CN114444582A - Mechanical equipment fault diagnosis method based on convolutional neural network and Bayesian network - Google Patents

Mechanical equipment fault diagnosis method based on convolutional neural network and Bayesian network Download PDF

Info

Publication number
CN114444582A
CN114444582A CN202210014436.0A CN202210014436A CN114444582A CN 114444582 A CN114444582 A CN 114444582A CN 202210014436 A CN202210014436 A CN 202210014436A CN 114444582 A CN114444582 A CN 114444582A
Authority
CN
China
Prior art keywords
fault
reason
bayesian network
convolutional neural
probability
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.)
Pending
Application number
CN202210014436.0A
Other languages
Chinese (zh)
Inventor
丁华
孙晓春
王焱
牛锐祥
吕彦宝
孟祥龙
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Taiyuan University of Technology
Original Assignee
Taiyuan University of Technology
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Taiyuan University of Technology filed Critical Taiyuan University of Technology
Priority to CN202210014436.0A priority Critical patent/CN114444582A/en
Publication of CN114444582A publication Critical patent/CN114444582A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/29—Graphical models, e.g. Bayesian networks
    • 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
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Biology (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Biomedical Technology (AREA)
  • Medical Informatics (AREA)
  • Computer Hardware Design (AREA)
  • Geometry (AREA)
  • Testing And Monitoring For Control Systems (AREA)

Abstract

The invention relates to the technical field of fault diagnosis, in particular to a mechanical equipment fault diagnosis method based on a convolutional neural network and a Bayesian network. The method comprises the following steps: s1: monitoring various parameters of the emulsion pump through a sensor; s2: collecting various monitoring data of the emulsion pump when various faults occur, and carrying out normalization and standardization preprocessing on the monitoring data to obtain a trained emulsion pump fault diagnosis model; s3: building a diagnostic Bayesian network comprising a fault layer and a reason layer; s4: embedding an emulsion pump fault diagnosis model formed by the trained lightweight convolutional neural network and the diagnosis Bayesian network into a cloud platform; s5: and monitoring all data of the emulsion pump in operation in real time, inputting the data into the cloud platform, giving out early warning if a fault is diagnosed, and inputting a diagnosis result into the Bayesian network to carry out conditional probability reasoning, so that the probability of each fault reason is obtained, and fault positioning is realized.

