CN111664083B - Nuclear power main pump fault diagnosis method based on Bayesian network - Google Patents
Nuclear power main pump fault diagnosis method based on Bayesian network Download PDFInfo
- Publication number
- CN111664083B CN111664083B CN202010474489.1A CN202010474489A CN111664083B CN 111664083 B CN111664083 B CN 111664083B CN 202010474489 A CN202010474489 A CN 202010474489A CN 111664083 B CN111664083 B CN 111664083B
- Authority
- CN
- China
- Prior art keywords
- main pump
- value
- probability
- bayesian network
- fluctuation
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 46
- 238000003745 diagnosis Methods 0.000 title claims abstract description 25
- 230000005856 abnormality Effects 0.000 claims abstract description 9
- 238000012544 monitoring process Methods 0.000 claims description 68
- 208000024891 symptom Diseases 0.000 claims description 44
- 238000004364 calculation method Methods 0.000 claims description 21
- 238000005070 sampling Methods 0.000 claims description 18
- 238000005259 measurement Methods 0.000 claims description 12
- 238000004458 analytical method Methods 0.000 claims description 5
- 230000007257 malfunction Effects 0.000 claims description 4
- 230000004888 barrier function Effects 0.000 claims description 3
- 230000002596 correlated effect Effects 0.000 claims description 3
- 230000000875 corresponding effect Effects 0.000 claims description 3
- 238000012216 screening Methods 0.000 claims description 2
- 238000012423 maintenance Methods 0.000 abstract description 2
- 230000002159 abnormal effect Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000002405 diagnostic procedure Methods 0.000 description 1
- 238000003912 environmental pollution Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
Classifications
-
- F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
- F04—POSITIVE - DISPLACEMENT MACHINES FOR LIQUIDS; PUMPS FOR LIQUIDS OR ELASTIC FLUIDS
- F04B—POSITIVE-DISPLACEMENT MACHINES FOR LIQUIDS; PUMPS
- F04B51/00—Testing machines, pumps, or pumping installations
Landscapes
- Engineering & Computer Science (AREA)
- Mechanical Engineering (AREA)
- General Engineering & Computer Science (AREA)
- Monitoring And Testing Of Nuclear Reactors (AREA)
- Testing And Monitoring For Control Systems (AREA)
- Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)
Abstract
The invention belongs to the field of intelligent diagnosis of equipment faults, and particularly relates to a nuclear power main pump fault diagnosis method based on a Bayesian network. The method comprises the following steps: determining information of each node of the Bayesian network; step two: determining the structure of a Bayesian network; step three: determining a conditional probability table of the Bayesian network; step four: input of a Bayesian diagnostic model; step five: application of diagnostic models. According to the method, three types of faults including main pump sensor faults, vibration abnormality caused by process parameter fluctuation and main pump body faults are diagnosed by combining main pump historical data and artificial experience, so that the problem of false report on site is reduced, the maintenance efficiency of the main pump is improved, and the safe and reliable operation of the main pump is ensured.
Description
Technical Field
The invention belongs to the field of intelligent diagnosis of equipment faults, and particularly relates to a nuclear power main pump fault diagnosis method based on a Bayesian network.
Background
The nuclear power main pump is one of key equipment of a pressurized water reactor nuclear power station, and once the main pump fails, the production is affected slightly, so that economic loss is caused, and the main pump is destroyed and killed, thereby causing nuclear leakage and causing serious environmental pollution. Therefore, the accurate and rapid diagnosis of the main pump fault plays an important role in the safe and efficient operation of the nuclear power station.
The nuclear power site monitors the state of the main pump by adopting a single parameter threshold alarm method, and the method has a large number of false alarms, wherein the main false alarms are derived from two points: the sensor faults and the vibration abnormality caused by the fluctuation of the technological parameters exceed the alarm threshold, and a method is necessary to identify the two faults. It is therefore necessary to establish a diagnostic method for sensor failure, vibration abnormality caused by fluctuation of process parameters, and main pump equipment body failure.
Considering that the main pump apparatus accumulates a large amount of history data, it would be effective to improve the accuracy of diagnosis if it could be applied to solve the problem. Based on the analysis, the invention provides a nuclear main pump fault diagnosis method based on a Bayesian network. According to the method, three types of faults including main pump sensor faults, vibration abnormality caused by process parameter fluctuation and main pump body faults are diagnosed by combining main pump historical data and artificial experience, so that the problem of false report on site is reduced, the maintenance efficiency of the main pump is improved, and the safe and reliable operation of the main pump is ensured.
