CN115510926B - Cross-machine type diesel engine combustion chamber fault diagnosis method and system - Google Patents
Cross-machine type diesel engine combustion chamber fault diagnosis method and system Download PDFInfo
- Publication number
- CN115510926B CN115510926B CN202211471504.2A CN202211471504A CN115510926B CN 115510926 B CN115510926 B CN 115510926B CN 202211471504 A CN202211471504 A CN 202211471504A CN 115510926 B CN115510926 B CN 115510926B
- Authority
- CN
- China
- Prior art keywords
- model
- diesel engine
- fault diagnosis
- combustion chamber
- module
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M15/00—Testing of engines
- G01M15/04—Testing internal-combustion engines
-
- 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
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Abstract
The invention discloses a cross-machine type diesel engine combustion chamber fault diagnosis method, which comprises the following steps: collecting signals related to a fault excitation source and a diesel engine of a reference model and a model to be diagnosed, comparing the similarity of the signals between the models, and if the signals are similar, taking a labeled source domain data sample of the reference model and a unlabeled data sample collected by the model to be diagnosed as a training sample set; constructing a fault diagnosis network model, wherein the fault diagnosis network model is obtained by training samples in a sample set and comprises a feature extraction module, a classification module and a domain self-adaptive module, wherein the feature extraction module comprises a plurality of convolution layers and a self-adaptive average pooling layer; and inputting a signal segment which is acquired by the diesel engine of the model to be diagnosed in real time and is related to the fault excitation source into the fault diagnosis network model to diagnose the fault of the combustion chamber of the diesel engine of the cross-model, so as to obtain a fault diagnosis result. The invention can realize the fault diagnosis of the combustion chamber of the cross-model diesel engine.
Description
Technical Field
The invention relates to the field of intelligent fault diagnosis of diesel engines, in particular to a domain-adaptive cross-machine type diesel engine combustion chamber fault diagnosis method and system.
Background
Economy, high efficiency, long service life and intellectualization are the main requirements of modern diesel engines. In recent years, monitoring and diagnosis of key components of diesel engines are increasingly gaining importance. The working environment of the combustion chamber of the diesel engine, such as parts of piston rings, gas valves, fuel injectors and the like, is severe, the failure rate is high, if serious failures occur, the economy of the diesel engine can be reduced, and malignant safety accidents can be caused. Therefore, monitoring and diagnosing key components of the combustion chamber of the diesel engine are of great significance.
Vibration/acoustic emission monitoring is a non-destructive inspection technique commonly used to identify the operating conditions of diesel combustion chamber components. The method is characterized in that the characteristics of cylinder cover vibration/acoustic emission signals have certain similarity but difference according to different models of diesel engines, and how to excavate the joint expression of the characteristics from the vibration/acoustic emission signals of different models and reduce the distribution difference of the characteristics is the key of the universality of the vibration/acoustic emission fault diagnosis model of the combustion chamber of the diesel engine.
With the accumulation of monitoring data, data-driven fault diagnosis methods get more and more attention and application, and the data-driven methods mainly include two types, namely machine learning and deep learning. The machine learning method relies on artificial feature engineering, time domain and frequency domain features are extracted from original signals through statistical analysis, abundant prior knowledge and expert experience are needed, the fault diagnosis precision of the models is easily interfered by subjective factors, the generalization capability is poor, and common feature expression is difficult to find in different diagnosis tasks.
The deep learning method adaptively learns the expression in the deep feature space from complex raw data through a group of nonlinear convolution layers, and the deep learning model can intelligently extract the features related to the equipment state from the raw data through training. Machine learning and deep learning both require training sets and test sets to be independently and identically distributed, which means that in practical engineering application, when the model corresponding to a diagnosed sample is different from the training sample, the diagnosis precision is reduced, even misdiagnosis is performed.
The fault signal characteristics of the diesel engine are slightly different due to the change of the structure and the operation mechanism of the diesel engine, and the common characteristic expression of signals among different types of engines is difficult to be mined by the existing fault diagnosis technologies such as signal processing, machine learning, deep learning and the like.
Disclosure of Invention
The invention mainly aims to provide a cross-model combustion chamber fault diagnosis method for a diesel engine.
