CN114264478A - Diesel engine crankshaft bearing wear degree prediction method and system - Google Patents
Diesel engine crankshaft bearing wear degree prediction method and system Download PDFInfo
- Publication number
- CN114264478A CN114264478A CN202111573265.7A CN202111573265A CN114264478A CN 114264478 A CN114264478 A CN 114264478A CN 202111573265 A CN202111573265 A CN 202111573265A CN 114264478 A CN114264478 A CN 114264478A
- Authority
- CN
- China
- Prior art keywords
- neural network
- wear degree
- classification accuracy
- judgment result
- diesel engine
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Landscapes
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The invention relates to a method and a system for predicting the wear degree of a crankshaft bearing of a diesel engine, which belong to the field of crankshaft bearings of diesel engines, and are characterized in that the characteristics of accelerated vibration signal samples corresponding to each wear degree are extracted by a method combining complementary set empirical mode decomposition and singular value decomposition, the complementary set empirical mode decomposition can better eliminate the influence generated by added white Gaussian noise compared with EMD and EEMD, so that the reconstruction error is smaller, the singular value decomposition has good stability, the complementary set empirical mode decomposition and the singular value decomposition are combined to well extract the characteristics of vibration signals, a cyclic neural network is optimized by utilizing a simulated annealing algorithm according to the extracted characteristics, the prediction accuracy of an original cyclic neural network model is improved, and the accuracy of the prediction of the wear degree of the crankshaft bearing of the diesel engine is further improved.
Description
Technical Field
The invention relates to the field of diesel engine crankshaft bearings, in particular to a method and a system for predicting the wear degree of a diesel engine crankshaft bearing.
Background
In recent years, military vehicles play an important role in earthquake relief, and in order to guarantee normal operation of the vehicles, fault detection on machinery at regular intervals becomes a necessary preventive measure, and the judgment of internal parts by experts inevitably requires disassembly of some parts, so that the requirements of detectors are high while manpower and material resources are consumed. The detection of machine faults by computer communication technology is the focus of research of scholars at home and abroad.
ChenY C and other [1], utilize Fast Fourier Transform (FFT) to analyze and process the vibration signal, but its characteristic has easy gathering characteristic, can't be very good to discern, need artificial intervention; zhangling et al [2], a wavelet packet-AR spectrum is adopted to perform feature extraction on the bearing fault of the transmission, but the method has the defect of difficulty in selecting wavelet base; cheng J et al [3], utilizes Empirical Mode Decomposition (EMD) to decompose the vibration signal, but EMD can generate Mode aliasing; xia, W, etc. [4], decomposing signals by using self-adaptive wavelets and Ensemble Empirical Mode Decomposition (EEMD) and analyzing the size of an oil injection advance angle by using a fault tree, but because actual operation is limited, the influence of added white noise cannot be well eliminated, and reconstruction errors can be caused; zhou Xiaolin et al [5], use supplementary set empirical mode decomposition, satin blue gardener optimization algorithm and Least square Support Vector Regression (LSSVT) to realize ultrashort-term wind power combination prediction, but Least square Support Vector Regression has poor effect on multi-classification problem processing.
Disclosure of Invention
The invention aims to provide a method and a system for predicting the wear degree of a crankshaft bearing of a diesel engine so as to improve the accuracy of predicting the wear degree of the crankshaft bearing of the diesel engine.
In order to achieve the purpose, the invention provides the following scheme:
a method for predicting wear of a crankshaft bearing of a diesel engine, the method comprising:
acquiring accelerated vibration signal samples of different wear degrees of a crankshaft bearing of the diesel engine during operation;
extracting the characteristics of the accelerated vibration signal sample corresponding to each wear degree by using a method of combining complementary set empirical mode decomposition and singular value decomposition to obtain a characteristic sample set corresponding to each wear degree;
taking the wear degrees as labels and the characteristic sample sets as input quantities, and forming a training set and a testing set by using all the wear degrees and the characteristic sample sets corresponding to all the wear degrees;
optimizing the recurrent neural network by using a simulated annealing algorithm based on the training set and the test set to obtain the optimized recurrent neural network;
acquiring a real-time acceleration vibration signal of a diesel engine crankshaft bearing to be detected in operation;
complementary set empirical mode decomposition and singular value decomposition are carried out on the real-time accelerated vibration signal to obtain a feature set;
and inputting the characteristic set into the optimized recurrent neural network, and outputting the wear degree of the crankshaft bearing of the diesel engine to be tested.
Optionally, the obtaining of samples of acceleration vibration signals of the diesel engine crankshaft bearing with different wear degrees during operation further includes:
and performing wavelet threshold denoising on the accelerated vibration signal sample by adopting a soft threshold method with a threshold as an unbiased likelihood estimation threshold.
Optionally, the extracting, by using a method combining empirical mode decomposition and singular value decomposition of a complementary set, features of the accelerated vibration signal sample corresponding to each wear degree to obtain a feature sample set corresponding to each wear degree specifically includes:
performing complementary set empirical mode decomposition on the accelerated vibration signal sample corresponding to each wear degree to obtain a plurality of intrinsic mode functions, and forming the first 6 intrinsic mode functions into an initial vector matrix corresponding to each wear degree;
and performing singular value decomposition on the initial vector matrix corresponding to each wear degree to obtain a plurality of singular values, and forming a characteristic sample set corresponding to each wear degree by the plurality of singular values.
