CN112270128B - Dynamic fault tree-based drilling pump hydraulic end fault diagnosis method - Google Patents
Dynamic fault tree-based drilling pump hydraulic end fault diagnosis method Download PDFInfo
- Publication number
- CN112270128B CN112270128B CN202011178173.4A CN202011178173A CN112270128B CN 112270128 B CN112270128 B CN 112270128B CN 202011178173 A CN202011178173 A CN 202011178173A CN 112270128 B CN112270128 B CN 112270128B
- Authority
- CN
- China
- Prior art keywords
- event
- failure
- gate
- fault
- state
- 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
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N5/00—Computing arrangements using knowledge-based models
- G06N5/04—Inference or reasoning models
- G06N5/042—Backward inferencing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N7/00—Computing arrangements based on specific mathematical models
- G06N7/01—Probabilistic graphical models, e.g. probabilistic networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
- G06F2119/02—Reliability analysis or reliability optimisation; Failure analysis, e.g. worst case scenario performance, failure mode and effects analysis [FMEA]
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- General Physics & Mathematics (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- Computational Mathematics (AREA)
- Pure & Applied Mathematics (AREA)
- Mathematical Optimization (AREA)
- Mathematical Analysis (AREA)
- Algebra (AREA)
- Probability & Statistics with Applications (AREA)
- Computational Linguistics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Computer Hardware Design (AREA)
- Geometry (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The invention discloses a drilling pump hydraulic end fault diagnosis method based on a dynamic fault tree, and relates to the field of fault diagnosis. The invention aims to improve a fault diagnosis method of a discrete-time Bayesian network based on a dynamic fault tree. The core of the invention is to construct a mapping relation of 'rate parameter lambda-division number n', and the time division number n can be selected more scientifically by using a fault diagnosis method of a discrete time Bayesian network based on a dynamic fault tree; the method solves the problem of inaccurate diagnosis caused by the fact that researchers select the time division number n subjectively by proposing a mode of establishing the corresponding relation of the rate parameter lambda-the division number n, improves the reliability of fault diagnosis, and can provide reference for other similar complex machines when using the fault diagnosis method.
Description
Technical Field
The invention relates to the field of fault diagnosis, in particular to an improved discrete time Bayesian network fault diagnosis method based on a dynamic fault tree.
Background
The fault tree analysis method is characterized in that on the premise that the structure and the principle of a system are comprehensively and accurately mastered, a top event of the system is determined, and then analysis is performed step by step layer according to a central event, so that all reasons causing the top event are found out until a bottom event, and a top-down diagnosis process is formed. Aiming at establishing the logical relationship among the top event, the middle event and the bottom event in the fault tree, the fault tree is continuously perfected. After the fault tree is determined, the fault tree is analyzed by a qualitative analysis method and a quantitative analysis method. But the failure of modern engineering is usually accompanied by complex dynamic characteristics of failure time, sequence and the like. The accuracy of the obtained fault is lower by simply using a conventional fault tree method. In the 90 s of the 20 th century, a dynamic fault tree analysis method is provided, and the analysis method has strong modeling and analysis capabilities for a system with dynamic fault logic. The dynamic fault tree differs from the conventional fault tree in that a series of dynamic logic gates are introduced to describe the timing rules and dynamic aging behavior of the system. The current algorithms for solving dynamic fault trees mainly include: a dynamic fault tree analysis algorithm based on a Markov chain, a dynamic fault tree analysis algorithm based on a Bayesian network and an approximation algorithm based on a trapezoidal formula dynamic fault tree. By contrast, an algorithm based on a Bayesian network is more suitable for dynamic fault tree analysis. However, this method has a disadvantage that the number of time divisions n is determined subjectively by the researcher or only by one division n. For complex machines, the state degradation of a component or system is gradual over time, and the degradation efficiency varies from object to object.
Disclosure of Invention
The invention aims to improve a fault diagnosis method of a discrete-time Bayesian network based on a dynamic fault tree. The core of the invention is to construct a mapping relation of 'rate parameter lambda-division number n', and the time division number n can be selected more scientifically by using the fault diagnosis method of the discrete time Bayesian network based on the dynamic fault tree, so that the fault diagnosis result is more reliable.