The technical scheme adopted by the invention is as follows:
the method for diagnosing the faults of the combustion chamber of the cross-model diesel engine comprises the following steps:
acquiring signals related to a fault excitation source in excitation signals of a reference model and a model diesel engine to be diagnosed, comparing the similarity of the signals between the models, if the signals are similar, taking a labeled data sample of the reference model as a source domain, taking an acquired unlabeled data sample of the model to be diagnosed as a target domain, and establishing a training sample set;
constructing a fault diagnosis network model, wherein the model is obtained by training samples in a training sample set and comprises a feature extraction module, a classification module and a domain self-adaptive module, wherein the feature extraction module comprises a plurality of convolution layers and a self-adaptive average pooling layer, feature extraction is carried out on the plurality of convolution layers, average pooling calculation is carried out on the self-adaptive average pooling layer, and features with consistent dimensions are formed; the extracted features with consistent dimensions are respectively input into a classification module and a domain self-adaptive module, classification loss and feature distribution difference loss of different models are calculated, and finally network parameters are automatically updated through an Adam optimization algorithm to reduce feature distribution difference and improve classification accuracy;
and inputting a signal section which is acquired by the machine type diesel engine to be diagnosed in real time and is related to the fault excitation source into the fault diagnosis network model to diagnose the fault of the cross-machine type diesel engine combustion chamber, so as to obtain a fault diagnosis result.
According to the technical scheme, the feature extraction module specifically comprises 5 convolution layers, and each convolution layer specifically performs 1-dimensional convolution, linear rectification, maximum pooling and 1-dimensional batch normalization on the input signal.
According to the technical scheme, the self-adaptive average pooling layer is embedded into the last layer of the feature extraction module and converts the features with inconsistent dimensions into the features with consistent dimensions.
According to the technical scheme, the similarity of signals among models is judged by calculating the signal attenuation coefficient and the correlation coefficient.
According to the technical scheme, a sample set comprises a source domain data sample set and a target domain data sample set, a reference label data set is constructed through fault simulation test and real machine data accumulation of a certain diesel engine type, and a vibration/sound emission time domain signal segment related to an excitation source of the cylinder is determined based on a diesel engine ignition sequence and a timing diagram to form a source domain data sample set for model training; and forming a target domain data sample set for model training and diagnosis by adopting the same method for the unlabeled data set of the unknown state acquired by the other diesel engine type to be diagnosed.
The invention also provides a cross-machine type diesel engine combustion chamber fault diagnosis system, which comprises:
the sample acquisition module is used for acquiring signals related to a fault excitation source in excitation signals of a reference model and a model diesel engine to be diagnosed, comparing the similarity of the signals between the models, if the signals are similar, taking a labeled data sample of the reference model as a source domain, taking a non-labeled data sample acquired in real time of the model to be diagnosed as a target domain, and establishing a training sample set;
the fault diagnosis network model building module is used for building a fault diagnosis network model, the fault diagnosis network model is obtained by training samples in a sample set, and comprises a feature extraction module, a classification module and a domain self-adaption module, wherein the feature extraction module comprises a plurality of convolution layers and a self-adaption average pooling layer, specifically, feature extraction is carried out on the plurality of convolution layers, and then average pooling calculation is carried out on the self-adaption average pooling layer to form features with consistent dimensions; the extracted features with consistent dimensions are respectively input into a classification module and a domain self-adaptive module, classification loss and feature distribution difference loss of different models are calculated, and finally network parameters are automatically updated through an Adam optimization algorithm to reduce feature distribution difference and improve classification accuracy;
and the diagnosis module is used for inputting a signal section which is acquired by the to-be-diagnosed model diesel engine in real time and is related to the fault excitation source into the fault diagnosis network model to carry out cross-model diesel engine combustion chamber fault diagnosis so as to obtain a fault diagnosis result.
According to the technical scheme, the feature extraction module specifically comprises 5 convolution layers, and each convolution layer specifically performs 1-dimensional convolution, linear rectification, maximum pooling and 1-dimensional batch normalization on the input signal.
According to the technical scheme, the self-adaptive average pooling layer is embedded into the last layer of the feature extraction module and converts the features with inconsistent dimensions into the features with consistent dimensions.
The invention also provides terminal equipment which comprises a memory and a processor, wherein the memory is internally stored with a computer program which can be executed by the processor, and the processor realizes the steps of the cross-model diesel engine combustion chamber fault diagnosis method in the technical scheme when executing the computer program.
The invention also provides a non-transitory computer readable storage medium, on which a computer program is stored, and the computer program, when executed by a processor, implements the steps of the cross-model diesel engine combustion chamber fault diagnosis method according to the above technical solution.
The invention has the following beneficial effects: the invention discloses an end-to-end diesel engine combustor type self-adaptive fault diagnosis method based on a domain self-adaptive theory, which realizes the direct connection from a data end to a diagnosis end, automatically extracts the deep-level characteristics of vibration/acoustic emission signals through a convolution network layer, and the input dimensionality of self-adaptive average pooling layer self-adaptive data in a characteristic extraction module can be fully utilized to train by fully utilizing data sources with various different data dimensionalities; the domain self-adaptive module reduces the feature distribution difference between the labeled source domain data (reference model) and the unlabeled target domain data (model to be diagnosed), and excavates common feature expression. Experimental verification shows that the method has higher accuracy rate and better generalization on fault diagnosis of the diesel engine combustion chamber, and can be used for diagnosing the label-free data samples of other types in a self-adaptive manner on the basis of the reference model label data sample.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the embodiments or the description of the prior art will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and other drawings can be obtained by those skilled in the art without creative efforts.
