CN115114955B - Equipment fault detection method based on sound and vibration signals - Google Patents
Equipment fault detection method based on sound and vibration signals Download PDFInfo
- Publication number
- CN115114955B CN115114955B CN202210637988.7A CN202210637988A CN115114955B CN 115114955 B CN115114955 B CN 115114955B CN 202210637988 A CN202210637988 A CN 202210637988A CN 115114955 B CN115114955 B CN 115114955B
- Authority
- CN
- China
- Prior art keywords
- loss
- data
- normal
- signal
- test
- 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
- 238000001514 detection method Methods 0.000 title claims abstract description 34
- 238000012549 training Methods 0.000 claims abstract description 42
- 238000013528 artificial neural network Methods 0.000 claims abstract description 33
- 238000012360 testing method Methods 0.000 claims abstract description 33
- 238000000034 method Methods 0.000 claims abstract description 21
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 17
- 238000005516 engineering process Methods 0.000 claims abstract description 17
- 230000008569 process Effects 0.000 claims abstract description 10
- 238000012545 processing Methods 0.000 claims description 10
- 230000006870 function Effects 0.000 claims description 9
- 238000003745 diagnosis Methods 0.000 claims description 8
- 238000009432 framing Methods 0.000 claims description 8
- 238000004364 calculation method Methods 0.000 claims description 6
- 230000005236 sound signal Effects 0.000 claims description 5
- 238000004458 analytical method Methods 0.000 claims description 3
- 239000013598 vector Substances 0.000 claims description 3
- 238000013473 artificial intelligence Methods 0.000 abstract description 6
- 238000004519 manufacturing process Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 108010076504 Protein Sorting Signals Proteins 0.000 description 1
- 230000006978 adaptation Effects 0.000 description 1
- 238000007792 addition Methods 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000009776 industrial production Methods 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
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
- G01M13/00—Testing of machine parts
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M99/00—Subject matter not provided for in other groups of this subclass
- G01M99/005—Testing of complete machines, e.g. washing-machines or mobile phones
-
- 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
- Y02T90/00—Enabling technologies or technologies with a potential or indirect contribution to GHG emissions mitigation
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)
Abstract
The invention provides a device fault detection method based on sound and vibration signals, and belongs to the technical field of device fault detection technology and artificial intelligence. Firstly, collecting sound and vibration signals in the running process of equipment to be tested through a sensor, and then analyzing and denoising the collected signals through technologies such as EMD (empirical mode decomposition) and wavelet transform to obtain denoised signals; randomly extracting data from the data in the normal state to form a normal data set, and randomly extracting data from the data in the fault state to form a fault data set; and establishing an equipment fault detection model based on the physical information neural network, training by utilizing partial data in the normal data set, fitting a wave equation of the normal data signal, testing by utilizing the residual normal data and the fault data set, and diagnosing the data signal by calculating error loss. The technology provided by the invention can accurately and efficiently detect the equipment state, and solves the problem of low equipment fault detection accuracy in the prior art.
Description
Technical Field
The invention relates to the technical field of equipment fault detection and artificial intelligence, in particular to an equipment fault detection method based on sound and vibration signals.
Background
In recent years, as the operation environment of mechanical equipment is gradually complicated in industrial production, the probability of mechanical equipment failure is significantly increased, and the health state of the equipment is closely related to production safety and production quality, especially for some key equipment, the equipment cannot be shut down for maintenance at any time in one production cycle, and if the equipment is shut down accidentally, serious production accidents are likely to be caused. The equipment state can be effectively detected through the equipment fault detection technology, and accidents are prevented, so that the equipment fault detection technology has a wide application prospect.
In the conventional equipment fault detection technology, the operation state of equipment is generally judged according to experience or related indexes by manually observing operation parameters of the equipment. The effect of such manual diagnosis is affected by the experience of the expert, and has the disadvantages of low accuracy and limited predictive power.
With the development of the artificial intelligence technology field, the intelligent fault detection of the equipment on line by installing a sensor on a machine and utilizing the artificial intelligence technology becomes a new approach for solving the problems, and a great deal of manpower can be saved in the mode, and the on-line detection can be continuously carried out. However, since artificial intelligence models need to be trained for different devices, and the sample tags trained are completely manually labeled, the detection capability for some human-made imperceptible potential faults is limited. Besides, the method has a large improvement space in the aspects of accuracy, robustness and the like of equipment fault detection technology under the influence of noise. Therefore, research on the application of the artificial intelligence technology in equipment fault detection has very important theoretical and application values.
Disclosure of Invention
The invention aims to provide a device fault detection method based on sound and vibration signals, so as to improve the accuracy and the robustness of a device fault detection technology and the detection capability of a potential fault.
