System fault probability calculation method

Abstract

The invention relates to a system fault probability calculation method, which comprises the following steps of: acquiring fault modes of a system, and representing a system fault state as a binary vector containing all fault modes; for any fault mode, calculating the prior probability of the fault mode and the triggering probability of each detection node in the fault mode; based on the Bayesian theorem, according to the prior probability of each occurred fault mode contained in the system fault state and the triggering probability of each triggered detection node in each occurred fault mode, establishing a function relationship between the occurrence probability of the system fault state and the triggering state of each detection node; and obtaining the trigger state of each detection node, substituting the trigger state into the function relationship, and calculating to obtain the occurrence probability of the set system fault state.

Description

Technical Field

[0001] The present invention relates to the technical field of fault detection, and in particular to a method for calculating system fault probability. Background Art

[0002] In complex electromechanical systems, aerospace equipment, and industrial automation control, fault diagnosis technology is crucial for ensuring reliable system operation. Existing fault diagnosis methods primarily rely on rule-based diagnostic systems based on empirical databases and condition monitoring techniques based on signal thresholds. However, when multiple faults couple or characteristic signals overlap, these methods can easily generate fuzzy fault groups, making it difficult to accurately distinguish between different fault modes. Furthermore, sensor noise and operating condition fluctuations often lead to false alarms, making it impossible to accurately locate faults and resulting in low maintenance efficiency for system equipment.

[0003] In recent years, deep learning-based fault classifiers (such as convolutional neural networks and LSTM time series models) have been widely used in the field of fault pattern recognition. However, such methods rely on massive amounts of labeled fault data for model training, while the cost of obtaining fault samples in actual engineering is extremely high. In addition, the black box nature of the model leads to a lack of interpretability in the diagnostic results, making it difficult to meet the review requirements of scenarios such as aviation airworthiness certification.

[0004] Therefore, a failure probability calculation method is needed to accurately calculate the probability of occurrence of possible failure modes, thereby ensuring the effectiveness and accuracy of fault diagnosis, especially fault diagnosis of aerospace systems and equipment. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a method for calculating the probability of system failure, which can accurately calculate the probability of occurrence of possible failure modes when a system or device fails.

[0006] The technical solution adopted by the present invention to solve the technical problem is to provide a method for calculating system failure probability, which is applied to a system including multiple detection nodes, including the following steps:

[0007] Obtaining the failure mode of the system, and then expressing the system failure state as a binary vector containing the occurrence states of all failure modes;

[0008] For any of the failure modes, calculating the prior probability of the failure mode and the triggering probability of each detection node under the failure mode;

[0009] Based on Bayes' theorem, a functional relationship between the probability of occurrence of the system fault state and the triggering state of each detection node is established according to the prior probability of each fault mode contained in the system fault state and the triggering probability of each triggered detection node under each fault mode.

[0010] The trigger state of each detection node is obtained and substituted into the functional relationship to calculate the probability of occurrence of the set system fault state.

[0011] Furthermore, the functional relationship is expressed as:

[0012]

[0013] Wherein, mc represents the trigger state of each detection node, fault_C represents the set system fault state, FC=0 represents the system fault state caused by unknown factors, P(fault_C|mc) represents the probability of occurrence of the system fault state fault_C under the trigger state mc, P(fault_C) represents the prior probability of the system fault state fault_C, P(mc|fault_C) represents the probability that each detection node is in the trigger state mc under the system fault state fault_C, P(FC=0|mc) represents the probability that the system fault state is caused by unknown factors under the trigger state mc, P(FC=0) represents the prior probability that the system fault state is caused by unknown factors, and P(mc|FC=0) represents the probability that each detection node is in the trigger state mc when the system fault state is caused by unknown factors.

[0014] Furthermore, the prior probability of the system fault state fault_C is calculated by the following formula:

[0015] P(fault_C)=Πα i

[0016] Among them, a i It represents the prior probability of each fault mode i contained in the system fault state fault_C.

[0017] Furthermore, the probability of each detection node being in the trigger state mc under the system fault state fault_C is calculated by the following formula:

[0018] P(mc|fault_C)=Π[1-(1-λ k )Π(1-β ki )

[0019] Among them, λ k represents the triggering probability of the triggered detection node k being triggered by unknown factors, β ki represents the triggering probability of each triggered detection node k under the fault mode i.

[0020] Furthermore, the probability that the system fault state in the trigger state mc is caused by an unknown factor is calculated by the following formula:

[0021] P(FC=0|mc)=Π[Π(1-β kx )]

[0022] Among them, β kx It represents the probability that the fault mode fx causes the detection node k to be triggered.

