CN117435981A - Method and device for diagnosing operation faults of machine pump equipment, storage medium and electronic equipment - Google Patents
Method and device for diagnosing operation faults of machine pump equipment, storage medium and electronic equipment Download PDFInfo
- Publication number
- CN117435981A CN117435981A CN202311777537.4A CN202311777537A CN117435981A CN 117435981 A CN117435981 A CN 117435981A CN 202311777537 A CN202311777537 A CN 202311777537A CN 117435981 A CN117435981 A CN 117435981A
- Authority
- CN
- China
- Prior art keywords
- pump equipment
- model
- machine
- fault diagnosis
- training
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 56
- 238000003860 storage Methods 0.000 title claims abstract description 7
- 238000003745 diagnosis Methods 0.000 claims abstract description 58
- 238000012549 training Methods 0.000 claims abstract description 50
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 46
- 230000002787 reinforcement Effects 0.000 claims abstract description 22
- 238000012545 processing Methods 0.000 claims abstract description 20
- 238000009826 distribution Methods 0.000 claims abstract description 14
- 230000008569 process Effects 0.000 claims abstract description 13
- 230000003044 adaptive effect Effects 0.000 claims abstract description 7
- 238000000605 extraction Methods 0.000 claims abstract description 7
- 238000012360 testing method Methods 0.000 claims description 20
- 230000006870 function Effects 0.000 claims description 7
- 238000010606 normalization Methods 0.000 claims description 4
- 238000004590 computer program Methods 0.000 claims description 3
- 238000010276 construction Methods 0.000 claims description 3
- 230000000694 effects Effects 0.000 description 17
- 230000009471 action Effects 0.000 description 14
- 238000007637 random forest analysis Methods 0.000 description 14
- 238000010586 diagram Methods 0.000 description 9
- 238000004364 calculation method Methods 0.000 description 5
- 239000003795 chemical substances by application Substances 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 4
- 230000010354 integration Effects 0.000 description 4
- 238000012423 maintenance Methods 0.000 description 4
- 239000013598 vector Substances 0.000 description 3
- 230000007547 defect Effects 0.000 description 2
- 230000001419 dependent effect Effects 0.000 description 2
- 230000007613 environmental effect Effects 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 238000005096 rolling process Methods 0.000 description 2
- 230000002159 abnormal effect Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000010523 cascade reaction Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 238000009776 industrial production Methods 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000003449 preventive effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M13/00—Testing of machine parts
- G01M13/04—Bearings
- G01M13/045—Acoustic or vibration analysis
-
- 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
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
-
- 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
- G06N3/08—Learning methods
- G06N3/092—Reinforcement learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Acoustics & Sound (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The application belongs to the field of machine pump equipment fault diagnosis, and discloses a machine pump equipment operation fault diagnosis method, a device, a storage medium and electronic equipment, wherein the method comprises the following steps: collecting vibration signals in the running process of pump equipment of the machine to be tested; carrying out standardization processing on the acquired vibration signals to obtain standardized vibration signals; constructing a machine pump equipment fault diagnosis model and training; and inputting the vibration signals subjected to standardized processing into a trained model for feature extraction, and selecting an adaptive base classifier combination according to the extracted features to realize fault diagnosis of pump equipment of the machine to be tested, wherein the base classifier combination is distributed with different weights. According to the method and the device, dynamic weight distribution is carried out on each base classifier forming the model through the reinforcement learning algorithm DQN, and accurate diagnosis on fault types of the machine pump equipment can be achieved.
Description
Technical Field
The application belongs to the field of machine pump equipment fault diagnosis, and particularly relates to a machine pump equipment operation fault diagnosis method, a device, a storage medium and electronic equipment.
Background
The mechanical equipment diagnosis technology is a comprehensive discipline, combines the sensor technology, the dynamic test technology and theoretical knowledge in different fields, collects and analyzes the operation state information of the mechanical equipment, discovers the abnormal operation and failure mechanism, comprehensively evaluates the whole operation condition of the equipment, and has important guiding function on the aspects of design, manufacture, maintenance and the like of the mechanical equipment. The machine pump fault diagnosis has the following significance: (1) The fault condition of the machine pump equipment can be timely and accurately mastered, the fault condition is judged and processed in advance, the potential faults are maintained in a preventive mode, and the maintenance cost and the expenditure are reduced. (2) Through real-time monitoring and regular detection to the machine pump, overhaul the machine pump equipment under the prerequisite that does not influence production, avoid the trouble to worsen, prolong mechanical equipment's life-span, maximize availability factor. (3) The change of the running state of the pump is known, the stability of continuous operation of the equipment is ensured, and the management and control of the unit are enhanced.
