Equipment part fault diagnosis method, device and equipment based on sensing data mixed noise suppression

Abstract

The invention discloses an equipment part fault diagnosis method, device and equipment based on sensing data mixed noise suppression, and relates to the field of fault diagnosis. The diagnosis method comprises the following steps: acquiring vibration signals of an equipment part under different fault types; performing data preprocessing on the vibration signal to obtain a sample, taking a fault type corresponding to the vibration signal as a real label, and constructing a training set and a verification set; using the training set and the verification set, and using a migratable Adaboost algorithm and an adaptive cost sensitive learning method to train a plurality of deep learning models to obtain a plurality of trained deep learning models; and performing fault diagnosis on the to-be-tested part of the equipment by using the plurality of trained deep learning models. According to the invention, the accuracy of equipment part fault diagnosis in the mixed noise environment is improved.

Description

Technical Field

[0001] The present invention relates to the field of fault diagnosis, and in particular to a method, a device and an apparatus for diagnosing equipment parts faults based on mixed noise suppression of sensor data. Background Art

[0002] Key parts of complex equipment such as bearings, gears, and motors are widely used in many fields of modern industry. However, many parts work poorly, which inevitably leads to failures, which may cause serious accidents and huge losses. Therefore, accurate and timely fault diagnosis of key parts of complex equipment and timely intervention when early signs of failure are found are of great significance in preventing major accidents and ensuring personal safety.

[0003] In academia and industry, fault diagnosis methods based on deep learning have received increasing attention, including convolutional neural networks (CNN), recurrent neural networks (RNN), autoencoders (AE), deep belief networks (DBN), etc. However, due to the harsh working environment of complex equipment, the data collected from many key parts of complex equipment usually contain a lot of noise, which reduces the quality of the data. The interference of noise makes it easy for valuable fault features to be submerged, and the feature extraction ability of deep learning tends to decline. It may not be able to learn fault-related features well, resulting in a decrease in the accuracy of fault diagnosis.

[0004] The most representative existing methods for the problem of noisy data are deep residual shrinkage networks and their various improved models. However, due to the influence of factors such as random noise, the noise distribution in actual industrial data is uneven, resulting in different signal-to-noise ratios for each sample. Most existing methods only implement fault diagnosis under multiple fixed signal-to-noise ratios. For fault diagnosis in mixed noise environments, the stronger the noise, the more difficult it is to train. Due to the different training difficulties of each sample, these methods are difficult to obtain a high diagnostic accuracy. Summary of the invention

[0005] The purpose of the present invention is to provide a method, device and equipment for equipment part fault diagnosis based on mixed noise suppression of sensor data, which can improve the accuracy of equipment part fault diagnosis in a mixed noise environment.

[0006] To achieve the above object, the present invention provides the following solutions:

[0007] In a first aspect, the present invention provides an equipment parts fault diagnosis method based on mixed noise suppression of sensor data, the diagnosis method comprising:

[0008] Obtain vibration signals of equipment parts under different fault types;

[0009] Perform data preprocessing on the vibration signal to obtain samples. The fault type corresponding to the vibration signal is used as the true label to construct the training set and the validation set.

[0010] Using the training set and the validation set, a plurality of deep learning models are trained using a transferable Adaboost algorithm and an adaptive cost-sensitive learning method to obtain a plurality of trained deep learning models;

[0011] Use multiple trained deep learning models to perform fault diagnosis on the equipment's parts to be tested.

[0012] Optionally, using the training set and the validation set, using a transferable Adaboost algorithm and an adaptive cost-sensitive learning method to train multiple deep learning models to obtain multiple trained deep learning models, specifically including:

[0013] Build M identical deep learning models;

[0014] Initialize the weight of each sample in the training set;

[0015] Let the value of m be 1;

[0016] Using the training set and the target loss function to train the m-th deep learning model, d trained m-th deep learning models are obtained; wherein the d trained m-th deep learning models are obtained by performing d iterations on the m-th deep learning model, when m is equal to 1, the loss function is a cross entropy loss function, and when m is greater than 1, the loss function is an adaptive weighted cross entropy loss function, and the number of training iterations is less than the number of iterations when m is equal to 1;

[0017] Use the validation set to validate the m-th deep learning model after d training, and obtain the m-th deep learning model after training with the highest validation accuracy as the m-th trained deep learning model; update the weight of each sample in the training set according to the training result when the m-th deep learning model has the highest validation accuracy;

[0018] Determine whether the value of m is less than M and obtain the determination result;

[0019] If the judgment result indicates yes, the parameters of the m-th trained deep learning model are transferred to the m+1-th deep learning model; the value of m is increased by 1, and the step of "training the m-th deep learning model using the training set and the target loss function to obtain d trained m-th deep learning models" is returned;

[0020] If the judgment result indicates no, M trained deep learning models are output.