Disclosure of Invention
Aiming at the problem that whether the main pump equipment body faults exist in a single-parameter threshold alarm method adopted on site, the invention provides a main pump fault diagnosis method based on a Bayesian network model, and the false alarm caused by sensor faults and technological parameter fluctuation is diagnosed. According to the invention, the most probable fault type is deduced by reasoning through a Bayesian network according to the characteristic indexes monitored by each main pump as symptom variables.
The main pump fault diagnosis method based on the Bayesian network comprises the following 5 steps:
Step 1: and determining the node information of the Bayesian network.
The main pump fault diagnosis model based on the Bayesian network comprises a barrier layer and a symptom layer.
In the step 1, "determining information of each node of the bayesian network", the content is: and determining the faults and corresponding symptoms of the main pump according to the related information of the main pump, so as to determine the fault layer nodes and the symptom layer nodes of the Bayesian network.
The fault layer node comprises a sensor fault, vibration abnormality caused by technological parameter fluctuation and main pump equipment body fault 3 major faults; the symptom layer nodes comprise fluctuation indexes, multi-parameter correlation indexes, experience indexes and vibration frequency domain indexes 4 major type characteristic indexes, and the symptom nodes of the 4 major type characteristic indexes are respectively described below.
The fluctuation index is used for measuring the fluctuation condition of each monitoring point, the calculation method of the fluctuation index is the absolute value of the pearson correlation coefficient obtained by monitoring point data and the time within one sampling time, the larger the value is in the range of [0,1], the more intense the fluctuation of the monitoring point is indicated, and the value is in the range of [0,1], so the value can be regarded as the probability value as the probability of occurrence of symptoms in the Bayesian network. The probability that there will be a fluctuation is expressed as a set P (a=1) as shown in equation 1.
P(A＝1)＝{P(A1＝1),P(A2＝1),…,P(Ai＝1),…,P(AN＝1)} (1)
In the formula 1, a i =1 indicates that the ith monitoring point of the main pump has fluctuation, P (a i =1) indicates the probability that the ith monitoring point of the main pump has fluctuation, N is the number of monitoring points of the main pump, and the calculation method of P (a i =1) is shown in the formula 2.
Note that, a i =0 indicates that there is no fluctuation at the i-th monitoring point of the main pump, and P (a i =0) indicates the probability that there is no fluctuation at the i-th monitoring point of the main pump, and P (a i＝0)+P(Ai =1) =1, so that description of P (a i =0) is not necessary again.
In the formula 2, X is vibration data with one sampling time length, and n sampling point data X j are contained in the vibration data; t is a time sequence formed by sampling duration, and any time point T j in the time sequence corresponds to the sampling time of X j in X; e (-) represents the calculated mean.
The multi-parameter correlation index is used for deducing a sensor fault according to the correlation among multi-sensor data, and specifically means that two sensors are installed at important parts of a main pump for monitoring, if the correlation of the data of monitoring points of the two sensors is larger within a certain time, the trend of the data collected by the two sensors is consistent, and the probability of the two sensors simultaneously faults is small, so that the probability of the sensor fault is smaller when the correlation is larger. The probability of a sensor failure is represented by a set P (b=1) as shown in equation 3.
P(B＝1)＝{P(B1＝1),P(B2＝1),…,P(Bk＝1),…,P(BM＝1)} (3)
In formula 3, B k =1 indicates that the monitoring point sensor at the kth part of the main pump fails, P (B k =1) indicates the probability that the monitoring point sensor at the kth part of the main pump fails, the value is [0,1], M is the number of parts with two monitoring points on the main pump, and the calculation method of P (B k =1) is shown in formula 4.
Note that, B k =0 indicates that the monitoring point sensor at the kth portion of the main pump does not malfunction, and P (B i =0) indicates the probability that the monitoring point sensor at the kth portion of the main pump does not malfunction, and P (B i＝0)+P(Bi =1) =1, so that description of P (B i =0) is not necessary again.
In the formula 4, Y is vibration data in a period of sampling time acquired by one of two identical sensors at a certain part of a main pump, and the vibration data comprises n sampling point data Y j; y 'is vibration data in a period of sampling time acquired by another sensor at the same part, and n pieces of sampling point data Y' j are contained in the vibration data; e (-) represents the calculated mean.
Where "experience index" is a symptom node added based on artificial experience, and may include one or more. For example, according to field experience, when the data of the monitoring point exceeds a preset alarm threshold value, it indicates that the main pump equipment body is likely to be failed, so that whether the monitoring point reaches the alarm threshold value set on the field can be set as a sign node; as another example, it is empirically likely that a sensor is malfunctioning when the sensor measurement is zero (or full scale), so it is also possible to set as a symptom node whether the sensor measurement is zero (or full scale). The monitoring point reaches an alarm threshold and is expressed by a set P (C=1), as shown in a formula 5; let the monitor point value be the end value (zero value or full scale value) represented by the set P (d=1).