FIG. 1 is a flow chart of a cross-model diesel engine combustion chamber fault diagnosis method according to an embodiment of the invention;
FIG. 2 is a general architecture of a cross-model diesel engine combustor fault diagnosis method based on domain adaptation according to another embodiment of the present invention;
FIG. 3 (a) is a schematic diagram of cross-model fault diagnosis feasibility analysis and adjacent cylinder interference analysis of a 6DK20 diesel engine according to an embodiment of the present invention;
FIG. 3 (b) is a schematic diagram of cross-model fault diagnosis feasibility analysis and adjacent cylinder interference analysis of a Z6170 diesel engine according to the embodiment of the invention;
FIG. 4 is a diagram of a domain adaptive network architecture according to an embodiment of the present invention;
fig. 5 is a schematic diagram of a visualization result of the migration between different diesel engine types according to an embodiment of the present invention, wherein (a) is a schematic diagram of a visualization result after PCA _ SVM extraction; (b) A visual result schematic diagram after 1D _CNNextraction; (c) Is a schematic diagram of the visualization result obtained by the method;
fig. 6 is a schematic diagram of a terminal device according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
As shown in fig. 1, the cross-model diesel engine combustion chamber fault diagnosis method according to the embodiment of the present invention includes the following steps:
s1, collecting signals related to a fault excitation source in excitation signals of a reference model and a model diesel engine to be diagnosed, comparing similarity of the signals between the models, if the signals are similar, taking a labeled data sample of the reference model as a source domain, taking an acquired unlabeled data sample of the model to be diagnosed as a target domain, and establishing a training sample set;
s2, constructing a fault diagnosis network model, wherein the model is obtained by training samples in a sample set and comprises a feature extraction module, a classification module and a domain self-adaptive module, the feature extraction module comprises a plurality of convolution layers and a self-adaptive average pooling layer, feature extraction is carried out on the plurality of convolution layers, average pooling calculation is carried out on the self-adaptive average pooling layer, and features with consistent dimensions are formed; the extracted features with consistent dimensions are respectively input into a classification module and a domain self-adaptive module, classification loss and feature distribution difference loss of different models are calculated, and finally network parameters are automatically updated through an Adam optimization algorithm to reduce feature distribution difference and improve classification accuracy;
and S3, inputting a signal section which is acquired by the to-be-diagnosed diesel engine in real time and is related to the fault excitation source into the fault diagnosis network model to perform cross-model diesel engine combustion chamber fault diagnosis, and obtaining a fault diagnosis result.
Because the models are different, the diesel engine structure and the installation position of the vibration/acoustic emission sensor are possibly different, the single-cylinder whole-period vibration/acoustic emission signal characteristics of different models are possibly greatly different, and the common characteristic expression of a source domain and a target domain cannot be effectively mined, the similarity of the vibration/acoustic emission signals among the models needs to be contrasted and analyzed, and the feasibility of cross-model diagnosis is judged. Cross-model diagnostic feasibility analysis is therefore required prior to collecting data samples. In the cross-machine type diagnosis feasibility analysis stage, vibration/sound emission full-cycle time domain signals are extracted, main excitation source excitation signals such as single-cylinder full-cycle air valve opening and closing and cylinder internal detonation pressure are identified, and the similarity of fault excitation source signals of different machine types is judged according to indexes such as attenuation coefficients and correlation coefficients so as to judge whether cross-machine type diagnosis can be carried out.
In the stage of collecting data samples, a reference label data set, namely a source domain, is constructed through fault simulation tests and real machine data accumulation of a certain diesel engine type, and a vibration/acoustic emission time domain signal segment related to the cylinder excitation source is determined based on a diesel engine ignition sequence and a timing diagram, so that a source domain data sample set for algorithm training is formed; and forming a target domain data sample set for algorithm training and diagnosis by adopting the same method for a non-label data set (namely a target domain) of an unknown state acquired by another diesel engine type to be diagnosed.
In the model training stage, the label sample of the source domain and the label-free sample of the target domain are jointly input into an algorithm for training, the characteristics of the samples are automatically extracted, the domain self-adaptive module is embedded into the 1D _CNN, and the difference of the characteristic distribution of the source domain and the target domain is effectively learned in the training process.
In the fault diagnosis stage, a trained fault diagnosis model is used for diagnosing a target domain unlabeled sample acquired in real time, and the generalization of cross-domain diagnosis is improved.