[0021] Optionally, the formula for initializing the weight of each sample in the training set is:

[0022]

[0023] in, is the weight of the i-th sample after initialization, and n is the number of samples in the training set.

[0024] Optionally, the adaptive weighted cross entropy loss function is:

[0025]

[0026] Among them, l (m) is the adaptive weighted cross entropy loss used to train the m-th deep learning model, g is the batch size set during training, Sigmoid() is the Sigmoid function, and are the weight of the i-th sample and the average of all sample weights used to train the m-th deep learning model, y i is the true label of the i-th sample, p (m )(x i ) is the result obtained after the i-th sample passes through the m-th deep learning model and Softmax operation during training.

[0027] Optionally, the formula for updating the weight of each sample in the training set according to the training result when the m-th deep learning model verification accuracy is the highest is:

[0028]

[0029] δ i =(a1,a2,···,a b ,···,a K ),1≤b≤K;

[0030]

[0031] in, is the weight of the i-th sample when training the m+1-th deep learning model, and are the weights of the i-th and j-th samples when training the m-th deep learning model, β is the stability coefficient, K is the number of fault types, and δ i and δ j are the category vectors obtained from the true labels of the i-th sample and the j-th sample, respectively. a1, a2, a b and a K is δ i The first, second, bth and Kth elements in p (m) (xi ) and p (m) (x j ) are the results of the i-th sample and the j-th sample after the m-th deep learning model and Softmax operation during training, and n is the number of samples in the training set.

[0032] Optionally, using multiple trained deep learning models, the formula for fault diagnosis of the parts to be tested of the equipment is:

[0033]

[0034] Among them, C(x) is the fault diagnosis result corresponding to the sample x of the tested part, and They are the probabilities of the sample x of the tested part belonging to the kth class and the k'th class after the mth trained deep learning model and Softmax operation, M is the number of trained deep learning models, and K is the number of fault types.

[0035] Optionally, the deep learning model is a deep residual shrinkage network with a selective kernel convolution module and a global parameterized ReLU.

[0036] Optionally, the selective kernel convolution module includes a first convolution layer, a second convolution layer, a first normalization layer, a second normalization layer, a first activation function module, a second activation function module, a first element-by-element addition module, a global average pooling layer, a first fully connected layer, a third batch normalization layer, a third activation function module, a second fully connected layer, a third fully connected layer, a Softmax module, a first channel-by-channel multiplication module, a second channel-by-channel multiplication module, and a second element-by-element addition module;

[0037] The first convolutional layer is connected to the first batch of normalized layers, and the first batch of normalized layers is connected to the first activation function module;

[0038] The second convolutional layer is connected to the second batch normalization layer, and the second batch normalization layer is connected to the second activation function module;

[0039] The first activation function module is respectively connected to the first input end of the first element-by-element addition module and the first input end of the first channel-by-channel multiplication module; the second activation function module is respectively connected to the second input end of the first element-by-element addition module and the first input end of the second channel-by-channel multiplication module;

[0040] The output end of the first element-by-element addition module is connected to the global average pooling layer, the global average pooling layer is connected to the first fully connected layer, the first fully connected layer is connected to the third batch normalization layer, the third batch normalization layer is connected to the third activation function module, and the third activation function module is connected to the second fully connected layer and the third fully connected layer respectively;

[0041] The second fully connected layer is connected to the first input terminal of the Softmax module, and the first output terminal of the Softmax module is connected to the second input terminal of the first channel-by-channel multiplication module;

[0042] The third fully connected layer is connected to the second input terminal of the Softmax module, and the second output terminal of the Softmax module is connected to the second input terminal of the second channel-by-channel multiplication module;

[0043] The output end of the first channel-by-channel multiplication module is connected to the first input end of the second element-by-element addition module, and the output end of the second channel-by-channel multiplication module is connected to the second input end of the second element-by-element addition module.