Optionally, the optimizing the recurrent neural network by using a simulated annealing algorithm based on the training set and the test set to obtain the optimized recurrent neural network specifically includes:
initializing parameters of a simulated annealing algorithm; the simulated annealing algorithm parameters comprise an initial temperature, iteration times and a termination temperature;
initializing an initial solution of a recurrent neural network; the initial solution comprises a sequence of length x1And the number of hidden layer nodes is x2；
Using training set to sequence length x1And the number of hidden layer nodes is x2Training the circulating neural network, and testing the trained circulating neural network by using a test set to obtain a first classification accuracy;
randomly generating a new solution of the recurrent neural network; the new solution comprises a sequence length of x1' and number of hidden layer nodes is x2'；
Using training set to sequence length x1' and number of hidden layer nodes is x2' the recurrent neural network is trained, and the trained recurrent neural network is tested by using a test set to obtain a second classification criterionDetermining the rate;
judging whether the increment of the second classification accuracy rate relative to the first classification accuracy rate is larger than or equal to zero or not, and obtaining a first judgment result;
if the first judgment result shows that the first classification accuracy rate is the first classification accuracy rate, receiving a new solution as a new initial solution, and assigning the second classification accuracy rate to the first classification accuracy rate;
if the first judgment result shows no, receiving a new solution according to the Metropolis criterion;
judging whether the current iteration times are greater than or equal to the iteration times to obtain a second judgment result;
if the second judgment result shows no, returning to the step of randomly generating a new solution of the recurrent neural network;
if the second judgment result shows yes, judging whether the current temperature is lower than the target temperature or not, and obtaining a third judgment result;
if the third judgment result shows no, reducing the current temperature, and returning to the step of randomly generating a new solution of the recurrent neural network;
and if the third judgment result shows that the current solution is the optimal solution, outputting the new current solution as the optimal solution, and outputting the second classification accuracy as the optimal classification accuracy.
A system for predicting wear of a crankshaft bearing of a diesel engine, the system comprising:
the acceleration vibration signal sample acquisition module is used for acquiring acceleration vibration signal samples of different wear degrees of the diesel engine crankshaft bearing during operation;
the characteristic sample set acquisition module is used for extracting the characteristics of the accelerated vibration signal sample corresponding to each wear degree by utilizing a method of combining complementary set empirical mode decomposition and singular value decomposition to obtain a characteristic sample set corresponding to each wear degree;
the training set and test set forming module is used for forming a training set and a test set by using the wear degrees as labels and the characteristic sample set as input quantities, wherein the characteristic sample set corresponds to all the wear degrees;
the cyclic neural network optimization module is used for optimizing the cyclic neural network by utilizing a simulated annealing algorithm based on the training set and the test set to obtain the optimized cyclic neural network;
the real-time acceleration vibration signal acquisition module is used for acquiring a real-time acceleration vibration signal of the diesel engine crankshaft bearing to be detected in the operation process;
the characteristic set obtaining module is used for carrying out complementary set empirical mode decomposition and singular value decomposition on the real-time acceleration vibration signal to obtain a characteristic set;
and the wear degree output module is used for inputting the feature set into the optimized recurrent neural network and outputting the wear degree of the crankshaft bearing of the diesel engine to be tested.
Optionally, the system further includes:
and the denoising module is used for performing wavelet threshold denoising on the accelerated vibration signal sample by adopting a soft threshold method with the threshold being an unbiased likelihood estimation threshold.
Optionally, the feature sample set obtaining module specifically includes:
the complementary set empirical mode decomposition submodule is used for carrying out complementary set empirical mode decomposition on the accelerated vibration signal sample corresponding to each wear degree to obtain a plurality of intrinsic mode functions, and the first 6 intrinsic mode functions form an initial vector matrix corresponding to each wear degree;
and the singular value decomposition submodule is used for carrying out singular value decomposition on the initial vector matrix corresponding to each wear degree to obtain a plurality of singular values, and meanwhile, the plurality of singular values form a characteristic sample set corresponding to each wear degree.
Optionally, the recurrent neural network optimization module specifically includes:
the parameter sub-initialization module is used for initializing the parameters of the simulated annealing algorithm; the simulated annealing algorithm parameters comprise an initial temperature, iteration times and a termination temperature;
the initial solution initialization submodule is used for initializing the initial solution of the recurrent neural network; the initial solution comprises a sequence of length x1And the number of hidden layer nodes is x2；
First classificationAn accuracy obtaining submodule for using the training set to obtain the length x of the sequence1And the number of hidden layer nodes is x2Training the circulating neural network, and testing the trained circulating neural network by using a test set to obtain a first classification accuracy;
a new solution generation submodule for randomly generating a new solution of the recurrent neural network; the new solution comprises a sequence length of x1' and number of hidden layer nodes is x2'；
A second classification accuracy obtaining submodule for using the training set to obtain the length x of the sequence1' and number of hidden layer nodes is x2The recurrent neural network of' is trained, and the trained recurrent neural network is tested by using a test set to obtain a second classification accuracy;
the first judgment result obtaining submodule is used for judging whether the increment of the second classification accuracy relative to the first classification accuracy is larger than or equal to zero or not and obtaining a first judgment result;
the assignment submodule is used for receiving a new solution as a new initial solution if the first judgment result shows that the first classification accuracy is correct, and assigning the second classification accuracy to the first classification accuracy;
a new solution receiving submodule, configured to receive a new solution according to the Metropolis criterion if the first determination result indicates no;
a second judgment result obtaining submodule for judging whether the current iteration times is greater than or equal to the iteration times to obtain a second judgment result;
a step returning submodule, configured to return to the step "randomly generate a new solution of the recurrent neural network" if the second determination result indicates no;
a third judgment result obtaining submodule, configured to, if the second judgment result indicates yes, judge whether the current temperature is lower than the target temperature, and obtain a third judgment result;
a temperature reduction submodule, configured to reduce the current temperature if the third determination result indicates no, and return to the step "randomly generate a new solution of the recurrent neural network";
and the result output submodule is used for outputting the new current solution as the optimal solution and outputting the second classification accuracy as the optimal classification accuracy if the third judgment result shows that the current solution is the optimal solution.