In another embodiment of the invention, taking two types of diesel engines, i.e., a Z6170 type and a 6DK20 type as examples, as shown in fig. 2, a cross-model fault diagnosis method for a combustion chamber of a diesel engine based on domain self-adaptation specifically comprises the following steps:
step 1: formation of data samples
The vibration/acoustic emission data sample in the source domain can be accumulated through a simulation test or a real machine running process, and an unknown state data sample collected for another machine type in the target domain to be diagnosed. In the embodiment, a data sample is obtained through acoustic emission fault simulation tests of two types of diesel engines (Z6170 type and 6DK20 type), a cylinder cover acoustic emission signal and a diesel engine top dead center signal are respectively collected, and the acoustic emission signal sampling rate is set to be 800kHz. The diesel engine combustion chamber faults simulated by the present example include: the exhaust valve was slightly worn and the exhaust valve was severely worn as shown in table 1. As shown in fig. 3 (a) and 3 (b), the influence of the valve seating and combustion of other cylinders on the monitored cylinder can be analyzed through a diesel engine timing diagram and a whole-cycle acoustic emission signal, so that a signal in a crank angle section related to a fault excitation source of the monitored cylinder is selected to form a data sample, and when an exhaust valve has a gas leakage fault, 20 ° CA before the ignition top dead center (BTDC) to 40 ° CA after the ignition top dead center (ATDC) of the combustion section can be selected as the signal sample by taking a 6DK20 type diesel engine as an example. Cross-model fault diagnosis scenes are simulated through fault data of Z6170 type and 6DK20 type diesel engines, and source domains are definedD S For tagged reference model data sample set, and target domainD T And the model is a data sample set of the model to be diagnosed without a label.
TABLE 1 simulated failure of diesel engine
| Fault code | Severity level | State description |
| Normal | \ | Is |
| F1 | ||
| 1 | Slight wear of the | |
| F2 | ||
| 2 | Exhaust valve severe wear |
Step 2: feasibility analysis for cross-model diagnostics
The data collected by Z6170 type and 6DK20 type diesel engines are analyzed in the embodiment; referring to fig. 3 (a) and 3 (b), the Z6170 type diesel engine and the 6DK20 type diesel engine are both 6-cylinder four-stroke diesel engines, the cylinder cover whole-cycle acoustic emission signals of the monitoring cylinders of the two types are respectively extracted, the excitation sources such as the single-cylinder valve seating excitation, the cylinder implosion pressure excitation and the like are identified according to the timing diagram, it can be seen that the excitation signals in the whole-cycle signals of the two types are obvious and have similar distribution rules, in the embodiment, the air valve air leakage fault is mainly diagnosed, the air valve air leakage excitation mainly occurs in the combustion section, and the two types are not easily interfered by the excitation sources of other cylinders in the combustion section by comparing with the timing diagram, and the attenuation coefficients of the attenuation acoustic emission signals of the Z6170 type diesel engine and the 6DK20 type diesel engine are respectively 7.93 × 10 based on the formulas (1) and (2) -3 And 7.67X 10 -3 (ii) a The envelope is taken to calculate the correlation coefficient value through the formula (3), the correlation coefficient value is 0.87, and the excitation signals are similar, so that the difference of exhaust valve air leakage fault characteristic distribution between two types of machines can be reduced by using the proposed method, and cross-machine type fault diagnosis is carried out.
Wherein adjacent wave crests are taken when the implosion pressure of the cylinder startsx 1 And trough of wavex 2 End adjacent tox n+1 Peak sumx n+2 The number of the wave troughs,n 1 andn 2 respectively detecting the number of wave crests of a reference model and a model to be diagnosed in the cylinder implosion pressure excitation period;δ 1 for the attenuation coefficient of the reference model,δ 2 for the attenuation coefficient of the model to be diagnosed,rpm 1 andrpm 2 the rotating speeds of the reference model and the model to be diagnosed are respectively eliminated based on the method, and the rotating speed variation caused by fixed sampling rate is eliminatednThe difference in (a).
WhereinCov(X,Y) Is the covariance of X and Y,Var[X]is the variance of X and is the sum of the differences,Var[Y]variance of Y. Wherein X and Y are combustion section acoustic emission time domain signals of a reference model and a model to be diagnosed respectively.
And step 3: algorithm training
Firstly, referring to fig. 4, a domain adaptive diagnostic algorithm is established, the domain adaptive diagnostic algorithm mainly includes a feature extraction module, a classification module and a domain adaptive module, in this example, the domain adaptive module is established by using MMD, and the detailed configuration parameters of each module are shown in table 2:
in the feature extraction module, features are extracted through a one-dimensional convolution network, input data samples are convoluted through convolution kernels, and feature mapping is output through an activation function. The formula for the convolutional layer is as follows:
whereinM j Representing an input mapping. Each output map has a deviationb，x j l Is the firstlFirst of a layerjThe output of the computer is used as the output,k i l j is a convolution kernel, selecting a linear rectification unit (ReLU) As a function of activation.