Therefore, the pump is very necessary for safety monitoring and fault diagnosis as a core device for industrial production. The fault diagnosis technology of the research machine pump can rapidly and accurately judge the fault type, shorten the equipment maintenance time, discover the early fault of the machine pump in time, improve the maintenance effect, delay the scrapping time of the equipment, prevent the cascade reaction and improve the safety and the durability of the whole system. The existing machine pump fault diagnosis method is too dependent on manual experience and inherent diagnosis rules, fault analysis is not comprehensive enough, and judgment basis is too single, so that the real situation of machine pump faults cannot be completely known, and therefore a new machine pump equipment fault diagnosis method is necessary to be provided, and the fault diagnosis accuracy of the machine pump equipment is improved.
Disclosure of Invention
Aiming at the defects in the prior art, the purpose of the application is to provide a machine pump equipment operation fault diagnosis method, which improves the accuracy and stability of machine pump equipment fault diagnosis by performing model integration weight distribution on vibration data through reinforcement learning.
In order to achieve the above purpose, the present application provides the following technical solutions:
a method of diagnosing an operational failure of a machine pump apparatus, the method comprising the steps of:
s100: collecting vibration signals in the running process of pump equipment of the machine to be tested;
s200: carrying out standardization processing on the acquired vibration signals to obtain standardized vibration signals;
s300: constructing a machine pump equipment fault diagnosis model and training;
the machine pump equipment fault diagnosis model comprises a plurality of base classifiers, wherein the base classifiers realize dynamic weight distribution through a reinforcement learning algorithm DQN so as to cope with different types of machine pump equipment fault diagnosis;
s400: and inputting the vibration signals subjected to standardized processing into a trained model for feature extraction, and selecting an adaptive base classifier combination according to the extracted features to realize fault diagnosis of pump equipment of the machine to be tested, wherein the base classifier combination is distributed with different weights.
Preferably, in step S200, the acquired vibration signal is normalized by using a normalization method.
Preferably, in step S300, the machine pump equipment fault diagnosis model is trained by the following steps:
s301: the method comprises the steps of collecting historical fault data of different types of machine pump equipment to obtain a sample data set, and dividing the data set into a training sample data set and a test sample data set after standardized processing;
s302: setting training parameters of each base classifier, training each base classifier by using a training sample data set, and completing model training when the maximum training times or cross entropy loss function convergence is reached, wherein the maximum training times are based on the number of fault types in the training sample data set;
s303: testing the trained model by using a test sample data set, wherein in the test process, when the classification precision of each base classifier reaches 95%, the model test passes; otherwise, training parameters of each base classifier are adjusted to retrain the model until the model test passes.
Preferably, in step S300, the weight allocation of the plurality of base classifiers by the reinforcement learning algorithm DQN includes the following steps:
s3001: initializing a reinforcement learning DQN algorithm;
s3002: and carrying out weight distribution on each base classifier according to an epsilon-greedy strategy and based on an initialized DQN algorithm.
The present disclosure also provides a machine pump equipment operation fault diagnosis device, the device comprising:
the acquisition unit is used for acquiring vibration signals in the running process of the pump equipment of the machine to be tested;
the standardized processing unit is used for carrying out standardized processing on the acquired vibration signals so as to obtain standardized processed vibration signals;
the model construction and training unit is used for constructing and training a fault diagnosis model of the pump equipment;
the machine pump equipment fault diagnosis model comprises a plurality of base classifiers, wherein the base classifiers realize dynamic weight distribution through a reinforcement learning algorithm DQN so as to cope with different types of machine pump equipment fault diagnosis;
the fault diagnosis unit is used for inputting the vibration signals subjected to standardized processing into a trained model for feature extraction, and selecting an adaptive base classifier combination according to the extracted features so as to realize fault diagnosis of pump equipment of the machine to be tested, wherein the base classifier combination is distributed with different weights.
The present disclosure also provides an electronic device, including:
a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein,
the processor, when executing the program, implements a method as described in any of the preceding.
The present disclosure also provides a computer storage medium storing computer-executable instructions for performing a method as described in any one of the preceding claims.
Compared with the prior art, the beneficial effect that this application brought is: according to the method and the device, the running process of the actual machine pump equipment can be combined, the weights are dynamically distributed to the base classifiers through the reinforcement learning algorithm DQN according to the fault characteristics of the vibration data to be tested, so that the fault types of the machine pump equipment can be accurately diagnosed, and fault diagnosis errors are reduced.
Drawings
FIG. 1 is a flow chart of a method for diagnosing operation faults of a pump device according to one embodiment of the present application;
FIG. 2 is a schematic diagram of the structure of the DQN algorithm;
fig. 3 is a schematic diagram of a fault recognition effect of a pump device obtained by using a Random Forest method according to another embodiment of the present application;
fig. 4 is a schematic diagram of a fault recognition effect of a pump device obtained by using a GBDT method according to another embodiment of the present application;
fig. 5 is a schematic diagram of a fault recognition effect of a pump device obtained by using a category lifting algorithm Catboost method according to another embodiment of the present application;
FIG. 6 is a schematic diagram of a fault recognition effect of a pump device obtained by using an extreme gradient lifting tree XGBoost method according to another embodiment of the present application;
fig. 7 is a schematic diagram of a fault recognition effect of a pump device obtained by using the method according to another embodiment of the present application.