The technical scheme of the invention is a drilling pump hydraulic end fault diagnosis method based on a dynamic fault tree, which comprises the following steps:
step 1: establishing a dynamic fault tree of a drilling pump hydraulic system by adopting a deductive reasoning method:
step 1.1: taking the fault of a hydraulic end of a drilling pump as a top event of a dynamic fault tree, taking the fault which is easy to occur in a sub-component in a main component in the hydraulic end as an intermediate event, and taking the fault of a component in the sub-component as a bottom event;
step 1.2: according to the working principle and the failure occurrence mechanism of the hydraulic end of the drilling pump, connecting each level of events with a top event by using a corresponding dynamic logic gate, and connecting each level of events from a bottom event to the top event by using the dynamic logic gate to obtain a dynamic failure tree of the hydraulic end of the drilling pump;
and 2, step: converting the dynamic fault tree into a Bayesian network model, and establishing a corresponding relation of a rate parameter lambda-a time division number n;
step 2.1: converting the dynamic fault tree into a Bayesian network model; the expression method of each logic gate in the dynamic fault tree of the dynamic fault tree corresponding to the root node, the middle node and the leaf node of the Bayesian network model respectively in the Bayesian network model comprises the following steps:
and gate:
let X = [ X ] 1 ,X 2 ,…,X m ]Where m is the number of input events of the AND gate, X i I =1,2, …, m is the state variable of the input event, and the number of state combinations is (N + 1) m (ii) a Let Y be the state variable output by AND gate, the state space of all variables is {1,2, …, N +1}, let k = max (X) 1 ,X 2 ,…,X m ) J is a judgment fixed value; the failure mechanism of the and gate is that all input events occur and then output events occur, and then the output events should be at the maximum of the state values of all events, so in any state combination of the input events, the conditional probability distribution of Y is:
or gate:
the failure mechanism of the OR gate is that the output event occurs as long as one element occurs in the input event, so that the state of the output event of the OR gate is the same as the minimum value of the state in the input event; let r = min (X) 1 ,X 2 ,…,X m ) Then the conditional probability distribution of the or gate output event is:
priority and gate:
assuming that an input event is A, B, an output event is Y, state values are a, B and Y respectively, and the failure mechanism of the priority AND gate is that when A and B both fail, A fails before B, and the event Y occurs, namely when a is less than B, the output event Y is in a state B; otherwise, the output event Y is in a state of N +1, which represents that Y does not have a fault; the conditional probability distribution of the priority and gate output event Y is as follows:
when the a is less than the b, the second step is carried out,
when a is more than or equal to b,
closing the doors sequentially:
the order dependent gates force their input events to occur in a particular order without failing in other orders. Sequential door closing, like priority door closing, represents the timing of basic events, which differ in that: input events in sequential door closing cannot fail in any order; while the priority and gates may fail in any order, failure of a particular order will only trigger failure of its output event. Therefore, the sequence correlation gate can be changed into a cascade form of a plurality of priority gates;
the functions are closed:
setting that only one relevant input event A exists when the function phase is closed, the triggering event is Tr, and the output event is T; when Tr occurs, the related event A occurs, and the output event T occurs; when A independently fails, an output event T also occurs; adding an intermediate node A' between the input event and the output event to represent the total failure event of A failure caused by Tr triggering or independent failure of A itself, wherein p and q are corresponding event states, the conditional probability distribution of the output event of the function-related gate is as follows:
step 2.2: setting a value range of n, and selecting different rate parameters lambda to calculate the reliability of the system by adopting a plurality of groups of data collected in advance; comparing the maximum difference ratio of the system reliability among all groups of data to obtain the most appropriate time division number n which should be selected by the rate parameter lambda in different ranges, and establishing the corresponding relation of the rate parameter lambda-the division number n;
and step 3: defining the state of nodes in the Bayesian network model;
dividing a time interval [0,t ] formed by the whole task into n subintervals with equal length, wherein the length of each subinterval is delta = t/n, and then the whole time axis [0, + ∞ ] is divided into n +1 subintervals; when a part corresponding to a certain node A fails in the ith time interval within the task time t, namely A fails in the time interval [ (i-1) delta, i delta ], the node A is in a state i; if A is not expired within the task time t, i.e. A is expired within [ t, ∞), then A is said to be in state n +1;
by the above definition, a time interval in which the state spaces of all nodes in the bayesian network are as follows is obtained: [0, delta ], (Delta, 2 Delta ], …, ((n-1) Delta, n Delta ], (n Delta, plus infinity), abbreviated as {1,2, …, n +1}, the failure time X of the system and the components always corresponds to one section i of n +1 sections, the sum of the probabilities of the system in the first n states is the failure rate of the system at the task time t, and the probability of the system in the n +1 state is the non-failure rate of the system at the task time t;
and 4, step 4: establishing probability distribution of all nodes according to the states of the nodes in the Bayesian network model, and completing quantitative calculation of dynamic fault tree fault diagnosis based on the Bayesian network;
and 5: obtaining a fault diagnosis result through reverse reasoning;
the conditional probability distribution of each node is denoted as P (X) i |pa(X i ) By usingTo express a quantitative relationship between a child node and a parent node; under the condition of giving prior probability distribution of root nodes and conditional probability distribution of non-root nodes, obtaining joint probability distribution containing all nodes, and further performing marginalization calculation on target nodes to obtain marginal probability distribution; and finally, calculating the fault probability of each bottom event by utilizing the reverse reasoning of the Bayesian network, and outputting a diagnosis result according to the magnitude of the fault probability.