To prevent overfitting and reduce the computational effort, downsampling is performed using a max-pooling layer to reduce the dimensionality of the input data, which in a one-dimensional convolution can be defined as follows:
whereinL in AndL out are the lengths of the data input and output respectively,paddingis the number of padding added on both sides of the data,dilationis the distance between the elements within the sliding window,size kernel is the size of the sliding window.
To reduce the offset of the internal covariates, the convolutional network layer is normalized using a bulk normalization layer. Can be represented by the following equation, whereinεIs to prevent the variance from being zero.
When the input data dimensions of the source domain and the target domain are inconsistent, the extracted feature dimensions are inconsistent, so that the classification module and the domain self-adaptive module cannot be trained due to the fixed number of the full-connection layer network neurons. The adaptive average pooling layer can be used to solve this problem, and it is embedded in the last layer of the feature extraction module, and features with inconsistent dimensions can be converted into consistent features. It is composed of a base, a cover and a coverFirst assume that the required output size issize output . The parameters in the adaptive average pooling layer are then obtained by the following formula to perform the average pooling calculation:
whereinpadding=0，floorIs taken away downwards.
Further, the classification module is used for identifying the state of the diesel engine and the domain self-adapting module. Meanwhile, based on the output of the domain self-adaptive module and the classification module, the MDD algorithm is selected to accurately measure the difference of the feature distribution.
The fully connected layer can be represented as follows:
whereinw j l Is the firstlThe first of the complete connection layeriAnd (4) a weight. The output label of the neural network can then be calculated as follows:
whereinx i last Is on the last fully-connected layeriAnd (6) outputting.
The classification loss selects a cross-entropy loss function, which is defined as follows:
whereinx last Is the output vector on the last fully-connected layer,
is a true label>
The corresponding label vector.
The loss term of the MDD measurement is used for measuring the domain adaptation distribution difference and is defined as follows:
whereiny s Andy t is the probabilistic output of the classification module on the source domain and the target domain, respectively.y s adv Andy t adv is the output of the domain adaptation module.γIs a parameter that is in excess of the parameter,h y is output when the probability is asyA predictive tag of time.
In the training process of the algorithm, the optimization of the whole algorithm consists of two parts: the classification loss of the source domain and the feature distribution difference loss between the source domain and the target domain. The optimization objective is given by the following formula:
wherein the coefficientsλWhich is used to adjust the balance between these two loss terms in the algorithm optimization process to promote positive migration.
In order to optimize the proposed diagnostic algorithm and update the parameters, an Adam optimization algorithm is used to minimize the loss function. The algorithm designs independent adaptive learning rates for different parameters by calculating first moment and second moment estimates of the gradient. It has higher computational efficiency and lower memory requirements.
Based on the above materials, the training process of the proposed method is summarized in table 3.
Table 3 training procedure for the proposed method:
the algorithm is as follows: training procedure of the proposed method
Inputting:
training setD train ϵData S ,T[] Test setD test ϵ Data T[] Source domain
，
Target domain
，
-weightλLearning rateεNumber of iterationsepochsAnd bytes size b
1 outputting the diagnosis result of the proposed methody
Weight and deviation ← random initial training parameters
3: for epoch from 1 to epochs do:
4: for batch from 1 to number of batches do:
5: # Forward propagation
6: G f （D train ) Oid step of computing extracted features based on expressions (4) to (9)
7: G y （G f ) AndG d （G f ) Oid label and MDD label are calculated based on equations (10) - (11)
8: err c +λD γ （S，T) ← calculating a classification loss and a feature distribution difference loss by equations (12) to (14)
9: # counter propagation
10 optimization of equation (14) by Adam's algorithmLTo update neural network parameters
11: end
12: end
13 rate of accuracyG y （G f （D test ) Oid.). No. Paid test setD test Verifying the Performance of the proposed method
And 4, step 4: fault diagnosis
Based on the step 3, training a domain adaptive diagnostic algorithm by taking the source domain sample and part of the target domain sample as a training set; the domain self-adaptive module in the algorithm is only used for reducing the feature distribution difference of a source domain and a target domain during training, so that a fault diagnosis module is only formed by the feature extraction module and the classification module during fault diagnosis of the diesel engine so as to diagnose a target domain label-free data sample acquired in real time; another part of the target domain unlabeled samples are the test set, verifying the accuracy of the proposed diagnostic method, for example, the fault diagnosis model can be obtained by training data samples from 0%, 25%, 50%, 75%, and 90% load (source domain) of the 6DK20 model and 25% load (target domain) condition of the Z6170 model, i.e., the task of migrating from the 6DK20 model to the 25% load condition of Z6170, noted as 0%/25%/50%/75%/90% (6 DK 20) → 25% (Z6170), as shown in table 4; the diagnostic effect can then be verified at 25% load (test set). Comparative tests were performed using PCA _ SVM (machine learning) and 1d _cnn (deep learning), in which samples of multiple conditions of one model were used for training and tested under another model. For example, a fault diagnosis model can be obtained by training 0%, 25%, 50%, 75% and 90% load (source domain) conditions of a 6DK20 model, and then the fault diagnosis effect of the model can be verified under 25% load conditions (target domain) of the Z6170 model.