[0044] In a second aspect, the present invention provides an equipment parts fault diagnosis device based on sensor data mixed noise suppression, the diagnostic device applies the above-mentioned diagnostic method, and the diagnostic device comprises:

[0045] A signal acquisition module, used to obtain vibration signals of equipment parts under different fault types;

[0046] The data preprocessing module is used to preprocess the vibration signal, obtain samples, and use the fault type corresponding to the vibration signal as the true label to construct the training set and the verification set;

[0047] A training module is used to train multiple deep learning models using the training set and the validation set, using a transferable Adaboost algorithm and an adaptive cost-sensitive learning method, to obtain multiple trained deep learning models;

[0048] The diagnosis module is used to use multiple trained deep learning models to perform fault diagnosis on the parts to be tested of the equipment.

[0049] In a third aspect, a computer device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-mentioned equipment part fault diagnosis method based on mixed noise suppression of sensor data.

[0050] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0051] The present invention provides a method, device and apparatus for equipment parts fault diagnosis based on mixed noise suppression of sensor data. The core of the diagnostic method provided by the present invention includes transferable Adaboost, adaptive cost-sensitive learning and a deep residual shrinkage network with a selective kernel convolution module and a global parameterized ReLU, which reduces the influence of mixed noise in sensor data on feature learning, solves the problem that it is difficult to well implement fault diagnosis in a mixed noise environment in the existing methods, and improves the accuracy of equipment parts fault diagnosis in a mixed noise environment.

[0052] Because the training difficulty of each sample in the mixed noise data is different, the present invention uses the principle of migratable Adaboost to increase the weight of difficult-to-train samples and reduce the weight of easy-to-train samples to update the weights, and proposes an adaptive weighted cross entropy loss function to achieve cost-sensitive learning. The sample weights and adaptive weighted cross entropy loss function obtained by migratable Adaboost enable the deep learning model to pay more attention to strong noise samples. In addition, migratable Adaboost improves the diagnostic performance in a mixed noise environment, and the number of iterations for training other deep learning models is less than the number of iterations for training the first deep learning model, which reduces the model training time while ensuring a better diagnostic effect.

[0053] The present invention proposes a deep residual shrinkage network with a selective kernel convolution module and a global parameterized ReLU, introduces a selective kernel convolution module, and sets different convolution kernel sizes in the two branches. Because large convolution kernels are suitable for processing strong noise samples, and small convolution kernels are suitable for processing weak noise samples, the selective kernel convolution module can effectively combine the respective advantages of large and small convolution kernels, thereby improving the fault diagnosis effect in a mixed noise environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0055] Figure 1 This is an application environment diagram of an equipment parts fault diagnosis method based on mixed noise suppression of sensor data in one embodiment of the present invention;

[0056] Figure 2 A schematic flow chart of a method for diagnosing equipment parts faults based on mixed noise suppression of sensor data provided by one embodiment of the present invention;

[0057] Figure 3A schematic diagram of a method for diagnosing equipment parts faults based on mixed noise suppression of sensor data provided by an embodiment of the present invention;

[0058] Figure 4 A structural diagram of a deep residual shrinkage network with a selective kernel convolution module and a global parameterized ReLU provided in one embodiment of the present invention;

[0059] Figure 5 A structural diagram of a residual block provided by an embodiment of the present invention;

[0060] Figure 6 A structural diagram of a selective kernel convolution module provided by an embodiment of the present invention;

[0061] Figure 7 A structural diagram of a global parameterized ReLU activation function module provided by an embodiment of the present invention;

[0062] Figure 8 A schematic diagram of the structure of a computer device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0063] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0064] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and understandable, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0065] The equipment parts fault diagnosis method based on sensor data mixed noise suppression provided by the embodiment of the present invention can be applied to Figure 1In the application environment shown. Among them, the terminal 101 communicates with the server 102 through the network. The data storage system can store the data that the server 102 needs to process. The data storage system can be set up separately, integrated on the server 102, or placed on the cloud or other servers. The terminal 101 can send the vibration signals of the equipment parts under different fault types to the server 102, and the server 102 receives the vibration signals of the equipment parts under different fault types; the server 102 performs data preprocessing on the vibration signal to obtain samples, and uses the fault type corresponding to the vibration signal as the true label to construct a training set and a verification set; using the training set and the verification set, use the transferable Adaboost algorithm and the adaptive cost-sensitive learning method to train multiple deep learning models to obtain multiple trained deep learning models; use multiple trained deep learning models to perform fault diagnosis on the parts to be tested of the equipment. The server 102 can feedback the obtained diagnosis results to the terminal 101. In addition, in some embodiments, the equipment parts fault diagnosis method based on mixed noise suppression of sensor data can also be implemented independently by the server 102 or the terminal 101. For example, the terminal 101 can directly process the vibration signals of the equipment parts under different fault types, or the server 102 can obtain the vibration signals of the equipment parts under different fault types from the data storage system and process the vibration signals of the equipment parts under different fault types.