Detailed Description
Specific embodiments of the present application will be described in detail below with reference to fig. 1 to 7. While specific embodiments of the present application are shown in the drawings, it should be understood that the present application may be embodied in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.
It should be noted that certain terms are used throughout the description and claims to refer to particular components. Those of skill in the art will understand that a person may refer to the same component by different names. The specification and claims do not identify differences in terms of components, but rather differences in terms of the functionality of the components. As used throughout the specification and claims, the terms "include" and "comprise" are used in an open-ended fashion, and thus should be interpreted to mean "include, but not limited to. The description hereinafter sets forth the preferred embodiment for carrying out the present application, but is not intended to limit the scope of the present application in general, as the description proceeds. The scope of the present application is defined by the appended claims.
For the purpose of facilitating an understanding of the embodiments of the present application, reference will now be made to the drawings, by way of example, and specific examples of which are illustrated in the accompanying drawings and are not intended to limit the embodiments of the present application.
In one embodiment, as shown in fig. 1, the present application provides a method for diagnosing operation failure of a pump device, including the following steps:
s100: collecting vibration signals in the running process of pump equipment of the machine to be tested;
s200: carrying out standardization processing on the acquired vibration signals to obtain standardized vibration signals;
s300: constructing a machine pump equipment fault diagnosis model and training;
the machine pump equipment fault diagnosis model comprises a plurality of base classifiers, wherein the base classifiers realize dynamic weight distribution through a reinforcement learning algorithm DQN so as to cope with different types of machine pump equipment fault diagnosis;
s400: and inputting the vibration signals subjected to standardized processing into a trained model for feature extraction, and selecting an adaptive base classifier combination according to the extracted features to realize fault diagnosis of pump equipment of the machine to be tested, wherein the base classifier combination is distributed with different weights.
The above embodiments constitute a complete technical solution of the present disclosure. In this embodiment, the reinforcement learning algorithm DQN can introduce diversity among different base classifiers by dynamically assigning weights to the different base classifiers for integration. Because each base classifier can perform better on different feature subspaces or data samples, different modes and rules in the data can be better captured by the model through introducing the diversity of different base classifiers, so that the accuracy is improved.
Anti-overfitting: by combining multiple basis classifiers, the reinforcement learning algorithm DQN can reduce the risk of overfitting. In a single base classifier, if the base classifier complexity is too high, it is easy to overfit the training data, resulting in poor generalization performance on new data. And the over-fitting tendency of each base classifier can be balanced through a reinforcement learning algorithm DQN, so that the generalization capability of the whole model is improved.
Error correction: when an error occurs in a single base classifier, other correct base classifiers can correct the error, thereby improving the stability. The overall performance of a multi-basis classifier tends to be more robust than a single-basis classifier because there may be cases where a certain basis classifier performs poorly in some cases, but these drawbacks can be remedied by enhancing integration.
Dynamic weight allocation: the reinforcement learning algorithm DQN dynamically distributes weights of different base classifiers according to different fault types, so that the base classifier with better performance can obtain larger weights in the overall model, and the model can have larger influence on a final fault diagnosis result, thereby improving accuracy.
Reducing noise effects: in real data, there may be some noise or outliers, resulting in an unstable output of a single model. The outputs of the plurality of base classifiers can be balanced with each other by the reinforcement learning algorithm DQN to reduce the effects of noise, thereby improving stability.
In summary, the reinforcement learning algorithm DQN fully utilizes the advantages of the plurality of base classifiers by combining the base classifiers and balances the defects of the base classifiers, so that the stability of the fault diagnosis model is improved while the accuracy is improved, and the model can show better performance under different conditions.
In another embodiment, in step S200, the collected vibration data is scaled by using a normalization (Standardization) method, so that the vibration data falls within a small specific interval (e.g., [0,1 ]), and the normalization process is as follows:
wherein,is a standard score->For the acquired vibration data, +.>Is the average of a set of vibration data, +.>Is the standard deviation of a set of vibration data.
In another embodiment, in step S300, the machine pump equipment fault diagnosis model includes a plurality of base classifiers and a reinforcement learning DQN algorithm.
In this embodiment, the plurality of base classifiers includes a Random Forest Random, a gradient lifting tree GBDT, a class lifting algorithm Catboost, and an extreme gradient lifting tree XGBoost. Further, the model dynamically distributes the weight of each base classifier through a reinforcement learning DQN algorithm according to different fault characteristics of the extracted vibration signals.