The invention adopts the following technical scheme:
a device fault detection method based on sound and vibration signals comprises the following steps:
step 1: collecting a sound signal or a vibration signal S, wherein S is time series data;
Step 2: processing the signal S acquired in the step 1 by using a windowing and framing technology, wherein the duration of a signal frame is set to be w, the starting time of an ith signal frame is t i, and S in a interception window [ t i,ti +w ] is recorded as follows: s_t i, i=1, 2,3, &..n;
Wherein N is the total number of signal frames, i is the signal frame number, and t i is the starting time of the ith signal frame; after framing, S is divided into N signal frames;
Step 3: performing EMD decomposition on the signal frame s_t i obtained in the step 2 to obtain a decomposed IMF component s_t i_IMFj, j=1, 2, 3. Analyzing the S_t i_IMFj to determine a noise-containing IMF component S_t i_IMFn needing wavelet threshold denoising in the high-frequency component and a noise-free low-frequency IMF component S_t i_IMFm, wherein n is E j, m is E j, and m is not equal to n; performing wavelet threshold denoising on the S_t i_IMFn to obtain a wavelet threshold denoised signal Will beCarrying out signal reconstruction with the S_t i_IMFm to obtain a denoised signal S_process_t i;
Step 4: dividing the denoised signal S_process_t i obtained in the step3 according to states to form a normal data set Nomal _set and a Fault data set fault_set; extracting partial data from Nomal _set to form a training data set train_set, wherein the data set of the rest part is denoted as Nomal _reserve_set; combining the fault_set with Nomal _result_set to obtain a Test data set test_set;
step 5: establishing an equipment fault detection model based on a physical information neural network, and determining the structure, the layer number, the node number and a vibration equation to be fitted of the equipment fault diagnosis model of the physical information neural network; performing unsupervised training on the established equipment fault detection model based on the physical information neural network by using the training data set obtained in the step 4 to obtain a training fitting error Loss train_loss, a fitted vibration equation and initial model parameters; calculating the maximum fitting error Loss Max_normal_Loss of the Normal state data; calculating a Normal error Loss range normal_range by utilizing the train_loss and the Max_normal_loss;
Step 6: diagnosing faults of equipment to be tested, inputting a Test set of test_set into the model obtained by training in the step 5, and calculating Test fitting error Loss test_loss j of each sample data; calculating the train_loss and the test_loss j obtained in the step 5 to obtain a test_range j of the Test error Loss; if test_range j exceeds the Normal error loss range normal_range, the vibration signal is considered as a fault condition.
Preferably, the sound signal or vibration signal S in step 1 is derived from the sound or vibration signal collected by the sensor during the operation of the device under test.
Preferably, the step 3 specifically includes the following sub-steps:
Step 3.1: performing EMD decomposition on the s_t i to obtain M decomposed IMF components s_t i_IMFj, j=1, 2, 3.
S_ti_IMFj=emd(S_ti); (1)
Step 3.2: analyzing the IMF component S_t i_IMFj to determine a noise-containing IMF component S_t i_IMFn which needs wavelet threshold denoising in the high-frequency component, wherein n is E j; the noise-free low frequency IMF component is denoted s_t i_IMFm, where m e j, m+.n;
Step 3.3: performing wavelet threshold denoising on the S_t i_IMFn, analyzing and removing unsteady signals to obtain signals subjected to wavelet threshold denoising N signal frames are processed to obtain N1-dimensional vectors, as shown in a formula (2):
Step 3.4: obtaining noise-reduced data The signal reconstruction is performed together with the noise-free low-frequency IMF component s_t i_IMFm to obtain a denoised signal s_process_t i, as shown in formula (3):
Preferably, the step 3.2 specifically comprises the following sub-steps:
Step 3.2.1: performing m-layer wavelet decomposition on the signal frame S_t i_IMFn, and finally outputting m+1 wavelet coefficient sets As shown in formulas (4) and (5):
Wherein: for decomposed low frequency information, approximation,/> For the decomposed high-frequency information, the detail part, func is the selected wavelet basis function, and m is the decomposition layer number;
Step 3.2.2: after the wavelet decomposition in step 3.2.1, the wavelet coefficient is obtained by selecting proper threshold value and threshold value function And (5) processing. Wavelet coefficients above the threshold are considered to be generated by the signal and should be preserved, while wavelet coefficients below the threshold are considered to be generated by noise and are set to 0 for denoising purposes. Finally, the wavelet coefficient after threshold processing is obtainedAs shown in formula (6):
Wherein: value is a threshold;
step 3.2.3: thresholding the wavelet coefficient subjected to the step 3.2.2 Performing wavelet reconstruction to obtain signal data/>, after wavelet denoisingAs shown in formula (7):
wherein: func is the same wavelet basis function as in step 3.2.1.
Preferably, the step 4 specifically includes the following sub-steps:
step 4.1: dividing the denoised signal S_process_t i obtained in the step 3 according to a normal state and a Fault state, randomly extracting data from the data in the normal state to form a normal data set Nomal _set, and randomly extracting data from the data in the Fault state to form a Fault data set fault_set;
Step 4.2: randomly extracting part of normal data from the normal data set Nomal _set obtained in the step 4.1 to form a training data set train_set, and marking the rest part of normal data set as Nomal _reserve_set;
Step 4.3: after combining the Fault data set fault_set from step 4.1 with Nomal _result_set in step 4.2, a Test data set test_set is obtained.