Further, the top event in step 1.1 is a hydraulic end fault, and the middle event is: suction pipe failure, valve failure, piston failure, hydraulic cylinder failure, discharge pipe failure, air bag failure, safety valve failure, the end event is: the suction pipeline is not tightly sealed, the suction filter screen is blocked, the valve is seriously abraded, the valve cover is not tightened, the guide sleeve is blocked, the piston is seriously abraded, the piston nut is loosened, the hydraulic cylinder enters air, the cylinder cover is not tightened, the cylinder cover gland is loosened, the discharge filter cylinder is blocked, the discharge pipeline is blocked, the inflation connector is blocked, the air bag inner bag is broken, the stop valve is not tightly sealed, and the safety valve is improperly arranged;
the relationship between events in step 1.2 is:
the suction pipeline is not tightly sealed, and the suction filter screen is blocked and is connected with the suction pipe through a door preferentially; the valve is seriously abraded, the valve cover is not tightened, and the guide sleeve is stuck to be connected with the valve through an OR gate; the piston is seriously worn, and the piston nut is loosened and is connected with the piston through an OR gate; the hydraulic cylinder enters air, a cylinder cover is not tightened, and a cylinder cover gland is loosened and is connected with the hydraulic cylinder through sequential door closing; the blockage of the discharge filter cylinder and the blockage of the discharge pipeline are connected with the discharge pipe through an OR gate in a fault way; the blockage of the inflation connector, the breakage of the air bag inner bag and the untight sealing of the stop valve are connected with the failure of the air bag through the functional phase door closing, the blockage of the inflation connector and the untight sealing of the stop valve are input of relevant events of the functional phase door closing, the breakage of the air bag inner bag is input of a trigger event of the functional phase door closing, and the failure of the air bag is output of the functional phase door closing; the safety valve is connected with the safety valve through the gate when the safety valve is improperly arranged.
Further, the specific method of step 5 is as follows:
containing m nodesEvent U for non-leaf node in time-varying Bayesian network DTBN i Wherein 1. Ltoreq. I.ltoreq.m-1,U i Has an occurrence interval of { [0, Δ ]],(Δ,2Δ],...,((n-1)Δ,nΔ](T, + ∞) }; if top event U T Occurring within the task time T, the occurrence time of the top event must be [0, delta ]],(Δ,2Δ],...,((n-1)Δ,nΔ]Within one of the intervals (n Δ, + ∞); thus, U T The probability of occurrence within the task time T can be directly calculated and expressed as:
in the formula u i Represents U i The occurrence interval of (u) i Belongs to { [0, Δ { ]],(Δ,2Δ],...,((n-1)Δ,nΔ],(T,+∞)};
And then the failure probability of each component when the system fails can be obtained by using a reverse reasoning method.
The invention has the advantages that: the invention eliminates the problem of inaccurate diagnosis caused by that researchers select the time division number n subjectively by proposing a mode of establishing the corresponding relation of the rate parameter lambda-the division number n, improves the reliability of fault diagnosis, and can provide reference for other similar complex machines when using the fault diagnosis method.
Drawings
FIG. 1 is a flow chart of the present invention.
FIG. 2 is a dynamic fault tree model of the hydraulic end of the wellbore pump.
FIG. 3 is a comparison of system reliability when n takes different values.
Fig. 4 is a diagram of the and/or/priority and gate conversion to bayesian network rules.
FIG. 5 is a graph of the rule for a Bayesian network in which sequential gates are closed, converted to a priority AND gate, and then to a Bayesian network.
FIG. 6 is a diagram of the conversion of function-dependent gates into Bayesian network rules.
FIG. 7 is a Bayesian network model transformed by a dynamic fault tree of a hydraulic end of a drilling pump.
Detailed Description
The invention will be further described with reference to the following examples and drawings:
step 1: establishing a dynamic fault tree of a hydraulic end of the drilling pump by adopting a deductive reasoning method:
1. the method comprises the steps of taking a fault of a hydraulic end system of a drilling pump as a top event of a dynamic fault tree of the hydraulic end of the drilling pump, taking a fault of the hydraulic end and a fault of a power end of the drilling pump as second-stage intermediate events M1 and M2, taking a fault which is easy to occur to each sub-component in each main component as a next-stage event, sequentially analogizing to a fault of a component as a bottom event, and coding the top event, each intermediate event and the bottom event, wherein the top event, each intermediate event and the bottom event are shown in table 1
TABLE 1 event code and event name correspondence
Event code | Event(s) | Event code | Event(s) | Event code | Event(s) |
T | Fluid end failure | X1 | Untight seal of suction pipeline | X9 | Unfinished cylinder head |
A | Suction tube failure | X2 | Suction filter screen blocking | X10 | Loosening of cylinder cover gland |
B | Valve failure | X3 | Severe valve wear | X11 | Drain cartridge plug |
C | Piston failure | X4 | The valve cover is not tightened | X12 | Discharge line block |
D | Hydraulic cylinder failure | X5 | Guide sleeve is locked | X13 | Inflation connector blocking |
E | Discharge pipe failure | X6 | Piston wear is severe | X14 | Rupture of air-bag inner bag |
F | Air bag failure | X7 | Loosening of piston nut | X15 | Untight seal of stop valve |
G | Failure of safety valve | X8 | Air entering the hydraulic cylinder | X16 | Improper setting of safety valve |
Wherein T is a top event, A, B, C, D, E, F, G is an intermediate event, and X1-X15 are bottom events.