In another embodiment, in step S300, the machine pump device fault diagnosis model is trained by:
s301: collecting different types of historical fault data (such as inner ring faults, outer ring faults, rolling body faults and the like) of the machine pump equipment to obtain a data sample set, and dividing the data sample set into a training data sample set and a test data sample set after standardized processing;
s302: setting training parameters of each base classifier, for example, setting the number of base estimators of Random Forest to 100, setting the learning rate of gradient lifting tree GBDT to 0.1, setting the tree depth of class lifting algorithm CatBOOST to 6, setting n_estimators of extreme gradient lifting tree XGBoost to 100, and setting the maximum depth to 6; when the model reaches the maximum number of training (e.g., the maximum number of training is dependent on the types of faults in the training data sample set, e.g., the training data sample set includes 30 types of faults, the maximum number of training is set to 30, the training data sample set includes 50 types of faults, the maximum number of training is set to 50) or the cross entropy loss function converges, the model training is completed;
in this step, the training data sample set includes various types of fault data known to the pump equipment, for example, in the first round of training, data mainly including fault data corresponding to a first type of fault (for example, an inner ring fault) in the training data sample set (which is mixed with other fault type data) is input to train the model, firstly, a characteristic value of the fault data corresponding to the first type of fault (the inner ring fault causes high-frequency vibration in a direction perpendicular to the pump axis to increase, the characteristic value represents a peak value of a specific frequency in a vibration signal), and secondly, a DQN algorithm assigns weights of the base classifiers according to the characteristic value corresponding to the first type of fault in the following manner:
the model is assumed to consist of four basic classifiers of Random Forest, gradient lifting tree GBDT, class lifting algorithm CatBoost and extreme gradient lifting tree XGBoost, and the four basic classifiers give the following probability predictions for the first class of faults (assuming three classes of faults in total, the first class of faults are inner ring faults, the second class of faults are outer ring faults, and the third class of faults are rolling body faults): the probability of the occurrence of the faults is 70% by Random Forest, 80% by gradient lifting tree GBDT, 40% by class lifting algorithm CatBoost, and 20% by extreme gradient lifting tree XGBoost. Meanwhile, the DQN algorithm distributes different weights for the four base classifiers according to an epsilon-greedy strategy, and the weights are respectively as follows: random Forest 30%, gradient lifting tree GBDT 50%, category lifting algorithm CatBoost 10%, extreme gradient lifting tree XGBoost 10%.
In order to calculate the total probability of occurrence of the first type of faults, the present disclosure adopts a weighted score calculation method, namely, the probability predicted by each base classifier is multiplied by the corresponding weight, specifically: the weighted score of Random Forest is 70% ×30% =0.21, the weighted score of gradient-lifted tree GBDT is 80% ×50% =0.40, the weighted score of gradient-lifted tree GBDT is 40% ×10% =0.04, and the weighted score of extreme gradient-lifted tree XGBoost is 20% ×10% =0.02. Adding these weighted scores gives a total weighted score for the first type of fault: 0.21 +0.40+0.04+0.02=0.67. The weighted score of 0.67 indicates that, considering the predictions and weights of the classifiers, the total probability of occurrence of the first class of faults is 67%, and meanwhile, according to the calculation method, the total probability of occurrence of the second class of faults and the third class of faults is 22% and 11%, and the judgment result of the model is obtained through comparison: the total probability of the first type of faults is highest, so that the prediction result output by the model is the first type of faults.
In the second training, the data which is mainly based on the fault data corresponding to the second type of faults (for example, the faults of the outer ring) in the training data sample set is input to train the model, firstly, the characteristic value of the fault data corresponding to the second type of faults (the faults of the outer ring can cause the vibration increase in the axial direction, the characteristic value is expressed as a specific frequency related to the natural vibration frequency of the outer ring in the vibration signal), and secondly, the DQN algorithm distributes the weight of each base classifier according to the characteristic value corresponding to the faults of the second type according to the following mode:
the assumption model consists of four base classifiers of Random Forest, gradient lifting tree GBDT, class lifting algorithm CatBoost and extreme gradient lifting tree XGBoost, and the four base classifiers give the following probability predictions for the second class of faults: the probability of the occurrence of the faults is 72% by Random Forest, 85% by gradient lifting tree GBDT, 98% by class lifting algorithm CatBoost, and 93% by extreme gradient lifting tree XGBoost. Meanwhile, the DQN algorithm distributes different weights for the four base classifiers according to an epsilon-greedy strategy, and the weights are respectively as follows: random Forest 2%, gradient lifting tree GBDT 5%, category lifting algorithm CatBoost 86%, extreme gradient lifting tree XGBoost 7%.
In order to calculate the total probability of occurrence of the second class of faults, the present disclosure adopts a weighted score calculation method, namely, the probability predicted by each base classifier is multiplied by the corresponding weight, specifically: the weighting score of Random Forest was 72% ×2% =0.0144, the weighting score of gradient-lifted tree GBDT was 85% ×5% =0.0425, the weighting score of class-lifted algorithm CatBoost was 98% ×86% = 0.8428, and the weighting score of extreme gradient-lifted tree XGBoost was 93% ×7% = 0.0651. Adding these weighted scores gives a total weighted score for the first type of fault: 0.0144 +0.0425+ 0.8428 + 0.0651 = 0.9648. The weighted score 0.9648 indicates that, considering the prediction and the weight of each classifier, the model considers that the total probability of occurrence of the second type of faults is 96.48%, the total probability of occurrence of other two types of faults is calculated according to the calculation method and is respectively 1.96% and 1.56%, the total probability of occurrence of the second type of faults is highest through comparison, and the model judges and outputs the prediction result as the second type of faults.