Preferably, the specific implementation of the step 5 includes the following sub-steps:
Step 5.1: and determining the structure, the layer number, the node number and the vibration equation to be fitted of the fault diagnosis model of the physical information neural network equipment. The neural network is structurally divided into an input layer of a first layer, a plurality of hidden layers in the middle and an output layer of a last layer, wherein the input of the input layer is data in a training data set train_set, the hidden layers are used for extracting different layer characteristics, the number of layers and the number of nodes of the hidden layers are determined according to training and testing analysis, and the output of the output layer is a signal time sequence predicted according to a fitted vibration equation. The network structure adopts a structural form that no connection exists in the layers and all adjacent layers are connected;
Step 5.2: and (3) obtaining a data set train_set from the step (4), selecting one sample data from the data set train_data=train_set i, performing unsupervised training on the model established in the step (5.1), obtaining a training fitting error Loss train_loss, and obtaining a fitted vibration equation and initial model parameters.
Step 5.3: according to step 5.2, data is obtained from a data set train_set j (j not equal to i), the data is input into the model trained in step 5.2, the model fitting error Loss of each Normal state data sample is calculated, the step is repeated, a plurality of groups of Normal state data fitting error Loss normal_loss j is obtained, and the maximum value is Max_normal_loss;
Step 5.4: calculating the train_loss obtained in the step 5.2 and the Max_normal_loss obtained in the step 5.3 to obtain a Normal error Loss range, as shown in a formula (8):
Normal_range=abs(Max_Normal_Loss-Train_Loss); (8)
preferably, the step 5.2 specifically comprises the following sub-steps:
Step 5.2.1: setting an initial position T 0 as a starting point of a fitting period, taking an integer multiple of a signal repetition period as a relative period T, and taking the relative period T as a unit to extract sample points as input for subsequent test working data;
step 5.2.2: this step is a data driven discovery process, defining f (t) as shown in equation (9):
f(t):=ut+Ν[u;λ]; (9)
Approximating u t by a deep neural network, where u t is a derivative operator N [ u; lambda forms a physical information neural network f (t) together, the parameter lambda in the differential operator is changed into the parameter f (t), and the corresponding lambda parameter in the normal state can be obtained through training, so that the vibration equation f (t) of the required fitting is obtained;
Step 5.2.3: calculating a mean square error Loss train_loss in the training process, and learning the neural network by minimizing the mean square error Loss, wherein the calculation process is shown in formulas (10) - (12):
Train_Loss=Train_Lossu+Train_Lossf; (10)
Wherein train_loss u is the mean square error Loss of the neural network approaching u t, train_loss f is the mean square error Loss of the sample points in formula (9), N is the number of the sampled points, u (t i) is the predicted value of u t, u i is the true value, and f (t i) is the error Loss of formula (9).
Preferably, the specific implementation of the step 6 includes the following sub-steps:
Step 6.1: inputting each signal sample data in the test_set into the model trained in the step 5, and obtaining a Test fitting error Loss test_loss j of each sample data through calculation;
Step 6.2: calculating the train_loss obtained in step 5.2 and the test_loss j obtained in step 6.1 to obtain a test_range j of the Test error Loss of each sample data, as shown in formula (13):
Test_rangej=abs(Test_Lossj-Train_Loss); (13)
Step 6.3: taking the normal_range obtained in the step 5.4 as a reference error range, if the test_range j exceeds the reference range normal_range, indicating that the vibration signal is not within the Normal range, the vibration signal is regarded as a fault state.
The invention has the beneficial effects that:
(1) The method provided by the invention improves the accuracy and the robustness of the equipment fault detection technology and the detection capability of the equipment fault detection technology on the potential faults.
(2) The invention obtains good noise reduction effect by utilizing the noise reduction method combining the fast Fourier transform and the wavelet transform.
(3) The physical information neural network equipment fault detection model established by the invention improves the detection accuracy, and can automatically fit a vibration equation of a signal according to the denoised signal sequence and the extracted signal characteristics. The model is not only suitable for analyzing sound and vibration signals, but also suitable for analyzing wave equation with higher dimension, and has wide applicability.
Drawings
FIG. 1 is a flow chart of a method of modeling the present invention;
Fig. 2 is a schematic diagram of a physical information neural network structure in the present invention.
Detailed Description
The following describes the embodiments of the present invention further with reference to the accompanying drawings:
referring to fig. 1 and 2, a method for detecting equipment failure based on sound and vibration signals includes the following steps:
step 1: collecting a sound signal or a vibration signal S, wherein S is time series data;
Step 2: processing the signal S acquired in the step 1 by using a windowing and framing technology, wherein the duration of a signal frame is set to be w, the starting time of an ith signal frame is t i, and S in a interception window [ t i,ti +w ] is recorded as follows: s_t i, i=1, 2,3, &..n;
Wherein N is the total number of signal frames, i is the signal frame number, and t i is the starting time of the ith signal frame; after framing, S is divided into N signal frames;
Step 3: performing EMD decomposition on the signal frame s_t i obtained in the step 2 to obtain a decomposed IMF component s_t i_IMFj, j=1, 2, 3. Analyzing the S_t i_IMFj to determine a noise-containing IMF component S_t i_IMFn needing wavelet threshold denoising in the high-frequency component and a noise-free low-frequency IMF component S_t i_IMFm, wherein n is E j, m is E j, and m is not equal to n; performing wavelet threshold denoising on the S_t i_IMFn to obtain a wavelet threshold denoised signal Will beCarrying out signal reconstruction with the S_t i_IMFm to obtain a denoised signal S_process_t i;
The specific implementation comprises the following substeps:
Step 3.1: performing EMD decomposition on the s_t i to obtain M decomposed IMF components s_t i_IMFj, j=1, 2, 3.