Description

Mechanical equipment fault diagnosis method based on convolutional neural network and Bayesian network
Technical Field
The invention relates to the technical field of fault diagnosis, in particular to a mechanical equipment fault diagnosis method based on a convolutional neural network and a Bayesian network.
Background
Coal is an important basic energy in China, and is predicted according to relevant reports of China engineering institute: by 2030, the coal accounts for 50% of the energy structure in China; until 2050, the leader was still high at 40% of the percent. Safe and efficient production of coal mines is one of important foundations for ensuring economic development and social stability of China. At present, coal is still the main energy source in China, the underground environment of a coal mine is complex and severe, so that the failure of coal mine equipment is frequent, the coal mining safety is seriously threatened, and the economic income of the coal mine is seriously lost.
In the daily production process of the fully mechanized coal mining face of a coal mine, various mechanical equipment play a key role, but due to the complex underground working condition and the severe environment, the daily maintenance of the equipment consumes time and labor, and has high technical requirements on workers of the first-line equipment. When a fault occurs, rapid and accurate fault diagnosis cannot be achieved, fault location is difficult, maintenance time is greatly prolonged, long-time shutdown is caused, and coal mine economic loss is aggravated. At present, the research on the fault diagnosis method of the mechanical equipment is less, and the parameters of the mechanical equipment are monitored on line mostly through a PLC (programmable logic controller) and the like, so that accurate prevention and diagnosis cannot be carried out. Most of mechanical equipment has complex fault mechanisms including various fault phenomena, fault coupling, inconsistent generation reasons, difficult fault positioning and the like, and the traditional fault diagnosis method cannot be competent for the fault diagnosis task of the mechanical equipment.
Therefore, the method for diagnosing the fault of the mechanical equipment based on the convolutional neural network and the Bayesian network has the advantages that the operation data of the mechanical equipment are monitored and diagnosed on line, the operation state of the mechanical equipment is analyzed and judged, reference is provided for daily maintenance of the mechanical equipment, when the fault occurs, the mechanical equipment can be diagnosed quickly and accurately by the aid of the diagnostic Bayesian network, the cause of the fault can be analyzed, and the method has important significance in practical application.
Disclosure of Invention
The invention provides a method for diagnosing the fault of mechanical equipment based on a convolutional neural network and a Bayesian network, which monitors various real-time data of the mechanical equipment in actual operation through a lightweight convolutional neural network, diagnoses in real time, can give an alarm in time when the fault occurs, and utilizes the Bayesian network to reason the fault reason.
The invention adopts the following technical scheme: a mechanical equipment fault diagnosis method based on a convolutional neural network and a Bayesian network comprises the following steps: s1: monitoring various parameters of the emulsion pump through a sensor; s2: collecting various monitoring data of the emulsion pump when various faults occur, carrying out normalization and standardization preprocessing on the monitoring data, and training a lightweight convolutional neural network model by utilizing the preprocessed various monitoring data to obtain a trained emulsion pump fault diagnosis model; s3: building a diagnostic Bayesian network comprising a fault layer and a reason layer; s4: embedding an emulsion pump fault diagnosis model formed by the trained lightweight convolutional neural network and the diagnosis Bayesian network into a cloud platform; s5: and monitoring all data of the emulsion pump in operation in real time, inputting the data into the cloud platform, giving out early warning if a fault is diagnosed, and inputting a diagnosis result into the Bayesian network to carry out conditional probability reasoning, so that the probability of each fault reason is obtained, and fault positioning is realized.
The step S2 takes the following approach,
s21: the method comprises the following steps of taking 100 samples in a normal state and each fault mode, wherein each sample comprises monitoring parameter data and a corresponding label, the monitoring data in each sample are collected according to a sampling frequency of 100Hz, the collection time is 1 minute, and finally all samples are divided into a training set, a verification set and a test set according to a ratio of 6:2: 2;
s22: inputting the divided training sets into a built lightweight convolutional neural network initial model for training, verifying by using a verification set, inputting the test set into the model for testing, and verifying whether an output result conforms to an actual fault type; in the training process, an Adam learning rate self-adaptive algorithm is adopted, an optimal model is found through a cross entropy loss function, and parameters and a structure of the optimal model are stored, so that a well-trained emulsion pump fault diagnosis model is obtained.
The initial model of the lightweight convolutional neural network comprises the following steps: a two-dimensional separable convolutional layer, a two-dimensional maximum pooling layer, a global average pooling layer, and a Softmax classifier.
The step S3 takes the following approach,
s31: the fault layer is formed by fault modes of an emulsion pump, and the reason layer is formed by reasons causing each fault;