[0023] Furthermore, the prior probability that the system fault state is caused by unknown factors is calculated by the following formula:

[0024] P(FC=0)=Π(1-α n )

[0025] Among them, a n Represents the prior probability of each failure mode n of the system.

[0026] Furthermore, the probability of each detection node being in the trigger state mc when the system fault state is caused by an unknown factor is calculated by the following formula:

[0027] P(mc|FC=0)=Πλ k

[0028] Among them, λ k It represents the triggering probability of the triggered detection node k being triggered by an unknown factor.

[0029] Furthermore, the system fault state is represented as a binary vector containing all fault modes, including:

[0030] Each failure mode of the system is taken as an element, and the corresponding element is assigned a value based on whether the failure mode occurs, thereby obtaining the system failure state represented as a binary vector.

[0031] Furthermore, the prior probability is obtained by calculating the frequency ratio of the current failure mode using FMECA.

[0032] Furthermore, the prior probability is obtained by calculating the ratio of the number of occurrences of the current failure mode to the total number of system failures using historical data.

[0033] Beneficial effects

[0034] Due to the adoption of the above-mentioned technical solution, the present invention has the following advantages and positive effects compared with the prior art: the present invention sorts out the logical relationship between the system failure mode and the monitoring unit, and calibrates the probability of failure occurrence and the probability of failure causing the monitoring unit to be triggered within the logical relationship framework based on the design parameters of the failure and historical monitoring data, and then uses Bayesian theorem to derive the functional relationship between the probability of occurrence of the system failure state and the triggering state of each detection node, thereby realizing the accurate calculation of the probability of occurrence of possible failure modes when a system or equipment failure is detected. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 is a flow chart of an embodiment of the present invention;

[0036] Figure 2 It is a schematic diagram of the system structure of an embodiment of the present invention;

[0037] Figure 3 It is a schematic diagram of the logical relationship between the fault mode and the monitoring unit in the embodiment of the present invention. DETAILED DESCRIPTION

[0038] Below in conjunction with specific embodiment, further set forth the present invention.Should be understood that these embodiments are only used to illustrate the present invention and are not used in limiting the scope of the present invention.In addition, should be understood that after reading the content taught by the present invention, those skilled in the art can make various changes or modifications to the present invention, and these equivalent forms fall equally within the scope limited by the appended claims of the application.

[0039] The embodiment of the present invention relates to a method for calculating system failure probability, which is applied to a system including multiple detection nodes, such as Figure 1 As shown, the following steps are included:

[0040] Obtaining the failure mode of the system, and then expressing the system failure state as a binary vector containing the occurrence states of all failure modes;

[0041] For any fault mode, calculate the prior probability of the fault mode and the triggering probability of each detection node under the fault mode;

[0042] Based on Bayes' theorem, according to the prior probability of each fault mode contained in the system fault state and the triggering probability of each triggered detection node under each fault mode, a functional relationship between the occurrence probability of the system fault state and the triggering state of each detection node is established;

[0043] The trigger state of each detection node is obtained and substituted into the above functional relationship to calculate the probability of occurrence of the set system fault state.

[0044] The functional relationship between the occurrence probability of the system fault state and the trigger state of each detection node can be expressed as:

[0045]

[0046] Wherein, mc represents the trigger state of each detection node, fault_C represents the set system fault state, FC=0 represents the system fault state caused by unknown factors, P(fault_C|mc) represents the probability of occurrence of the system fault state fault_C under the trigger state mc, P(fault_C) represents the prior probability of the system fault state fault_C, P(mc|fault_C) represents the probability that each detection node is in the trigger state mc under the system fault state fault_C, P(FC=0|mc) represents the probability that the system fault state is caused by unknown factors under the trigger state mc, P(FC=0) represents the prior probability that the system fault state is caused by unknown factors, and P(mc|FC=0) represents the probability that each detection node is in the trigger state mc when the system fault state is caused by unknown factors.

[0047] The present embodiment will be further described below in conjunction with a specific system and derivation process.

[0048] like Figure 2 The system S shown is equipped with a monitoring unit M, which is responsible for monitoring the status of the system. S can be a component or a complex system, such as an aircraft, a bicycle, an avionics system, an electromechanical system, etc. The monitoring system uses various types of sensors and test equipment within the system to monitor the status of the system S.

[0049] System S has multiple failure mode sets:

[0050] F=[f1,f2,f3,……,f N ]

[0051] Failure mode f i Contains two states:

[0052] f n =0, failure mode not occurring

[0053] f n =1, failure mode occurs

[0054] The fault state vector corresponding to the fault mode set F is a binary vector (a vector consisting of 0 and 1):

[0055] FC=[fc1,fc2,fc3,……,fc N ]

[0056] The occurrence of each failure mode is a random event and independent of each other.