According to the specific embodiment provided by the invention, the invention discloses the following technical effects:
the invention discloses a method and a system for predicting the wear degree of a crankshaft bearing of a diesel engine, wherein the characteristics of an accelerated vibration signal sample corresponding to each wear degree are extracted by a method combining complementary set empirical mode decomposition and singular value decomposition, the complementary set empirical mode decomposition can better eliminate the influence generated by added white Gaussian noise compared with EMD and EEMD (empirical mode decomposition), so that the reconstruction error is smaller, the singular value decomposition has good stability, the complementary set empirical mode decomposition and the singular value decomposition are combined to well extract the characteristics of a vibration signal, a cyclic neural network is optimized by using a simulated annealing algorithm according to the extracted characteristics, the prediction accuracy of an original cyclic neural network model is improved, and the accuracy of the prediction of the wear degree of the crankshaft bearing of the diesel engine is further improved.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings without inventive exercise.
FIG. 1 is a flow chart of a method for predicting wear of a crankshaft bearing of a diesel engine according to the present invention;
FIG. 2 is a schematic diagram of a method for predicting the degree of wear of a crankshaft bearing of a diesel engine according to the present invention;
FIG. 3 is a schematic diagram of a simulated annealing algorithm optimized recurrent neural network provided by the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The invention aims to provide a method and a system for predicting the wear degree of a crankshaft bearing of a diesel engine so as to improve the accuracy of predicting the wear degree of the crankshaft bearing of the diesel engine.
In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in further detail below.
The invention provides a method for predicting the wear degree of a crankshaft bearing of a diesel engine, which comprises the following steps of:
The research object is a certain type of diesel engine, a sensor is placed at a specified position, and an acceleration vibration signal is collected to serve as an initial signal of an experiment.
Due to the interference of noise existing outside, wavelet threshold denoising is carried out on the acquired vibration signals x (t), and a soft threshold method with a threshold value being an unbiased likelihood estimation threshold value is selected for processing to obtain x' (t).
102, extracting the characteristics of the accelerated vibration signal sample corresponding to each wear degree by using a method of combining complementary set empirical mode decomposition and singular value decomposition to obtain a characteristic sample set corresponding to each wear degree, wherein the specific implementation process is as follows:
102-1, performing complementary set empirical mode decomposition on accelerated vibration signal samples corresponding to each wear degree to obtain a plurality of eigenmode functions, and forming the first 6 eigenmode functions into an initial vector matrix corresponding to each wear degree;
taking a vibration signal as an example, a Complementary Ensemble Empirical Mode Decomposition (CEEMD) is used for decomposition, and the steps are as follows: randomly equal length and given standard deviationWhite noise of opposite sign
And added to the signal: to obtain
Is the ith1Decomposing the signal after adding white noise for the first time by using EMD algorithm
For example, the following steps are carried out:
i＝1，h0＝ri-1(t)，j＝1。
obtaining the ith IMF
(a) Find hj-1(t) local extreme points.
(b) To hj-1And (t) respectively carrying out cubic spline function interpolation on the maximum value point and the minimum value point to form an upper envelope line and a lower envelope line.
(c) Calculate the mean m of the upper and lower envelopesj-1(t)。
(d) Let hj(t)＝hj-1(t)-mj-1(t)。
(e) If hj(t) is an IMF function, then IMFi(t)＝hj(t); otherwise, j equals j +1, go to (a)
Let r bei(t)＝ri-1(t)-imfi(t)。
If r isi(t) if the number of extreme points is still more than 2, turning to 2 if i is i + 1; otherwise, decomposition ends, ri(t) is the residual component. The algorithm finally yields:
obtaining the IMF components and the Re components with preset numbers.
By analogy, the following steps are carried out
J (j) decomposed by empirical mode decomposition1Each IMF component is
And
the component Re is respectively
And
summing each IMF component and Re and averaging to obtain:
to the j th final product1An intrinsic mode function; r is the decomposed remainder component.
And 102-2, performing singular value decomposition on the initial vector matrix corresponding to each wear degree to obtain a plurality of singular values, and forming a characteristic sample set corresponding to each wear degree by the plurality of singular values.