2. According to the working principle and the failure occurrence mechanism of the hydraulic end of the drilling pump, connecting each level of events with a top event by using corresponding dynamic logic gates, and connecting each level of events from a bottom event to the top event by using the dynamic logic gates to obtain a dynamic failure tree of the hydraulic end of the drilling pump, as shown in the attached figure 2.
Step 2: establishing a corresponding relation of 'rate parameter lambda-division number n' according to a failure distribution model of the system
1. Firstly, the topological structure of the dynamic fault tree model is converted into the network structure of the Bayesian network model. And according to the conditional probability table construction method in the formula 1-6, matlab is used for compiling the conditional probability table construction functions of various logic gates, and a BNT tool box and a variable elimination algorithm are used for solving the BN model.
2. Considering the problem of calculation amount, the value range of n is determined within [1,10], the reliability of different rate parameter lambda calculation systems is selected, the maximum difference ratio of the system reliability among various groups of data is compared, the most appropriate time division number n which is selected by the rate parameter lambda in different ranges can be obtained, and the corresponding relation of the rate parameter lambda to the division number n is established.
With lambda epsilon (0.1 × 10) -6 h -1 ,4.5×10 -6 h -1 ) Assuming that the task time t =50000h, when n is 2, 3, or 4, the time curve of the system reliability is as shown in fig. 3. From the analysis of the calculated data: at each time point, the maximum difference ratio of the reliability of the data obtained by taking n as 3 to the reliability of the data obtained by taking n as 2 is 0.0521%, and the maximum difference ratio of the reliability of the data obtained by taking n as 4 to the reliability of the data obtained by taking n as 3 is 0.0297%, and thus is (0.1 × 10) -6 h -1 ,4.5×10 -6 h -1 ) The number of time divisions n corresponding to the rate parameter λ in the interval is 4.
And step 3: defining states of nodes in a network
The time interval [0,t ] of the entire task is divided into n subintervals of equal length, each subinterval having a length Δ = t/n, and the entire time axis [0, + ∞) is divided into n +1 subintervals. When a part corresponding to a certain node A fails in the ith time interval within the task time t, namely A fails in the time interval [ (i-1) delta, i delta ], the node A is called to be in the state i. If A has not failed within task time t, i.e., A has failed within [ t, ∞), then A is said to be in state n +1.
By the above definition, a time interval in which the state spaces of all nodes in the bayesian network are as follows is obtained:
[0, delta ], (Delta, 2 Delta ], …, ((n-1) Delta, n Delta ], (n Delta, plus infinity). Abbreviated as {1,2, …, n +1}, the sum of the probabilities that the system and the invisible failure time X always correspond to one of n +1 sections i, the probability that the system is in the first n states is the failure rate of the system at the task time t, and the probability that the system is in the n +1 state is the non-failure rate of the system at the task time t.
And 4, step 4: converting the dynamic fault tree model into a Bayesian network model;
converting the dynamic fault tree event into a Bayesian network node according to rules that a top event, a middle event AND a bottom event of the dynamic fault tree respectively correspond to a root node, a middle node AND a leaf node of a Bayesian network model, converting an AND gate (AND), an OR gate (OR) AND a prior AND gate (PAND) into a Bayesian network model rule as shown in figure 4, converting a sequential gate close (SEQ) into a prior AND gate AND then into a Bayesian network model as shown in figure 5, AND converting a functional phase close (FDEP) into a Bayesian network model as shown in figure 6. According to the rules, the dynamic fault tree of the hydraulic end of the drilling pump is converted into a Bayesian network model as shown in the attached figure 7. In the bayesian network model, an edge Probability Distribution (MPD) table is listed beside each root node, and all states of the node and their corresponding probabilities are listed respectively. Each non-root node is provided with a Conditional Probability Distribution table (CPD) which records the Conditional Probability Distribution of the node given the state combination of its parent node.
And 5: establishing a probability distribution for all nodes
Namely, the quantitative analysis of the dynamic fault tree fault diagnosis based on the Bayesian network is completed.
