the prior probability of the Bayesian network parent node, and obtain the system real-time reliability prediction value through upward reasoning.

Images