[0066] The terminal 101 may be, but is not limited to, various desktop computers, laptop computers, smart phones, tablet computers, IoT devices, and portable wearable devices. The IoT devices may be smart speakers, smart TVs, smart air conditioners, smart vehicle-mounted devices, etc. The portable wearable devices may be smart watches, smart bracelets, head-mounted devices, etc. The server 102 may be implemented as an independent server or a server cluster consisting of multiple servers, or may be a cloud server.

[0067] In an exemplary embodiment, Figure 2 As shown, a method for equipment parts fault diagnosis based on mixed noise suppression of sensor data is provided. The method is executed by a computer device, and can be executed by a computer device such as a terminal or a server alone, or by a terminal and a server together. In an embodiment of the present invention, the method is applied to Figure 1 The server 102 in the example is used as an example to illustrate, including the following steps 201 to 204. Among them:

[0068] Step 201, obtaining vibration signals of equipment parts under different fault types.

[0069] Step 202 , preprocess the vibration signal to obtain samples, use the fault type corresponding to the vibration signal as the true label, and construct a training set and a validation set.

[0070] Step 203: Use the training set and the validation set to train multiple deep learning models using a transferable Adaboost algorithm and an adaptive cost-sensitive learning method to obtain multiple trained deep learning models.

[0071] Step 204: Utilize multiple trained deep learning models to perform fault diagnosis on the parts to be tested of the equipment.

[0072] Implementing the above steps 201 to 204 can improve the accuracy of equipment parts fault diagnosis in a mixed noise environment. The above technical solution of the present invention can be applied to the fault diagnosis of key parts of complex equipment.

[0073] In another exemplary embodiment of the present invention, in order to implement data preprocessing, as Figure 3 As shown, the above steps 201 and 202 are replaced by the following steps S1-S2:

[0074] S1: The initial vibration signal of the key parts of complex equipment during operation is obtained through sensors. The vibration signal contains mixed noise.

[0075] This embodiment uses the Case Western Reserve University (CWRU) rolling bearing data set to experimentally verify the equipment parts fault diagnosis method based on sensor data mixed noise suppression proposed in the present invention.

[0076] The test bench for collecting the CWRU data set includes a motor, a torque sensor / encoder, a power tester, and an electronic controller. Single-point faults are introduced into the test bearing using electrospark machining, the vibration signal of the bearing is collected by an acceleration sensor, and the speed and horsepower data are collected using a torque sensor / encoder. The CWRU data set includes normal baseline data, drive-end bearing fault data at a sampling frequency of 12kHz, drive-end bearing fault data at a sampling frequency of 48kHz, and fan-end bearing fault data at a sampling frequency of 12KHz. The CWRU data set includes 4 working conditions.

[0077] In the embodiment, normal baseline data and drive end bearing fault data at a sampling frequency of 48kHz are selected for experiment, and the selected working conditions are 2 horsepower and 1750r / min. The selected data description is shown in Table 1.

[0078] Table 1 is a description of the selected data

[0079] describe Label normal 0 Inner race fault with a fault diameter of 7 mils 1 Inner race fault with a fault diameter of 21 mils 2 Rolling element failure with a failure diameter of 7 mils 3 Rolling element failure with a failure diameter of 21 mils 4 Outer ring fault with a fault diameter of 7 mils 5 Outer ring fault with a fault diameter of 21 mils 6

[0080] S2: After data preprocessing of the initial vibration signal, samples of the training set and the validation set are obtained, and the training set and the validation set are composed of the samples of the training set and the validation set and the corresponding fault types.