And gate
Let X = [ X ] 1 ,X 2 ,…,X m ]Where m is the number of input events of the AND gate, X i I =1,2, …, m is the state variable of the input event, and the number of state combinations is (n + 1) m And n is the number of time divisions. Let Y be the state variable output by AND gate, the state space of all variables is {1,2, …, n +1}, let k = max (X) 1 ,X 2 ,…,X m ). The failure mechanism of the and gate is that if all input events occur and then output events occur, the output events should be at the maximum value of the state values of all events, so in any state combination of the input events, the conditional probability distribution of Y is:
and the values of the elements in the CPD table of the output node of the AND gate are 0 or 1, the element in each row corresponding to the maximum value of the input event state is 1, and other elements are 0.
OR gate
Similar to an AND gate, except that the failure mechanism of an OR gate is as long asAn output event occurs when one of the input events occurs, so that the state of the output event of the or gate is the same as the minimum value of the state in the input event. Let r = min (X) 1 ,X 2 ,…,X m ) Then the conditional probability distribution of the OR gate output event is
The distribution is consistent with the form of an and gate, except that the element in each row on the column corresponding to the minimum value of the input event state is 1, and the other elements are 0.
Priority AND gate
Assume an input event of A, B, an output event of Y, and state values of a, b, and Y, respectively. The failure mechanism of the priority AND gate is that when A and B both fail, A fails before B, and the event Y occurs, namely when a is less than B, the output event Y is in a state B; otherwise, the output event Y is in state n +1, which represents that Y has not failed. The conditional probability distribution of Y is as follows:
when a is less than b, the first and second groups,
when a is more than or equal to b,
Closing the doors sequentially:
the order dependent gates force their input events to occur in a particular order without failing in other orders. Sequential door closing, like priority door closing, represents the timing of basic events, which differ in that: input events in sequential closing of the doors cannot fail in any order; while the priority and gates may fail in any order, failure in a particular order will only trigger failure of its output event. The sequential dependent gates are thus converted into a cascade of a plurality of priority gates during the computation.
Functional door
Suppose that the function-related door has only one related input event a, the trigger event is Tr, and the output event is T. When Tr occurs, the related event A occurs, and the output event T occurs; when a fails independently, an output event T also occurs. That is, event a, whether an independent failure or a dependent failure, will result in the occurrence of output event T. In order to determine the CPD table (conditional probability distribution table) of the node more intuitively and conveniently, an intermediate node a' is added between the input event and the output event to represent the total failure event of a failure triggered by Tr or caused by an independent failure of a itself (if the node is not added, the failure probability distribution of node a is not an edge distribution, but a conditional probability distribution affected by Tr). Therefore, the conditional probability distribution table is an identity matrix E, and the conditional probability distribution is as follows:
step 6: reverse reasoning to obtain fault diagnosis result
From the conditional independence of the Bayesian network, the conditional probability distribution of each node can be represented as P (X) i |pa(X i ) To express a quantitative relationship between a node and a parent node. Under the condition of giving prior probability distribution of root nodes and conditional probability distribution of non-root nodes, joint probability distribution containing all the nodes can be obtained, and then marginal calculation is carried out on target nodes to obtain marginal probability distribution of the target nodes. And finally, calculating the fault probability of each bottom event by utilizing the reverse reasoning of the Bayesian network, and outputting a diagnosis result according to the magnitude of the fault probability.
Event U for non-leaf node in DTBN containing m nodes i (1. Ltoreq. I. Ltoreq. M-1) represents, U i Has an occurrence interval of { [0, Δ ]],(Δ,2Δ],...,((n-1)Δ,nΔ](T, + ∞). If the top event U T Occurring within the task time T, the occurrence time of the top event must be [0, delta ]],(Δ,2Δ],...,((n-1)Δ,nΔ]And (n Δ, + ∞) in one of the intervals. Thus, U T The probability of occurrence within the task time T can be directly calculated and expressed as:
in the formula u i Represents U i The occurrence interval of (u) i Belongs to { [0, Δ { ]],(Δ,2Δ],...,((n-1)Δ,nΔ],(T,+∞)}。
And then the failure probability of each component when the system fails can be obtained by using a reverse reasoning method.
The rate parameter lambda of exponential distribution approximate to the fault distribution function of the hydraulic end of the drilling pump is in an interval (0.1 multiplied by 10) -6 h -1 ,4.5×10 -6 h -1 ) Therefore, the time division number n is 4. When the system fails, that is, when the leaf node state is 5, the failure probability of each bottom event is calculated by using the reverse reasoning of the bayesian network, as shown in table 2.
TABLE 2 probability table of failure probability of each bottom event when top event occurs
Event code | Probability of failure | Event code | Probability of failure |
X1 | 0.0231 | X9 | 0.0561 |
X2 | 0.0261 | X10 | 0.0472 |
X3 | 0.1090 | X11 | 0.0842 |
X4 | 0.0354 | X12 | 0.0233 |
X5 | 0.0261 | X13 | 0.0463 |
X6 | 0.0116 | X14 | 0.0321 |
X7 | 0.1562 | X15 | 0.0569 |
X8 | 0.1008 | X16 | 0.0103 |
From this table, the probability of failure of X16 is the smallest, while the probability of failure of X7, X3, X8 is the largest, so the probability of failure of piston nut loosening, valve wear is severe, and cylinder intake air is high.