S_ti_IMFj=emd(S_ti); (1)
Step 3.2: analyzing the IMF component S_t i_IMFj to determine a noise-containing IMF component S_t i_IMFn which needs wavelet threshold denoising in the high-frequency component, wherein n is E j; the noise-free low frequency IMF component is denoted s_t i_IMFm, where m e j, m+.n;
The specific implementation comprises the following substeps:
Step 3.2.1: performing m-layer wavelet decomposition on the signal frame S_t i_IMFn, and finally outputting m+1 wavelet coefficient sets As shown in formulas (2) and (3):
Wherein: for decomposed low frequency information, approximation,/> For the decomposed high-frequency information, the detail part, func is the selected wavelet basis function, and m is the decomposition layer number;
Step 3.2.2: after the wavelet decomposition in step 3.2.1, the wavelet coefficient is obtained by selecting proper threshold value and threshold value function And (5) processing. Wavelet coefficients above the threshold are considered to be generated by the signal and should be preserved, while wavelet coefficients below the threshold are considered to be generated by noise and are set to 0 for denoising purposes. Finally, the wavelet coefficient after threshold processing is obtainedAs shown in formula (4):
Wherein: value is a threshold;
step 3.2.3: thresholding the wavelet coefficient subjected to the step 3.2.2 Performing wavelet reconstruction to obtain signal data/>, after wavelet denoisingAs shown in formula (5):
wherein: func is the same wavelet basis function as in step 3.2.1.
Step 3.3: performing wavelet threshold denoising on the S_t i_IMFn, analyzing and removing unsteady signals to obtain signals subjected to wavelet threshold denoisingN signal frames are processed to obtain N1-dimensional vectors, as shown in a formula (6):
Step 3.4: obtaining noise-reduced data The signal reconstruction is performed together with the noise-free low-frequency IMF component s_t i_IMFm to obtain a denoised signal s_process_t i, as shown in formula (7):
Step 4: dividing the denoised signal S_process_t i obtained in the step3 according to states to form a normal data set Nomal _set and a Fault data set fault_set; extracting partial data from Nomal _set to form a training data set train_set, wherein the data set of the rest part is denoted as Nomal _reserve_set; combining the fault_set with Nomal _result_set to obtain a Test data set test_set;
The specific implementation comprises the following substeps:
step 4.1: dividing the denoised signal S_process_t i obtained in the step 3 according to a normal state and a Fault state, randomly extracting data from the data in the normal state to form a normal data set Nomal _set, and randomly extracting data from the data in the Fault state to form a Fault data set fault_set;
Step 4.2: randomly extracting part of normal data from the normal data set Nomal _set obtained in the step 4.1 to form a training data set train_set, and marking the rest part of normal data set as Nomal _reserve_set;
Step 4.3: after combining the Fault data set fault_set from step 4.1 with Nomal _result_set in step 4.2, a Test data set test_set is obtained.
Step 5: establishing an equipment fault detection model based on a physical information neural network, and determining the structure, the layer number, the node number and a vibration equation to be fitted of the equipment fault diagnosis model of the physical information neural network; performing unsupervised training on the established equipment fault detection model based on the physical information neural network by using the training data set obtained in the step 4 to obtain a training fitting error Loss train_loss, a fitted vibration equation and initial model parameters; calculating the maximum fitting error Loss Max_normal_Loss of the Normal state data; calculating a Normal error Loss range normal_range by utilizing the train_loss and the Max_normal_loss;
The specific implementation comprises the following substeps:
Step 5.1: and determining the structure, the layer number, the node number and the vibration equation to be fitted of the fault diagnosis model of the physical information neural network equipment. The neural network is structurally divided into an input layer of a first layer, a plurality of hidden layers in the middle and an output layer of a last layer, wherein the input of the input layer is data in a training data set train_set, the hidden layers are used for extracting different layer characteristics, the number of layers and the number of nodes of the hidden layers are determined according to training and testing analysis, and the output of the output layer is a signal time sequence predicted according to a fitted vibration equation. The network structure adopts a structural form that no connection exists in the layers and all adjacent layers are connected;
Step 5.2: and (3) obtaining a data set train_set from the step (4), selecting one sample data from the data set train_data=train_set i, performing unsupervised training on the model established in the step (5.1), obtaining a training fitting error Loss train_loss, and obtaining a fitted vibration equation and initial model parameters.