s32: collecting fault maintenance records of the emulsion pump within a period of time, and finding 50 cases of each fault form according to the records, wherein each case has a corresponding specific reason;
s33: determining the connection relation between a failure layer and a reason layer of a diagnostic Bayesian network structure according to the corresponding relation between the failure form and the reason of each case, thereby determining the structure of the Bayesian network;
s34: the conditional probability between the fault node and the reason node in the Bayesian network structure and the prior probability of each fault are determined by combining the established Bayesian network structure and the collected case data, and finally a conditional probability table of the Bayesian network is formed.
Step S34 the conditional probability table is determined in detail by first screening the relevant data of the cause of a failure from the cases collected in a failure mode, and then carrying out quantitative analysis to determine the prior probability P of various causesA priori(ii) a Secondly, counting the probability of faults caused by various reasons, and taking the probability as the conditional probability of the faults and the reasons; by analogy, each fault is processed according to the steps so as to determine the conditional probability table P for diagnosing the Bayesian networkCondition。
The specific process of the bayesian network performing inference of conditional probabilities in step S5 is,
s51: the probability of occurrence of a fault due to various reasons is calculated by the following formula,
Figure 100002_DEST_PATH_IMAGE002
in the formula PReasonProbability of causing the cause of the fault, PA prioriA priori probability for various reasons, PConditionConditional probability tables, P, for diagnosing Bayesian networksFault ofIs the probability of failure; will be P in various causesReasonThe fault form with the maximum probability value is used as a diagnosis conclusion;
s52: according to the fault form diagnosed by the convolutional neural network as the input of the diagnostic bayesian network model, the following three situations occur:
1. if the fault is found to be caused by the reason corresponding to the node through overhauling and the reason is located in the bottom-level event in the fault tree, indicating that the reason is the reason causing the fault;
2. if the overhaul result shows that the fault is caused by the reason corresponding to the node and the reason is the middle-level event of the fault tree, the fault tree is indicated to be caused by the bottom-level event, and further reasoning is needed; setting the node as 100 percent occurrence, continuously carrying out deep reasoning, finding out a father node with the maximum posterior probability and carrying out maintenance again;
3. if the overhaul result shows that the fault is not caused by the reason corresponding to the node, and the reason represented by the node does not occur, the node needs to be set to be 100% and does not occur, and the node is used as evidence input to continue reasoning, and a father node with the maximum posterior probability is found out to overhaul; and repeating the related steps corresponding to the maintenance result until the cause of the fault is maintained.
In step S1, the parameters of the emulsion pump include: motor current, motor torque, motor speed, emulsion pump system pressure, emulsion flow, emulsion pump vibration, emulsion concentration, emulsion temperature, emulsion level, lubricant temperature, fuel injection pressure, winding temperature, bearing temperature, tank level, and acoustic signals.
In step S2, the failure mode of the emulsion pump includes: the pump can not be started, the pressure or the pressure can not be increased after the pump is started, the pressure pulsation large flow is insufficient or no flow exists, the temperature of a crankcase is overhigh, the pressure of a system can not be adjusted and slowly drops, the pressure of the pump suddenly rises to exceed the set pressure of the unloading valve, the noise is large during the operation of the pump, the unloading valve is frequently operated when a support stops supplying liquid, the pump does not discharge liquid during the operation, the proportion concentration is not adjusted high, and the motor fails in 11 fault modes.
Compared with the prior art, the method has the advantages that the convolutional neural network and the Bayesian network are combined, respective advantages are fully exerted, separable convolutions adopted by the convolutional layers reduce a large number of parameters compared with common convolutions, the diagnosis efficiency of the model is greatly improved, and input data are fused with multi-source data, so that the diagnosis accuracy can be improved. Meanwhile, global average pooling is adopted in the last pooling layer part, and compared with a full-connection layer, parameter explosion of the full-connection layer is avoided, so that the problem of overfitting of a model is caused. And after the fault diagnosis, the Bayesian network is used for carrying out uncertain reasoning on the fault reasons, and compared with an expert system, the system not only can reason out all possible fault reasons, but also can obtain the probability of each reason.
Drawings
FIG. 1 is a flow chart of the present invention;
FIG. 2 is a flow chart of the fusion of multi-source monitoring data according to the present invention (taking an emulsion pump as an example);
FIG. 3 is a schematic structural diagram of a conventional convolutional layer;
FIG. 4 is a schematic diagram of a separable convolutional layer structure in accordance with an embodiment of the present invention;
FIG. 5 is a diagram of a lightweight convolutional neural network architecture provided by the present invention;
FIG. 6 is a flow chart of a lightweight convolutional neural network training and testing provided by the present invention;
FIG. 7 is a diagram of a diagnostic Bayesian network model architecture (taking an emulsion pump as an example) provided by the present invention;