And the method is analogically performed until a weight distribution mode of the corresponding base classifier is given for fault data corresponding to all types of faults contained in the training data sample set, and meanwhile, a fault type prediction result is given, so that model training is completed.
S303: testing the trained model by using a test sample data set, wherein in the test process, when the classification precision of each base classifier reaches 95%, the model test passes; otherwise, training parameters of each base classifier are adjusted (for example, the number of base estimators of Random Forest is set to 150, the learning rate of gradient-lifted tree GBDT is set to 0.05, the tree depth of class-lifted algorithm CatBOOST is set to 8, the n_optimators of extreme gradient-lifted tree XGBoost is set to 200, and the maximum depth is set to 10) and the model is retrained until the model test passes.
After model training is completed and the test is passed, the model can rapidly judge the main fault type of the pump equipment of the machine to be tested according to the characteristic value of the vibration signal extracted from the operation process of the input pump equipment of the machine to be tested, thereby realizing accurate fault diagnosis.
In another embodiment, in step S300, the weight distribution of the plurality of base classifiers by the reinforcement learning algorithm DQN includes the following steps:
s3001: the DQN algorithm is initialized as shown in fig. 2 (parameters in fig. 2)sThe (status) represents the environmental status in which the agent is located, which in the context of machine pump equipment fault diagnosis represents a specific characteristic of the vibration signal or the current condition of the system. Parameters (parameters)aThe (action) represents the action taken by the agent in a given state, in the case of fault diagnosis, this is the selection of a particular set of base classifier weights. Parameters (parameters)r(rewards) means that when an agent takes an action, it gets rewards or penalties according to the effect of the action, which are usually a number, which is used to measure the quality of the action, in fault diagnosis this is based on the effect of the action (weight allocation) on the accuracy of the diagnosis. The parameter s' (the next state) represents a new state in which the environmental state changes after the agent performs an action.a'The next action is indicated and is indicated,Q（s，a) Is shown in the state sExecute action downwardsaIs also referred to asQA value;Q（s'，a') Indicating in the next states'Next executing next actiona'Is a predicted jackpot for (1); initializing an experience playback pool, an estimation network and a target network of the DQN algorithm;
in this step, initialization of the empirical playback pool refers to creating an empty data structure for storing empirical data collected as the DQN algorithm interacts with the environment, and randomly sampling the data for updating the neural network as it is trained.
Initialization of the estimated network and the target network typically involves setting the weights of the two structurally identical networks to random values, wherein the weights of the target network are initially copied from the estimated network to provide an initial learning point for the algorithm.
S3002: acquiring states from an environment (the environment is a dynamically updated queue and consists of input feature vectors, probability prediction vectors of a baseline model and cross entropy loss functions of the model), and selecting the allocation weights required by each base classifier integration according to an epsilon-greedy strategy (an agent) by selecting random actions according to the probability epsilon so as to discover new better strategies, and selecting actions currently considered to be optimal according to the probability 1-epsilon so as to utilize the actions, thereby helping to prevent the local optimal solutions from being trapped and facilitating the discovery of the better strategies);
in this step, the states include:
1. inputting a feature vector;
2. the class probability can help explain the prediction result of the model, and can also be used as the basis for integrating a plurality of models, and the formula is as follows:
wherein,outputting a probability function for the base classifier; />For base classifier in sample->The following prediction results;is->A radix classifier; />Is->A sample number; />Indicate->The sample is at->Prediction probability under the individual basis classifier.
3. By calculating cross entropy loss functionsThe accuracy of the classification task can be optimized in model training, and the prediction probability distribution of the model can be evaluated and compared, and the cross entropy loss function is +.>The calculation formula is as follows:
wherein,is the number of samples, +.>Is the category number->Is->The sample belongs to->True tags of individual categories->Is model predictive +.>The sample belongs to->Probability of individual categories.
S3003: executing the action of integrating new prediction results according to the weights, integrating the prediction results of all base classifiers according to the weights of all base classifiers, setting rewards as values of model cross entropy loss, adding a feature matrix of a current sample, the prediction results and a set of the current sample cross entropy loss values into a current state, and deleting the forefront set of the current state to form a next state;
s3004: storing the current state, action, reward, and next state in an experience playback pool;
s3005: randomly extracting a batch of samples from the experience playback pool, and training a Q network and updating a target network;
s3006: steps S3002-S3005 are repeated until a maximum number of iterations is reached or the DQN algorithm converges.
Next, the technical effects of the present method will be described with reference to the drawings.