The specific implementation comprises the following substeps:
Step 5.2.1: setting an initial position T 0 as a starting point of a fitting period, taking an integer multiple of a signal repetition period as a relative period T, and taking the relative period T as a unit to extract sample points as input for subsequent test working data;
step 5.2.2: this step is a data driven discovery process, defining f (t) as shown in equation (9):
f(t):=ut+Ν[u;λ]; (9)
Approximating u t by a deep neural network, where u t is a derivative operator N [ u; lambda forms a physical information neural network f (t) together, the parameter lambda in the differential operator is changed into the parameter f (t), and the corresponding lambda parameter in the normal state can be obtained through training, so that the vibration equation f (t) of the required fitting is obtained;
step 5.2.3: calculating a mean square error Loss train_loss in the training process, and learning the neural network by minimizing the mean square error Loss, wherein the calculation process is shown in formulas (9) - (11):
Train_Loss=Train_Lossu+Train_Lossf; (9)
Wherein train_loss u is the mean square error Loss of the neural network approaching u t, train_loss f is the mean square error Loss of the sample points in formula (8), N is the number of the sampled points, u (t i) is the predicted value of u t, u i is the true value, and f (t i) is the error Loss of formula (8).
Step 5.3: according to step 5.2, data is obtained from a data set train_set j (j not equal to i), the data is input into the model trained in step 5.2, the model fitting error Loss of each Normal state data sample is calculated, the step is repeated, a plurality of groups of Normal state data fitting error Loss normal_loss j is obtained, and the maximum value is Max_normal_loss;
Step 5.4: calculating the train_loss obtained in the step 5.2 and the max_normal_loss obtained in the step 5.3 to obtain a Normal error Loss range, as shown in a formula (12):
Normal_range=abs(Max_Normal_Loss-Train_Loss); (12)
Step 6: diagnosing faults of equipment to be tested, inputting a Test set of test_set into the model obtained by training in the step 5, and calculating Test fitting error Loss test_loss j of each sample data; calculating the train_loss and the test_loss j obtained in the step 5 to obtain a test_range j of the Test error Loss; if test_range j exceeds the Normal error loss range normal_range, the vibration signal is considered as a fault condition.
The specific implementation comprises the following substeps:
Step 6.1: inputting each signal sample data in the test_set into the model trained in the step 5, and obtaining a Test fitting error Loss test_loss j of each sample data through calculation;
Step 6.2: calculating the train_loss obtained in step 5.2 and the test_loss j obtained in step 6.1 to obtain a test_range j of the Test error Loss of each sample data, as shown in formula (13):
Test_rangej=abs(Test_Lossj-Train_Loss); (13)
Step 6.3: taking the normal_range obtained in the step 5.4 as a reference error range, if the test_range j exceeds the reference range normal_range, indicating that the vibration signal is not within the Normal range, the vibration signal is regarded as a fault state.
Example 1
The following are specific examples of the application of the present invention:
Step 1: the method is validated using a vibration signal S in a kesi Chu Da (CWRU) bearing dataset, where S is time series data;
Step 2: processing the signal S acquired in the step 1 by using a windowing and framing technology, wherein the duration of a signal frame is set to be w, the starting time of an ith signal frame is t i, and S in a interception window [ t i,ti +w ] is recorded as follows: s_t i, i=1, 2,3, &..n;
Wherein N is the total number of signal frames, i is the signal frame number, and t i is the starting time of the ith signal frame; after framing, S is divided into N signal frames;
Step 3: performing EMD decomposition on the signal frame s_t i obtained in the step 2 to obtain a decomposed IMF component s_t i_IMFj, j=1, 2, 3. Analyzing the S_t i_IMFj to determine a noise-containing IMF component S_t i_IMFn needing wavelet threshold denoising in the high-frequency component and a noise-free low-frequency IMF component S_t i_IMFm, wherein n is E j, m is E j, and m is not equal to n; performing wavelet threshold denoising on the S_t i_IMFn to obtain a wavelet threshold denoised signal Will beCarrying out signal reconstruction with the S_t i_IMFm to obtain a denoised signal S_process_t i;
Step 4: dividing the denoised signal S_process_t i obtained in the step3 according to states to form a normal data set Nomal _set and a Fault data set fault_set; extracting partial data from Nomal _set to form a training data set train_set, wherein the data set of the rest part is denoted as Nomal _reserve_set; combining the fault_set with Nomal _result_set to obtain a Test data set test_set;
step 5: establishing an equipment fault detection model based on a physical information neural network, and determining the structure, the layer number, the node number and a vibration equation to be fitted of the equipment fault diagnosis model of the physical information neural network; performing unsupervised training on the established equipment fault detection model based on the physical information neural network by using the training data set obtained in the step 4 to obtain a training fitting error Loss train_loss, a fitted vibration equation and initial model parameters; calculating the maximum fitting error Loss Max_normal_Loss of the Normal state data; calculating a Normal error Loss range normal_range by utilizing the train_loss and the Max_normal_loss;
A differential operator n [ u ]; lambda is shown in formula (14):
Ν[u;λ]=λ1u′+λ2u″; (14)
wherein lambda 1=e-5.73,λ2=e-11.52;
model parameters are shown in table 1:
TABLE 1 model parameter settings
Step 6: diagnosing faults of equipment to be tested, inputting a Test set of test_set into the model obtained by training in the step 5, and calculating Test fitting error Loss test_loss j of each sample data; calculating the train_loss and the test_loss j obtained in the step 5 to obtain a test_range j of the Test error Loss; if test_range j exceeds the Normal error loss range normal_range, the vibration signal is considered as a fault condition.