FIG. 8 is a flow chart of a diagnostic Bayesian network model provided by the present invention.
Detailed Description
The present invention will be described in detail with reference to specific examples. The following examples will assist those skilled in the art in further understanding the invention, but are not intended to limit the invention in any way. It should be noted that it would be obvious to those skilled in the art that various changes and modifications can be made without departing from the spirit of the invention. All falling within the scope of the present invention.
Referring to fig. 1, the present embodiment discloses a method for diagnosing a fault of a mechanical device based on a convolutional neural network and a bayesian network, which is described in detail below by taking an emulsion pump as an example, and mainly includes the following steps:
s1, in order to comprehensively and accurately diagnose the faults of the emulsion pump, various parameters of the emulsion pump are monitored through various sensors by utilizing a mode of fusing multiple monitoring parameters.
S2, collecting various monitoring data of the emulsion pump when various faults occur, wherein the monitoring data are collected manually, and when the emulsion pump is started and has no pressure, a certain amount of data are collected; when the temperature of the crankcase of the emulsion pump is too high, a certain amount of data is collected; when the pressure of the emulsion pump system is not adjusted up and slowly drops, a certain amount of data is also collected, and the like, and the normal state is added to the data to obtain 12 fault forms.
For data acquisition, as shown in table 1, 100 samples including normal state are taken for each fault form, and a total of 1200 samples, where a and b.. M in the table are 16 monitoring parameters, each sample includes the 16 monitoring parameters, and the monitoring data in each sample is acquired according to a sampling frequency of 100Hz for 1 minute, and finally, a matrix formed by time series data of each sample becomes an input of the convolutional neural network model.
TABLE 1
Figure DEST_PATH_IMAGE004
The collected data are subjected to normalization and standardization preprocessing, a training set, a verification set and a test set are set according to the number of the collected data sets and the ratio of 6:2:2, an initial lightweight convolutional neural network model is trained by the preprocessed training set and verified by the verification set, overfitting of the model is prevented, then the test set is input into the model for testing, and whether the output result is consistent with the actual fault type is verified. In the training process, an Adam learning rate self-adaptive algorithm is adopted, an optimal model is found through a cross entropy loss function, and parameters and a structure of the optimal model are stored, so that an emulsion pump fault diagnosis model with a good effect is obtained.
As shown in fig. 5, the initial model of the lightweight convolutional neural network includes: a two-dimensional separable convolutional layer, a two-dimensional maximum pooling layer, a global average pooling layer, and a Softmax classifier. Setting the number of the convolutional layers, the pooling layers and the global average pooling layer in the convolutional neural network model, the size and the number of the convolutional cores, the step length of operation and other hyper-parameters, selecting proper model training parameters including training batches, batch sizes, initial learning rate and the like, and training by using the data in S2.
And (3) rolling layers: the convolutional layer performs a convolution operation on a local region of the input signal using a convolution kernel and generates a corresponding feature. The convolution layer has the characteristic of weight sharing, namely, the same convolution kernel traverses once input with fixed step length. The invention selects separable convolution which can be regarded as dividing common convolution into two parts: the spatial convolution and the channel convolution, as shown in FIG. 4, first pass
Figure DEST_PATH_IMAGE006
Is spatially convolved and then passed through
Figure DEST_PATH_IMAGE008
The convolution kernel of (a) performs channel convolution and finally outputs the same result as the ordinary convolution of fig. 3. By calculating the number of parameters of the ordinary convolution and the separable convolution, the parameter quantity of the ordinary convolution can be obtained as follows:
Figure DEST_PATH_IMAGE010
（1）
the parameters of the separable convolution are:
Figure DEST_PATH_IMAGE012
（2）
in the formulae (1) and (2),
Figure DEST_PATH_IMAGE014
in order to input the size of the layer,
Figure DEST_PATH_IMAGE016
is the size of the convolution layer, and is,
Figure DEST_PATH_IMAGE018
in order to input the number of characteristic channels,
Figure DEST_PATH_IMAGE020
is the number of convolution kernels.
By comparing the parameter quantities of the separable convolution with the normal convolution:
Figure DEST_PATH_IMAGE022
（3）
it can be found that: compared with the common convolution, the separable convolution can greatly reduce parameters for the same input layer and output layer, thereby shortening the training time and improving the training efficiency.
In the calculation of convolution layer, the convolution adopts the input matrix and the corresponding point in convolution kernel to make dot multiplication, then summing, then adding an offset, and the calculation formula is:
Figure DEST_PATH_IMAGE024
（4）
where K represents the number of channels, M is the number of rows of convolution kernels per channel, and N is the number of columns of convolution kernels per channel. y isnRepresenting the convolution output result, bnRepresenting the offset in the linear calculation, ai,j,kAs weighting coefficients in linear operations, xi,j,kRepresenting the characteristic element value of the original input or the output result of the convolution layer on the previous layer.