FIG. 3 is a diagram showing the effect of machine pump equipment fault identification obtained by solely adopting a Random Forest method; FIG. 4 is a graph showing the effect of machine pump equipment fault identification obtained by the gradient lift tree GBDT method alone; FIG. 5 is a diagram showing the effect of machine pump equipment fault identification obtained by the class lifting algorithm Catboost method alone; FIG. 6 is a diagram of the effect of machine pump equipment fault identification obtained by solely employing the extreme gradient lifted tree XGBoost method; fig. 7 shows the effect of identifying the failure of the pump equipment obtained by the method of the present application. In fig. 3 to 7, circles represent predicted values, x represents actual values, and when circles coincide with the symbol x, it is indicated that the fault diagnosis is accurate. From the graph, the Random Forest method has the lowest accuracy rate of only 73%; the accuracy of the GBDT method of the gradient lifting tree is improved to a certain extent, and the accuracy reaches 82%; the class lifting algorithm Catboost method and the extreme gradient lifting tree XGBoost are 79% and 87% respectively; the best diagnostic effect is that the accuracy of the method reaches 95 percent.
In another embodiment, the present disclosure further provides an apparatus for diagnosing an operation failure of a pump device, the apparatus comprising:
the acquisition unit is used for acquiring vibration signals in the running process of the pump equipment of the machine to be tested;
the standardized processing unit is used for carrying out standardized processing on the acquired vibration signals so as to obtain standardized processed vibration signals;
the model construction and training unit is used for constructing and training a fault diagnosis model of the pump equipment;
the machine pump equipment fault diagnosis model comprises a plurality of base classifiers, wherein the base classifiers realize dynamic weight distribution through a reinforcement learning algorithm DQN so as to cope with different types of machine pump equipment fault diagnosis;
the fault diagnosis unit is used for inputting the vibration signals subjected to standardized processing into a trained model for feature extraction, and selecting an adaptive base classifier combination according to the extracted features so as to realize fault diagnosis of pump equipment of the machine to be tested, wherein the base classifier combination is distributed with different weights.
In another embodiment, the present disclosure further provides an electronic device, including:
a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein,
the processor, when executing the program, implements a method as described in any of the preceding.
The present disclosure also provides a computer storage medium storing computer-executable instructions for performing a method as described in any one of the preceding claims.
The above embodiments are provided to illustrate the technical concept and features of the present invention and are intended to enable those skilled in the art to understand the content of the present invention and implement the same, and are not intended to limit the scope of the present invention. All equivalent changes or modifications made in accordance with the spirit of the present invention should be construed to be included in the scope of the present invention.
Claims (7)
1. A method for diagnosing an operation failure of a machine pump apparatus, comprising the steps of:
s100: collecting vibration signals in the running process of pump equipment of the machine to be tested;
s200: carrying out standardization processing on the acquired vibration signals to obtain standardized vibration signals;
s300: constructing a machine pump equipment fault diagnosis model and training;
the machine pump equipment fault diagnosis model comprises a plurality of base classifiers, wherein the base classifiers realize dynamic weight distribution through a reinforcement learning algorithm DQN so as to cope with different types of machine pump equipment fault diagnosis;
s400: and inputting the vibration signals subjected to standardized processing into a trained model for feature extraction, and selecting an adaptive base classifier combination according to the extracted features to realize fault diagnosis of pump equipment of the machine to be tested, wherein the base classifier combination is distributed with different weights.
2. The method according to claim 1, characterized in that in step S200, the acquired vibration signal is normalized by using a normalization method.
3. The method according to claim 1, wherein in step S300, the machine pump equipment failure diagnosis model is trained by:
s301: the method comprises the steps of collecting historical fault data of different types of machine pump equipment to obtain a sample data set, and dividing the data set into a training sample data set and a test sample data set after standardized processing;
s302: setting training parameters of each base classifier, training each base classifier by using a training sample data set, and completing model training when the maximum training times or cross entropy loss function convergence is reached, wherein the maximum training times are based on the number of fault types in the training sample data set;
s303: testing the trained model by using a test sample data set, wherein in the test process, when the classification precision of each base classifier reaches 95%, the model test passes; otherwise, training parameters of each base classifier are adjusted to retrain the model until the model test passes.
4. The method according to claim 1, wherein in step S300, the weight distribution of the plurality of base classifiers by the reinforcement learning algorithm DQN comprises the steps of:
s3001: initializing a reinforcement learning DQN algorithm;
s3002: and carrying out weight distribution on each base classifier according to an epsilon-greedy strategy and based on an initialized DQN algorithm.
5. An apparatus for diagnosing an operation failure of a pump device, said apparatus comprising:
the acquisition unit is used for acquiring vibration signals in the running process of the pump equipment of the machine to be tested;
the standardized processing unit is used for carrying out standardized processing on the acquired vibration signals so as to obtain standardized processed vibration signals;
the model construction and training unit is used for constructing and training a fault diagnosis model of the pump equipment;
the machine pump equipment fault diagnosis model comprises a plurality of base classifiers, wherein the base classifiers realize dynamic weight distribution through a reinforcement learning algorithm DQN so as to cope with different types of machine pump equipment fault diagnosis;
the fault diagnosis unit is used for inputting the vibration signals subjected to standardized processing into a trained model for feature extraction, and selecting an adaptive base classifier combination according to the extracted features so as to realize fault diagnosis of pump equipment of the machine to be tested, wherein the base classifier combination is distributed with different weights.
