Fault frequency analysis method and label description method for nuclear power bearing / angular contact deep groove ball bearing

Abstract

The invention belongs to the technical field of bearing fault diagnosis, and particularly relates to a fault frequency analysis method and a label description method of a nuclear power bearing / angular contact deep groove ball bearing. Inspired by'envelope demodulation 'of a signal processing method, according to the method, bits far less than those of the first two are used for label description of data, and after data enhancement operation of the conditional generative adversarial network, improvement of a diagnosis task is also obtained; the method occupies little space, does not additionally consume computing resources, and can enable the conditional generative adversarial network to have the capability of generating position data when describing the data training conditional generative adversarial network; according to the scheme, the fault diagnosis requirement of the bearing in the nuclear power field can be well met, and the operation safety and reliability of mechanical equipment are guaranteed.

Description

Technical Field

[0001] The present invention belongs to the technical field of bearing fault diagnosis, and specifically relates to a method for analyzing fault frequencies and a method for describing tags of bearings / angle contact deep groove ball bearings used in nuclear power plants. Background Art

[0002] Rolling bearings are currently widely used in the nuclear power field, and their health status will directly affect the operation safety and reliability of mechanical equipment. If the minor defects are allowed to develop continuously, it will gradually lead to the performance degradation of the rolling bearings, and further cause faults in the key equipment of the nuclear power system, resulting in significant economic, environmental and other impacts, and ultimately frustrating the development of the nuclear power field. Therefore, studying the fault diagnosis method of bearings has very important practical significance.

[0003] With the rapid development of artificial intelligence technology, the field of fault diagnosis has also entered the era of intelligent diagnosis. The current artificial intelligence technology is restricted by training data, and the quality of its training data determines the upper limit of the model performance. In terms of fault diagnosis, taking the automated production line in our nuclear power field as an example, in order to collect sufficient labeled data, theoretically, it is necessary for the automated production line of the nuclear power plant to frequently fail to meet the data collection requirements for training the artificial intelligence model. In fact, any nuclear power plant has an active maintenance strategy for the automated production line. If the line still frequently fails after maintenance, then ultimately it may be scrapped or even discarded. Thus, in actual production, it is impossible to meet the sufficient labeled data required for the fault diagnosis task. The common situation is that normal fault-free data occupies the vast majority of the collected data, while fault samples only occupy a minority class, which naturally forms an imbalanced data set. When using such a data set to train an artificial intelligence algorithm, it will cause the final diagnosis result to be biased towards the majority class, that is, the normal fault-free condition.

[0004] To address the problem of data imbalance, the most classical approach currently is the Synthetic Minority Over-sampling Technique (SMOTE) and its variant methods. Based on real samples, this method synthesizes new samples through linear interpolation, alleviating the overfitting problem brought about by random sampling methods. However, the data synthesized based on the SMOTE method has low quality and is prone to reducing the overall stability of data features. Adaptive Synthetic Sampling (ADASYN) improves the quality of synthesized samples by refining the sample synthesis rules of the original SMOTE, but it is easily affected by the synthesis rules. Using Generative Adversarial Network (GAN) to generate new samples is another major category of methods currently. GAN generates fake data similar to real data, capturing the distribution of the given real data as much as possible. The generated signal can have the same distribution as the original signal, alleviating the problem of insufficient data to a certain extent. Currently, GAN has become the mainstream method for solving the problem of insufficient data in the field of fault diagnosis. Among them, the conditional Generative Adversarial Network (cGAN) has been widely applied because it can specify the generated data category by leveraging the additional information provided by labels.

[0005] Currently, the commonly used label description methods in artificial intelligence technologies are the One-Hot encoding method and the Embedding encoding. Among them, the One-Hot encoding converts label information into a vector, with only one element being 1 and the rest being 0. Each element corresponds to a category, indicating the category to which the sample belongs. This method is simple and direct and is suitable for multi-classification tasks. The Embedding representation maps labels to a low-dimensional real-valued vector, usually implemented using an Embedding Layer, thus providing more representational flexibility. This method provides a flexible label representation, can better capture the relationships between labels and convey information, and is generally used in natural language processing (NLP) tasks.