A pooling layer: the invention selects the largest pooling layer as the mainThe method has the functions of downsampling, dimensionality reduction, redundant information removal, feature compression, network complexity simplification, calculation amount reduction, memory consumption reduction, and has the characteristics of realizing nonlinearity, expanding a perception field, realizing invariance (translation invariance, rotation invariance and scale invariance) and the like. For data T e
Figure DEST_PATH_IMAGE026
And after pooling, outputting:
Figure DEST_PATH_IMAGE028
（5）
in the formula, n: the part of the input vector that is divided,
Figure DEST_PATH_IMAGE030
the ith feature tensor is expressed, W: size of pooling window, S: step size.
Global Average Pooling (GAP): the reason for replacing the fully connected layer with the global average pooling layer after the convolutional layer is: the global average pooling is simpler and more natural to convert between the characteristic diagram and the final classification, and meanwhile, the global average pooling does not need a large amount of training and tuning parameters like a full connection layer, so that the spatial parameters are reduced, the model is more robust, and the over-fitting resisting effect is better.
A classifier layer: the activation function employed by the output layer is the Softmax function. By the action of the Softmax function, the nodes are mapped to (0,1) values, and the summation of the values is 1 (the property of probability is satisfied), so that the probability can be understood, and when the output node is selected finally, the node with the highest probability (namely, the node with the highest value) can be selected as the final prediction target.
Figure DEST_PATH_IMAGE032
（6）
Loss function in the invention, a cross entropy loss function is adopted:
Figure DEST_PATH_IMAGE034
（7）
in the formula: e is an objective function, n is the number of samples, y is the true value, and t is the predicted value.
And S3, constructing a diagnostic Bayesian network comprising a fault layer and a reason layer. And collecting historical records of historical reasons of the emulsion pump faults to make a statistical table, determining nodes of a fault layer and nodes of a reason layer, further establishing a connection relation between the fault layer and the reason layer, and determining a conditional probability table of the Bayesian network according to the collected historical records to complete the steps, namely completing the construction of the diagnostic Bayesian network.
The bayesian network is a probabilistic network, which is a graphical network based on probabilistic reasoning. The Bayesian network is a mathematical model based on probabilistic reasoning, the probabilistic reasoning is a process of acquiring other probabilistic information through information of some variables, and the Bayesian network (Bayesian network) based on probabilistic reasoning is proposed for solving the problems of uncertainty and incompleteness, has good advantages for solving the faults caused by the uncertainty and relevance of complex equipment, and is widely applied in multiple fields.
Wherein the step S3 further comprises:
s31, the failure layer is composed of 16 failure modes of the emulsion pump, and the reason layer is composed of the reasons causing each failure.
And S32, collecting fault maintenance records of the emulsion pump in a period of time, and finding 50 cases of each fault form according to the records, wherein the cases have specific corresponding reasons in 550 cases.
And S33, determining the connection relation between the failure layer and the reason layer of the diagnostic Bayesian network structure according to the corresponding relation between the failure form and the reason of each case, thereby determining the structure of the Bayesian network.
And S34, determining the conditional probability between the fault node and the reason node in the Bayesian network structure and the prior probability of the fault caused by each reason by combining the constructed Bayesian network structure and the collected case data, and finally forming a conditional probability table of the Bayesian network. For theThe detailed steps of the conditional probability table determination are: for example, for a fault that the pressure is not available or cannot be increased after the pump is started, relevant data of the cause of the fault are screened from collected cases, quantitative analysis is carried out, and the prior probability P of various causes is determinedA priori(ii) a Secondly, counting the probability of fault occurrence caused by each reason, and taking the probability as the conditional probability of the fault and the reason; by analogy, each fault is processed according to the steps, thereby determining a conditional probability table P for diagnosing the Bayesian networkCondition。
And S4, embedding the emulsion pump fault diagnosis model formed by the trained lightweight convolutional neural network and the Bayesian network into a smart cloud platform of the coal mine fully mechanized mining face.
S5, monitoring all data of the emulsion pump in real time during operation, inputting the data into a smart cloud platform of the coal mine fully mechanized coal mining face, and continuing monitoring if normal diagnosis is made; if the fault is diagnosed, an early warning is sent out, the fault diagnosis result is input into the Bayesian network to carry out conditional probability reasoning, so that the probability of each fault reason is obtained, the reason corresponding to the node with the highest posterior probability value is used as the diagnosis result, fault positioning is realized, and quick maintenance is carried out by first-line equipment maintenance personnel.