It should be understood that the above description is not intended to limit the invention to the particular embodiments disclosed, but to limit the invention to the particular embodiments disclosed, and that the invention is not limited to the particular embodiments disclosed, but is intended to cover modifications, adaptations, additions and alternatives falling within the spirit and scope of the invention.
Claims (6)
1. The equipment fault detection method based on the sound and vibration signals is characterized by comprising the following steps of:
step 1: collecting a sound signal or a vibration signal S, wherein S is time series data;
Step 2: processing the signal S acquired in the step 1 by using a windowing and framing technology, wherein the duration of a signal frame is set to be w, the starting time of an ith signal frame is t i, and S in a interception window [ t i,ti +w ] is recorded as follows: s_t i, i=1, 2,3, &..n;
Wherein N is the total number of signal frames, i is the signal frame number, and t i is the starting time of the ith signal frame; after framing, S is divided into N signal frames;
Step 3: performing EMD decomposition on the signal frame s_t i obtained in the step 2 to obtain a decomposed IMF component s_t i_IMFj, j=1, 2, 3. Analyzing the S_t i_IMFj to determine a noise-containing IMF component S_t i_IMFn needing wavelet threshold denoising in the high-frequency component and a noise-free low-frequency IMF component S_t i_IMFm, wherein n is E j, m is E j, and m is not equal to n; performing wavelet threshold denoising on the S_t i_IMFn to obtain a wavelet threshold denoised signal Will beCarrying out signal reconstruction with the S_t i_IMFm to obtain a denoised signal S_process_t i;
Step 4: dividing the denoised signal S_process_t i obtained in the step3 according to states to form a normal data set Nomal _set and a Fault data set fault_set; extracting partial data from Nomal _set to form a training data set train_set, wherein the data set of the rest part is denoted as Nomal _reserve_set; combining the fault_set with Nomal _result_set to obtain a Test data set test_set;
step 5: establishing an equipment fault detection model based on a physical information neural network, and determining the structure, the layer number, the node number and a vibration equation to be fitted of the equipment fault diagnosis model of the physical information neural network; performing unsupervised training on the established equipment fault detection model based on the physical information neural network by using the training data set obtained in the step 4 to obtain a training fitting error Loss train_loss, a fitted vibration equation and initial model parameters; calculating the maximum fitting error Loss Max_normal_Loss of the Normal state data; calculating a Normal error Loss range normal_range by utilizing the train_loss and the Max_normal_loss;
step 5.1: determining the structure, the layer number, the node number and a vibration equation to be fitted of a fault diagnosis model of the physical information neural network equipment; the neural network structure is divided into a first input layer, a plurality of middle hidden layers and a last output layer, wherein the input of the input layer is data in a training data set train_set, the hidden layers are used for extracting different layer characteristics, the number of layers and the number of nodes of the hidden layers are determined according to training and test analysis, and the output of the output layer is a signal time sequence predicted according to a fitted vibration equation; the network structure adopts a structural form that no connection exists in the layers and all adjacent layers are connected;
step 5.2: acquiring a data set train_set from the step 4, selecting one sample data from the data set train_set as train_data=train_set i, performing unsupervised training on the model established in the step 5.1, acquiring a training fitting error Loss train_loss, and acquiring a fitted vibration equation and initial model parameters;
Step 5.2.1: setting an initial position T 0 as a starting point of a fitting period, taking an integer multiple of a signal repetition period as a relative period T, and taking the relative period T as a unit to extract sample points as input for subsequent test working data;
step 5.2.2: this step is a data driven discovery process, defining f (t) as shown in equation (9):
f(t):=ut+Ν[u;λ]; (9)
Approximating u t by a deep neural network, where u t is a derivative operator N [ u; lambda forms a physical information neural network f (t) together, the parameter lambda in the differential operator is changed into the parameter f (t), and the corresponding lambda parameter in the normal state can be obtained through training, so that the vibration equation f (t) of the required fitting is obtained;
Step 5.2.3: calculating a mean square error Loss train_loss in the training process, and learning the neural network by minimizing the mean square error Loss, wherein the calculation process is shown in formulas (10) - (12):
Train_Loss=Train_Lossu+Train_Lossf; (10)
Wherein train_loss u is the mean square error Loss of the neural network approaching u t, train_loss f is the mean square error Loss of the sample points in formula (9), N is the number of the sampled points, u (t i) is a predicted value of u t, u i is a true value, and f (t i) is the error Loss of formula (9);
Step 5.3: according to step 5.2, data is obtained from a data set train_set j (j not equal to i), the data is input into the model trained in step 5.2, the model fitting error Loss of each Normal state data sample is calculated, the step is repeated, a plurality of groups of Normal state data fitting error Loss normal_loss j is obtained, and the maximum value is Max_normal_loss;
Step 5.4: calculating the train_loss obtained in the step 5.2 and the Max_normal_loss obtained in the step 5.3 to obtain a Normal error Loss range, as shown in a formula (8):
Normal_range=abs(Max_Normal_Loss-Train_Loss); (8)
Step 6: diagnosing faults of equipment to be tested, inputting a Test set of test_set into the model obtained by training in the step 5, and calculating Test fitting error Loss test_loss j of each sample data; calculating the train_loss and the test_loss j obtained in the step 5 to obtain a test_range j of the Test error Loss; if test_range j exceeds the Normal error loss range normal_range, the vibration signal is considered as a fault condition.