Claims (3)
1. A drilling pump hydraulic end fault diagnosis method based on a dynamic fault tree comprises the following steps:
step 1: establishing a dynamic fault tree of a drilling pump hydraulic system by adopting a deductive reasoning method:
step 1.1: taking the fault of a hydraulic end of a drilling pump as a top event of a dynamic fault tree, taking the fault which is easy to occur in a sub-component in a main component in the hydraulic end as an intermediate event, and taking the fault of a component in the sub-component as a bottom event;
step 1.2: according to the working principle and the failure occurrence mechanism of the hydraulic end of the drilling pump, connecting each level of events with a top event by using a corresponding dynamic logic gate, and connecting each level of events from a bottom event to the top event by using the dynamic logic gate to obtain a dynamic failure tree of the hydraulic end of the drilling pump;
step 2: converting the dynamic fault tree into a Bayesian network model, and establishing a corresponding relation of a rate parameter lambda-a time division number n;
step 2.1: converting the dynamic fault tree into a Bayesian network model; the top event, the middle event and the bottom event of the dynamic fault tree respectively correspond to a root node, a middle node and a leaf node of a Bayesian network model, and the expression method of each logic gate in the dynamic fault tree in the Bayesian network model comprises the following steps:
and gate:
let X = [ X = 1 ,X 2 ,…,X m ]Where m is the number of input events of the AND gate, X i I =1,2, …, m is the state variable of the input event, and the number of state combinations is (n + 1) m (ii) a Let Y be the state variable output by AND gate, the state space of all variables is {1,2, …, n +1}, let k = max (X) 1 ,X 2 ,…,X m ) J is a judgment fixed value; andthe failure mechanism of the gate is that all input events occur and then output events occur, and then the output events should be at the maximum of the state values of all events, so in any state combination of input events, the conditional probability distribution of Y is:
or gate:
the failure mechanism of the OR gate is that the output event occurs as long as one element occurs in the input event, so that the state of the output event of the OR gate is the same as the minimum value of the state in the input event; let r = min (X) 1 ,X 2 ,…,X m ) Then the conditional probability distribution of the or gate output event is:
priority and gate:
assuming that an input event is A, B, an output event is Y, state values are a, B and Y respectively, and the failure mechanism of the priority AND gate is that when A and B both fail, A fails before B, and the event Y occurs, namely when a is less than B, the output event Y is in a state B; otherwise, the output event Y is in a state n +1, which represents that Y does not have a fault; the conditional probability distribution of the priority and gate output event Y is as follows:
when a is less than b, the first and second groups,
when a is more than or equal to b,
sequentially closing the doors:
the order dependent gate forces its input events to occur in a particular order, but not to fail in other orders; sequential door closing, like priority door closing, represents the timing of basic events, which differ in that: input events in sequential closing of the doors cannot fail in any order; the priority AND gate can fail in any sequence, and the failure of the output event can be triggered only by the failure in a specific sequence; therefore, the sequence correlation gate can be changed into a cascade form of a plurality of priority gates;
the functions are closed:
setting that only one relevant input event A exists when the function phase is closed, the triggering event is Tr, and the output event is T; when Tr occurs, the related event A occurs, and the output event T occurs; when A independently fails, an output event T also occurs; adding an intermediate node A' between the input event and the output event to represent the total failure event of A failure caused by Tr triggering or independent failure of A itself, wherein p and q are corresponding event states, the conditional probability distribution of the output event of the function-related gate is as follows:
step 2.2: setting the value range of n, and selecting different rate parameters lambda to calculate the reliability of the system by adopting a plurality of groups of data acquired in advance; comparing the maximum difference ratio of the system reliability among all groups of data to obtain the most appropriate time division number n which should be selected by the rate parameter lambda in different ranges, and establishing the corresponding relation of the rate parameter lambda-the division number n;
and 3, step 3: defining the state of nodes in the Bayesian network model;
dividing a time interval [0,t ] formed by the whole task into n subintervals with equal length, wherein the length of each subinterval is delta = t/n, and then the whole time axis [0, + ∞ ] is divided into n +1 subintervals; when a part corresponding to a certain node A fails in the ith time interval within the task time t, namely A fails in the time interval [ (i-1) delta, i delta ], the node A is in a state i; if A is not expired within the task time t, i.e. A is expired within [ t, ∞), then A is said to be in state n +1;
by the above definition, a time interval in which the state spaces of all nodes in the bayesian network are as follows is obtained:
[0, delta ], (Delta, 2 Delta ], …, ((n-1) Delta, n Delta ], (n Delta, plus infinity), abbreviated as {1,2,. The, n +1}, the failure time X of the system and the components always corresponds to a certain interval i of n +1 intervals, the sum of the probabilities of the system in the first n states is the failure rate of the system at the task time t, and the probability of the system in the n +1 th state is the non-failure rate of the system at the task time t;
and 4, step 4: establishing probability distribution of all nodes according to the states of the nodes in the Bayesian network model, and completing quantitative calculation of dynamic fault tree fault diagnosis based on the Bayesian network;
and 5: obtaining a fault diagnosis result through reverse reasoning;
the conditional probability distribution of each node is denoted as P (X) i |pa(X i ) To express a quantitative relationship between a child node and a parent node; under the condition of giving prior probability distribution of root nodes and conditional probability distribution of non-root nodes, obtaining joint probability distribution containing all nodes, and further performing marginalization calculation on target nodes to obtain marginal probability distribution; and finally, calculating the fault probability of each bottom event by utilizing the reverse reasoning of the Bayesian network, and outputting a diagnosis result according to the magnitude of the fault probability.