6. An electronic device, the electronic device comprising:
a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein,
the processor, when executing the program, implements the method of any one of claims 1 to 4.
7. A computer storage medium having stored thereon computer executable instructions for performing the method of any of claims 1 to 4.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311777537.4A CN117435981B (en) | 2023-12-22 | 2023-12-22 | Method and device for diagnosing operation faults of machine pump equipment, storage medium and electronic equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311777537.4A CN117435981B (en) | 2023-12-22 | 2023-12-22 | Method and device for diagnosing operation faults of machine pump equipment, storage medium and electronic equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117435981A true CN117435981A (en) | 2024-01-23 |
| CN117435981B CN117435981B (en) | 2024-03-01 |
Family
ID=89546617
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311777537.4A Active CN117435981B (en) | 2023-12-22 | 2023-12-22 | Method and device for diagnosing operation faults of machine pump equipment, storage medium and electronic equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117435981B (en) |
Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20090304290A1 (en) * | 2008-06-09 | 2009-12-10 | Denso Corporation | Image recognition apparatus utilizing plurality of weak classifiers for evaluating successive sub-images extracted from an input image |
| CN105574547A (en) * | 2015-12-22 | 2016-05-11 | 北京奇虎科技有限公司 | Integrated learning method and device adapted to weight of dynamically adjustable base classifier |
| CN105738109A (en) * | 2016-02-22 | 2016-07-06 | 重庆大学 | Bearing fault classification diagnosis method based on sparse representation and ensemble learning |
| CN107180140A (en) * | 2017-06-08 | 2017-09-19 | 中南大学 | Shafting fault recognition method based on dual-tree complex wavelet and AdaBoost |
| CN110363230A (en) * | 2019-06-27 | 2019-10-22 | 华南理工大学 | Stacking integrated sewage handling failure diagnostic method based on weighting base classifier |
| US20220004811A1 (en) * | 2021-03-25 | 2022-01-06 | Beijing Baidu Netcom Science Technology Co., Ltd. | Method and apparatus of training model, device, medium, and program product |
| CN114861719A (en) * | 2022-04-29 | 2022-08-05 | 西安理工大学 | High-speed train bearing fault diagnosis method based on ensemble learning |
| CN115221973A (en) * | 2022-07-29 | 2022-10-21 | 中国航空工业集团公司上海航空测控技术研究所 | Aviation bearing fault diagnosis method based on enhanced weighted heterogeneous ensemble learning |
| CN115496108A (en) * | 2022-10-08 | 2022-12-20 | 航天智控(北京)监测技术有限公司 | Fault monitoring method and system based on manifold learning and big data analysis |
| CN116026588A (en) * | 2022-11-10 | 2023-04-28 | 浙江工商大学 | Bearing fault diagnosis and early warning method based on ensemble learning |
| CN116520140A (en) * | 2022-12-07 | 2023-08-01 | 华润水泥投资有限公司 | Permanent magnet direct drive motor fault detection method integrating feature extraction and small sample classification |
| CN116561517A (en) * | 2023-05-24 | 2023-08-08 | 桂林电子科技大学 | Unbalanced intelligent fault quantitative diagnosis method based on reward optimization deep reinforcement learning |
-
2023
- 2023-12-22 CN CN202311777537.4A patent/CN117435981B/en active Active
Patent Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20090304290A1 (en) * | 2008-06-09 | 2009-12-10 | Denso Corporation | Image recognition apparatus utilizing plurality of weak classifiers for evaluating successive sub-images extracted from an input image |
| CN105574547A (en) * | 2015-12-22 | 2016-05-11 | 北京奇虎科技有限公司 | Integrated learning method and device adapted to weight of dynamically adjustable base classifier |
| CN105738109A (en) * | 2016-02-22 | 2016-07-06 | 重庆大学 | Bearing fault classification diagnosis method based on sparse representation and ensemble learning |
| CN107180140A (en) * | 2017-06-08 | 2017-09-19 | 中南大学 | Shafting fault recognition method based on dual-tree complex wavelet and AdaBoost |
| CN110363230A (en) * | 2019-06-27 | 2019-10-22 | 华南理工大学 | Stacking integrated sewage handling failure diagnostic method based on weighting base classifier |
| US20220004811A1 (en) * | 2021-03-25 | 2022-01-06 | Beijing Baidu Netcom Science Technology Co., Ltd. | Method and apparatus of training model, device, medium, and program product |
| CN114861719A (en) * | 2022-04-29 | 2022-08-05 | 西安理工大学 | High-speed train bearing fault diagnosis method based on ensemble learning |
| CN115221973A (en) * | 2022-07-29 | 2022-10-21 | 中国航空工业集团公司上海航空测控技术研究所 | Aviation bearing fault diagnosis method based on enhanced weighted heterogeneous ensemble learning |
| CN115496108A (en) * | 2022-10-08 | 2022-12-20 | 航天智控(北京)监测技术有限公司 | Fault monitoring method and system based on manifold learning and big data analysis |
| CN116026588A (en) * | 2022-11-10 | 2023-04-28 | 浙江工商大学 | Bearing fault diagnosis and early warning method based on ensemble learning |