In the formula 5, C i =1 represents that the ith monitoring point of the main pump exceeds the alarm threshold, and P (C i =1) represents the probability that the ith monitoring point of the main pump exceeds the alarm threshold; d i =1 denotes that the i-th monitoring point value is an end value, and P (D i =1) denotes the probability that the i-th monitoring point value is an end value; n is the number of monitoring points of the main pump; p (C i =1) and P (D i =1) are both state quantities 0 or 1,0 indicating no occurrence, 0,1 indicating occurrence, and 100%.
In addition, C i =0 indicates that the i-th monitoring point of the main pump does not exceed the alarm threshold, and D i =0 indicates that the i-th monitoring point value is not an end value, and since P (C i＝1)+P(Ci＝0)＝1、P(Di＝1)+P(Di =0) =1, it is not necessary to explain P (C i =0) and P (D i =0) again.
The vibration frequency domain index refers to that main frequency is obtained by carrying out frequency domain analysis on the vibration data of the main pump, and whether the main frequency is frequency doubling is judged. Let the main frequency component of the vibration frequency domain be a frequency multiplication expressed by a set P (e=1), as shown in equation 6.
P(E＝1)＝{P(E1＝1),P(E2＝1),…,P(Eh＝1),…,P(EH＝1)}(6)
In formula 6, E h =1 represents that the main frequency component is a frequency multiplication, P (E h =1) represents that the main frequency of the h vibration measuring point is a frequency multiplication probability, and the probability is represented by a state quantity of 0 or 1, 0 represents that the frequency multiplication is not a frequency multiplication, the probability is 0,1 represents that the frequency multiplication is a frequency multiplication, and the probability is 100%; h is the number of vibration stations.
In addition, E h =0 indicates that the main frequency component is not a frequency multiplication, and P (e=1) +p (e=0) =1, so that it is not necessary to explain P (e=0) again.
Step 2: the structure of the bayesian network is determined. The bayesian network structure is determined to establish a connection between the barrier layer and the symptom layer. And (2) establishing connection between all sign nodes and fault nodes according to the class 4 feature indexes in the step (1), wherein the connection line represents that the corresponding fault nodes and the sign nodes are associated, as shown in fig. 2, which is a schematic diagram of the established Bayesian network structure.
Step 3: a conditional probability table of the bayesian network is determined. The purpose of determining the conditional probability table of the bayesian network is to determine the conditional probability between the interconnected fault node and the symptom node and the prior probability of each fault occurrence in the bayesian network structure established in step 2, namely the conditional probability between the connecting lines in fig. 3.
The method comprises the following specific steps: for the fluctuation index, firstly screening data related to vibration abnormality caused by sensor faults and process parameter fluctuation and main pump equipment body faults from historical monitoring data, and determining prior probability P prior of occurrence of various faults by quantitatively analyzing the screened data; secondly, counting the probability of fluctuation of each monitoring point when each fault occurs, and taking the probability as the fluctuation index of each monitoring point and the conditional probability of the fault; the other three types of feature indexes are set according to the manual experience and literature data, and the conditional probability of the feature nodes which cannot be determined is set as the reciprocal of the number of all feature nodes (P guess), so that the conditional probability table P condition of the Bayesian network is determined.
Step 4: input of a bayesian diagnostic model. Based on the above three steps, the establishment of the bayesian diagnostic model is completed, and the symptom variable values of each symptom node are calculated and used as the input of the bayesian diagnostic model.
The calculation methods for different characteristic indexes are different, and the following description will be given respectively:
(1) For the "fluctuation index", the sign variable value of the node reflects whether the fluctuation exists at the monitoring point and is represented by a probability value, the calculation method is as shown in the formula 2 of the step 1, the value is a numerical value between [0,1], 0 represents no fluctuation, 1 represents fluctuation, the numerical value between (0, 1) represents the probability value of fluctuation, and the closer the value is to 1, the more severe the fluctuation is considered.
(2) For the "multi-parameter correlation coefficient index", the variables of the node are the correlations of the two sensor monitoring points at the same position, the calculation method is as shown in the formula 4 of the step 1, the values of the variables are numerical values between [0,1], 0 represents that the two parameters are correlated, 1 represents that the two parameters are uncorrelated, and the numerical values between (0, 1) represent that the probability values of the two parameters are uncorrelated, so that the closer the values are to 1, the uncorrelated values can be considered.