[0057] The monitoring unit includes various sensor detection nodes, detection node collection:

[0058] M=[M1,M2,M3,……,MK ]

[0059] M k Contains two states:

[0060] M k =0, detection node k is not triggered

[0061] M k =1, detection node K triggers

[0062] The state vector MC corresponding to M is a binary vector (a vector composed of 0 and 1):

[0063] MC=[MC1,MC2,MC3,……,MC K ].

[0064] If MC k = 0, then all detection nodes are not triggered; if any MC k ≠0 indicates that the monitoring unit is triggered.

[0065] like Figure 3 As shown in the figure, there is a cross-linked relationship between fault modes and sensor nodes. That is, a fault mode can be detected by one sensor node or by multiple other sensor nodes. At the same time, a sensor node can detect one fault mode or multiple other fault modes. For example, f1 can be monitored by M2 and M4, and M2 can monitor f1 and f2. In other words, the logical relationship model between fault modes and monitoring units is a many-to-many model.

[0066] Assume that the state of the system S monitored by the monitoring unit is C, which contains two states: fault-free state and faulty state; its corresponding fault vector C=[f1,f2,f3,……,f N ], the state vector SC corresponding to the fault vector C is:

[0067] SC=[0,0,0,……,0] 1×N is a fault-free state, that is, any fault mode f n =0;

[0068] SC=[f1,f2,f3,……,f N ], there exists at least one f n ≠0, indicating a fault state.

[0069] Setting fault f n =1, that is, the failure mode f n Prior probability α n, which is determined by FMECA analysis during the product or system design and production process and can be replaced by the failure mode frequency ratio in FMECA. When the system enters the large-scale service period, the prior probability can be calculated from its historical data:

[0070]

[0071] Among them, C is the total number of system failures, C n Failure mode f n Number of occurrences.

[0072] f n Not all detection nodes are associated with it, so the detection node set FM associated with it is set to [FM1, FM2, FM3, ..., FM N ], where FM n =[M n1 ,M n2 ,M n3 ,……,M nn ], where M nn Belongs to the monitoring unit set M.

[0073] Set the detection node M of the monitoring unit M k The probability of triggering is β. The detection node is triggered by the fault mode f or some factors not considered by the system design. The monitoring unit M is not related to every fault mode. Therefore, a set of fault modes associated with the monitoring unit MF is set. k =[f k1 ,f k2 ,f k3 ,……,f km ], where f km Belongs to the failure mode set F.

[0074] Detection node M k The trigger is caused by unknown factors or failure mode set MF k The failure mode in the set unknown factors (including the unknown factors that can trigger the detection node M k The probability of the detection node being triggered by all unknown factors is λ k , the probability that the fault mode causes the detection node to trigger is β km (Failure mode f km Causes detection node M k The probability of triggering, f km ∈MF k ), whose probabilities sum to 1:

[0075] λ k +β k1 +β k2 +β k3+……+β km =1,f km ∈MF k

[0076] β km It is a theoretical design value or obtained by statistics based on the historical data of the system. The following is obtained by statistics based on the historical operation data of the system:

[0077]

[0078] Among them, N k is the total number of times the kth detection node triggers the state; N km The trigger state of the kth node is determined by the fault mode f km The number of times it was triggered.

[0079] When the detection node is triggered, that is, there is a state mc in MC, and there is at least one MC in its state vector k ≠0, the fault state in this state is considered to be fault_C, and there is at least one f in its state vector n ≠0; in this state, calculating the fault probability becomes the probability of fault_C existing under the condition of state mc, that is, P(fault_C|mc); based on Bayesian probability theory:

[0080]

[0081] Therefore, calculating the fault probability P(fault_C|mc) becomes calculating three probabilities: P(fault_C), P(mc|fault_C), and P(mc).

[0082] The state mc can be caused by the fault state fault_C (at least one fc n ≠0), or it can be caused by unknown factors (in this state FC=0), so:

[0083] P(FC=0|mc)P(mc)=P(FC=0)·P(mc|FC=0)

[0084] Right now

[0085]

[0086] Therefore

[0087]

[0088] In the above formula, P(fault_C|mc) consists of five parts, among which:

[0089] ①P(fault_C)=Πa i, i is f n =1, that is, the number of the fault mode that has occurred in the system fault state fault_C;

[0090] ②P(mc|fault_C)=ΠP(mc k |fault_C), k is the number of the triggered detection node, P(mc k |fault_C) indicates that in the fault state fault_C (there is at least one fc n ≠0) the probability of the kth detection node being triggered:

[0091] P(mc k |fault_C)=1-(1-λ k )·(1-β k1 )·(1-β k2 )·……·(1-β kN )

[0092] =1-(1-λ k )Π(1-β kn )

[0093] Where i is the index of the fault mode fn=1; the above formula can be further expressed as:

[0094] P(mc|fault_C)=∏[1-(1-λ k )∏(1-β ki )]

[0095] ③ The triggering of detection nodes is an independent event, so:

[0096] P(FC=0|mc)=ΠP(FC=0|mc k )

[0097] =Π[(1-P(f k1 |mc k ))·(1-P(f k2 |mc k ))·……·(1-P(f kx |mc k ))]

[0098] =Π[Π(1-β kx )]

[0099] where f kx =1, and the failure mode f kx and detection node MC k associated, i.e., failure mode f kx The probability of causing the detection node to trigger is β kx ;

[0100] ④P(FC=0)=Π(1-a n ),n∈[1,N]

[0101] ⑤P(mc|FC=0)=ΠP(mc k |FC=0)=Πλ k

[0102] The above calculation is for the probability of any fault state fault_C under the monitoring unit state mc. To calculate the probability of a single fault mode under the monitoring unit state mc, specify the condition that only one fault mode occurs in fault_C.

Claims

1. A method for calculating system failure probability, applied to a system including multiple detection nodes, characterized in that: The following steps are involved: Obtaining the failure mode of the system, and then expressing the system failure state as a binary vector containing the occurrence states of all failure modes; For any of the failure modes, calculating the prior probability of the failure mode and the triggering probability of each detection node under the failure mode; Based on Bayes' theorem, a functional relationship between the probability of occurrence of the system fault state and the triggering state of each detection node is established according to the prior probability of each fault mode contained in the system fault state and the triggering probability of each triggered detection node under each fault mode. The trigger state of each detection node is obtained and substituted into the functional relationship to calculate the probability of occurrence of the set system fault state.

2. The method according to claim 1, characterized in that The functional relationship is expressed as: Wherein, mc represents the trigger state of each detection node, fault_C represents the set system fault state, FC=0 represents the system fault state caused by unknown factors, P(fault_C|mc) represents the probability of occurrence of the system fault state fault_C under the trigger state mc, P(fault_C) represents the prior probability of the system fault state fault_C, P(mc|fault_C) represents the probability that each detection node is in the trigger state mc under the system fault state fault_C, P(FC=0|mc) represents the probability that the system fault state is caused by unknown factors under the trigger state mc, P(FC=0) represents the prior probability that the system fault state is caused by unknown factors, and P(mc|FC=0) represents the probability that each detection node is in the trigger state mc when the system fault state is caused by unknown factors.

3. The method according to claim 2, characterized in that The prior probability of the system fault state fault_C is calculated by the following formula: P(fault_C)=Pa i Among them, a i It represents the prior probability of each fault mode i contained in the system fault state fault_C.

4. The method according to claim 2, characterized in that The probability of each detection node being in the trigger state mc under the system fault state fault_C is calculated by the following formula: P(mc|fault_C)=Π[1-(1-λ k )Π(1-b ki )] Among them, λ k represents the triggering probability of the triggered detection node k being triggered by unknown factors, β ki represents the triggering probability of each triggered detection node k under the fault mode i.

5. The method according to claim 2, characterized in that The probability that the system fault state under the trigger state mc is caused by unknown factors is calculated by the following formula: P(FC=0|mc)=Π[Π(1-β kx )] Among them, β kx It represents the probability that the fault mode fx causes the detection node k to be triggered.

6. The method according to claim 2, characterized in that The prior probability that the system fault state is caused by unknown factors is calculated by the following formula: P(FC=0)=Π(1-α n ) Among them, a n Represents the prior probability of each failure mode n of the system.

7. The method according to claim 2, characterized in that When the system fault state is caused by unknown factors, the probability of each detection node being in the trigger state mc is calculated by the following formula: P(mc|FC=0)=Πλ k Among them, λ k It represents the triggering probability of the triggered detection node k being triggered by an unknown factor.

8. The method according to claim 1, characterized in that The system fault state is represented as a binary vector containing all fault modes, including: Each failure mode of the system is taken as an element, and the corresponding element is assigned a value based on whether the failure mode occurs, thereby obtaining the system failure state represented as a binary vector.

9. The method according to claim 1, characterized in that The prior probability is obtained by calculating the frequency ratio of the current failure mode using FMECA.

10. The method according to claim 1, characterized in that The prior probability is obtained by calculating the ratio of the number of occurrences of the current failure mode to the total number of system failures using historical data.

Images