2. The method for detecting equipment failure based on sound and vibration signals according to claim 1, wherein the sound signal or vibration signal S in step 1 is derived from a sound or vibration signal collected by a sensor during operation of the equipment to be detected.
3. The method for detecting equipment failure based on sound and vibration signals according to claim 1, wherein said step 3 comprises the following concrete implementation steps:
Step 3.1: performing EMD decomposition on the s_t i to obtain M decomposed IMF components s_t i_IMFj, j=1, 2, 3.
S_ti_IMFj=emd(S_ti); (1)
Step 3.2: analyzing the IMF component S_t i_IMFj to determine a noise-containing IMF component S_t i_IMFn which needs wavelet threshold denoising in the high-frequency component, wherein n is E j; the noise-free low frequency IMF component is denoted s_t i_IMFm, where m e j, m+.n;
Step 3.3: performing wavelet threshold denoising on the S_t i_IMFn, analyzing and removing unsteady signals to obtain signals subjected to wavelet threshold denoising N signal frames are processed to obtain N1-dimensional vectors, as shown in a formula (2):
Step 3.4: obtaining noise-reduced data The signal reconstruction is performed together with the noise-free low-frequency IMF component s_t i_IMFm to obtain a denoised signal s_process_t i, as shown in formula (3):
4. a method for detecting equipment failure based on sound and vibration signals according to claim 3, wherein said step 3.2 is implemented by the following sub-steps:
Step 3.2.1: performing m-layer wavelet decomposition on the signal frame S_t i_IMFn, and finally outputting m+1 wavelet coefficient sets As shown in formulas (4) and (5):
Wherein: for decomposed low frequency information, approximation,/> For the decomposed high-frequency information, the detail part, func is the selected wavelet basis function, and m is the decomposition layer number;
Step 3.2.2: after the wavelet decomposition in step 3.2.1, the wavelet coefficient is obtained by selecting proper threshold value and threshold value function Processing; wavelet coefficients greater than a threshold are considered to be generated by the signal, should be preserved, and wavelet coefficients less than the threshold are considered to be generated by noise, set to 0 for denoising purposes; finally, the wavelet coefficient after threshold processing is obtainedAs shown in formula (6):
Wherein: value is a threshold;
step 3.2.3: thresholding the wavelet coefficient subjected to the step 3.2.2 Performing wavelet reconstruction to obtain signal data/>, after wavelet denoisingAs shown in formula (7):
wherein: func is the same wavelet basis function as in step 3.2.1.
5. The method for detecting equipment failure based on sound and vibration signals according to claim 1, wherein said step 4 comprises the following concrete implementation steps:
step 4.1: dividing the denoised signal S_process_t i obtained in the step 3 according to a normal state and a Fault state, randomly extracting data from the data in the normal state to form a normal data set Nomal _set, and randomly extracting data from the data in the Fault state to form a Fault data set fault_set;
Step 4.2: randomly extracting part of normal data from the normal data set Nomal _set obtained in the step 4.1 to form a training data set train_set, and marking the rest part of normal data set as Nomal _reserve_set;
Step 4.3: after combining the Fault data set fault_set from step 4.1 with Nomal _result_set in step 4.2, a Test data set test_set is obtained.