Application to diagnosing bayesian networks. The failure mode diagnosed by the convolutional neural network is used as the input of the Bayesian network model for diagnosis, namely representing that the failure node occurs 100%. The method for calculating the probability of the occurrence of the fault caused by various reasons is shown in the formula 8, and the maximum probability value is taken as a diagnosis conclusion.
Figure DEST_PATH_IMAGE035
（8）
In the formula PReasonProbability of causing the cause of the fault, PA prioriA priori probability for various reasons, PConditions ofConditional probability tables, P, for diagnosing Bayesian networksFault ofIs the probability of failure.
The invention utilizes GeNIe2.0 software to diagnose the construction of Bayesian Network (BN). The 11 faults are used as child nodes of the diagnosis Bayesian network, and the reason of each fault is used as a parent node of the diagnosis Bayesian network. For example, a failure diagnosis bayesian network without or without pressure increase after the pump is started is shown in fig. 7, which has 17 nodes in total, and the structure diagram of other failure modes and causes is similar to this.
The specific reasoning process takes the fault that the pressure is not available or cannot be increased after the pump in fig. 7 is started as an example, when the bayesian network receives the fault form, the reason corresponding to the node with the maximum posterior probability value is taken as a diagnosis result, and the bayesian network is overhauled, the following three conditions will occur:
1. and finding that the fault is caused by the reason corresponding to the node through overhauling, wherein the reason is located in the bottom-level event in the fault tree, and the reason is the reason causing the fault.
2. If the overhaul result shows that the fault is caused by the reason corresponding to the node and the reason is the middle-level event of the fault tree, the fault tree is indicated to be caused by the bottom-level event, and further reasoning is needed; and setting the node as 100 percent, continuously carrying out deep reasoning, and finding out the father node with the maximum posterior probability for overhauling again.
3. And if the overhaul result shows that the fault is not caused by the reason corresponding to the node, which indicates that the reason represented by the node does not occur, setting the node as 100% to not occur, inputting the evidence to continue reasoning, and finding out the parent node with the maximum posterior probability for overhaul.
And repeating the related steps corresponding to the maintenance result until the cause of the fault is maintained.
Taking the convolution neural network as an example of a fault that the pressure is not available or the pressure cannot be increased after the emulsion pump is started, as shown in fig. 7, firstly, the output of the convolution neural network is used as the input of the bayesian network, that is, the node of "no pressure or pressure cannot be increased after the pump is started" is set to be 100%; the attributes of each node in the BN are defined next. "Observations" means observation nodes, typically representing overhead events, accident types, fault types, etc.; "Targets" means target nodes, generally representing nodes for which a posterior probability is desired by evidence input. The node of "no pressure or no pressure rise after the pump is started" is defined as "occurrences", and the other nodes are defined as "Targets"; and finally, selecting the Cluster Algorithm (combined tree Algorithm) in GeNIe2.0 software, updating BN, finishing the inference result, obtaining the posterior probability of each target node, comparing the posterior probabilities of the target nodes, and supposing that the 'main valve fault' with the maximum posterior probability value is selected as a diagnosis result for maintenance. The overhaul results show that there is indeed a main valve failure, but since it corresponds to an intermediate event of the fault tree, whose occurrence must be caused by the underlying event, further diagnostics are still needed. And inputting the main valve fault as known information into the diagnostic Bayesian network, namely setting a 'main valve fault' node as 100% occurrence, updating BN again, and if 'main valve blockage' with the maximum posterior probability value is selected as a diagnostic result, carrying out maintenance, wherein the maintenance result shows that the main valve blockage really exists and the fault diagnosis is finished because the main valve blockage is positioned in a bottom layer event in an accident tree. And (4) maintaining according to the diagnosed fault reason, so that the normal work of the equipment can be ensured.
According to the main principle of the method, firstly, a lightweight convolutional neural network model is used for identifying the fault type, so that the powerful data feature extraction capability of the Convolutional Neural Network (CNN) is exerted, the parameters are reduced by using the separable convolutional layer, and the training efficiency of the model is improved; then, a global average pooling layer is adopted in the diagnosis model, so that parameters are reduced, and overfitting of the model is avoided; finally, the diagnosis Bayesian network is used for carrying out uncertainty reasoning on the fault reasons to obtain the probability of each fault reason, and compared with an expert system, the diagnosis efficiency is improved, the interference of human factors is reduced, and the intelligent level of diagnosis is integrally improved.
The foregoing description of specific embodiments of the present invention has been presented. It is to be understood that the present invention is not limited to the specific embodiments described above, and that various changes or modifications may be made by one skilled in the art within the scope of the appended claims without departing from the spirit of the invention. The embodiments and features of the embodiments of the present application may be combined with each other arbitrarily without conflict.