(3) For "empirical index" such nodes include "monitor whether the measurement point exceeds an alarm threshold" and "sensor measurement value reaches an end value. For 'whether the monitoring point exceeds the alarm threshold', determining the alarm threshold of each monitoring point according to a main pump manual, and when the amplitude of the monitoring point is larger than the alarm threshold, the sign variable value of the sign node is 1, otherwise, is 0; for the 'sensor measured value reaches the end value', the measuring range is determined according to each sensor parameter, and when the value of the monitoring point is the end value, the value of the sign variable of the sign node is 1, otherwise, the value of the sign variable of the sign node is 0.
(4) For the "vibration frequency domain index", such node variables are whether the main frequency component is a frequency doubling. And carrying out frequency domain analysis on the vibration data, wherein if the main characteristic of the vibration frequency domain is frequency doubling, the sign variable value of the sign node is 0, and otherwise, is 1.
The inputs to the bayesian network are thus:
P(I)＝(P(A),P(B),P(C),P(D),P(E))
Step 5: application of diagnostic models. And (3) according to the symptom variable values of the symptom nodes obtained in the step (4) as input, carrying out reasoning through a Bayesian network diagnosis model, calculating a probability calculation method for various faults, wherein the probability calculation method is shown in a formula (7), taking the maximum probability value as a diagnosis conclusion, and the formula (7) is relatively abstract, and the specific calculation method is described in the embodiment.
The invention provides a nuclear main pump fault diagnosis method based on a Bayesian network, which combines historical monitoring data and manual experience, and can realize three types of faults including main pump sensor faults, vibration anomalies caused by process parameter fluctuation and main pump equipment body faults. In addition, compared with a method for determining whether the main pump equipment has the body fault based on a single parameter threshold, the method can effectively reduce the false alarm rate.
Drawings
FIG. 1 is a flow chart of a method for diagnosing a nuclear main pump fault based on a Bayesian network
FIG. 2 is a schematic diagram of a Bayesian network diagnostic model constructed in accordance with the present invention
FIG. 3 is a structure of a detailed Bayesian diagnostic model constructed in accordance with the present invention
FIG. 4 is a structure of a Bayesian diagnostic model constructed in accordance with an embodiment
FIG. 5 is a graph showing the trend of vibration data of measurement point 1 of pump shaft vibration in the embodiment
FIG. 6 is a graph showing the trend of vibration data of measurement point No. 2 of pump shaft vibration in the embodiment
Detailed Description
The invention will be further described with reference to the accompanying drawings by taking a main pump of a nuclear power plant as an example.
The implementation step flow chart of the nuclear main pump fault diagnosis method based on the Bayesian network is shown in fig. 1, and the specific content of the implementation step flow chart comprises the following steps of 5 steps:
Step 1: and determining the node information of the Bayesian network.
According to the related information of the nuclear main pump, the nodes of the fault layer are three main categories, namely sensor faults, vibration anomalies caused by process parameter fluctuation and main pump body faults. The symptom layer node comprises four types, namely: fluctuation index, multi-parameter correlation index, experience index and vibration frequency domain index. Each type of symptom node comprises a plurality of main pump monitoring points, such as pressure, temperature, flow and the like, and only pump shaft vibration monitoring parameters are selected for illustration in order to more clearly express the method.
Taking two vibration measuring points contained in a pump shaft as an example, namely n=2, three types of faults of a main pump are diagnosed, and the change trend graphs of the vibration measuring point 1 of the pump shaft and the vibration measuring point 2 of the pump shaft are respectively shown in fig. 5 and 6, wherein for the two vibration measuring points, the on-site preset alarm threshold value is 250, the horizontal line in the graph is the alarm threshold value, and the symptom layer contains 7 symptom nodes, which are respectively as follows:
(1) 2 "fluctuation indices": fluctuation index P of pump shaft No. 1 vibration measurement point (a 1 =1), fluctuation index P of pump shaft No. 2 vibration measurement point (a 2 =1);
(2) 1 "multi-parameter correlation index": correlation P of pump shaft No. 1 vibration measurement point and No. 2 vibration measurement point (B 1 =1);
(3) For "experience index", choose whether the alarm threshold is exceeded, thus contain 2 "experience index": whether the amplitude of the vibration measuring point of the pump shaft No. 1 exceeds an alarm threshold value P (C 1), and whether the amplitude of the vibration measuring point of the pump shaft No. 2 exceeds the alarm threshold value P (C 2 =1);
(4) 2 "vibration frequency index": whether the main frequency of the vibration measuring point 1 of the pump shaft is frequency multiplication P (E 1 =1), and whether the main frequency of the vibration measuring point 2 of the pump shaft is frequency multiplication P (E 2 =1).
Contains 7 sign nodes in total, thenThe node information is shown in fig. 4.
Step 2: and determining the association relation of each node.
According to the method of step 2 of the present disclosure, a bayesian network structure is established as shown in fig. 4.
Step 3: a conditional probability table of the bayesian network is determined.