Supervised training and testing was performed using PCA _ SVM and 1d _cnn, and the proposed diagnostic method was used for domain adaptive training and testing, resulting in diagnostic accuracy rates as shown in table 5. Because the diagnosis between different diesel engine models is carried out, the difference of the characteristic distribution is large, and the diagnosis accuracy is low when the fault diagnosis model trained from one diesel engine model is directly applied to another diesel engine model by supervised learning. The proposed domain adaptive method can find the common feature distribution between the two models and make the cross-model fault diagnosis possible. As can be seen from the figure, the proposed method is superior to other methods with an average accuracy of 100% and 97.26% in the migration tasks of 6DK20 → Z6170 and Z6170 → 6DK20, respectively. The diagnostic accuracy of the migration task 6DK20 → Z6170 is better than the migration task Z6170 → 6DK20, which may be due to the fact that the samples in the source domain in the migration task 6DK20 → Z6170 contain more condition data, which allows the target domain to find a more common feature distribution in the source domain when performing domain adaptive training.
TABLE 5 Accuracy of different methods in the test set
Fig. 5 shows a two-dimensional distribution of features extracted by different diagnostic methods, where the time-domain and frequency-domain features extracted by PCA-SVM are very dispersed in the feature distribution, with very close inter-class distances. 1D _CNN, because the characteristics extracted by the method are trained together by using samples of a plurality of working conditions in the source domain, the characteristic distribution difference between the working conditions is eliminated, but the characteristic distribution difference between the source domain and the target domain is obvious. The proposed method reduces the difference in feature distribution by finding a common feature expression in a labeled source domain (one model) and an unlabeled target domain (another model). As can be seen from fig. 5, the proposed method not only has distinct classification boundaries, but also has relatively tight feature distribution clusters within the same class.
As shown in fig. 6, the terminal device according to the embodiment of the present invention includes a memory, a processor, and a communication bus, where the memory and the processor complete communication with each other through the communication bus. The communication bus may communicate with the outside through a communication interface. The processor may invoke logic instructions in the memory to perform the cross-model diesel engine combustor fault diagnosis method steps of the embodiments described above.
The present application also provides a non-transitory computer readable storage medium, such as a flash memory, a hard disk, a multimedia card, a card type memory (e.g., SD or DX memory, etc.), a Random Access Memory (RAM), a Static Random Access Memory (SRAM), a read-only memory (ROM), an electrically erasable programmable read-only memory (EEPROM), a programmable read-only memory (PROM), a magnetic memory, a magnetic disk, an optical disk, a server, an App, etc., having stored thereon a computer program that, when executed by a processor, implements the corresponding functions. The computer readable storage medium of the present embodiment is configured to implement the cross-model diesel engine combustion chamber fault diagnosis method of the method embodiment when executed by a processor.
It will be appreciated that modifications and variations are possible to those skilled in the art in light of the above teachings, and it is intended to cover all such modifications and variations as fall within the scope of the appended claims.
Claims (9)
1. A fault diagnosis method for a combustion chamber of a cross-model diesel engine is characterized by comprising the following steps:
acquiring signals related to a fault excitation source in excitation signals of a reference model and a model diesel engine to be diagnosed, comparing the similarity of the signals between the models, if the signals are similar, taking a labeled data sample of the reference model as a source domain, taking an acquired unlabeled data sample of the model to be diagnosed as a target domain, and establishing a training sample set; the method comprises the steps that a training sample set comprises a source domain data sample set and a target domain data sample set, a reference label data set is constructed through fault simulation tests and real machine data accumulation of a certain diesel engine type, a vibration/acoustic emission time domain signal segment related to an excitation source of a cylinder is determined based on a diesel engine ignition sequence and a timing diagram, and a source domain data sample set for model training is formed; forming a target domain data sample set for model training and diagnosis by adopting the same method for a non-label data set of an unknown state acquired by another diesel engine type to be diagnosed;
constructing a fault diagnosis network model, wherein the model is obtained by training samples in a training sample set and comprises a feature extraction module, a classification module and a domain self-adaptive module, wherein the feature extraction module comprises a plurality of convolution layers and a self-adaptive average pooling layer, feature extraction is carried out on the plurality of convolution layers, average pooling calculation is carried out on the self-adaptive average pooling layer, and features with consistent dimensions are formed; the extracted features with consistent dimensions are respectively input into a classification module and a domain self-adaptive module, classification loss and feature distribution difference loss of different models are calculated, and finally network parameters are automatically updated through an Adam optimization algorithm to reduce feature distribution difference and improve classification accuracy;
and inputting a signal section which is acquired by the machine type diesel engine to be diagnosed in real time and is related to the fault excitation source into the fault diagnosis network model to diagnose the fault of the cross-machine type diesel engine combustion chamber, so as to obtain a fault diagnosis result.