2. The method for diagnosing the hydraulic end fault of the drilling pump based on the dynamic fault tree as claimed in claim 1, wherein the top event in the step 1.1 is the hydraulic end fault, and the middle events are: suction pipe failure, valve failure, piston failure, hydraulic cylinder failure, discharge pipe failure, air bag failure, safety valve failure, the end event is: the suction pipeline is not tightly sealed, the suction filter screen is blocked, the valve is seriously abraded, the valve cover is not tightened, the guide sleeve is blocked, the piston is seriously abraded, the piston nut is loosened, the hydraulic cylinder enters air, the cylinder cover is not tightened, the cylinder cover gland is loosened, the discharge filter cylinder is blocked, the discharge pipeline is blocked, the inflation connector is blocked, the air bag inner bag is broken, the stop valve is not tightly sealed, and the safety valve is improperly arranged;
the relationship between events in step 1.2 is:
the suction pipeline is not tightly sealed, and the suction filter screen is blocked and is connected with the suction pipe through a door preferentially; the valve is seriously abraded, the valve cover is not tightened, and the guide sleeve is stuck to be connected with the valve through an OR gate; the piston is seriously worn, and the piston nut is loosened and is connected with the piston through an OR gate; the hydraulic cylinder enters air, the cylinder cover is not tightened, and the cylinder cover gland is loosened and is connected with the hydraulic cylinder through sequential door closing; the blockage of the discharge filter cylinder and the blockage of the discharge pipeline are connected with the fault of the discharge pipe through an OR gate; the blockage of the inflation connector, the breakage of the air bag inner bag and the untight sealing of the stop valve are connected with the failure of the air bag through the functional phase door closing, the blockage of the inflation connector and the untight sealing of the stop valve are input of relevant events of the functional phase door closing, the breakage of the air bag inner bag is input of a trigger event of the functional phase door closing, and the failure of the air bag is output of the functional phase door closing; the safety valve is connected with the safety valve through the gate when the safety valve is improperly arranged.
3. The method for diagnosing the fault of the hydraulic end of the drilling pump based on the dynamic fault tree as claimed in claim 1, wherein the concrete method of the step 5 is as follows:
event U for non-leaf node in discrete time Bayesian network DTBN containing m nodes i Wherein 1. Ltoreq. I.ltoreq.m-1,U i Has an occurrence interval of { [0, Δ ]],(Δ,2Δ],...,((n-1)Δ,nΔ](T, + ∞) }; if the top event U T If the event occurs within the task time T, the occurrence time of the top event must be [0, delta ]],(Δ,2Δ],...,((n-1)Δ,nΔ]Within one of the intervals (n Δ, + ∞); thus, U T The probability of occurrence within the task time T can be directly calculated and expressed as:
in the formula u i Represents U i The occurrence interval of (u) i Belongs to { [0, Δ { ]],(Δ,2Δ],...,((n-1)Δ,nΔ],(T,+∞)};
And then the failure probability of each component when the system fails can be obtained by using a reverse reasoning method.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011178173.4A CN112270128B (en) | 2020-10-29 | 2020-10-29 | Dynamic fault tree-based drilling pump hydraulic end fault diagnosis method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011178173.4A CN112270128B (en) | 2020-10-29 | 2020-10-29 | Dynamic fault tree-based drilling pump hydraulic end fault diagnosis method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN112270128A CN112270128A (en) | 2021-01-26 |
CN112270128B true CN112270128B (en) | 2022-10-11 |
Family
ID=74344882
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202011178173.4A Active CN112270128B (en) | 2020-10-29 | 2020-10-29 | Dynamic fault tree-based drilling pump hydraulic end fault diagnosis method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112270128B (en) |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114925865A (en) * | 2022-06-08 | 2022-08-19 | 山西科达自控股份有限公司 | Fault diagnosis method for electromechanical braking system of mine electric locomotive and storage medium |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6336108B1 (en) * | 1997-12-04 | 2002-01-01 | Microsoft Corporation | Speech recognition with mixtures of bayesian networks |
CN104063586A (en) * | 2014-06-11 | 2014-09-24 | 西北工业大学 | Polymorphic failure tree-based bayesian network failure prediction method |