According to the method of step 3 of the present disclosure, the conditional probability table of the bayesian network is determined by analyzing the nuclear main pump annual history monitoring data, as shown in table 1.
Table 1 conditional probability tables for bayesian networks
Wherein the meaning of each node code is as follows:
(1) Y 1 represents a sensor failure, Y 2 represents a vibration abnormality caused by fluctuation of a process parameter, and Y 3 represents a main pump equipment body failure;
(2) For the fault node Y 1～Y3, variable 0 indicates that the fault did not occur, and variable 1 indicates that the fault occurred;
(3) For the symptom node, the variable 0 indicates that the symptom does not occur, and the variable 1 indicates that the symptom occurs, for example, at a 1＝0,Y1,Y2,Y3 =0, indicating that the probability of no fluctuation of pump shaft vibration data is 0.98 and the probability of fluctuation of pump shaft vibration data is 0.02 under the condition that none of the three faults of Y 1～Y3 occur.
Step 4: using the pump shaft vibration data shown in the calculation fig. 5 and 6, and adopting the present disclosure formula 2 to calculate, the symptom variable P (a 1 =1) =0.893 of the symptom node a 1; symptom variable P of symptom node a 2 (a 2 =1) =0.912; calculated according to the present disclosure, equation 4, symptom variable P (a 3 =1) =0.067 for symptom node B 1; the values of the symptom variables of the rest symptom nodes are shown in table 2, the meaning of the values is shown in the step 1 of the content of the invention, and the symptoms in table 2 can be expressed as formula 8.
I＝(A1＝1,A2＝1,B1＝1,C1＝1,C2＝0,E1＝0,E2＝0)(8)
Table 2 sign node input values
Step 5: diagnostic model application
The Bayesian network is adopted for calculation, and the calculation process is shown in the following for the concrete description, but in actual use, the artificial calculation is not needed, and after the Bayesian network is established by a calculation mechanism in actual use, the diagnosis result can be directly output by inputting the symptom variable.
Using the Bayesian theorem, there are
Wherein,
Similarly, P (a 1＝1|Y2 =0) =0.188 can be obtained.
Since P (a 1 =1) =0.893, interpolation can be used to obtain
P(A1＝1|Y2＝1)＝0.188+0.893×(0.812-0.188)＝0.745。
Similarly, the conditional probability of the other symptom at Y 2 =1 can be obtained, and the result is shown in table 3, which is the conditional probability of the related symptom obtained at Y 2 =0 or Y 2 =1 in table 3
From the conditional probabilities found in Table 3, one can obtain
Probability of occurrence of vibration abnormal faults caused by fluctuation of process parameters:
P(Y2＝1|I)＝0.751
Similarly, the probability of occurrence of sensor failure P (Y 1 = 1|I) =0.219, and the probability of occurrence of main pump body failure P (Y 3 = 1|I) =0.028.
According to the calculation result of the Bayesian network diagnosis model, under the condition that the symptoms of the table 2 appear, the probability of occurrence of the vibration abnormal fault caused by the fluctuation of the process parameters is highest, and the fault is the most likely to occur, and accords with the conclusion of manual reasoning.
Claims (5)
1. A nuclear power main pump fault diagnosis method based on a Bayesian network is characterized by comprising the following steps of: the method comprises the following steps: determining information of each node of the Bayesian network; step two: determining the structure of a Bayesian network; step three: determining a conditional probability table of the Bayesian network; step four: input of a Bayesian diagnostic model; step five: application of a diagnostic model;
Wherein, step one: determining information of each node of the Bayesian network specifically comprises the following steps: according to the related information of the main pump, determining the faults and corresponding symptoms of the main pump, thereby determining the fault layer nodes and the symptom layer nodes of the Bayesian network;
The fault layer node comprises a sensor fault, vibration abnormality caused by technological parameter fluctuation and main pump equipment body fault 3 major faults; the symptom layer node comprises a fluctuation index, a multi-parameter correlation index, an experience index and a vibration frequency domain index 4 major characteristic index;
The fluctuation index is used for measuring the fluctuation condition of each monitoring point, the calculation method of the fluctuation index is that the absolute value of the pearson correlation coefficient obtained by monitoring point data in one sampling time period and the time period is in the range of [0,1], and the value is in the range of [0,1], so that the value can be regarded as probability value as the probability of occurrence of symptoms in the Bayesian network;
The probability of fluctuation to exist is expressed as a set P (a=1) as shown in equation 1:
P(A＝1)＝{P(A1＝1),P(A2＝1),…,P(Ai＝1),…,P(AN＝1)} (1)