[0081] In this embodiment, the sliding window length is set to 1024, and the sliding step size is set to 1024; part of the vibration signals are selected for dividing the samples, the number of samples in each category in the training set is 120, and the number of samples in each category in the validation set is 90; because the noise intensity of the vibration signal in the CWRU data set is very small and unknown, Gaussian white noise is added to the training set samples and the validation set samples to simulate the mixed noise data, and 10 signal-to-noise ratios from -6dB to 3dB are set, and the number of samples corresponding to each signal-to-noise ratio accounts for 10% of the total number of samples.

[0082] In another exemplary embodiment of the present invention, in order to realize the construction and training of the deep learning model, as Figure 3 As shown, the above step 203 is replaced by the following steps S31-S39:

[0083] S31: Build M (M>1) identical and noise-resistant deep learning models.

[0084] S32: Initialize the parameters of the first deep learning model; initialize the weights of the training set samples by the following formula:

[0085]

[0086] in, is the weight of the i-th sample after initialization, and n is the number of samples in the training set.

[0087] S33: Set a larger number of iterations, use the cross entropy loss function to train the first deep learning model, and select the model with the highest verification accuracy for subsequent model parameter migration and model testing.

[0088] S34: When the verification accuracy is the highest, the sample weight is updated by the following formula for training the next deep learning model.

[0089]

[0090] δ i =(a1,a2,···,a b ,···,a K ),1≤b≤K.

[0091]

[0092] in, is the weight of the i-th sample used to train the second deep learning model, β is the stability coefficient, K is the number of fault types, and δ i is the category vector obtained from the true label of the i-th sample, a b is δ i The bth element in p(1) (x i ) is the result obtained after the i-th sample passes through the first deep learning model and Softmax operation during training.

[0093] S35: Migrate model parameters to the next deep learning model.

[0094] S36: Update the adaptive weighted cross entropy loss function according to the current sample weight; the adaptive weighted cross entropy loss function formula is as follows:

[0095]

[0096] Among them, l (m) is the adaptive weighted cross entropy loss used to train the m-th deep learning model, g is the batch size set during training, Sigmoid() is the Sigmoid function, and are the weight of the i-th sample and the average of all sample weights used to train the m-th deep learning model, y i is the true label of the i-th sample, p (m) (x i ) is the result obtained after the i-th sample passes through the m-th deep learning model and Softmax operation during training.

[0097] S37: Set a smaller number of iterations, use the adaptive weighted cross entropy loss function to train the current deep learning model, and select the model with the highest verification accuracy for subsequent model parameter migration and model testing.

[0098] S38: When the verification accuracy is the highest, the sample weight is updated by the following formula for training the next deep learning model.

[0099]

[0100] in, is the weight of the i-th sample when training the m+1-th deep learning model. At this time, m is greater than 1. and are the weights of the i-th and j-th samples when training the m-th deep learning model, β is the stability coefficient, K is the number of fault types, and δ i and δ j are the category vectors obtained from the true labels of the i-th sample and the j-th sample, respectively, (m) (x i ) and p (m) (x j) are the results of the i-th sample and the j-th sample after the m-th deep learning model and Softmax operation during training, and n is the number of samples in the training set.

[0101] S39: Repeat S35-S38 until all M deep learning models have completed training.

[0102] The noise-resistant deep learning model in S31 is specifically based on the deep residual contraction network with global parameterized ReLU (GPReLU), and the front convolution layer of the model is replaced with a selective kernel convolution module; the deep residual contraction network with selective kernel convolution module and GPReLU is as follows Figure 4 As shown in, the residual block is as follows Figure 5 shown.

[0103] Figure 6 Where C represents the number of feature channels, W1 represents the length of the sample along the time dimension, W2 represents the length of the feature along the time dimension, and M represents the number of neurons in the fully connected (FC) layer. Figure 6 As shown, the selective kernel convolution module includes:

[0104] The first convolution layer, the second convolution layer, the first batch normalization (BN) layer, the second BN layer, the first activation function module, the second activation function module, the first element-by-element addition module, the global average pooling (GAP) layer, the first FC layer, the third BN layer, the third activation function module, the second FC layer, the third FC layer, the Softmax module, the first channel-by-channel multiplication module, the second channel-by-channel multiplication module, and the second element-by-element addition module;

[0105] The first convolutional layer is connected to the first BN layer, and the first BN layer is connected to the first activation function module;

[0106] The second convolutional layer is connected to the second BN layer, and the second BN layer is connected to the second activation function module;