6. The method for detecting equipment failure based on sound and vibration signals according to claim 1, wherein said step 6 comprises the following concrete implementation steps:
Step 6.1: inputting each signal sample data in the test_set into the model trained in the step 5, and obtaining a Test fitting error Loss test_loss j of each sample data through calculation;
Step 6.2: calculating the train_loss obtained in step 5.2 and the test_loss j obtained in step 6.1 to obtain a test_range j of the Test error Loss of each sample data, as shown in formula (13):
Test_rangej=abs(Test_Lossj-Train_Loss); (13)
Step 6.3: taking the normal_range obtained in the step 5.4 as a reference error range, if the test_range j exceeds the reference range normal_range, indicating that the vibration signal is not within the Normal range, the vibration signal is regarded as a fault state.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210637988.7A CN115114955B (en) | 2022-06-07 | 2022-06-07 | Equipment fault detection method based on sound and vibration signals |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210637988.7A CN115114955B (en) | 2022-06-07 | 2022-06-07 | Equipment fault detection method based on sound and vibration signals |
Publications (2)
Publication Number | Publication Date |
---|---|
CN115114955A CN115114955A (en) | 2022-09-27 |
CN115114955B true CN115114955B (en) | 2024-04-30 |
Family
ID=83326791
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210637988.7A Active CN115114955B (en) | 2022-06-07 | 2022-06-07 | Equipment fault detection method based on sound and vibration signals |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN115114955B (en) |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111428685A (en) * | 2020-04-14 | 2020-07-17 | 北京华控智加科技有限公司 | Machine fault detection, classification and grading method based on neural network unified modeling |
CN111523659A (en) * | 2020-04-14 | 2020-08-11 | 北京华控智加科技有限公司 | Machine fault prediction diagnosis method based on three-level neural network modeling |
CN114234361A (en) * | 2021-12-14 | 2022-03-25 | 北京工业大学 | Central air-conditioning sensor fault detection method based on double noise reduction and fuzzy indexes |
Family Cites Families (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107832687A (en) * | 2017-10-27 | 2018-03-23 | 武汉大学 | Fault diagnostic method for transformer winding based on wireless identification sensing |
CN109933881A (en) * | 2019-03-06 | 2019-06-25 | 武汉大学 | A kind of Fault Diagnosis of Power Electronic Circuits method based on optimization deepness belief network |
-
2022
- 2022-06-07 CN CN202210637988.7A patent/CN115114955B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111428685A (en) * | 2020-04-14 | 2020-07-17 | 北京华控智加科技有限公司 | Machine fault detection, classification and grading method based on neural network unified modeling |
CN111523659A (en) * | 2020-04-14 | 2020-08-11 | 北京华控智加科技有限公司 | Machine fault prediction diagnosis method based on three-level neural network modeling |
CN114234361A (en) * | 2021-12-14 | 2022-03-25 | 北京工业大学 | Central air-conditioning sensor fault detection method based on double noise reduction and fuzzy indexes |
Non-Patent Citations (2)
Title |
---|
基于提升小波变换与EEMD的神经网络齿轮故障诊断方法;宋萌萌;肖顺根;陈肇祥;;组合机床与自动化加工技术(09);全文 * |
基于自适应模糊神经网络的发动机故障诊断;马继昌;司景萍;牛嘉骅;王二毛;;噪声与振动控制(02);全文 * |
Also Published As
Publication number | Publication date |
---|---|
CN115114955A (en) | 2022-09-27 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US11630034B2 (en) | Method for diagnosing and predicting operation conditions of large-scale equipment based on feature fusion and conversion | |
CN111238814B (en) | Rolling bearing fault diagnosis method based on short-time Hilbert transform | |
CN110795843B (en) | Method and device for identifying faults of rolling bearing | |
CN112101174A (en) | LOF-Kurtogram-based mechanical fault diagnosis method | |
CN108444696A (en) | A kind of gearbox fault analysis method | |
CN110987434A (en) | Rolling bearing early fault diagnosis method based on denoising technology | |
CN116226646B (en) | Method, system, equipment and medium for predicting health state and residual life of bearing | |
CN105424366A (en) | Bearing fault diagnosis method based on EEMD adaptive denoising | |
CN113569990B (en) | Strong noise interference environment-oriented performance equipment fault diagnosis model construction method | |
CN115424635B (en) | Cement plant equipment fault diagnosis method based on sound characteristics | |
Zheng et al. | Faults diagnosis of rolling bearings based on shift invariant K-singular value decomposition with sensitive atom nonlocal means enhancement | |
CN111291918B (en) | Rotating machine degradation trend prediction method based on stationary subspace exogenous vector autoregression | |
CN113345399A (en) | Method for monitoring sound of machine equipment in strong noise environment | |
Li et al. | Optimal symbolic entropy: An adaptive feature extraction algorithm for condition monitoring of bearings | |
Mubaraali et al. | Intelligent fault diagnosis in microprocessor systems for vibration analysis in roller bearings in whirlpool turbine generators real time processor applications | |
CN111025100A (en) | Transformer ultrahigh frequency partial discharge signal mode identification method and device | |
CN114263621A (en) | Test method and system for diagnosing and simulating cavitation fault of centrifugal pump | |
Yu et al. | Sparse time–frequency representation for the transient signal based on low-rank and sparse decomposition | |
CN110222390B (en) | Gear crack identification method based on wavelet neural network | |
CN115114955B (en) | Equipment fault detection method based on sound and vibration signals | |
CN112033656A (en) | Mechanical system fault detection method based on broadband spectrum processing | |
CN116482526A (en) | Analysis system for non-fault phase impedance relay | |
CN116484176A (en) | Bearing fault diagnosis method, system and storage medium based on ultra-wavelet | |
Zhang et al. | A novel hybrid compound fault pattern identification method for gearbox based on NIC, MFDFA and WOASVM | |
CN115563480A (en) | Gear fault identification method for screening octave geometric modal decomposition based on kurtosis ratio coefficient |
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 |