A rolling bearing fault diagnosis method with unbalanced and biased data sets

Abstract

The present invention provides a rolling bearing fault diagnosis method with unbalanced data and biased data sets. The present invention includes a generator G, a discriminator D and a cross-domain fault classifier C. The entire fault diagnosis model performs alternating adversarial training, the purpose of which is to use synthetic fault samples to rebalance the training set data, and use the rebalanced training set data and unlabeled test set samples to train the fault classifier C. The trained cross-domain fault classifier C is used to perform fault diagnosis on the test set fault samples. The present invention has a high fault diagnosis accuracy and is very suitable for rolling bearing fault diagnosis with unbalanced fault data and biased data sets.

Description

Technical Field

[0001] The present invention relates to the technical field of rolling bearing fault processing, and in particular to a rolling bearing fault diagnosis method with unbalanced data and deviation in data sets. Background Art

[0002] In industrial applications, rolling bearings are in normal condition most of the time, with a low probability of failure and their operating conditions often changing. This leads to two problems in the collected rolling bearing fault samples: (1) imbalanced fault data and (2) data set bias, which seriously reduces the accuracy of traditional intelligent fault diagnosis methods. Summary of the invention

[0003] According to the technical problems raised above, a rolling bearing fault diagnosis method with unbalanced data and data set deviation is provided. The present invention is used for rolling bearing fault diagnosis with unbalanced fault data and data set deviation. The principle is to automatically generate low-proportion fault samples without data set deviation. The low-proportion fault samples without data set deviation balance the fault data for fault model training. The technical means adopted by the present invention are as follows:

[0004] A rolling bearing fault diagnosis method with unbalanced data and biased data sets comprises the following steps:

[0005] Step 1: Collect rolling bearing vibration signals as fault data, divide the fault data into a training set and a test set, and there is a deviation between the fault samples in the training set and the fault samples in the test set;

[0006] Step 2: Establish a fault diagnosis model, which includes a rebalanced fault sample generator G for generating small deviations and a discriminator D for identifying true and false fault samples, initialize weights and thresholds, input noise and labels into the generator G, generate low-proportion fault samples without data set deviation, and input both the generated fault samples and the real fault samples into the discriminator D for judgment until the discriminator D cannot distinguish whether the sample is a real sample or a generated sample;

[0007] Step 3: Establish a cross-domain fault diagnosis classifier C, initialize the weights and thresholds, and input the rebalanced fault samples of the training set and the fault samples of the test set into the fault classifier C at the same time;

[0008] Step 4: Generator G simultaneously shrinks L g Error and fault classifier error L C , the discriminator D shrinks L D Error, fault classifier C while reducing the cross entropy loss L cls and CORAL loss L CORAL , reduce the deviation between the generated fault samples and the target domain sample data set;

[0009] Step 5: Use the trained fault classifier C to diagnose the faults in the test set.

[0010] Further, in step 4,

[0011] L G =L g +L C

[0012]

[0013]

[0014] L C =αL cls +βL CORAL

[0015] Among them, α and β are constants between 0 and 1, i represents the i-th sample, K is the number of samples in the training set, and x i is the i-th true sample, z i is the i-th noise sample, c i The label of the i-th sample.

[0016] The fault diagnosis of the present invention has high accuracy and is very suitable for rolling bearing fault diagnosis where the fault data is unbalanced and the data set has deviations. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] 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 or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0018] Figure 1 is a rolling bearing fault diagnosis model diagram for data imbalance and data set deviation in an embodiment of the present invention;

[0019] Figure 2 is a schematic diagram of the structure of the generator G in an embodiment of the present invention;

[0020] Figure 3 is a schematic diagram of the structure of the discriminator D in an embodiment of the present invention;

[0021] Figure 4 is a schematic diagram of the structure of a fault classifier C in an embodiment of the present invention;

[0022] Figure 5 is a rolling bearing fault diagnosis flow chart for data imbalance and data set deviation in an embodiment of the present invention;

[0023] Figure 6 is the fault diagnosis result diagram; DETAILED DESCRIPTION

[0024] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are 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.

[0025] like Figure 5 As shown, the embodiment of the present invention discloses a rolling bearing fault diagnosis method with data imbalance and data set deviation, including: Figure 2 A generator G is shown, such as Figure 3 A discriminator D shown and Figure 4 A cross-domain fault classifier C is shown as Figure 1 As shown, the fault diagnosis model is an end-to-end fault diagnosis model. It includes the following steps:

[0026] Step 1: Collect rolling bearing vibration signals as fault data, and divide the fault data into a training set and a test set. There is a deviation between the fault samples in the training set and the fault samples in the test set. The training set data contains labels and is unbalanced; the test set data is balanced and does not contain labels.