Taking a vibration signal as an example, the first 6 IMF components after CEEMD decomposition in the previous step form an initial vector matrix Ee×gThe rank of the matrix is r', singular value decomposition is carried out, E ═ UDWT，Ue×e＝[u1，u2，...，ue]，
Δr′×r′＝diag(σ1，σ2，...，σr′)，
λ1≥λ2≥...≥λr′≧ 0 is the matrix ETCharacteristic value of E. To obtain Ee×gSingular value ofi(i ═ 1, 2.., r'), and the obtained feature component σiAs a set of features for a neural network. I of n signal2A characteristic component of
Take the first feature set as an example. n is 1, i21, 2.. 6, the first piece of data of the feature set is: sigma12、σ12、...、σ16The nth data is σn2、σn2、...、σn6。
And extracting features by using SVD to obtain a feature set which is used as an input of the neural network.
And 103, forming a training set and a testing set by using the wear degrees as labels and the characteristic sample sets as input quantities, wherein all the wear degrees and the characteristic sample sets corresponding to all the wear degrees.
And 104, optimizing the recurrent neural network by using a simulated annealing algorithm based on the training set and the test set to obtain the optimized recurrent neural network.
Taking a cyclic neural network as an objective function f () of a Simulated Annealing (SA) algorithm, and setting the number of nodes of the hidden layer as an independent variable x of the objective function1Setting the sequence length as the argument x of the objective function2The basic steps of the simulated annealing algorithm are as follows:
1. initialization: initial temperature T0(high enough), initial solution S (iteration start), iteration number L, termination temperature T1。
2. And (3) carrying out steps 3 to 6 on k which is 1, … and L.
3. A new solution S' is generated.
4. The increment Δ C ═ f (S') -f (S), f () is calculated as the objective function.
5. If the delta C is larger than or equal to 0, receiving S' as a new current solution; otherwise, generating a random number epsilon in the interval of [0,1] and comparing the random number epsilon with the probability P, if epsilon is less than P, receiving a new solution, otherwise, keeping S.
6. If the iteration times are not met, turning to the step 3;
see if the target temperature is met, if not, T is gradually reduced, and T>T1And then turning to the step 3.
7. And obtaining the optimal classification prediction accuracy S.
The training and testing process of the Recurrent Neural Network (RNN) is as follows:
dividing the obtained characteristic set T (n) into a training set T1And test set T2Wherein the training set T1Is given by the label y1The number of test sets is N, and the test set T2Is given by the label y2. The basic steps of the recurrent neural network are: training set T with training set1And label y1For input, training is performed, inputting test set T2Making network test and outputting prediction label y2′If the correct number is N0Let N stand for0-0, if y'2(a)-y2(a) 0(a 1, 2.., N) is N0＝N0+1, cycle from 1 to N, from formula
Where s is the accuracy of the classification prediction. The parameters and weights of the recurrent neural network are derived as follows:
let x 'be the input of the neural network, h be the hidden layer unit, o be the output, L be the loss function, and y' be the label of the training set.
In the formula: f. of1() G () is an activation function. In the training and parameter tuning process of the RNN, only W, U, V parameters are required to be tuned and optimized. Partial derivatives of V are solved for the loss function L:
since the loss of the network is proportional to time, the recurrent neural network requiresFrom the beginning to t1Time partial derivation:
due to the fact that
Can obtain
And
related, observations and findings
There are also W and U, i.e. the solution of the partial derivatives of W and U is also related to the data at all historical times.
Thus, the following steps are obtained:
referring to fig. 3, the specific steps of the simulated annealing algorithm for optimizing the recurrent neural network are as follows:
initializing parameters of a simulated annealing algorithm; the simulated annealing algorithm parameters comprise an initial temperature, iteration times and a termination temperature;
initializing an initial solution of a recurrent neural network; the initial solution includes a sequence of length x1And the number of hidden layer nodes is x2；
Using training set to sequence length x1And the number of hidden layer nodes is x2Training the circulating neural network, and testing the trained circulating neural network by using a test set to obtain a first classification accuracy;
randomly generating a new solution of the recurrent neural network; the new solution includes a sequence length of x1' and number of hidden layer nodes is x2'；
Using training set to sequence length x1' and number of hidden layer nodes is x2The recurrent neural network of' is trained, and the trained recurrent neural network is tested by using a test set to obtain a second classification accuracy;
judging whether the increment of the second classification accuracy rate relative to the first classification accuracy rate is larger than or equal to zero or not, and obtaining a first judgment result;
if the first judgment result shows that the first classification accuracy rate is the first classification accuracy rate, receiving a new solution as a new initial solution, and assigning the second classification accuracy rate to the first classification accuracy rate;
if the first judgment result shows no, receiving a new solution according to the Metropolis criterion;
judging whether the current iteration times are greater than or equal to the iteration times (10 times) or not, and obtaining a second judgment result;
if the second judgment result shows no, returning to the step of randomly generating a new solution of the recurrent neural network;
if the second judgment result shows yes, judging whether the current temperature is lower than the target temperature or not, and obtaining a third judgment result;
if the third judgment result shows no, reducing the current temperature (by 10 percent), and returning to the step of randomly generating a new solution of the recurrent neural network;
and if the third judgment result shows that the current solution is the optimal solution, outputting the new current solution as the optimal solution, and outputting the second classification accuracy as the optimal classification accuracy.
And 105, acquiring a real-time acceleration vibration signal of the diesel engine crankshaft bearing to be detected in the operation process.