in the formula 1, a i =1 indicates that the ith monitoring point of the main pump has fluctuation, P (a i =1) indicates the probability that the ith monitoring point of the main pump has fluctuation, and N is the number of the monitoring points of the main pump;
A i =0 indicates that there is no fluctuation at the i-th monitoring point of the main pump, and P (a i =0) indicates the probability that there is no fluctuation at the i-th monitoring point of the main pump, and since P (a i＝0)+P(Ai =1) =1, it is not necessary to explain P (a i =0) again;
In the formula 2, x is vibration data of one sampling duration, and n sampling point data x j are contained in the vibration data; t is a time sequence formed by sampling duration, and any time point T j in the time sequence corresponds to the sampling time of x j in x; e (-) represents the calculated mean;
The "multi-parameter correlation index" is used for deducing a sensor fault according to the correlation among the multi-sensor data, and the probability of the sensor fault is represented by a set P (b=1), as shown in formula 3:
P(B＝1)＝{P(B1＝1),P(B2＝1),…,P(Bk＝1),…,P(BM＝1)} (3)
In the formula 3, B k =1 indicates that the monitoring point sensor at the kth part of the main pump fails, P (B k =1) indicates the probability that the monitoring point sensor at the kth part of the main pump fails, the value is [0,1], M is the number of parts with two monitoring points on the main pump, and the calculation method of P (B k =1) is shown in the formula 4;
Note that, B k =0 indicates that the monitoring point sensor at the kth part of the main pump does not malfunction, and P (B i =0) indicates the probability that the monitoring point sensor at the kth part of the main pump does not malfunction, and since P (B i＝0)+P(Bi =1) =1, it is not necessary to explain P again (B i =0);
In the formula 4, Y is vibration data in a period of sampling time acquired by one of two identical sensors at a certain part of a main pump, and the vibration data comprises n sampling point data Y j; y 'is vibration data in a period of sampling time acquired by another sensor at the same part, and n pieces of sampling point data Y' j are contained in the vibration data; e (-) represents the calculated mean;
the experience index is a sign node added according to manual experience; the monitoring point reaches an alarm threshold and is represented by a set P (C=1); let the monitor point value be the end value (zero value or full scale value) represented by the set P (d=1), as shown in equation 5:
In the formula 5, C i =1 represents that the ith monitoring point of the main pump exceeds the alarm threshold, and P (C i =1) represents the probability that the ith monitoring point of the main pump exceeds the alarm threshold; d i =1 denotes that the i-th monitoring point value is an end value, and P (D i =1) denotes the probability that the i-th monitoring point value is an end value; n is the number of monitoring points of the main pump; p (C i =1) and P (D i =1) are both state quantities 0 or 1,0 indicating no occurrence, 0,1 indicating occurrence, 100%;
In addition, C i =0 indicates that the ith monitoring point of the main pump does not exceed the alarm threshold, D i =0 indicates that the value of the ith monitoring point is not an end value, and P (C i＝1)+P(Ci＝0)＝1、P(Di＝1)+P(Di =0) =1, so that it is not necessary to explain P (C i =0) and P (D i =0) again;
The vibration frequency domain index refers to that main frequency is obtained by carrying out frequency domain analysis on the vibration data of the main pump, and whether the main frequency is one frequency multiplication or not is judged; let the main frequency component of the vibration frequency domain be a frequency multiplication expressed by a set P (e=1), as shown in equation 6:
P(E＝1)＝{P(E1＝1),P(E2＝1),…,P(Eh＝1),…,P(EH＝1)}(6)
In formula 6, E h =1 represents that the main frequency component is a frequency multiplication, P (E h =1) represents that the main frequency of the h vibration measuring point is a frequency multiplication probability, and the probability is represented by a state quantity of 0 or 1, 0 represents that the frequency multiplication is not a frequency multiplication, the probability is 0,1 represents that the frequency multiplication is a frequency multiplication, and the probability is 100%; h is the number of vibration stations;
In addition, E h =0 indicates that the main frequency component is not a frequency multiplication, and P (e=1) +p (e=0) =1, so that it is not necessary to explain P (e=0) again.
2. The method for diagnosing a nuclear power main pump fault based on a bayesian network according to claim 1, wherein the method comprises the steps of: step 2: the method for determining the structure of the Bayesian network specifically comprises the following steps: determining a Bayesian network structure to establish a connection relationship between the barrier layer and the symptom layer; and (3) establishing connection between all sign nodes and fault nodes according to the 4-class characteristic indexes in the step one.