[0107] The first activation function module is respectively connected to the first input end of the first element-by-element addition module and the first input end of the first channel-by-channel multiplication module; the second activation function module is respectively connected to the second input end of the first element-by-element addition module and the first input end of the second channel-by-channel multiplication module;

[0108] The output end of the first element-by-element addition module is connected to the GAP layer, the GAP layer is connected to the first FC layer, the first FC layer is connected to the third BN layer, the third BN layer is connected to the third activation function module, and the third activation function module is connected to the second FC layer and the third FC layer respectively;

[0109] The second FC layer is connected to the first input terminal of the Softmax module, and the first output terminal of the Softmax module is connected to the second input terminal of the first channel-by-channel multiplication module;

[0110] The third FC layer is connected to the second input terminal of the Softmax module, and the second output terminal of the Softmax module is connected to the second input terminal of the second channel-by-channel multiplication module;

[0111] The output end of the first channel-by-channel multiplication module is connected to the first input end of the second element-by-element addition module, and the output end of the second channel-by-channel multiplication module is connected to the second input end of the second element-by-element addition module.

[0112] The activation function module is the GPReLU activation function module, such as Figure 7 shown.

[0113] In the specific implementation, the structural hyperparameters of the deep residual shrinkage network with selective kernel convolution module and GPReLU are shown in Table 2; the convolution kernel sizes of the first convolution layer and the second convolution layer in the selective kernel convolution module are 7 and 31 respectively; the number of neurons in the 4 FC layers in the activation function module is equal to the number of channels of the module input features.

[0114] Table 2 shows the structural hyperparameters of the deep residual shrinkage network with selective kernel convolution module and GPReLU.

[0115] Network layer name Number of units Output size Selective Kernel Convolution Module 1 4×512×1 Residual block (4,3, / 2) 1 4×256×1 Residual Block (4,3) 3 4×256×1 Residual block (8,3, / 2) 1 8×128×1 Residual Block (8,3) 3 8×128×1 Residual block (16,3, / 2) 1 16×64×1 Residual Block (16,3) 3 16×64×1 BN, GPReLU, GAP 1 16 FC 1 7

[0116] In this embodiment, the batch size is 32, the optimizer is Adam, the learning rate is 0.002, the learning rate adjustment strategy uses CosineAnnealingLR, and the stability coefficient is 0.1; when training the first deep learning model, the loss function uses the cross entropy loss function, and the number of iterations is 100; when training other deep learning models, the loss function uses the adaptive weighted cross entropy loss function, and the number of iterations is 20.

[0117] In another exemplary embodiment of the present invention, in order to implement fault diagnosis, as Figure 3 As shown, the above step 204 is replaced by the following step S4:

[0118] S4: After preprocessing the vibration signal of the part to be tested, the test sample is obtained and input into multiple trained deep learning models. Through the model test based on the transferable Adaboost, the fault type of the current part is obtained.

[0119] The model test based on transferable Adaboost is specifically to obtain the fault type of the current part through the following formula:

[0120]

[0121] Among them, C(x) is the fault diagnosis result corresponding to the sample x of the tested part, and They are the probabilities of the sample x of the tested part belonging to the kth class and the k'th class after the mth trained deep learning model and Softmax operation, M is the number of trained deep learning models, and K is the number of fault types.

[0122] In the specific implementation, the sliding window length is set to 1024, and the sliding step size is set to 1024; some vibration signals are selected to divide the samples, the number of samples in each category in the test set is 90, and they will not overlap with the training set samples and the validation set samples; because the noise intensity of the vibration signal in the CWRU dataset is very small and unknown, Gaussian white noise is added to the training set samples and the validation set samples to simulate the mixed noise data, and a total of 10 signal-to-noise ratios from -6dB to 3dB are set, and the number of samples corresponding to each signal-to-noise ratio accounts for 10% of the total number of samples.

[0123] The experiment set 4 different numbers of deep learning models. In order to reduce the influence of random factors, 10 experiments were conducted under each number of deep learning models. The experimental results are shown in Table 3. It can be seen that the accuracy rate obtained in the experiment is relatively high, which verifies the effectiveness of the method proposed in the present invention.

[0124] Table 3 shows the fault diagnosis test accuracy (%) under different numbers of deep learning models

[0125]

[0126]

[0127] The above embodiment is a tuning result of the present invention on the CWRU bearing data set, but the specific implementation of the present invention is not limited to the above embodiment. The description of the above embodiment is only used to help understand the method and core idea of ​​the present invention; at the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation of the present invention.