And 106, performing complementary set empirical mode decomposition and singular value decomposition on the real-time acceleration vibration signal to obtain a feature set.
Other optimization algorithms, such as a particle swarm algorithm, an ant colony algorithm, an exercise gardener algorithm, a genetic algorithm, a bat algorithm, a flower pollination algorithm, and the like, can also be selected for optimizing the RNN.
And step 107, inputting the feature set into the optimized recurrent neural network, and outputting the wear degree of the crankshaft bearing of the diesel engine to be tested.
Finally, the accuracy S of the classification prediction of the wear degree of the crankshaft bearing of the diesel engine by using the recurrent neural network optimized by the simulated annealing algorithm reaches 97.45762712%, and is improved by more than 5% compared with the prediction accuracy of the basic RNN network. The model can effectively carry out fault prediction on the obtained vibration signals.
The invention has the following advantages:
1) the CEEMD can improve the mode confusion phenomenon of the EMD and the EEMD, and can better eliminate the influence generated by the added white Gaussian noise, so that the reconstruction error is smaller and the completeness is better.
2) SVD singular value decomposition has good stability, CEEMD-SVD can well extract the characteristics of the vibration signal.
3) The RNN recurrent neural network can process input with any length, is also suitable for multi-output problems, and is suitable for predicting the wear degree of a crankshaft bearing.
4) The recurrent neural network optimized by the SA-RNN simulated annealing algorithm improves the prediction accuracy of the original recurrent neural network model.
The invention also provides a system for predicting the wear degree of the crankshaft bearing of the diesel engine, which comprises:
the acceleration vibration signal sample acquisition module is used for acquiring acceleration vibration signal samples of different wear degrees of the diesel engine crankshaft bearing during operation;
the characteristic sample set acquisition module is used for extracting the characteristics of the accelerated vibration signal sample corresponding to each wear degree by utilizing a method of combining complementary set empirical mode decomposition and singular value decomposition to obtain a characteristic sample set corresponding to each wear degree;
the training set and test set forming module is used for forming a training set and a test set by using the wear degrees as labels and the characteristic sample set as input quantities, wherein the characteristic sample set corresponds to all the wear degrees;
the cyclic neural network optimization module is used for optimizing the cyclic neural network by utilizing a simulated annealing algorithm based on the training set and the test set to obtain the optimized cyclic neural network;
the real-time acceleration vibration signal acquisition module is used for acquiring a real-time acceleration vibration signal of the diesel engine crankshaft bearing to be detected in the operation process;
the characteristic set obtaining module is used for carrying out complementary set empirical mode decomposition and singular value decomposition on the real-time acceleration vibration signal to obtain a characteristic set;
and the wear degree output module is used for inputting the feature set into the optimized recurrent neural network and outputting the wear degree of the crankshaft bearing of the diesel engine to be tested.
The system further comprises:
and the denoising module is used for performing wavelet threshold denoising on the acceleration vibration signal sample by adopting a soft threshold method with the threshold being an unbiased likelihood estimation threshold.
The feature sample set obtaining module specifically comprises:
the complementary set empirical mode decomposition submodule is used for carrying out complementary set empirical mode decomposition on the accelerated vibration signal sample corresponding to each wear degree to obtain a plurality of intrinsic mode functions, and the first 6 intrinsic mode functions form an initial vector matrix corresponding to each wear degree;
and the singular value decomposition submodule is used for carrying out singular value decomposition on the initial vector matrix corresponding to each wear degree to obtain a plurality of singular values, and meanwhile, the plurality of singular values form a characteristic sample set corresponding to each wear degree.
The recurrent neural network optimization module specifically comprises:
the parameter sub-initialization module is used for initializing the parameters of the simulated annealing algorithm; the simulated annealing algorithm parameters comprise an initial temperature, iteration times and a termination temperature;
the initial solution initialization submodule is used for initializing the initial solution of the recurrent neural network; the initial solution includes a sequence of length x1And the number of hidden layer nodes is x2；
A first classification accuracy obtaining submodule for using the training set to obtain the length x of the sequence1And the number of hidden layer nodes is x2Training the circulating neural network, and testing the trained circulating neural network by using a test set to obtain a first classification accuracy;
a new solution generation submodule for randomly generating a new solution of the recurrent neural network; the new solution includes a sequence length of x1' and number of hidden layer nodes is x2'；
A second classification accuracy obtaining submodule for using the training set to obtain the length x of the sequence1' and number of hidden layer nodes is x2The recurrent neural network of' is trained, and the trained recurrent neural network is tested by using a test set to obtain a second classification accuracy;
the first judgment result obtaining submodule is used for judging whether the increment of the second classification accuracy relative to the first classification accuracy is larger than or equal to zero or not and obtaining a first judgment result;
the assignment submodule is used for receiving a new solution as a new initial solution if the first judgment result shows that the first classification accuracy is correct, and assigning the second classification accuracy to the first classification accuracy;
the new solution receiving submodule is used for receiving a new solution according to the Metropolis criterion if the first judgment result shows that the first judgment result does not indicate that the first judgment result indicates that the first judgment result does not indicate that the first judgment result indicates that the first judgment result does not indicate that the first judgment result indicates that the first judgment module does not indicate that the first judgment module;
a second judgment result obtaining submodule for judging whether the current iteration times is greater than or equal to the iteration times to obtain a second judgment result;
a step returning submodule for returning to the step of randomly generating a new solution of the recurrent neural network if the second judgment result indicates no;
a third judgment result obtaining submodule, configured to, if the second judgment result indicates yes, judge whether the current temperature is less than the target temperature, and obtain a third judgment result;
a temperature reduction submodule, configured to reduce the current temperature if the third determination result indicates no, and return to the step "randomly generate a new solution of the recurrent neural network";
and the result output submodule is used for outputting the new current solution as the optimal solution and outputting the second classification accuracy as the optimal classification accuracy if the third judgment result shows that the current solution is the optimal solution.