Claims (8)

1. A mechanical equipment fault diagnosis method based on a convolutional neural network and a Bayesian network is characterized in that: comprises the following steps of (a) carrying out,
s1: monitoring various parameters of the emulsion pump through a sensor;
s2: collecting various monitoring data of the emulsion pump when various faults occur, carrying out normalization and standardization preprocessing on the monitoring data, and training a lightweight convolutional neural network model by utilizing the preprocessed various monitoring data to obtain a trained emulsion pump fault diagnosis model;
s3: building a diagnostic Bayesian network comprising a fault layer and a reason layer;
s4: embedding an emulsion pump fault diagnosis model formed by the trained lightweight convolutional neural network and the diagnosis Bayesian network into a cloud platform;
s5: and monitoring all data of the emulsion pump in operation in real time, inputting the data into the lightweight convolutional neural network, giving out early warning if a fault is diagnosed, and inputting a diagnosis result into the Bayesian network to carry out conditional probability reasoning, so that the probability of each fault reason is obtained, and fault positioning is realized.
2. The method for diagnosing the fault of the mechanical equipment based on the convolutional neural network and the Bayesian network as recited in claim 1, wherein: the step S2 takes the following approach,
s21: the method comprises the following steps of taking 100 samples in a normal state and each fault mode, wherein each sample comprises monitoring parameter data and a corresponding label, the monitoring data in each sample are collected according to a sampling frequency of 100Hz, the collection time is 1 minute, and finally all samples are divided into a training set, a verification set and a test set according to a ratio of 6:2: 2;
s22: inputting the divided training sets into a built lightweight convolutional neural network initial model for training, verifying by using a verification set, inputting the test set into the model for testing, and verifying whether an output result conforms to an actual fault type; in the training process, an Adam learning rate self-adaptive algorithm is adopted, an optimal model is found through a cross entropy loss function, and parameters and a structure of the optimal model are stored, so that a well-trained emulsion pump fault diagnosis model is obtained.
3. The method for diagnosing the fault of the mechanical equipment based on the convolutional neural network and the Bayesian network as recited in claim 2, wherein: the initial model of the lightweight convolutional neural network comprises the following steps: a two-dimensional separable convolutional layer, a two-dimensional maximum pooling layer, a global average pooling layer, and a Softmax classifier.
4. The method for diagnosing the fault of the mechanical equipment based on the convolutional neural network and the Bayesian network as recited in claim 3, wherein: the step S3 takes the following approach,
s31: the fault layer is formed by fault modes of an emulsion pump, and the reason layer is formed by reasons causing each fault;
s32: collecting fault maintenance records of the emulsion pump within a period of time, and finding 50 cases of each fault form according to the records, wherein each case has a corresponding specific reason;
s33: determining the connection relation between a failure layer and a reason layer of a diagnostic Bayesian network structure according to the corresponding relation between the failure form and the reason of each case, thereby determining the structure of the Bayesian network;
s34: the conditional probability between the fault node and the reason node in the Bayesian network structure and the prior probability of each fault are determined by combining the established Bayesian network structure and the collected case data, and finally a conditional probability table of the Bayesian network is formed.
5. The method for diagnosing the fault of the mechanical equipment based on the convolutional neural network and the Bayesian network as recited in claim 4, wherein: the detailed steps determined by the conditional probability table of step S34Firstly, screening out the related data of the cause causing the fault from the case collected in a fault form, carrying out quantitative analysis, and determining the prior probability P of various causesA priori(ii) a Secondly, counting the probability of faults caused by various reasons, and taking the probability as the conditional probability of the faults and the reasons; by analogy, each fault is processed according to the steps so as to determine the conditional probability table P for diagnosing the Bayesian networkConditions of。
6. The method for diagnosing the fault of the mechanical equipment based on the convolutional neural network and the Bayesian network as recited in claim 5, wherein: the specific process of the bayesian network performing conditional probability inference in step S5 is that,
s51: the probability of occurrence of a fault due to various reasons is calculated by the following formula,
Figure DEST_PATH_IMAGE002
in the formula PReasonProbability of causing the cause of the fault, PA prioriA priori probability for various reasons, PConditionConditional probability tables, P, for diagnosing Bayesian networksFault ofIs the probability of failure; will be P in various causesReasonThe fault form with the maximum probability value is used as a diagnosis conclusion;
s52: according to the fault form diagnosed by the convolutional neural network as the input of the diagnostic bayesian network model, the following three situations occur:
1. if the fault is found to be caused by the reason corresponding to the node through overhauling and the reason is located in the bottom-level event in the fault tree, indicating that the reason is the reason causing the fault;
2. if the overhaul result shows that the fault is caused by the reason corresponding to the node and the reason is the middle-level event of the fault tree, the fault tree is indicated to be caused by the bottom-level event, and further reasoning is needed; setting the node as 100 percent occurrence, continuously carrying out deep reasoning, finding out a father node with the maximum posterior probability and carrying out maintenance again;
3. if the overhaul result shows that the fault is not caused by the reason corresponding to the node, and the reason represented by the node does not occur, the node needs to be set to be 100% and does not occur, and the node is used as evidence input to continue reasoning, and a father node with the maximum posterior probability is found out to overhaul; and repeating the related steps corresponding to the maintenance result until the cause of the fault is maintained.
7. The method for diagnosing the fault of the mechanical equipment based on the convolutional neural network and the Bayesian network as recited in claim 6, wherein: in step S1, the parameters of the emulsion pump include: motor current, motor torque, motor speed, emulsion pump system pressure, emulsion flow, emulsion pump vibration, emulsion concentration, emulsion temperature, emulsion level, lubricant temperature, fuel injection pressure, winding temperature, bearing temperature, tank level, and acoustic signals.
8. The method for diagnosing the fault of the mechanical equipment based on the convolutional neural network and the Bayesian network as recited in claim 7, wherein: in step S2, the failure mode of the emulsion pump includes: the pump can not be started, the pressure or the pressure can not be increased after the pump is started, the pressure pulsation large flow is insufficient or no flow exists, the temperature of a crankcase is overhigh, the pressure of a system can not be adjusted and slowly drops, the pressure of the pump suddenly rises to exceed the set pressure of the unloading valve, the noise is large during the operation of the pump, the unloading valve is frequently operated when a support stops supplying liquid, the pump does not discharge liquid during the operation, the proportion concentration is not adjusted high, and the motor fails in 11 fault modes.
CN202210014436.0A 2022-01-07 2022-01-07 Mechanical equipment fault diagnosis method based on convolutional neural network and Bayesian network Pending CN114444582A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210014436.0A CN114444582A (en) 2022-01-07 2022-01-07 Mechanical equipment fault diagnosis method based on convolutional neural network and Bayesian network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210014436.0A CN114444582A (en) 2022-01-07 2022-01-07 Mechanical equipment fault diagnosis method based on convolutional neural network and Bayesian network