[0027] Specifically, the original fault data is normalized to the [0, 255] interval and the original signal is converted into a gray image.

[0028] Step 2: Establish a fault diagnosis model, which includes a rebalanced fault sample generator G for generating small deviations and a discriminator D for identifying true and false fault samples. Initialize weights and thresholds, input noise and labels into the generator G, generate low-proportion fault samples without dataset deviation, and input both the generated fault samples and the real fault samples into the discriminator D for judgment. If the termination condition is not met, continue to train the generator G until the discriminator D cannot distinguish whether the sample is a real sample or a generated sample. The loss function of the generator G is L G , the loss function of the discriminator D is L D Among them, the generator G is used to generate fault samples with the smallest deviation from the test set, and the generated fault samples are as close as possible to the distribution of the real samples in the test set; the discriminator D is used to discriminate whether the samples are from the generated samples or the real samples, and rebalance the original training fault samples.

[0029] The generated samples and real samples constitute the rebalanced training set fault samples, and the rebalanced training set fault samples and test set fault data are transmitted to the fault classifier C. The generator G and the discriminator D are trained in an adversarial manner to generate qualified rebalanced training set fault samples.

[0030] Step 3: Establish a cross-domain fault diagnosis classifier C, initialize the weights and thresholds, and input the rebalanced fault samples of the training set and the fault samples of the test set into the fault classifier C at the same time. The loss function of the fault classifier C is L C , the generator loss function contains the fault classifier error L C ,The purpose is to reduce the dataset deviation between the generated fault ,samples and the target domain samples;

[0031] Step 4: Generator G simultaneously shrinks L g Error and fault classifier error L C , the discriminator D shrinks L D Error, fault classifier C while reducing the cross entropy loss L cls and CORAL (Correlation Alignment) loss L CORAL , the fault classifier C reduces the dataset deviation between the rebalanced fault samples and the target domain samples, and the fault classifier error LC is fed back to the generator G to generate low-proportion fault data with a small dataset deviation. The fault diagnosis model of the present invention uses an end-to-end model instead of a traditional two-stage model to directly generate training set fault samples with a small dataset deviation;

[0032] Step 5: Use the trained fault classifier C to diagnose the faults in the test set.

[0033] Further, in step 4,

[0034] L G =L g +L C

[0035]

[0036]

[0037] L C =αL cls +βL CORAL

[0038] Among them, α and β are constants between 0 and 1, i represents the i-th sample, K is the number of samples in the training set, and x i is the i-th true sample, z i is the i-th noise sample, c i The label of the i-th sample.

[0039] The present invention is used to perform fault diagnosis tests on the CWRU (Case Western Reserve University) rolling bearing database. The test results are as follows: Figure 6 The experimental results verify the effectiveness of the proposed method. The fault diagnosis accuracy of the present invention is high and it is very suitable for rolling bearing fault diagnosis with unbalanced fault data and biased data sets.

[0040] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A rolling bearing fault diagnosis method with unbalanced data and biased data set, characterized in that: The steps include: Step 1: Collect rolling bearing vibration signals as fault data, divide the fault data into a training set and a test set, and there is a deviation between the fault samples in the training set and the fault samples in the test set; Step 2: Establish a fault diagnosis model, which includes a rebalanced fault sample generator G for generating small deviations and a discriminator D for identifying true and false fault samples, initialize weights and thresholds, input noise and labels into the generator G, generate low-proportion fault samples without data set deviation, and input both the generated fault samples and the real fault samples into the discriminator D for judgment until the discriminator D cannot distinguish whether the sample is a real sample or a generated sample; Step 3: Establish a cross-domain fault diagnosis classifier C, initialize the weights and thresholds, and input the rebalanced fault samples of the training set and the fault samples of the test set into the fault classifier C at the same time; Step 4: Generator G simultaneously shrinks L g Error and fault classifier error L C , the discriminator D shrinks L D Error, fault classifier C while reducing the cross entropy loss L cls and CORAL loss L CORAL , reduce the deviation between the generated fault samples and the target domain sample data set; Step 5: Perform fault diagnosis on the test set faults through the trained fault classifier C; In step 4, L G =L g +L C L C =αL cls +βL CORAL Among them, α and β are constants between 0 and 1, i represents the i-th sample, K is the number of samples in the training set, and x i is the i-th true sample, z i is the i-th noise sample, c i The label of the i-th sample.

Images