The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. For the system disclosed by the embodiment, the description is relatively simple because the system corresponds to the method disclosed by the embodiment, and the relevant points can be referred to the method part for description.
The principles and embodiments of the present invention have been described herein using specific examples, which are provided only to help understand the method and the core concept of the present invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, the specific embodiments and the application range may be changed. In view of the above, the present disclosure should not be construed as limiting the invention.
Claims (8)
1. A method for predicting the wear degree of a crankshaft bearing of a diesel engine is characterized by comprising the following steps:
acquiring accelerated vibration signal samples of different wear degrees of a crankshaft bearing of the diesel engine during operation;
extracting the characteristics of the accelerated vibration signal sample corresponding to each wear degree by using a method of combining complementary set empirical mode decomposition and singular value decomposition to obtain a characteristic sample set corresponding to each wear degree;
taking the wear degrees as labels and the characteristic sample sets as input quantities, and forming a training set and a testing set by using all the wear degrees and the characteristic sample sets corresponding to all the wear degrees;
optimizing the recurrent neural network by using a simulated annealing algorithm based on the training set and the test set to obtain the optimized recurrent neural network;
acquiring a real-time acceleration vibration signal of a diesel engine crankshaft bearing to be detected in operation;
complementary set empirical mode decomposition and singular value decomposition are carried out on the real-time accelerated vibration signal to obtain a feature set;
and inputting the characteristic set into the optimized recurrent neural network, and outputting the wear degree of the crankshaft bearing of the diesel engine to be tested.
2. The method for predicting the wear degree of the crankshaft bearing of the diesel engine according to claim 1, wherein the step of obtaining samples of the accelerated vibration signals of the crankshaft bearing of the diesel engine with different wear degrees during operation further comprises the following steps:
and performing wavelet threshold denoising on the accelerated vibration signal sample by adopting a soft threshold method with a threshold as an unbiased likelihood estimation threshold.
3. The method for predicting the wear degree of the crankshaft bearing of the diesel engine according to claim 1, wherein the method for extracting the features of the acceleration vibration signal sample corresponding to each wear degree by combining the empirical mode decomposition and the singular value decomposition of the complementary set is used to obtain the feature sample set corresponding to each wear degree, and specifically includes:
performing complementary set empirical mode decomposition on the accelerated vibration signal sample corresponding to each wear degree to obtain a plurality of intrinsic mode functions, and forming the first 6 intrinsic mode functions into an initial vector matrix corresponding to each wear degree;
and performing singular value decomposition on the initial vector matrix corresponding to each wear degree to obtain a plurality of singular values, and forming a characteristic sample set corresponding to each wear degree by the plurality of singular values.
4. The method for predicting the wear degree of the crankshaft bearing of the diesel engine according to claim 1, wherein the optimizing the recurrent neural network by using a simulated annealing algorithm based on the training set and the test set to obtain the optimized recurrent neural network specifically comprises:
initializing parameters of a simulated annealing algorithm; the simulated annealing algorithm parameters comprise an initial temperature, iteration times and a termination temperature;
initializing an initial solution of a recurrent neural network; the initial solution comprises a sequence of length x1And the number of hidden layer nodes is x2；
Using training set to sequence length x1And the number of hidden layer nodes is x2Training the circulating neural network, and testing the trained circulating neural network by using a test set to obtain a first classification accuracy;
randomly generating a new solution of the recurrent neural network; the new solution comprises a sequence length of x1' and number of hidden layer nodes is x2'；
Using training set to sequence length x1' and number of hidden layer nodes is x2The recurrent neural network of' is trained, and the trained recurrent neural network is tested by using a test set to obtain a second classification accuracy;
judging whether the increment of the second classification accuracy rate relative to the first classification accuracy rate is larger than or equal to zero or not, and obtaining a first judgment result;
if the first judgment result shows that the first classification accuracy rate is the first classification accuracy rate, receiving a new solution as a new initial solution, and assigning the second classification accuracy rate to the first classification accuracy rate;
if the first judgment result shows no, receiving a new solution according to the Metropolis criterion;
judging whether the current iteration times are greater than or equal to the iteration times to obtain a second judgment result;
if the second judgment result shows no, returning to the step of randomly generating a new solution of the recurrent neural network;
if the second judgment result shows yes, judging whether the current temperature is lower than the target temperature or not, and obtaining a third judgment result;
if the third judgment result shows no, reducing the current temperature, and returning to the step of randomly generating a new solution of the recurrent neural network;
and if the third judgment result shows that the current solution is the optimal solution, outputting the new current solution as the optimal solution, and outputting the second classification accuracy as the optimal classification accuracy.