[0006] The above two label description methods are widely used, but they both have the following characteristics: 1. Consume additional computing resources: The length of the One-Hot encoding depends on the number of categories in the classification task. When the number of categories is large, it will occupy a large amount of data capacity. That is, assuming a 100-classification task, an additional 100 dimensions need to be added to the input dimension when training an artificial intelligence model. Taking a deep neural network as an example, this will add a huge task volume to the parameters of the entire subsequent model. Similarly, the Embedding method itself uses a neural network for dimensionality reduction of label information. To better extract features, a more complex network structure is required, which will increase the computing resources consumed by the model.

[0007] 2. Poor malleability: One-Hot depends on the number of categories in the classification task. When the number of categories is large, it will occupy a large number of data bits. Similar to the Embedding method, when modifying the number of classification categories, the network structure needs to be reset and the model needs to be retrained, so the scalability is poor. Summary of the Invention

[0008] Aiming at the above problems existing in the prior art, the purpose of the present invention is to provide a fault frequency analysis method and a label description method for bearings / angular contact deep groove ball bearings used in nuclear power, which can represent multi-category data with fewer bits, reduce resource consumption at the same time, and have stronger generalization ability.

[0009] In order to solve the above problems, the technical solutions adopted by the present invention are as follows: A fault feature frequency analysis method for fault diagnosis of bearings used in nuclear power. This method first obtains the envelope signal of the original signal. Assume that the collected original signal is , and after Hilbert transform, the signal obtained is: ; Furthermore, the analytic signal is obtained: ; The envelope signal of the analytic signal is obtained by calculating the root mean square of the real part and the imaginary part of the analytic signal: ; Thus, the low-speed information of the bearing is separated by envelope demodulation, and the fault feature frequency analysis at a specific position is carried out.

[0010] A fault feature frequency analysis method for angular contact deep groove ball bearings used in nuclear power, adopting the aforementioned fault feature frequency analysis method applied to bearing fault diagnosis, This method further includes the following steps: Express the pitch diameter of the angular contact deep groove ball bearing as: ; The inner and outer raceway diameters of the angular contact deep groove ball bearing are expressed by the pitch diameter, diameter and contact angle: ; ; The rotational speed of the angular contact deep groove ball bearing is expressed by the radius and rotational angular frequency. The rotational frequency of the inner ring is , and the rotational speed is expressed as: ; The rotational frequency of the outer diameter is , the rotational speed of the outer ring is expressed as: ; The rotational speed of the cage In the case of no-slip, it is the average speed of the inner ring and the outer ring speed, expressed as: ; Furthermore, the theoretical rotational frequency FTF of the cage is: ; When the outer ring is fixed and the inner ring rotates with the main shaft, the rotational frequency of the cage is expressed as: ; At this time, the rotational frequency of the cage relative to the inner ring is: ; Assume the number of balls in the angular contact deep groove ball bearing is n, and the frequency of the balls passing through the inner ring is expressed as: ; At this time, the outer ring is fixed, so , and the frequency of the balls passing through the inner ring at this time is: ; where the negative sign indicates that the calculated frequency component is opposite to the preset rotational direction; The rotational frequency of the cage passing through the outer ring is expressed as: ; When the outer ring is fixed, the frequency of the balls in the angular contact deep groove ball bearing passing through the outer diameter is: ; The rotational frequency of the rolling elements The ratio to the frequency of the cage relative to the inner ring is: ; In the case of the outer ring being fixed, it can be obtained that: ; If the cage fails, the vibration signal will contain the cage rotational frequency ; When the inner ring raceway fails, the vibration signal will contain the inner ring fault frequency ; When the outer ring raceway fails, the vibration signal will contain the outer ring fault frequency ; If there is a fault in the rolling element itself, its fault frequency is twice the basic rolling element frequency .

[0011] A data augmentation label description method for bearing fault diagnosis in nuclear power. The aforementioned fault feature frequency analysis method applied to angular contact deep groove ball bearings. In this method, five - digit coding is used for label description. The first digit represents load information, the second digit represents fault size, and the third, fourth, and fifth digits respectively represent envelope information, that is, the bearing characteristic frequency , , ; Among them, the load is represented by horsepower, the fault diameter is represented by the information in the data set, and the third, fourth, and fifth digits are represented by one - hot coding; The method also includes the following steps: S1: Determine whether the data comes from a rotating machine with a cage. If not, directly end; if so, proceed to the next step; S2: Determine whether the fault type is a rolling element fault. If not, directly end; if so, proceed to the next step; S3: Determine whether the data comes from a fault - free condition. If so, the label is "load, fault size, 1, 0, 0" at this time; if not, proceed to the next step; S4: Determine whether the fault type is an inner ring fault of the bearing. If so, the label is "load, fault size, 0, 1, 0" at this time; if not, the label is "load, fault size, 0, 0, 1".