2. The cross-model diesel engine combustion chamber fault diagnosis method according to claim 1, wherein the feature extraction module specifically comprises 5 convolution layers, and each convolution layer specifically performs 1-dimensional convolution, linear rectification, maximum pooling and 1-dimensional batch normalization on the input signal.
3. The cross-model diesel engine combustion chamber fault diagnosis method according to claim 1, wherein an adaptive average pooling layer is embedded in the last layer of the feature extraction module, and converts features with inconsistent dimensions into features with consistent dimensions.
4. The cross-model diesel engine combustion chamber fault diagnosis method according to claim 1, wherein similarity of signals between models is specifically judged by calculating a signal attenuation coefficient and a correlation coefficient.
5. A cross-model diesel engine combustion chamber fault diagnosis system is characterized by comprising:
the sample acquisition module is used for acquiring signals related to a fault excitation source in excitation signals of a reference model and a model diesel engine to be diagnosed, comparing the similarity of the signals between the models, if the signals are similar, taking a labeled data sample of the reference model as a source domain, taking an unlabeled data sample acquired by the model diesel engine to be diagnosed as a target domain, and establishing a training sample set; the method comprises the steps that a training sample set comprises a source domain data sample set and a target domain data sample set, a reference label data set is constructed through fault simulation test and real machine data accumulation of a certain diesel engine type, a vibration/sound emission time domain signal segment related to a cylinder excitation source is determined based on a diesel engine ignition sequence and a timing diagram, and a source domain data sample set for model training is formed; forming a target domain data sample set for model training and diagnosis by adopting the same method for a tag-free data set of an unknown state acquired by another diesel engine type to be diagnosed;
the fault diagnosis network model building module is used for building a fault diagnosis network model, the fault diagnosis network model is obtained by training samples in a training sample set, and the fault diagnosis network model comprises a feature extraction module, a classification module and a domain self-adaption module, wherein the feature extraction module comprises a plurality of convolution layers and a self-adaption average pooling layer, specifically, feature extraction is carried out through the plurality of convolution layers, and then average pooling calculation is carried out through the self-adaption average pooling layer to form features with consistent dimensions; the extracted features with consistent dimensions are respectively input into a classification module and a domain self-adaptive module, classification loss and feature distribution difference loss of different models are calculated, and finally, network parameters are automatically updated through an Adam optimization algorithm to reduce feature distribution difference and improve classification accuracy;
and the diagnosis module is used for inputting a signal section which is acquired by the machine type diesel engine to be diagnosed in real time and is related to the fault excitation source into the fault diagnosis network model to carry out cross-machine type diesel engine combustion chamber fault diagnosis so as to obtain a fault diagnosis result.
6. The cross-model diesel engine combustion chamber fault diagnosis system according to claim 5, wherein the feature extraction module specifically comprises 5 convolution layers, and each convolution layer specifically performs 1-dimensional convolution, linear rectification, maximum pooling and 1-dimensional batch normalization on the input signal.
7. The cross-model diesel engine combustion chamber fault diagnosis system according to claim 5, wherein an adaptive average pooling layer is embedded in the last layer of the feature extraction module, and converts features with inconsistent dimensions into features with consistent dimensions.
8. Terminal equipment, comprising a memory and a processor, wherein the memory stores a computer program executable by the processor, and the terminal equipment is characterized in that the processor implements the steps of the cross-model diesel engine combustion chamber fault diagnosis method according to any one of claims 1 to 4 when executing the computer program.