5. A system for predicting wear of a crankshaft bearing of a diesel engine, the system comprising:
the acceleration vibration signal sample acquisition module is used for acquiring acceleration vibration signal samples of different wear degrees of the diesel engine crankshaft bearing during operation;
the characteristic sample set acquisition module is used for extracting the characteristics of the accelerated vibration signal sample corresponding to each wear degree by utilizing a method of combining complementary set empirical mode decomposition and singular value decomposition to obtain a characteristic sample set corresponding to each wear degree;
the training set and test set forming module is used for forming a training set and a test set by using the wear degrees as labels and the characteristic sample set as input quantities, wherein the characteristic sample set corresponds to all the wear degrees;
the cyclic neural network optimization module is used for optimizing the cyclic neural network by utilizing a simulated annealing algorithm based on the training set and the test set to obtain the optimized cyclic neural network;
the real-time acceleration vibration signal acquisition module is used for acquiring a real-time acceleration vibration signal of the diesel engine crankshaft bearing to be detected in the operation process;
the characteristic set obtaining module is used for carrying out complementary set empirical mode decomposition and singular value decomposition on the real-time acceleration vibration signal to obtain a characteristic set;
and the wear degree output module is used for inputting the feature set into the optimized recurrent neural network and outputting the wear degree of the crankshaft bearing of the diesel engine to be tested.
6. The method of predicting the degree of wear of a crankshaft bearing of a diesel engine as set forth in claim 5, wherein the system further comprises:
and the denoising module is used for performing wavelet threshold denoising on the accelerated vibration signal sample by adopting a soft threshold method with the threshold being an unbiased likelihood estimation threshold.
7. The system for predicting the wear degree of the crankshaft bearing of the diesel engine according to claim 5, wherein the feature sample set obtaining module specifically comprises:
the complementary set empirical mode decomposition submodule is used for carrying out complementary set empirical mode decomposition on the accelerated vibration signal sample corresponding to each wear degree to obtain a plurality of intrinsic mode functions, and the first 6 intrinsic mode functions form an initial vector matrix corresponding to each wear degree;
and the singular value decomposition submodule is used for carrying out singular value decomposition on the initial vector matrix corresponding to each wear degree to obtain a plurality of singular values, and meanwhile, the plurality of singular values form a characteristic sample set corresponding to each wear degree.
8. The system for predicting the wear degree of the crankshaft bearing of the diesel engine according to claim 5, wherein the recurrent neural network optimization module specifically comprises:
the parameter sub-initialization module is used for initializing the parameters of the simulated annealing algorithm; the simulated annealing algorithm parameters comprise an initial temperature, iteration times and a termination temperature;
the initial solution initialization submodule is used for initializing the initial solution of the recurrent neural network; the initial solution comprises a sequence of length x1And the number of hidden layer nodes is x2；
A first classification accuracy obtaining submodule for using the training set to obtain the length x of the sequence1And the number of hidden layer nodes is x2Training the circulating neural network, and testing the trained circulating neural network by using a test set to obtain a first classification accuracy;
a new solution generation submodule for randomly generating a new solution of the recurrent neural network; the new solution comprises a sequence length of x1' and number of hidden layer nodes is x2'；
A second classification accuracy obtaining submodule for using the training set to obtain the length x of the sequence1' and number of hidden layer nodes is x2' the recurrent neural network is trained, and the training is paired with the test setTesting the trained recurrent neural network to obtain a second classification accuracy;
the first judgment result obtaining submodule is used for judging whether the increment of the second classification accuracy relative to the first classification accuracy is larger than or equal to zero or not and obtaining a first judgment result;
the assignment submodule is used for receiving a new solution as a new initial solution if the first judgment result shows that the first classification accuracy is correct, and assigning the second classification accuracy to the first classification accuracy;
a new solution receiving submodule, configured to receive a new solution according to the Metropolis criterion if the first determination result indicates no;
a second judgment result obtaining submodule for judging whether the current iteration times is greater than or equal to the iteration times to obtain a second judgment result;
a step returning submodule, configured to return to the step "randomly generate a new solution of the recurrent neural network" if the second determination result indicates no;
a third judgment result obtaining submodule, configured to, if the second judgment result indicates yes, judge whether the current temperature is lower than the target temperature, and obtain a third judgment result;
a temperature reduction submodule, configured to reduce the current temperature if the third determination result indicates no, and return to the step "randomly generate a new solution of the recurrent neural network";