3. The method for diagnosing a nuclear power main pump fault based on a bayesian network according to claim 1, wherein the method comprises the steps of: and step 3: determining a conditional probability table of the Bayesian network; the purpose of determining the conditional probability table of the bayesian network is to determine the conditional probability between the fault node and the sign node which are connected with each other and the prior probability of each fault occurrence in the bayesian network structure established in the step 2, which specifically comprises the following steps:
for the fluctuation index, firstly screening data related to vibration abnormality caused by sensor faults and process parameter fluctuation and main pump equipment body faults from historical monitoring data, and determining prior probability P prior of occurrence of various faults by quantitatively analyzing the screened data;
Secondly, counting the probability of fluctuation of each monitoring point when each fault occurs, and taking the probability as the fluctuation index of each monitoring point and the conditional probability of the fault;
The other three types of feature indexes are set according to the manual experience and literature data, and the conditional probability of the feature nodes which cannot be determined is set as the reciprocal of the number of all feature nodes (P guess), so that the conditional probability table P condition of the Bayesian network is determined.
4. The method for diagnosing a nuclear power main pump fault based on a bayesian network according to claim 1, wherein the method comprises the steps of: step 4: the input of the Bayesian diagnosis model specifically comprises the following steps:
Based on the above three steps, the establishment of the Bayesian diagnosis model is completed, and the symptom variable values of each symptom node are calculated and used as the input of the Bayesian diagnosis model:
The calculation method is different for different characteristic indexes, and the specific method is as follows:
(1) For the fluctuation index, the sign variable value of the node reflects whether the fluctuation exists in the monitoring point and is represented by a probability value, the calculation method is as shown in a formula 2 in the step 1, wherein the value is a numerical value between 0 and 1, 0 represents no fluctuation, 1 represents fluctuation, the numerical value between 0 and 1 represents the probability value of fluctuation, and the closer the value is to 1, the more severe the fluctuation is considered;
(2) For the multi-parameter correlation coefficient index, the variables of the node are the correlation of two sensor monitoring points at the same position, the calculation method is as shown in the formula 4 in the step 1, the value is a numerical value between [0,1], 0 represents that the two parameters are correlated, 1 represents that the two parameters are uncorrelated, and the numerical value between (0, 1) represents that the probability value of the uncorrelated two parameters is not correlated, so that the closer the value is to 1, the uncorrelated two parameters are considered to be uncorrelated;
(3) For "experience indicators", such nodes include "monitoring whether the measurement point exceeds an alarm threshold" and "whether the sensor measurement value reaches an end value"; for 'whether the monitoring point exceeds the alarm threshold', determining the alarm threshold of each monitoring point according to a main pump manual, and when the amplitude of the monitoring point is larger than the alarm threshold, the sign variable value of the sign node is 1, otherwise, is 0; for the 'sensor measured value reaches an end value', determining a measuring range according to each sensor parameter, and when the value of the monitoring point is the end value, the value of the sign variable of the sign node is 1, otherwise, the value of the sign variable of the sign node is 0;
(4) For the vibration frequency domain index, whether the node variable is frequency doubling of the main frequency component or not, carrying out frequency domain analysis on vibration data, and if the main characteristic of the analyzed vibration frequency domain is frequency doubling, setting the sign variable value of the sign node to be 0 and otherwise setting the sign variable value to be 1;
the inputs to the bayesian network are thus:
P(I)＝(P(A),P(B),P(C),P(D),P(E))。
5. The method for diagnosing a nuclear power main pump fault based on a bayesian network according to claim 4, wherein the method comprises the steps of: step 5: and (3) applying a diagnosis model, taking the symptom variable value of each symptom node obtained in the step (4) as input, carrying out reasoning by using a Bayesian network diagnosis model, calculating the probability calculation method of various faults as shown in the formula (7), and taking the maximum probability value as a diagnosis conclusion:
Wherein, the prior probability of occurrence of various faults of P prior; a conditional probability table for the P condition bayesian network; p fault is various occurrence probabilities.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010474489.1A CN111664083B (en) | 2020-05-29 | 2020-05-29 | Nuclear power main pump fault diagnosis method based on Bayesian network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010474489.1A CN111664083B (en) | 2020-05-29 | 2020-05-29 | Nuclear power main pump fault diagnosis method based on Bayesian network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111664083A CN111664083A (en) | 2020-09-15 |