9. A non-transitory computer readable storage medium having a computer program stored thereon, wherein the computer program when executed by a processor implements the steps of the cross-model diesel engine combustion chamber fault diagnosis method according to any one of claims 1 to 4.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211471504.2A CN115510926B (en) | 2022-11-23 | 2022-11-23 | Cross-machine type diesel engine combustion chamber fault diagnosis method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211471504.2A CN115510926B (en) | 2022-11-23 | 2022-11-23 | Cross-machine type diesel engine combustion chamber fault diagnosis method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115510926A CN115510926A (en) | 2022-12-23 |
| CN115510926B true CN115510926B (en) | 2023-04-18 |
Family
ID=84513936
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211471504.2A Active CN115510926B (en) | 2022-11-23 | 2022-11-23 | Cross-machine type diesel engine combustion chamber fault diagnosis method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115510926B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117669388A (en) * | 2024-01-30 | 2024-03-08 | 武汉理工大学 | Fault sample generation method, device and computer medium |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104454785A (en) * | 2014-09-25 | 2015-03-25 | 中南大学 | Engineering machinery hydraulic cylinder fault diagnosis system and fault sample signal acquisition method applicable to same |
| CN107438398A (en) * | 2015-01-06 | 2017-12-05 | 大卫·伯顿 | Portable wearable monitoring system |
Family Cites Families (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR20200075344A (en) * | 2018-12-18 | 2020-06-26 | 삼성전자주식회사 | Detector, method of object detection, learning apparatus, and learning method for domain transformation |
| WO2020256732A1 (en) * | 2019-06-21 | 2020-12-24 | Siemens Aktiengesellschaft | Domain adaptation and fusion using task-irrelevant paired data in sequential form |
| US11580405B2 (en) * | 2019-12-26 | 2023-02-14 | Sap Se | Domain adaptation of deep neural networks |
| US11526807B2 (en) * | 2020-03-20 | 2022-12-13 | Beijing Didi Infinity Technology And Development Co., Ltd. | Machine learning systems and methods with source-target adaptation |
| CN112308147B (en) * | 2020-11-02 | 2024-02-09 | 西安电子科技大学 | Rotary machinery fault diagnosis method based on multi-source domain anchor adapter integrated migration |
| CN113095229B (en) * | 2021-04-14 | 2024-04-12 | 中国矿业大学 | Self-adaptive pedestrian re-identification system and method for unsupervised domain |
| CN114398992A (en) * | 2022-01-18 | 2022-04-26 | 安徽大学 | Intelligent fault diagnosis method based on unsupervised domain adaptation |
| CN114580288A (en) * | 2022-03-09 | 2022-06-03 | 杭州电子科技大学 | Intelligent fault diagnosis method based on DSECJAN |
| CN114755017B (en) * | 2022-04-13 | 2022-12-20 | 湖南大学 | Variable-speed bearing fault diagnosis method of cross-domain data driving unsupervised field shared network |
-
2022
- 2022-11-23 CN CN202211471504.2A patent/CN115510926B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104454785A (en) * | 2014-09-25 | 2015-03-25 | 中南大学 | Engineering machinery hydraulic cylinder fault diagnosis system and fault sample signal acquisition method applicable to same |
| CN107438398A (en) * | 2015-01-06 | 2017-12-05 | 大卫·伯顿 | Portable wearable monitoring system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115510926A (en) | 2022-12-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110954312B (en) | Reciprocating mechanical fault diagnosis method based on keyless phase whole-cycle signal | |
| Qin et al. | DTCNNMI: A deep twin convolutional neural networks with multi-domain inputs for strongly noisy diesel engine misfire detection | |
| CN110849626B (en) | Self-adaptive sparse compression self-coding rolling bearing fault diagnosis system | |
| Flett et al. | Fault detection and diagnosis of diesel engine valve trains | |
| Kumar et al. | Improved CNN for the diagnosis of engine defects of 2-wheeler vehicle using wavelet synchro-squeezed transform (WSST) | |
| CN106761668B (en) | Oil well failure intelligent analysis decision system and method | |
| Qin et al. | Anti‐noise diesel engine misfire diagnosis using a multi‐scale CNN‐LSTM neural network with denoising module | |
| Wang et al. | Research on the fault monitoring method of marine diesel engines based on the manifold learning and isolation forest | |
| Shahid et al. | Real-time abnormality detection and classification in diesel engine operations with convolutional neural network | |
| CN112016251B (en) | Nuclear power device fault diagnosis method and system | |
| CN115510926B (en) | Cross-machine type diesel engine combustion chamber fault diagnosis method and system | |
| CN112528519A (en) | Method, system, readable medium and electronic device for engine quality early warning service | |
| CN113591553B (en) | Turbine pump transfer learning fault intelligent judgment method based on small sample weight optimization | |
| CN111597760B (en) | Method for obtaining gas path parameter deviation value under small sample condition | |
| CN114297918A (en) | Aero-engine residual life prediction method based on full-attention depth network and dynamic ensemble learning | |
| Montazeri-Gh et al. | Gas path component fault diagnosis of an industrial gas turbine under different load condition using online sequential extreme learning machine | |
| Ye et al. | A deep learning-based method for automatic abnormal data detection: Case study for bridge structural health monitoring | |
| Kang et al. | Hierarchical level fault detection and diagnosis of ship engine systems | |
| CN117009814A (en) | Engine fault diagnosis method and system based on multi-source information fusion | |
| CN112487709A (en) | Marine diesel engine fault tracing method based on sensitivity analysis | |
| CN115791174B (en) | Rolling bearing abnormality diagnosis method, system, electronic equipment and storage medium | |
| CN112215281A (en) | Fan blade icing fault detection method | |
| Yu et al. | Remaining useful life prediction based on multi-scale residual convolutional network for aero-engine | |
| CN115597879A (en) | Method for positioning and identifying fault information in data-driven diesel engine vibration signal | |
| CN112560252A (en) | Prediction method for residual life of aircraft engine |
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 |