and the result output submodule is used for outputting the new current solution as the optimal solution and outputting the second classification accuracy as the optimal classification accuracy if the third judgment result shows that the current solution is the optimal solution.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111573265.7A CN114264478A (en) | 2021-12-21 | 2021-12-21 | Diesel engine crankshaft bearing wear degree prediction method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111573265.7A CN114264478A (en) | 2021-12-21 | 2021-12-21 | Diesel engine crankshaft bearing wear degree prediction method and system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114264478A true CN114264478A (en) | 2022-04-01 |
Family
ID=80828915
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111573265.7A Pending CN114264478A (en) | 2021-12-21 | 2021-12-21 | Diesel engine crankshaft bearing wear degree prediction method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114264478A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115326397A (en) * | 2022-07-28 | 2022-11-11 | 沈阳顺义科技有限公司 | Method for establishing crankshaft bearing wear degree prediction model and prediction method and related device |
| CN117250004A (en) * | 2023-11-13 | 2023-12-19 | 山东省宇捷轴承制造有限公司 | Bearing fault diagnosis method and system based on digital signal processing |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109975656A (en) * | 2019-03-06 | 2019-07-05 | 河南理工大学 | Based on flexible direct current power distribution network high resistance earthing fault detection method |
| CN111079710A (en) * | 2019-12-31 | 2020-04-28 | 江苏理工学院 | Multilayer noise reduction method based on improved CEEMD rolling bearing signal |
| CN111340221A (en) * | 2020-02-25 | 2020-06-26 | 北京百度网讯科技有限公司 | Method and device for sampling neural network structure |
| CN111368379A (en) * | 2020-04-30 | 2020-07-03 | 上海工程技术大学 | Rolling bearing reliability evaluation method and system for improving HWPSO-WPHM model |
-
2021
- 2021-12-21 CN CN202111573265.7A patent/CN114264478A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109975656A (en) * | 2019-03-06 | 2019-07-05 | 河南理工大学 | Based on flexible direct current power distribution network high resistance earthing fault detection method |
| CN111079710A (en) * | 2019-12-31 | 2020-04-28 | 江苏理工学院 | Multilayer noise reduction method based on improved CEEMD rolling bearing signal |
| CN111340221A (en) * | 2020-02-25 | 2020-06-26 | 北京百度网讯科技有限公司 | Method and device for sampling neural network structure |
| CN111368379A (en) * | 2020-04-30 | 2020-07-03 | 上海工程技术大学 | Rolling bearing reliability evaluation method and system for improving HWPSO-WPHM model |
Non-Patent Citations (3)
| Title |
|---|
| 吴漫: ""基于CEEMD-SVD和ELM的滚动轴承故障诊断"", 《化工自动化及仪表》 * |
| 周奇才: ""基于LSTM循环神经网络的盾构机故障预测***设计"", 《现代机械》 * |
| 张明: ""贝叶斯分类器改进的SA-RNN网络及CFD在MBR中的应用研究"", 《中国优秀硕士学位论文全文库 工程科技I辑》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115326397A (en) * | 2022-07-28 | 2022-11-11 | 沈阳顺义科技有限公司 | Method for establishing crankshaft bearing wear degree prediction model and prediction method and related device |
| CN115326397B (en) * | 2022-07-28 | 2023-10-27 | 沈阳顺义科技有限公司 | Method and related device for establishing crankshaft bearing wear degree prediction model and prediction method |
| CN117250004A (en) * | 2023-11-13 | 2023-12-19 | 山东省宇捷轴承制造有限公司 | Bearing fault diagnosis method and system based on digital signal processing |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110428004B (en) | Mechanical part fault diagnosis method based on deep learning under data imbalance | |
| CN110657985B (en) | Gearbox fault diagnosis method and system based on singular value spectrum manifold analysis | |
| CN109711609B (en) | Photovoltaic power station output power prediction method based on wavelet transformation and extreme learning machine | |
| CN110261109B (en) | Rolling bearing fault diagnosis method based on bidirectional memory cyclic neural network | |
| CN106248368B (en) | Combustion engine turbine blade fault detection method based on deep learning | |
| CN114264478A (en) | Diesel engine crankshaft bearing wear degree prediction method and system | |
| CN113281048B (en) | Rolling bearing fault diagnosis method and system based on relational knowledge distillation | |
| CN111412977A (en) | Preprocessing method for vibration sensing data of mechanical equipment | |
| CN110503075B (en) | Planetary gearbox fault diagnosis method and system based on PSO-EEMD method | |
| CN111368892A (en) | Generalized S transformation and SVM electric energy quality disturbance efficient identification method | |
| CN111582551A (en) | Method and system for predicting short-term wind speed of wind power plant and electronic equipment | |
| CN111914705A (en) | Signal generation method and device for improving health state evaluation accuracy of reactor | |
| CN111060316A (en) | Rolling bearing state monitoring method and system based on convolutional neural network model | |
| CN117473411A (en) | Bearing life prediction method based on improved transducer model | |
| CN114266416A (en) | Photovoltaic power generation power short-term prediction method and device based on similar days and storage medium | |
| Yang et al. | A novel multiple feature-based engine knock detection system using sparse Bayesian extreme learning machine | |
| CN116403032A (en) | Breaker fault evaluation method based on multi-domain information fusion and deep learning | |
| CN115409052A (en) | Fault diagnosis method and system for wind generating set bearing under data imbalance | |
| CN113657244B (en) | Fan gearbox fault diagnosis method and system based on improved EEMD and speech spectrum analysis | |
| Hussain et al. | Gearbox fault detection using real coded genetic algorithm and novel shock response spectrum features extraction | |
| CN117473414A (en) | Bearing fault position identification method based on low-noise time-frequency image | |
| CN112729825A (en) | Method for constructing bearing fault diagnosis model based on convolution cyclic neural network | |
| CN110044619B (en) | Multi-fault feature identification method based on sparse multi-cycle group lasso | |
| CN116610990A (en) | Method and device for identifying hidden danger of breaker based on characteristic space differentiation | |
| CN115326397A (en) | Method for establishing crankshaft bearing wear degree prediction model and prediction method and related 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 |