[0128] Based on the same inventive concept, the embodiment of the present invention also provides an equipment part fault diagnosis device for implementing the equipment part fault diagnosis method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more embodiments of the equipment part fault diagnosis device based on sensor data mixed noise suppression provided below can refer to the above limitations on the equipment part fault diagnosis method based on sensor data mixed noise suppression, and will not be repeated here.

[0129] In an exemplary embodiment, there is provided an equipment parts fault diagnosis device based on sensor data hybrid noise suppression, comprising:

[0130] The signal acquisition module is used to obtain vibration signals of equipment parts under different fault types.

[0131] The data preprocessing module is used to perform data preprocessing on the vibration signal, obtain samples, and use the fault type corresponding to the vibration signal as the true label to construct the training set and the verification set.

[0132] The training module is used to use the training set and the validation set to train multiple deep learning models using a transferable Adaboost algorithm and an adaptive cost-sensitive learning method to obtain multiple trained deep learning models.

[0133] The diagnosis module is used to use multiple trained deep learning models to perform fault diagnosis on the parts to be tested of the equipment.

[0134] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Figure 8 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data determined by complementarity. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for diagnosing equipment parts faults based on mixed noise suppression of sensor data is implemented.

[0135] Those skilled in the art will understand that Figure 8 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0136] In an exemplary embodiment, a computer device is also provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the steps in the above-mentioned embodiment of the equipment part fault diagnosis method based on mixed noise suppression of sensor data are implemented.

[0137] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided by the present invention can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0138] The database involved in each embodiment provided by the present invention may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., but is not limited thereto. The processor involved in each embodiment provided by the present invention may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, etc., but is not limited thereto.

[0139] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0140] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A method for equipment parts fault diagnosis based on mixed noise suppression of sensor data, characterized in that: The diagnostic method comprises: Obtain vibration signals of equipment parts under different fault types; Perform data preprocessing on the vibration signal to obtain samples. The fault type corresponding to the vibration signal is used as the true label to construct the training set and the validation set. Using the training set and the validation set, a plurality of deep learning models are trained using a transferable Adaboost algorithm and an adaptive cost-sensitive learning method to obtain a plurality of trained deep learning models; Use multiple trained deep learning models to perform fault diagnosis on the equipment's parts to be tested.

2. The equipment parts fault diagnosis method based on sensor data mixed noise suppression according to claim 1 is characterized in that: Using the training set and the validation set, multiple deep learning models are trained using a transferable Adaboost algorithm and an adaptive cost-sensitive learning method to obtain multiple trained deep learning models, specifically including: Build M identical deep learning models; Initialize the weight of each sample in the training set; Let the value of m be 1; Using the training set and the target loss function to train the m-th deep learning model, d trained m-th deep learning models are obtained; wherein the d trained m-th deep learning models are obtained by performing d iterations on the m-th deep learning model, when m is equal to 1, the loss function is a cross entropy loss function, and when m is greater than 1, the loss function is an adaptive weighted cross entropy loss function, and the number of training iterations is less than the number of iterations when m is equal to 1; Use the validation set to validate the m-th deep learning model after d training, and obtain the m-th deep learning model after training with the highest validation accuracy as the m-th trained deep learning model; update the weight of each sample in the training set according to the training result when the m-th deep learning model has the highest validation accuracy; Determine whether the value of m is less than M and obtain the determination result; If the judgment result indicates yes, the parameters of the m-th trained deep learning model are transferred to the m+1-th deep learning model; the value of m is increased by 1, and the step of "training the m-th deep learning model using the training set and the target loss function to obtain d trained m-th deep learning models" is returned; If the judgment result indicates no, M trained deep learning models are output.

3. The equipment parts fault diagnosis method based on sensor data mixed noise suppression according to claim 2 is characterized in that: The formula for initializing the weight of each sample in the training set is: in, is the weight of the i-th sample after initialization, and n is the number of samples in the training set.

4. The equipment parts fault diagnosis method based on sensor data mixed noise suppression according to claim 2 is characterized in that: The adaptive weighted cross entropy loss function is: Among them, l (m) is the adaptive weighted cross entropy loss used to train the m-th deep learning model, g is the batch size set during training, Sigmoid() is the Sigmoid function, and are the weight of the i-th sample and the average of all sample weights used to train the m-th deep learning model, y i is the true label of the i-th sample, p (m) (x i ) is the result obtained after the i-th sample passes through the m-th deep learning model and Softmax operation during training.