CN106401597A (en) * | 2016-10-27 | 2017-02-15 | 华中科技大学 | Failure prediction and diagnosis control method applicable to shield tunneling machine |
CN110442833A (en) * | 2019-06-10 | 2019-11-12 | 内蒙古工业大学 | One kind assessing Wind turbines health state evaluation method based on various dimensions SCADA data |
CN111275195A (en) * | 2020-02-13 | 2020-06-12 | 辽宁石油化工大学 | Dynamic Bayesian network modeling method based on coal gasification equipment |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8832497B2 (en) * | 2012-02-07 | 2014-09-09 | A.L.D. Advanced Logistics Development Ltd | Methods, apparatus and systems for performing dynamic fault tree analysis |
-
2020
- 2020-10-29 CN CN202011178173.4A patent/CN112270128B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6336108B1 (en) * | 1997-12-04 | 2002-01-01 | Microsoft Corporation | Speech recognition with mixtures of bayesian networks |
CN104063586A (en) * | 2014-06-11 | 2014-09-24 | 西北工业大学 | Polymorphic failure tree-based bayesian network failure prediction method |
CN106401597A (en) * | 2016-10-27 | 2017-02-15 | 华中科技大学 | Failure prediction and diagnosis control method applicable to shield tunneling machine |
CN110442833A (en) * | 2019-06-10 | 2019-11-12 | 内蒙古工业大学 | One kind assessing Wind turbines health state evaluation method based on various dimensions SCADA data |
CN111275195A (en) * | 2020-02-13 | 2020-06-12 | 辽宁石油化工大学 | Dynamic Bayesian network modeling method based on coal gasification equipment |
Non-Patent Citations (3)
Title |
---|
Dynamic Fault-tree Models for Fault-Tolerant Computer Systems;Joanne Bechta Dugan等;《IEEE TRANSACTIONS ON RELIABILITY》;19920930;第41卷(第3期);363-377 * |
基于动态故障树的钻井泵故障诊断专家系统的设计与实现;何旋;《中国优秀硕士学位论文全文数据库 工程科技Ⅰ辑》;20220115(第(2022)01期);B019-160 * |
复杂系统动态故障树分析的新方法及其应用研究;李彦锋;《中国博士学位论文全文数据库 工程科技Ⅱ辑》;20140615(第(2014)06期);C029-27 * |
Also Published As
Publication number | Publication date |
---|---|
CN112270128A (en) | 2021-01-26 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN112270128B (en) | Dynamic fault tree-based drilling pump hydraulic end fault diagnosis method | |
Deng et al. | A remaining useful life prediction method with long-short term feature processing for aircraft engines | |
CN110018670A (en) | A kind of industrial process unusual service condition prediction technique excavated based on dynamic association rules | |
CN108764551A (en) | Group preventive maintenance decision-making technique based on system-level life information | |
CN108491860A (en) | Shale gas pressure break equips Life cycle reliability in time dynamic assessment method and device | |
Shiozaki et al. | Fault diagnosis of chemical processes utilizing signed directed graphs-improvement by using temporal information | |
CN105043776A (en) | Aircraft engine performance monitoring and fault diagnosis method | |
CN114925865A (en) | Fault diagnosis method for electromechanical braking system of mine electric locomotive and storage medium | |
CN110333694B (en) | Fuzzy Petri network-based numerical control equipment fault diagnosis method | |
CN105629959A (en) | Industrial process fault detection method | |
Zhang et al. | Trend-augmented and temporal-featured Transformer network with multi-sensor signals for remaining useful life prediction | |
CN113029619A (en) | Underground scraper fault diagnosis method based on C4.5 decision tree algorithm | |
Kamal | An approach to the diagnosis of intermittent faults | |
CN116388769B (en) | Computer storage optimization method for environment detection data | |
Kenyon et al. | On evaluating boolean functions with unreliable tests | |
CN105718240A (en) | Any input signal bit stream adder based on sigma-delta modulation | |
CN117236428A (en) | Bayesian network-based fault diagnosis method for complex system in nuclear power plant | |
CN108694293B (en) | Active modeling and prediction method for flow rate of reciprocating type mixed transportation pump | |
Bobbio et al. | Reliability analysis of systems with dynamic dependencies | |
Mojarad et al. | Two effective anomaly correction methods in embedded systems | |
JP7375564B2 (en) | Vehicle information storage device | |
Wentao et al. | Fault diagnosis for steam turbine based on flow graphs and naïve Bayesian classifier | |
Beutner et al. | Consistent semiparametric estimators for recurrent event times models with application to virtual age models | |
CN117592571B (en) | Air conditioning unit fault type diagnosis method and system based on big data | |
CN111008438A (en) | Reliability analysis method and device for multi-stage system of umbilical cable |
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 |