Publications (1)

Publication Number Publication Date
CN114444582A true CN114444582A (en) 2022-05-06

Family

ID=81366990

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210014436.0A Pending CN114444582A (en) 2022-01-07 2022-01-07 Mechanical equipment fault diagnosis method based on convolutional neural network and Bayesian network

Country Status (1)

Country Link
CN (1) CN114444582A (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115093190A (en) * 2022-07-29 2022-09-23 长兴贝斯德邦建材科技有限公司 Aerogel inorganic heat-insulating paste and intelligent production system thereof
CN116030063A (en) * 2023-03-30 2023-04-28 同心智医科技(北京)有限公司 Classification diagnosis system, method, electronic device and medium for MRI image
CN116028849A (en) * 2022-12-30 2023-04-28 西安重装智慧矿山工程技术有限公司 Emulsion pump fault diagnosis method based on depth self-coding network

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115093190A (en) * 2022-07-29 2022-09-23 长兴贝斯德邦建材科技有限公司 Aerogel inorganic heat-insulating paste and intelligent production system thereof
CN116028849A (en) * 2022-12-30 2023-04-28 西安重装智慧矿山工程技术有限公司 Emulsion pump fault diagnosis method based on depth self-coding network
CN116028849B (en) * 2022-12-30 2024-05-14 西安重装智慧矿山工程技术有限公司 Emulsion pump fault diagnosis method based on depth self-coding network
CN116030063A (en) * 2023-03-30 2023-04-28 同心智医科技(北京)有限公司 Classification diagnosis system, method, electronic device and medium for MRI image

Similar Documents

Publication Publication Date Title
CN114444582A (en) Mechanical equipment fault diagnosis method based on convolutional neural network and Bayesian network
US11840998B2 (en) Hydraulic turbine cavitation acoustic signal identification method based on big data machine learning
DE102022201761A1 (en) Method, system and storage medium for automatically diagnosing devices
CN101950382B (en) Method for optimal maintenance decision-making of hydraulic equipment with risk control
CN111340282B (en) DA-TCN-based method and system for estimating residual service life of equipment
Roemer et al. An overview of selected prognostic technologies with application to engine health management
CN110957015A (en) Missing value filling method for electronic medical record data
Mosallam et al. Component based data-driven prognostics for complex systems: Methodology and applications
Ayodeji et al. Causal augmented ConvNet: A temporal memory dilated convolution model for long-sequence time series prediction
CN110705812A (en) Industrial fault analysis system based on fuzzy neural network
Son et al. Deep learning-based anomaly detection to classify inaccurate data and damaged condition of a cable-stayed bridge
CN115791174B (en) Rolling bearing abnormality diagnosis method, system, electronic equipment and storage medium
CN116384224A (en) Aero-engine life prediction method based on conditional parameter dynamic convolutional neural network
CN115290326A (en) Rolling bearing fault intelligent diagnosis method
CN116383722A (en) Fracturing measure process monitoring method based on gate control circulation unit neural network
CN111126477A (en) Learning and reasoning method of hybrid Bayesian network
Chang et al. Temporal convolution-based sorting feature repeat-explore network combining with multi-band information for remaining useful life estimation of equipment
Amin et al. Fuzzy inference and fusion for health state diagnosis of hydraulic pumps and motors
CN114169091A (en) Method for establishing prediction model of residual life of engineering mechanical part and prediction method
Liu et al. Deep learning approach for sensor data prediction and sensor fault diagnosis in wind turbine blade
Bond et al. A hybrid learning approach to prognostics and health management applied to military ground vehicles using time-series and maintenance event data
Raptodimos et al. An artificial neural network approach for predicting the performance of ship machinery equipment
Zhang et al. A Flexible Monitoring Framework via Dynamic-Multilayer Graph Convolution Network
Dui et al. Reliability Evaluation and Prediction Method with Small Samples.
KR20230075150A (en) Method for managing system health

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