5. The equipment parts fault diagnosis method based on sensor data mixed noise suppression according to claim 2 is characterized in that: The formula for updating the weight of each sample in the training set according to the training result with the highest verification accuracy of the mth deep learning model is: δ i =(a1,a2,···,a b ,···,a K ),1≤b≤K; in, is the weight of the i-th sample when training the m+1-th deep learning model, and are the weights of the i-th and j-th samples when training the m-th deep learning model, β is the stability coefficient, K is the number of fault types, and δ i and δ j are the category vectors obtained from the true labels of the i-th sample and the j-th sample, respectively. a1, a2, a b and a K is δ i The first, second, bth and Kth elements in p (m) (x i ) and p (m) (x j ) are the results of the i-th sample and the j-th sample after the m-th deep learning model and Softmax operation during training, and n is the number of samples in the training set.

6. The equipment parts fault diagnosis method based on sensor data mixed noise suppression according to claim 1 is characterized in that: Using multiple trained deep learning models, the formula for fault diagnosis of the equipment parts to be tested is: Among them, C(x) is the fault diagnosis result corresponding to the sample x of the tested part, and They are the probabilities of the sample x of the tested part belonging to the kth class and the k'th class after the mth trained deep learning model and Softmax operation, M is the number of trained deep learning models, and K is the number of fault types.

7. The equipment parts fault diagnosis method based on sensor data mixed noise suppression according to claim 1 is characterized in that: The deep learning model is a deep residual shrinkage network with a selective kernel convolution module and a global parameterized ReLU.

8. The equipment parts fault diagnosis method based on sensor data mixed noise suppression according to claim 7 is characterized in that: The selective kernel convolution module includes a first convolution layer, a second convolution layer, a first normalization layer, a second normalization layer, a first activation function module, a second activation function module, a first element-by-element addition module, a global average pooling layer, a first fully connected layer, a third batch normalization layer, a third activation function module, a second fully connected layer, a third fully connected layer, a Softmax module, a first channel-by-channel multiplication module, a second channel-by-channel multiplication module, and a second element-by-element addition module; The first convolutional layer is connected to the first batch of normalized layers, and the first batch of normalized layers is connected to the first activation function module; The second convolutional layer is connected to the second batch normalization layer, and the second batch normalization layer is connected to the second activation function module; The first activation function module is respectively connected to the first input end of the first element-by-element addition module and the first input end of the first channel-by-channel multiplication module; the second activation function module is respectively connected to the second input end of the first element-by-element addition module and the first input end of the second channel-by-channel multiplication module; The output end of the first element-by-element addition module is connected to the global average pooling layer, the global average pooling layer is connected to the first fully connected layer, the first fully connected layer is connected to the third batch normalization layer, the third batch normalization layer is connected to the third activation function module, and the third activation function module is connected to the second fully connected layer and the third fully connected layer respectively; The second fully connected layer is connected to the first input terminal of the Softmax module, and the first output terminal of the Softmax module is connected to the second input terminal of the first channel-by-channel multiplication module; The third fully connected layer is connected to the second input terminal of the Softmax module, and the second output terminal of the Softmax module is connected to the second input terminal of the second channel-by-channel multiplication module; The output end of the first channel-by-channel multiplication module is connected to the first input end of the second element-by-element addition module, and the output end of the second channel-by-channel multiplication module is connected to the second input end of the second element-by-element addition module.

9. An equipment parts fault diagnosis device based on mixed noise suppression of sensor data, characterized in that: The diagnostic device applies the diagnostic method described in any one of claims 1 to 8, and the diagnostic device comprises: A signal acquisition module, used to obtain vibration signals of equipment parts under different fault types; The data preprocessing module is used to preprocess the vibration signal, obtain samples, and use the fault type corresponding to the vibration signal as the true label to construct the training set and the verification set; A training module is used to train multiple deep learning models using the training set and the validation set, using a transferable Adaboost algorithm and an adaptive cost-sensitive learning method, to obtain multiple trained deep learning models; The diagnosis module is used to use multiple trained deep learning models to perform fault diagnosis on the parts to be tested of the equipment.

10. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the equipment parts fault diagnosis method based on mixed noise suppression of sensor data as described in any one of claims 1-8.

Images