[0012] Compared with the prior art, the beneficial effects of the present invention are: (1) The method proposed by the present invention can use fewer digits to represent more categories of data, reducing the consumption of computing resources; (2) The conditional generative adversarial network trained by the present invention has the ability to generate unknown data, effectively improving the generalization ability; (3) The present invention can better meet the fault diagnosis requirements of bearings in the field of nuclear power, ensuring the safety and reliability of mechanical equipment operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 Schematic diagram of amplitude - modulated signal and spectral frequency change in the prior art; Figure 2 Schematic diagram of the envelope signal obtained by solving the modulated signal using the method of the present invention; Figure 3 Schematic diagram of the encoding representation of the label description method of the present invention; Figure 4 Example diagrams of four different cases of encoding representation using the label description method of the present invention in this embodiment; Figure 5Schematic diagram of the steps of the method according to the present invention. Detailed implementation manners

[0014] The present invention will be further described below in conjunction with specific embodiments.

[0015] Inspired by the envelope modulation method, the present invention integrates envelope information, load, and fault size uniformly as the tag information of the rolling bearing data with a cage, which brings convenience to the cGAN training.

[0016] As Figure 1 shown, when the rotating equipment is working, low-frequency vibration signals will be generated. These vibration signals contain rich information for diagnosing the health status of the bearing. However, when the rotating equipment is working, resonance will occur, exciting the natural frequency of the equipment. These frequency components are generally much higher than the frequency band of the vibration signals, which will cause a modulation phenomenon. The seven high-frequency components act as carrier signals, and the low-frequency components act as modulation signals. Fourier transform of the modulated carrier signal can show that the low-frequency components appear as sidebands of the high-frequency components in the spectrum, which is not conducive to observing and judging the health status of the bearing.

[0017] Therefore, a fault feature frequency analysis method for bearing fault diagnosis in nuclear power plants according to the present invention first obtains the envelope signal of the original signal. Assuming that the collected original signal is , after Hilbert transform, the signal obtained is: ; Furthermore, the analytic signal is obtained: ; The envelope signal of the analytic signal is obtained by calculating the root mean square of the real part and the imaginary part of the analytic signal: ; As Figure 2 shown, in this embodiment, the schematic diagram of obtaining the envelope signal by solving the modulation signal. It can be seen from the figure that the blue solid line represents the original amplitude modulation signal, and the orange dashed line represents the Hilbert signal. Their time-domain curves are the same, but it should be noted that their frequency-domain components are not the same. The envelope curve of the analytic signal is calculated to obtain the green curve. Combining the previous analysis, the green curve is both the envelope of the analytic signal and the modulation signal in the original amplitude modulation signal.

[0018] It can be seen from this that the envelope modulation method using Hilbert transform can analyze the modulation signal with slower frequency components in the amplitude modulation signal. When analyzing the faults of rolling bearings, the envelope modulation means can be used to separate the low-speed information of the bearings and perform fault feature frequency analysis at specific positions.

[0019] Furthermore, we perform fault frequency analysis by means of the structural parameters of the bearing itself. A method for analyzing the fault characteristic frequency of an angular contact deep groove ball bearing according to the present invention adopts the aforementioned method for analyzing the fault characteristic frequency applied to bearing fault diagnosis. This method further includes the following steps: For a ball bearing with a contact angle, its pitch diameter is expressed as: ; The inner and outer raceway diameters of the angular contact deep groove ball bearing are expressed by the pitch diameter, diameter, and contact angle: ; ; The rotational speed of the angular contact deep groove ball bearing is expressed by the radius and rotational angular frequency. The rotational frequency of the inner ring is , and the rotational speed of the inner ring is expressed as: ; The rotational frequency of the outer diameter is , and the rotational speed of the outer ring is expressed as: ; The rotational speed of the cage is the average speed of the inner and outer ring speeds in the case of no slip. The cage is a component in the bearing that plays a role in fixing the position of the rolling elements and guiding. The average value of its rotational speed is of great significance during the operation of the bearing. When the bearing is operating, the inner and outer rings may rotate at different speeds. However, due to the presence of the cage, the rolling elements will be guided by the cage, making the relative movement between the inner and outer rings relatively stable. Therefore, the rotational speed of the cage is usually defined as the average value of the inner and outer ring speeds to reflect the average speed received by the rolling elements. In this case, it is expressed as: ; Furthermore, the theoretical rotational frequency FTF (Fundamental Train Frequency) of the cage is obtained as: ; In the case where the outer ring is fixed and the inner ring rotates with the main shaft, the rotational frequency of the cage is expressed as: ; At this time, the rotational frequency of the cage relative to the inner ring is: ; Let the number of balls in the angular contact deep groove ball bearing be \(n\). Since the balls are fixed by the cage during the operation of the bearing, when the cage rotates one week, the \(n\) balls also rotate one week. Therefore, the frequency of the balls passing through the inner ring (Ball Pass Frequency Inner Race) is expressed as: ; At this time, the outer ring is fixed. Therefore , and the frequency of the balls passing through the inner ring at this time is: ; The negative sign indicates that the calculated frequency component is opposite to the preset rotation direction; The rotational frequency of the cage passing through the outer ring is expressed as: ; When the outer ring is fixed, the frequency of the balls passing through the outer diameter in the angular contact deep groove ball bearing is: ; Assume that there is no slip of the rolling elements. Since the rolling elements rotate about their own axes and the cage rotates about the center of the bearing, the spin frequency of the rolling elements can be obtained according to the principle that the linear velocities of their contact points are equal. On the other hand, there is relative motion between the cage and the inner ring. Therefore, the frequency used at this time should be the frequency of the cage passing through the inner ring , so the spin frequency of the rolling elements (Ball Spin Frequency) and the ratio of the frequency of the cage relative to the inner ring is: ; In the case of the outer ring being fixed, we can get: ; In these frequency formulas, it is assumed that there is no slip. If the rotation of both the inner and outer rings is considered, the equation is the general form of the bearing characteristic frequency. In reality, slip always exists, and these expected theoretical frequencies are always adjusted in an appropriate way. In many cases, the outer ring is fixed, and at this time the outer ring is stationary, and the corresponding frequency formula will be simplified.

[0020] In the fault diagnosis of bearings, analyzing each of the above frequency components helps to determine possible faults. If the cage fails, the vibration signal will contain the cage rotation frequency ; When a fault occurs in the inner ring raceway, such as spalling, indentation or imbalance, the vibration signal will contain the inner ring fault frequency ; When a fault occurs in the outer ring raceway, the vibration signal will contain the outer ring fault frequency ; If there is a fault in the rolling element itself, its fault frequency is twice the basic rolling element frequency .

[0021] Since the envelope information has no intuitive effect on the ball fault, the present invention is only applicable to three cases: inner ring fault of the bearing, outer ring fault of the bearing, and no fault of the bearing.

[0022] Among the above parameters, represents the rotational angular frequency of the outer ring; represents the rotational angular frequency of the inner ring; represents the rotational speed of the outer ring; represents the rotational speed of the inner ring; represents the rotational speed of the ball; is the ball diameter; is the outer raceway diameter; is the inner raceway diameter; is the ball contact angle; is the bearing pitch diameter.

[0023] In the data enhancement label description method for bearing fault diagnosis in nuclear power plants described in the present invention, the aforementioned fault characteristic frequency analysis method applied to angular contact deep groove ball bearings, as Figure 3 shown, in this method, five - digit coding is used for label description. The first digit is the load information, the second digit is the fault size, and the third, fourth, and fifth digits of the coding respectively represent the envelope information, that is, the bearing characteristic frequencies , , ; Among them, as Figure 4 shown, the load is represented by horsepower, the fault diameter is represented by the information in the data set, and the third, fourth, and fifth digits of the coding are represented by one - hot coding; As Figure 5 shown, the method further includes the following steps: S1: Determine whether the data comes from a rotating machine with a cage. If not, directly end; if so, proceed to the next step; S2: Determine whether the fault type is a rolling element fault. If not, directly end; if so, proceed to the next step; S3: Determine whether the data comes from a fault - free working condition. If so, the label is "load, fault size, 1, 0, 0" at this time; if not, proceed to the next step; S4: Determine whether the fault type is an inner ring fault of the bearing. If so, the label is "load, fault size, 0, 1, 0" at this time; if not, the label is "load, fault size, 0, 0, 1" at this time.

[0024] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) The method proposed by the present invention can use fewer bits to represent more categories of data, reducing the consumption of computing resources; (2) The conditional generative adversarial network trained by the present invention has the ability to generate unknown data, effectively improving the generalization ability; (3) The present invention can preferably meet the fault diagnosis requirements of bearings in the field of nuclear power, ensuring the safety and reliability of the operation of mechanical equipment.

Claims

1. A fault characteristic frequency analysis method for fault diagnosis of nuclear power bearings, characterized in that: This method first obtains the envelope signal of the original signal. Assuming that the collected original signal is , the signal is transformed by Hilbert: ； Further analysis signal is obtained: ； The envelope signal of the analytical signal is obtained by calculating the root mean square formula for the real and imaginary parts of the analytical signal: ； Therefore, the low-speed information of the bearing is separated through envelope mediation, so as to perform fault characteristic frequency analysis at a specific location.

2. A method for analyzing the fault characteristic frequency of angular contact deep groove ball bearings for nuclear power, characterized in that: Using the fault characteristic frequency analysis method for bearing fault diagnosis as described in claim 1, The method further comprises the following steps: The pitch diameter of the angular contact deep groove ball bearing is expressed as: ； The inner and outer raceway diameters of the angular contact deep groove ball bearing are represented by the pitch diameter, diameter and contact angle: ； ； The rotation speed of the angular contact deep groove ball bearing is represented by the radius and the rotation angular frequency. The rotation frequency of the inner ring is , the speed of the inner ring It is expressed as: ； The rotation frequency of the outer diameter is , the outer ring speed It is expressed as: ； Cage rotation speed In the absence of sliding, it is the average speed of the inner and outer rings, expressed as: ； The theoretical rotation frequency FTF of the cage is further obtained as: ； When the outer ring is fixed and the inner ring rotates with the main shaft, the rotation frequency of the cage is expressed as: ； At this time, the rotation frequency of the cage relative to the inner ring is: ； Assuming the number of balls in the angular contact deep groove ball bearing is n, the frequency of the balls passing through the inner ring is expressed as: ； At this time, the outer ring is fixed, so , at this time the frequency of the ball passing through the inner ring is: ； The negative sign indicates that the calculated frequency component is opposite to the preset rotation direction; The frequency of rotation of the cage through the outer ring is expressed as: ； When the outer ring is fixed, the frequency of the ball passing through the outer diameter in the angular contact deep groove ball bearing is: ； Rolling element rotation frequency The ratio of the frequency of the cage relative to the inner ring is: ； When the outer ring is fixed, we can get: ； If the cage is faulty, the vibration signal will contain the cage rotation frequency ; When the inner ring raceway fails, the vibration signal will contain the inner ring fault frequency ; When the outer ring raceway fails, the vibration signal will include the outer ring fault frequency ; If the rolling element itself is faulty, the fault frequency is twice the basic rolling element frequency. .

3. A data enhancement label description method for bearing fault diagnosis for nuclear power, characterized in that: The fault characteristic frequency analysis method for angular contact deep groove ball bearings as described in claim 2 is adopted. In this method, a five-digit code is used for label description, the first digit is the load information, the second digit is the fault size, and the third, fourth, and fifth digits represent the envelope information, that is, the bearing characteristic frequency , , ; The load is represented by horsepower, the fault diameter is represented by the information in the data set, and the third, fourth, and fifth digits are represented by one-hot encoding; The method further comprises the following steps: S1: Determine whether the data is from a rotating machine with a cage. If not, end directly; if yes, proceed to the next step; S2: Determine whether the fault type is a rolling element fault. If not, end the process directly; if yes, proceed to the next step. S3: Determine whether the data is from a fault-free condition. If so, the label is "load, fault size, 1, 0, 0"; if not, proceed to the next step; S4: Determine whether the fault type is a bearing inner ring fault. If so, the label is "load, fault size, 0, 1, 0"; if not, the label is "load, fault size, 0, 0, 1".

Images