| CN116520140A (en) * | 2022-12-07 | 2023-08-01 | 华润水泥投资有限公司 | Permanent magnet direct drive motor fault detection method integrating feature extraction and small sample classification |
| CN116561517A (en) * | 2023-05-24 | 2023-08-08 | 桂林电子科技大学 | Unbalanced intelligent fault quantitative diagnosis method based on reward optimization deep reinforcement learning |
Non-Patent Citations (6)
| Title |
|---|
| PAN SHANG 等: "A new ensemble deep graph reinforcement learning network for spatio-temporal traffic volume forecasting in a freeway network", 《DIGITAL SIGNAL PROCESSING》, 29 January 2022 (2022-01-29), pages 1 - 8 * |
| YU DING 等: "Intelligent fault diagnosis for rotating machinery using deep Q-network based health state classification: A deep reinforcement learning approach", 《ADVANCED ENGINEERING INFORMATICS》, vol. 42, 21 August 2019 (2019-08-21), pages 1 - 11, XP085914994, DOI: 10.1016/j.aei.2019.100977 * |
| YUCHEN HE 等: "A dual attribute weighted decision fusion system for fault classification based on an extended analytic hierarchy process", 《ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE》, vol. 114, 29 June 2022 (2022-06-29), pages 1 - 4, XP087140755, DOI: 10.1016/j.engappai.2022.105066 * |
| 周海军 等: "基于随机森林的离心泵滚动轴承故障诊断", 《中国舰船研究》, vol. 15, no. 03, 30 June 2020 (2020-06-30), pages 129 - 135 * |
| 崔宇 等: "考虑不平衡案例样本的电力变压器故障诊断方法", 《高电压技术》, vol. 46, no. 01, 31 January 2020 (2020-01-31), pages 33 - 41 * |
| 徐凯 等: "改进麻雀算法和Q-Learning优化集成学习轨道电路故障诊断", 《铁道科学与工程学报》, vol. 20, no. 11, 30 November 2023 (2023-11-30), pages 4426 - 4437 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117435981B (en) | 2024-03-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111340238B (en) | Fault diagnosis method, device, equipment and storage medium of industrial system | |
| Zhao et al. | Deep multi-scale convolutional transfer learning network: A novel method for intelligent fault diagnosis of rolling bearings under variable working conditions and domains | |
| CN116757534B (en) | Intelligent refrigerator reliability analysis method based on neural training network | |
| CN113899577A (en) | Abnormal sound detection device, abnormal sound generation device, and abnormal sound generation method | |
| He et al. | Deep variational autoencoder classifier for intelligent fault diagnosis adaptive to unseen fault categories | |
| CN106886660A (en) | EEMD Hilbert envelope spectrums are with Rolling Bearing Status recognition methods under the varying load that DBN is combined | |
| CN110609524B (en) | Industrial equipment residual life prediction model and construction method and application thereof | |
| CN113837000A (en) | Small sample fault diagnosis method based on task sequencing meta-learning | |
| CN112904810B (en) | Process industry nonlinear process monitoring method based on effective feature selection | |
| CN109623489B (en) | Improved machine tool health state evaluation method and numerical control machine tool | |
| CN109813542A (en) | The method for diagnosing faults of air-treatment unit based on production confrontation network | |
| CN111079926B (en) | Equipment fault diagnosis method with self-adaptive learning rate based on deep learning | |
| CN114155397B (en) | Small sample image classification method and system | |
| JP7331369B2 (en) | Abnormal Sound Additional Learning Method, Data Additional Learning Method, Abnormality Degree Calculating Device, Index Value Calculating Device, and Program | |
| Cheng et al. | A Koopman operator approach for machinery health monitoring and prediction with noisy and low-dimensional industrial time series | |
| Lu et al. | Joint training of a predictor network and a generative adversarial network for time series forecasting: A case study of bearing prognostics | |
| CN108961460B (en) | Fault prediction method and device based on sparse ESGP (Enterprise service gateway) and multi-objective optimization | |
| CN116561517A (en) | Unbalanced intelligent fault quantitative diagnosis method based on reward optimization deep reinforcement learning | |
| CN115290326A (en) | Rolling bearing fault intelligent diagnosis method | |
| CN113177577A (en) | Bearing fault diagnosis method based on improved convolutional neural network | |
| Cui et al. | Intelligent fault quantitative identification via the improved deep deterministic policy gradient (DDPG) algorithm accompanied with imbalanced sample | |
| CN117056678B (en) | Machine pump equipment operation fault diagnosis method and device based on small sample | |
| CN117435981B (en) | Method and device for diagnosing operation faults of machine pump equipment, storage medium and electronic equipment | |
| CN113126489A (en) | CNN-GRU-BINN-based intelligent BIT design method for heavy-duty gas turbine control system | |
| CN115931359A (en) | Turbine pump bearing fault diagnosis method and device |
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 |