| CN111664083B true CN111664083B (en) | 2024-06-11 |
Family
ID=72385132
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010474489.1A Active CN111664083B (en) | 2020-05-29 | 2020-05-29 | Nuclear power main pump fault diagnosis method based on Bayesian network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111664083B (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112687407B (en) * | 2020-12-28 | 2022-05-17 | 山东鲁能软件技术有限公司 | Nuclear power station main pump state monitoring and diagnosing method and system |
| CN112801171B (en) * | 2021-01-25 | 2024-07-26 | 中国商用飞机有限责任公司北京民用飞机技术研究中心 | Sensor false alarm identification method and device, computer equipment and storage medium |
| CN113780566A (en) * | 2021-06-23 | 2021-12-10 | 核动力运行研究所 | Bayesian network parameter initialization method |
| CN115095534B (en) * | 2022-04-11 | 2024-07-16 | 中核核电运行管理有限公司 | CANDU6 reactor main pump fault diagnosis method based on KPCA |
| CN116992392B (en) * | 2023-09-27 | 2023-12-08 | 中国长江电力股份有限公司 | Simulation sample generation method oriented to equipment failure mechanism |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6785636B1 (en) * | 1999-05-14 | 2004-08-31 | Siemens Corporate Research, Inc. | Fault diagnosis in a complex system, such as a nuclear plant, using probabilistic reasoning |
| CN101915234A (en) * | 2010-07-16 | 2010-12-15 | 西安交通大学 | Method for diagnosing compressor-associated failure based on Bayesian network |
| CN106372330A (en) * | 2016-08-31 | 2017-02-01 | 北京化工大学 | Application of dynamic Bayesian network to intelligent diagnosis of mechanical equipment failure |
| CN110008350A (en) * | 2019-03-06 | 2019-07-12 | 杭州哲达科技股份有限公司 | A kind of pump Ankang knowledge base lookup method based on Bayesian inference |
-
2020
- 2020-05-29 CN CN202010474489.1A patent/CN111664083B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6785636B1 (en) * | 1999-05-14 | 2004-08-31 | Siemens Corporate Research, Inc. | Fault diagnosis in a complex system, such as a nuclear plant, using probabilistic reasoning |
| CN101915234A (en) * | 2010-07-16 | 2010-12-15 | 西安交通大学 | Method for diagnosing compressor-associated failure based on Bayesian network |
| CN106372330A (en) * | 2016-08-31 | 2017-02-01 | 北京化工大学 | Application of dynamic Bayesian network to intelligent diagnosis of mechanical equipment failure |
| CN110008350A (en) * | 2019-03-06 | 2019-07-12 | 杭州哲达科技股份有限公司 | A kind of pump Ankang knowledge base lookup method based on Bayesian inference |
Non-Patent Citations (1)
| Title |
|---|
| 贝叶斯网络用于屏蔽泵系统故障分析方法的研究;鲍依兰等;《核科学与工程》;第32卷(第02期);103-109页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111664083A (en) | 2020-09-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111664083B (en) | Nuclear power main pump fault diagnosis method based on Bayesian network | |
| CN106872657B (en) | A kind of multivariable water quality parameter time series data accident detection method | |
| CN104573850B (en) | A kind of Power Plant Equipment state evaluating method | |
| JP2009053938A (en) | Equipment diagnosing system and equipment-diagnosing method on the basis of multiple model | |
| KR102005138B1 (en) | Device abnormality presensing method and system using thereof | |
| MX2013000066A (en) | System, method, and apparatus for oilfield equipment prognostics and health management. | |
| CN110348150A (en) | A kind of fault detection method based on dependent probability model | |
| CN113987908A (en) | Natural gas pipe network leakage early warning method based on machine learning method | |
| CN112580858A (en) | Equipment parameter prediction analysis method and system | |
| CN115081647A (en) | Industrial intelligent instrument fault pre-diagnosis method based on Bayesian network model | |
| CN116483054A (en) | Industrial robot running state monitoring and early warning system and method | |
| CN117556347A (en) | Power equipment fault prediction and health management method based on industrial big data | |
| CN114201825B (en) | Evaluation method and system for equipment performance degradation state based on combined characteristics | |
| CN117560300B (en) | Intelligent internet of things flow prediction and optimization system | |
| CN117972600A (en) | Wind turbine generator set key component abnormality detection method based on multidimensional fault feature learning | |
| CN117744874A (en) | Equipment fault prediction method and device and electronic equipment | |
| KR102108975B1 (en) | Apparatus and method for condition based maintenance support of naval ship equipment | |
| CN117330948A (en) | Online monitoring method and system for mechanical characteristics of circuit breaker | |
| CN115600695B (en) | Fault diagnosis method for metering equipment | |
| CN115034094B (en) | Prediction method and system for operation state of metal processing machine tool | |
| Tao et al. | A state and fault prediction method based on RBF neural networks | |
| CN116701985A (en) | Fault diagnosis method, device and equipment for oil gas production instrument and related system | |
| Singer et al. | A pattern-recognition-based, fault-tolerant monitoring and diagnostic technique | |
| CN109886292A (en) | A kind of abnormal cause diagnostic method based on abnormal associated diagram | |
| CN116342110B (en) | Intelligent fault diagnosis and fault tolerance measurement method for multiple temperature measurement loops of train |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |