Data processing method, device and equipment for accessory gearbox bearing and medium
By performing Fourier transform and unique thermal coding label processing on the bearing data of the gas turbine accessory gearbox, a fault identification model was established, solving the problem of complex fault diagnosis and realizing accurate identification and diagnosis of faults in the gas turbine accessory gearbox.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CHINA NAT PETROLEUM CORP
- Filing Date
- 2022-12-05
- Publication Date
- 2026-06-05
AI Technical Summary
Existing technologies are insufficient for accurately diagnosing complex faults in gas turbine accessory gearboxes, and traditional signal analysis methods cannot effectively handle multi-component modulated unstable signals, leading to difficulties in fault diagnosis.
By acquiring the dataset of the gearbox bearings in the accessory gearbox, performing Fourier transform processing, adding one-hot encoded labels, randomly shuffling the sample data and dividing it into training and validation samples, a fault identification model is established for diagnosis.
It enables accurate identification and diagnosis of accessory gearbox faults, especially the effective identification of complex faults, thus improving the accuracy and efficiency of fault diagnosis.
Smart Images
Figure CN118152795B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of machine learning technology, and in particular to a data processing method, apparatus, device, and medium for an accessory gearbox bearing. Background Technology
[0002] The accessory gearbox is a crucial mechanical component in a gas turbine, playing a vital role in variable speed and torque transmission. Gears and rolling bearings are the most important parts of the gearbox. In practical engineering, gear and bearing failures in the gearbox often occur frequently, sometimes even in the form of compound failures. Compound failures typically refer to the simultaneous presence of two or more faults in mechanical equipment. In complex systems, the occurrence of faults is random and concurrent, and as the system scales up, the relationships between detected faults become increasingly complex.
[0003] The gearbox has a complex working structure, and when a fault occurs, it may be a single fault or a compound fault caused by a single fault. Its signal is generally a multi-component modulated unstable signal coupled with different fault characteristics. Traditional signal analysis and processing methods often cannot fully achieve the diagnostic goal in fault diagnosis.
[0004] Therefore, how to diagnose the cause of a failure in the accessory gearbox in a timely manner is currently an unsolved problem. Summary of the Invention
[0005] This application provides a data processing method, apparatus, equipment, and medium for accessory gearbox bearings to solve the problem of difficulty in accurately diagnosing the cause of failure when an accessory gearbox fails.
[0006] In a first aspect, embodiments of this application provide a data processing method for an accessory gearbox bearing, including:
[0007] Obtain the dataset of the accessory gearbox bearings, wherein the dataset includes a bearing dataset or a gearbox composite fault dataset;
[0008] Based on the number of fault types, the dataset is divided into multiple data sequences of a preset length; each data sequence contains data of different fault types.
[0009] The multiple data sequences are subjected to Fourier transform processing to obtain multiple frequency domain data.
[0010] For each piece of frequency domain data, a tag represented by one-hot encoding is added to the frequency domain data according to the fault type corresponding to the frequency domain data to obtain the corresponding sample data;
[0011] The sample data corresponding to all fault types are shuffled, and the shuffled sample data is divided into training sample data and validation sample data according to a preset ratio.
[0012] The model is trained using the training sample data, and the trained model is validated using the validation sample data to obtain a fault identification model.
[0013] In conjunction with the first aspect, in some embodiments, the dataset is a bearing dataset, then obtaining the dataset of the accessory gearbox bearing includes:
[0014] Fault experiments were conducted on the accessory gearbox to obtain the bearing dataset, which includes: normal baseline data, 12kHz drive-end fault data, 48kHz drive-end fault data, and 12kHz fan-end fault data.
[0015] In conjunction with the first aspect, in some embodiments, the method further includes:
[0016] Based on the operating data of the accessory gearbox bearing and the fault identification model, the faults of the accessory gearbox bearing are diagnosed.
[0017] In conjunction with the first aspect, in some embodiments, the dataset is a gearbox composite fault dataset, then obtaining the dataset of the accessory gearbox bearing includes:
[0018] A composite fault experiment was conducted on the accessory gearbox to obtain the gearbox composite fault dataset. The gearbox composite fault dataset includes: normal bearing data, bearing rolling element defect data, bearing inner ring defect data, bearing outer ring defect data, gear tooth breakage data, gear tooth breakage and bearing outer ring defect data, gear tooth breakage and bearing inner ring defect data, gear missing tooth defect data, gear missing tooth and bearing outer ring defect data, and gear missing tooth and bearing inner ring defect data.
[0019] In conjunction with the first aspect, in some embodiments, the method further includes:
[0020] Based on the operating data of the accessory gearbox and the fault identification model, the faults of the accessory gearbox are diagnosed.
[0021] In conjunction with the first aspect, in some embodiments, dividing the shuffled sample data into training sample data and validation sample data according to a preset ratio includes:
[0022] 90% of the shuffled sample data is determined as the training sample data, and the other 10% is determined as the validation sample data, with the preset ratio being 9:1.
[0023] In conjunction with the first aspect, in some embodiments, the heat labels added to the frequency domain data corresponding to each fault type are different.
[0024] Secondly, embodiments of this application provide a data processing device for an accessory gearbox bearing, comprising:
[0025] The receiving module is used to acquire the dataset of the accessory gearbox bearing, the dataset including the bearing dataset or the gearbox composite fault dataset;
[0026] The segmentation module is used to segment the dataset into multiple data sequences of a preset length based on the number of fault types; each data sequence contains data of different fault types.
[0027] The transformation module is used to perform Fourier transform processing on the multiple data sequences to obtain multiple frequency domain data.
[0028] The tagging module is used to add a tag represented by one-hot encoding to each piece of frequency domain data according to the fault type corresponding to the frequency domain data, so as to obtain the corresponding sample data.
[0029] The sample module is used to shuffle the sample data corresponding to all fault types and divide the shuffled sample data into training sample data and validation sample data according to a preset ratio.
[0030] The training module is used to train the model based on the training sample data and to verify the trained model using the verification sample data to obtain a fault identification model.
[0031] Thirdly, embodiments of this application also provide an electronic device, including:
[0032] The processor, the memory communicatively connected to the processor, and the communication interface for interacting with other devices;
[0033] The memory stores computer-executed instructions;
[0034] The processor executes computer execution instructions stored in the memory to implement the data processing method for the accessory gearbox bearing.
[0035] Fourthly, embodiments of this application also provide a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the data processing method for the accessory gearbox bearing as described in the first aspect.
[0036] This application provides a data processing method, apparatus, device, and medium for accessory gearbox bearings. It involves collecting a dataset of accessory gearbox bearing data, converting the data into frequency domain data, adding heat labels to the converted frequency domain data, shuffling the resulting samples, using the shuffled sample data for model training, and validating the model. Once validated, the model can be used for fault diagnosis of accessory gearboxes. This solution enables accurate diagnosis of both mechanical and complex faults in accessory gearboxes. Attached Figure Description
[0037] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.
[0038] Figure 1 An application scenario diagram of the data processing method for gearbox bearings provided in the appendix of this application;
[0039] Figure 2 A flowchart of an embodiment of a data processing method for an accessory gearbox bearing provided in this application;
[0040] Figure 3 A flowchart of Embodiment 2 of a data processing method for an accessory gearbox bearing provided in this application;
[0041] Figure 4 A flowchart of Embodiment 3 of a data processing method for an accessory gearbox bearing provided in this application;
[0042] Figure 5 This application provides a waveform diagram of normal bearing data in the embodiments;
[0043] Figure 6 This application provides a waveform diagram of a faulty rolling element at the drive end in an embodiment.
[0044] Figure 7 This application provides a waveform diagram of a fault in the inner ring of the drive end in an embodiment;
[0045] Figure 8 This application provides a waveform diagram of a fault on the outer ring of the drive end in an embodiment;
[0046] Figure 9 This application provides a waveform diagram of a fan-end rolling element failure in an embodiment;
[0047] Figure 10 This application provides a waveform diagram of a fault in the inner ring of the fan end in an embodiment;
[0048] Figure 11 This application provides a waveform diagram of a fault on the outer ring of the fan end in an embodiment;
[0049] Figure 12 The present application provides a normal composite data waveform diagram in the embodiments;
[0050] Figure 13 This application provides waveform diagrams of bearing rolling element defects in its embodiments;
[0051] Figure 14 This application provides waveform diagrams of gear tooth breakage defects in its embodiments.
[0052] Figure 15 This application provides waveform diagrams of gear tooth breakage and bearing outer ring defects in the embodiments.
[0053] Figure 16 This application provides a waveform diagram of gear tooth breakage and bearing inner ring defect data in the embodiments;
[0054] Figure 17 This application provides a waveform diagram of gear tooth missing defect data in the embodiments;
[0055] Figure 18 This application provides a waveform diagram of gear tooth loss and bearing outer ring defect data in the embodiments;
[0056] Figure 19 This application provides a waveform diagram of gear tooth loss and bearing inner ring defect data in the embodiments;
[0057] Figure 20 This application provides a loss / accuracy graph for the predicted fault classification in the embodiments;
[0058] Figure 21 The confusion matrix diagram in the embodiments of this application is provided;
[0059] Figure 22 A schematic diagram of the structure of the data processing device for the gearbox bearing provided in this application, as shown in Embodiment 1.
[0060] Figure 23 A schematic diagram of the electronic device structure provided in this application.
[0061] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation
[0062] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0063] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented, for example, in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.
[0064] The accessory gearbox is a crucial mechanical component in a gas turbine, playing a vital role in variable speed and torque transmission. Gears and rolling bearings are the most important parts of the gearbox. In practical engineering, gear and bearing failures in the gearbox often occur, sometimes even in the form of compound failures. Compound failures typically refer to the simultaneous presence of two or more faults in mechanical equipment. Different faults may occur simultaneously, or a failure in one component, if not addressed promptly, may trigger failures in other components. In complex systems, the occurrence of faults is random and concurrent, and as the system scales up, the relationships between multiple detected faults become increasingly complex.
[0065] The gearbox has a complex working structure, and when a fault occurs, it may be a single fault or a compound fault caused by a single fault. Its signal is generally a multi-component modulated unstable signal coupled with different fault characteristics. Traditional signal analysis and processing methods often cannot fully achieve the diagnostic goal in fault diagnosis.
[0066] To address the aforementioned issues, this application provides a data processing method for accessory gearbox bearings. The method processes collected fault data, converts it into machine-readable data, and trains the data of each fault using machine learning to establish a model that can accurately identify accessory gearbox faults for fault diagnosis.
[0067] Figure 1An application scenario diagram of the data processing method for gearbox bearings provided in the appendix of this application. (Example:) Figure 1 As shown, the technical solution of this application can be applied to the fault detection of the accessory gearbox, and can detect bearing faults, bearing fan end faults, bearing drive end faults, gear faults, and compound faults, etc.
[0068] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.
[0069] Figure 2 This is a flowchart of an embodiment of a data processing method for an accessory gearbox bearing provided in this application. Figure 2 As shown, the specific steps include:
[0070] S101. Obtain the dataset of the attached gearbox bearings. The dataset may include the bearing dataset or the gearbox composite fault dataset.
[0071] In this step, in practical applications, accessory gearboxes often malfunction, requiring timely diagnosis and analysis of the causes of failure. This solution uses a fault identification model to predict the causes of failure. However, establishing a fault identification model requires a large amount of data for training, so it is necessary to obtain the accessory gearbox dataset for model training.
[0072] The accessory gearbox mainly consists of bearings and gears, and the failure frequencies of these two parts differ. Therefore, the dataset collected for model training includes a bearing failure dataset and a combined gearbox failure dataset. The required failure data was obtained through laboratory experiments, specifically by acquiring the bearing dataset and the combined gearbox dataset through experiments.
[0073] S102. Based on the number of fault types, the dataset is divided into multiple data sequences of a preset length; the data in each data sequence is data of different fault types.
[0074] In this step, the gearbox bearing dataset obtained in the experiment is used for fault model training. Before training, the dataset needs to be processed by cutting it into data sequences of appropriate length for model training. The gearbox dataset includes various fault data. The dataset is cut according to the amount of data in each fault dataset. One fault dataset for each fault type is cut into multiple data sequences of preset length, and each fault data sequence contains fault features.
[0075] Specifically, the dataset can be cut into the required length according to the actual situation, for example, data sequences of length 500, 1000, 1500, or 2000. In this step, the time series data of the bearing dataset can be cut into a sequence of length 500; the time series data of the auxiliary gearbox composite fault dataset can be cut into a sequence of length 1000. This application embodiment does not limit this length.
[0076] S103. Perform Fourier transform on multiple data sequences to obtain multiple frequency domain data.
[0077] In this step, to facilitate the extraction of data features for model training, the segmented time-series data is processed by Fast Fourier Transform (FFT) to transform the time-series data into frequency domain data for model training.
[0078] S104. For each piece of frequency domain data, according to the fault type corresponding to the frequency domain data, add a tag represented by one-hot encoding to the frequency domain data to obtain the corresponding sample data.
[0079] In this step, machine learning requires representing the label values of data samples. One-hot encoding is one way to represent sample label values; it can be used when sample labels are discrete values. To enable machine learning training, the transformed frequency domain data needs to be labeled with tags to identify different fault types. In this scheme, labels represented by heat encoding are added to the frequency domain data for different fault types, resulting in labeled sample data for model training.
[0080] Optionally, the unique hot tags added to the frequency domain data corresponding to each fault type are different.
[0081] In one specific implementation, a one-hot coded label is added to the bearing fault data. For example, the normal data label is 0, the drive end rolling element fault label is 1, the drive end inner ring fault label is 2, the drive end outer ring fault label is 3, the fan end rolling element fault label is 4, the fan end inner ring fault label label is 5, and the fan end outer ring fault label is 6. Correspondingly, one-hot labels are generated. In the generated one-hot labels, the normal data one-hot label is [1, 0, 0, 0, 0, 0, 0].
[0082] S105. Shuffle the sample data corresponding to all fault types, and divide the shuffled sample data into training sample data and validation sample data according to a preset ratio.
[0083] In this step, to make the model training results more accurate, the labeled fault sample data is shuffled. Furthermore, to verify the accuracy of the fault model, a predetermined proportion of the fault sample data is retained for model validation, while the remaining fault sample data is used as training data for model training.
[0084] Optionally, the preset ratio of training sample data to validation sample data can be 9:1, 8:2, 7:3, or 6:4, depending on the specific circumstances. In one specific implementation, 90% of the shuffled sample data is determined as training sample data, and the other 10% is determined as validation sample data, with a preset ratio of 9:1. This ratio is not limited in the embodiments of this application.
[0085] S106. Train the model based on the training sample data, and validate the trained model using validation sample data to obtain the fault identification model.
[0086] In this step, the training sample data used for model training is used to train the model. The trained model is then validated using reserved validation sample data to verify its accuracy, ultimately yielding a fault identification model. This model is used for fault diagnosis of accessory gearboxes in practical applications.
[0087] The data processing method for accessory gearbox bearings provided in this embodiment involves collecting a dataset of accessory gearbox bearings, segmenting and Fourier transforming the fault data in the dataset to convert it into frequency domain data, adding labels to the frequency domain data according to different faults, using a portion of the labeled sample data for model training, validating the obtained model, and finally obtaining the fault identification model. The fault identification model obtained through this scheme can accurately identify faults occurring in the accessory gearbox.
[0088] The following section uses specific accessory gearbox data as an example to illustrate the specific implementation of the data processing method for accessory gearbox bearings through several embodiments.
[0089] Figure 3 A flowchart of a second embodiment of a data processing method for an accessory gearbox bearing provided in this application is shown below. Figure 3 As shown, the specific steps include:
[0090] S201. Conduct a fault test on the accessory gearbox to obtain a bearing dataset, which includes: normal baseline data, 12kHz drive-end fault data, 48kHz drive-end fault data, and 12kHz fan-end fault data.
[0091] In this step, to predict bearing failures in the accessory gearbox, it is necessary to obtain a known accessory gearbox bearing dataset for model training. Therefore, a failure experiment is conducted on the accessory gearbox. Time-series data is collected in the accessory gearbox using sensors at different frequencies. The collected bearing dataset includes: normal baseline data, 12kHz drive-end failure data, 48kHz drive-end failure data, and 12kHz fan-end failure data. Taking the 12kHz drive-end failure data as an example, the time-series data from the 12kHz drive-end sensor is divided into seven categories: normal data, fan-end inner race failure data, fan-end outer race failure data, fan-end rolling element failure data, drive-end inner race failure data, drive-end outer race failure data, and drive-end rolling element failure data.
[0092] S202. Based on the number of fault types, the dataset is divided into multiple data sequences of a preset length; the data in each data sequence is data of different fault types.
[0093] S203. Perform Fourier transform on multiple data sequences to obtain multiple frequency domain data.
[0094] S204. For each piece of frequency domain data, according to the fault type corresponding to the frequency domain data, add a tag represented by one-hot encoding to the frequency domain data to obtain the corresponding sample data.
[0095] Optionally, the unique hot tags added to the frequency domain data corresponding to each fault type are different.
[0096] S205. Shuffle the sample data corresponding to all fault types, and divide the shuffled sample data into training sample data and validation sample data according to a preset ratio.
[0097] Optionally, the preset ratio of training sample data to validation sample data can be 9:1, 8:2, 7:3, or 6:4, depending on the specific circumstances. In one specific implementation, 90% of the shuffled sample data is determined as training sample data, and the other 10% is determined as validation sample data, with a preset ratio of 9:1. This ratio is not limited in the embodiments of this application.
[0098] S206. Train the model based on the training sample data, and validate the trained model using validation sample data to obtain the fault identification model.
[0099] Optionally, faults in the accessory gearbox bearings can be diagnosed based on the operating data and fault identification model of the accessory gearbox bearings.
[0100] In one specific implementation, the resulting fault training model can be used in practical applications. Through this fault identification model, faults occurring in the accessory gearbox can be diagnosed based on the operating data of the accessory gearbox bearings, which can quickly help analyze the causes of the faults.
[0101] Steps S202-S206 are implemented in the same way as S102-S106 in Embodiment 1. The implementation process in the aforementioned embodiments can be referred to, and will not be repeated here.
[0102] The data processing method for accessory gearbox bearings provided in this embodiment collects time-series data of the gearbox bearings in the accessory gearbox using sensors of different frequencies. The collected normal data, drive-end fault data, and fan-end fault data are segmented and transformed, and labels are added for model training to obtain the final fault identification model. The fault identification model obtained by this method can identify faults in the fan-end and drive-end data of the accessory gearbox bearings.
[0103] Figure 4 A flowchart of Embodiment 3 of a data processing method for an accessory gearbox bearing provided in this application is shown below. Figure 4 As shown, the specific steps include:
[0104] S301. Conduct a composite fault experiment on the accessory gearbox to obtain a gearbox composite fault dataset. The gearbox composite fault dataset includes: normal bearing data, bearing rolling element defect data, bearing inner ring defect data, bearing outer ring defect data, gear tooth breakage data, gear tooth breakage and bearing outer ring defect data, gear tooth breakage and bearing inner ring defect data, gear missing tooth defect data, gear missing tooth and bearing outer ring defect data, and gear missing tooth and bearing inner ring defect data.
[0105] In this step, to predict the compound faults occurring in the accessory gearbox, it is necessary to obtain a known dataset of compound faults in the accessory gearbox for model training. Therefore, fault experiments are conducted on the accessory gearbox. By using a combination of normal bearings and gears with faulty bearings and gears, compound fault data are collected at a frequency of 16384Hz. The resulting gearbox compound fault dataset includes the following ten categories: normal bearing data, bearing rolling element defect data, bearing inner ring defect data, bearing outer ring defect data, gear tooth breakage data, gear tooth breakage and bearing outer ring defect data, gear tooth breakage and bearing inner ring defect data, gear missing tooth defect data, gear missing tooth and bearing outer ring defect data, and gear missing tooth and bearing inner ring defect data.
[0106] In one specific implementation, several types of data from the gearbox composite fault dataset can be selected and combined for model training.
[0107] S302. Based on the number of fault types, the dataset is divided into multiple data sequences of preset length; the data in each data sequence are data of different fault types.
[0108] S303. Perform Fourier transform on multiple data sequences to obtain multiple frequency domain data.
[0109] S304. For each piece of frequency domain data, according to the fault type corresponding to the frequency domain data, add a tag represented by one-hot encoding to the frequency domain data to obtain the corresponding sample data.
[0110] Optionally, the unique hot tags added to the frequency domain data corresponding to each fault type are different.
[0111] S305. Shuffle the sample data corresponding to all fault types, and divide the shuffled sample data into training sample data and validation sample data according to a preset ratio.
[0112] Optionally, depending on the specific circumstances, the preset ratio of training sample data to validation sample data can be 9:1, 8:2, 7:3, and 6:4. In one specific implementation, 90% of the shuffled sample data is determined as training sample data, and the other 10% is determined as validation sample data, with a preset ratio of 9:1. This ratio is not limited in the embodiments of this application.
[0113] S306. Train the model based on the training sample data, and verify the trained model using the verification sample data to obtain the fault identification model.
[0114] Optionally, faults in the accessory gearbox bearings can be diagnosed based on the operating data and fault identification model of the accessory gearbox bearings.
[0115] In one specific implementation, the obtained fault training model can be used in practical applications. Through this fault identification model, faults occurring in the accessory gearbox can be diagnosed based on the operating data of the accessory gearbox bearings, which can quickly help analyze the causes of the faults.
[0116] Steps S302-S306 are implemented in the same way as S102-S106 in Embodiment 1. The implementation process in the aforementioned embodiments can be referred to, and will not be repeated here.
[0117] The data processing method for accessory gearbox bearings provided in this embodiment involves collecting a dataset of composite faults in the manufactured gearbox, segmenting and transforming the dataset, adding labels, and training a model to obtain the final fault identification model. The fault identification model obtained through this method can identify composite faults occurring in the accessory gearbox.
[0118] Based on the above embodiments, the implementation scheme of the data processing method for the accessory gearbox will be described in detail below through several specific examples.
[0119] Example 1
[0120] Step 1: In the laboratory, the Case Western Reserve University (CWRU) bearing dataset for the accessory gearbox was collected at different speeds. The CWRU bearing dataset includes: normal baseline data, 12kHz drive-end fault data, 48kHz drive-end fault data, and 12kHz fan-end fault data.
[0121] In the specific implementation of this scheme, based on the structural parameters of the rolling bearing, the characteristic frequencies generated by the outer raceway fault, the inner raceway fault, and the rolling element fault can be calculated. The specific calculation formula is shown in Table 1, where d represents the rolling element diameter, D represents the rolling bearing pitch diameter, α represents the bearing contact angle, z represents the number of rolling elements, and f represents the rotational speed of the shaft.
[0122] Table 1. Formulas for calculating the failure frequency of rolling bearings
[0123]
[0124] The CWRU bearing dataset includes normal baseline data, 12kHz drive-end fault data, 48kHz drive-end fault data, and 12kHz fan-end fault data. The fan-end bearing is model SKF6203, and the drive-end bearing is model SKF6205. The bearing geometry parameters are shown in Table 2. Since both the drive-end and fan-end bearings are ball bearings, α is 0.
[0125] Table 2 CWRU Bearing Geometric Parameters
[0126]
[0127] The data in the CWRU dataset were collected at four different speeds. The speeds and the fault characteristic frequencies of the drive-end bearing and fan-end bearing at different speeds are shown in Table 3. The fault frequency refers to a multiple of the rotational frequency (Hz).
[0128] Table 3 CWRU bearing failure frequency
[0129]
[0130] As can be seen from Table 3, the frequency of each fault is similar at different speeds, so the data at different speeds can be treated as a category for fault identification.
[0131] In this embodiment, time-series data from a drive-end sensor with a frequency of 12kHz is used. The data is divided into seven categories: normal, fan-end inner ring fault, fan-end outer ring fault, fan-end rolling element fault, drive-end inner ring fault, drive-end outer ring fault, and drive-end rolling element fault.
[0132] Step 2: After reading the data files in the CWRU bearing dataset, cut the seven types of fault timing data into data sequences of length 500.
[0133] Step 3: Convert the sequence into a frequency domain signal using FFT transform. The FFT transform process is as follows:
[0134] (1) Let the time series of length N=2m be represented as x(n), n=1,2,…N, then the Discrete Fourier Transform (DFT) of x(n) is:
[0135] Equation (2-1)
[0136] Where k = 0, 1, ..., N-1,
[0137] (2) Decompose x(n) into the sum of two sequences of even and odd numbers, i.e.:
[0138] x(n) = x1(n) + x2(n) Equation (2-2)
[0139] (3) According to equation (2-2), the lengths of x1(n) and x2(n) are both N / 2. Assuming that x1(n) is an even sequence and x2(n) is an odd sequence, then:
[0140] (k = 0, 1, ..., N-1)
[0141] Equation (2-3)
[0142] so:
[0143] (k = 0, 1, ..., N-1)
[0144] Equation (2-4)
[0145] And because ,but:
[0146] (k = 0, 1, ..., N-1)
[0147] Equation (2-5)
[0148] Where X1(k) and X2(k) are the N / 2-point DFTs of x1(n) and x2(n), respectively. Since both X1(k) and X2(k) have a period of N / 2, and Therefore, X(k) can also be expressed as:
[0149] (k = 0, 1, ..., N-1) Equation (2-6)
[0150] (k = 0, 1, ..., N-1) Equation (2-7)
[0151] The shape of the FFT operation is called a butterfly operation. Following this pattern, after m-1 decompositions, the N-point DFT is finally decomposed into N / 2 two-point DFTs. Figures 5 to 11 The frequency domain waveforms of typical fault states for seven types of faults in the CWRU bearing dataset provided in this application are shown. Figure 5 This application provides a waveform diagram of normal bearing data in the embodiments; Figure 6 This application provides a waveform diagram of a faulty rolling element at the drive end in an embodiment. Figure 7 This application provides a waveform diagram of a fault in the inner ring of the drive end in an embodiment; Figure 8 This application provides a waveform diagram of a fault on the outer ring of the drive end in an embodiment; Figure 9 This application provides a waveform diagram of a fan-end rolling element failure in an embodiment; Figure 10 This application provides a waveform diagram of a fault in the inner ring of the fan end in an embodiment; Figure 11 The following waveform diagrams are provided for the failure of the outer ring of the fan end in the embodiments of this application; each diagram represents the data obtained by the acceleration sensor at the drive end.
[0152] Step 4: Add one-hot labels to these seven types of data. Let the normal data number label be 0, the drive end rolling element fault number label be 1, the drive end inner ring fault number label be 2, the drive end outer ring fault number label be 3, the fan end rolling element fault number label be 4, the fan end inner ring fault number label be 5, and the fan end outer ring fault number label be 6. Then generate one-hot labels accordingly. For example, the normal data one-hot label is [1, 0, 0, 0, 0, 0, 0].
[0153] Step 5: Shuffle the sample data corresponding to all fault types, and divide the shuffled sample data into training sample data and validation sample data according to a preset ratio.
[0154] In one specific implementation, 90% of the shuffled sample data is determined as the training sample data, and the other 10% is determined as the validation sample data, with the preset ratio being 9:1.
[0155] Step 6: Train the model based on the training sample data, and validate the trained model using the validation sample data to obtain the fault identification model.
[0156] Step 7: Based on the operating data and fault identification model of the accessory gearbox bearing, diagnose the faults of the accessory gearbox bearing.
[0157] This example provides a data processing method for accessory gearbox bearings. It collects CWRU datasets at different speeds, using the similarity of fault characteristic frequencies across different speeds as a feature fault. The collected bearing dataset is segmented into fixed-length sequences. The time-series data sequences are then transformed into frequency domain data using Fourier transform, and one-hot labels are added. A portion of the labeled sample data is selected for model training, and the remaining validation sample data is used to validate the trained model. The validated fault identification model can then be used for model diagnosis of accessory gearbox bearings. This approach enables the establishment of a fault identification model for diagnosing faults in accessory gearbox bearings.
[0158] Example 2
[0159] Step 1: Conduct a compound fault experiment on the accessory gearbox. The faulty bearing model is ER-16K, and the number of faulty gears is Z1=29. The data acquisition frequency is 16384Hz. A compound fault dataset for the gearbox is obtained, which includes: normal bearing data, bearing rolling element defect data, bearing inner ring defect data, bearing outer ring defect data, gear tooth breakage data, gear tooth breakage and bearing outer ring defect data, gear tooth breakage and bearing inner ring defect data, gear missing tooth defect data, gear missing tooth and bearing outer ring defect data, and gear missing tooth and bearing inner ring defect data.
[0160] Data was collected at five different rotational speeds: 1200 rpm, 1500 rpm, 1800 rpm, 2100 rpm, and 2400 rpm, corresponding to frequencies of 20 Hz, 25 Hz, 30 Hz, 35 Hz, and 40 Hz, respectively. This embodiment uses data from each channel collected at a rotational speed of 1200 rpm.
[0161] After reading the composite fault dataset centralized file, select eight types of fault data: normal bearing data, bearing rolling element defect data, gear broken tooth + bearing outer ring defect data, gear broken tooth + bearing inner ring defect data, gear broken tooth defect data, gear missing tooth + bearing outer ring defect data, gear missing tooth + bearing inner ring defect data, and gear missing tooth defect data.
[0162] Step 2: Cut the time-series data of these eight types of faults into sequences of length 1000.
[0163] Step 3: Use FFT to convert the time series into a frequency domain signal.
[0164] The transformation process is the same as in Example 1, and will not be repeated here.
[0165] Step 4: Add one-hot labels to these eight data categories. Let the normal bearing data be labeled as 0, the bearing rolling element defect data as 1, the gear tooth breakage defect data as 2, the gear tooth breakage + bearing outer ring defect data as 3, the gear tooth breakage + bearing inner ring defect data as 4, the gear missing tooth defect data as 5, the gear missing tooth + bearing outer ring defect data as 6, and the gear missing tooth + bearing inner ring defect data as 7. Then generate one-hot labels accordingly. For example, the one-hot label for normal data is [1, 0, 0, 0, 0, 0, 0, 0].
[0166] Specifically, Figures 12 to 19 Frequency domain waveforms of typical fault states in the composite fault dataset of eight types of faults provided in this application. Figure 12 The present application provides a normal composite data waveform diagram in the embodiments; Figure 13 This application provides waveform diagrams of bearing rolling element defects in its embodiments; Figure 14 This application provides waveform diagrams of gear tooth breakage defects in its embodiments. Figure 15 This application provides waveform diagrams of gear tooth breakage and bearing outer ring defects in the embodiments. Figure 16 This application provides a waveform diagram of gear tooth breakage and bearing inner ring defect data in the embodiments; Figure 17 This application provides a waveform diagram of gear tooth missing defect data in the embodiments; Figure 18 This application provides a waveform diagram of gear tooth loss and bearing outer ring defect data in the embodiments; Figure 19 The following waveform diagrams are provided for the embodiments of this application, showing the data of missing gear teeth and bearing inner ring defects; each diagram represents the data obtained by the channel acceleration sensor.
[0167] Step 5: Shuffle the sample data corresponding to all fault types, and divide the shuffled sample data into training sample data and validation sample data according to a preset ratio.
[0168] In one specific implementation, 90% of the shuffled sample data is determined as the training sample data, and the other 10% is determined as the validation sample data, with the preset ratio being 9:1.
[0169] Step 6: Train the model based on the training sample data, and validate the trained model using the validation sample data to obtain the fault identification model.
[0170] Optionally, the model can be trained using a self-attention model, as detailed below:
[0171] The output shape and number of parameters of the self-attention network are shown in Table 4.
[0172] Table 4. Output shape and number of parameters of each layer of the self-attention network on the test composite fault dataset.
[0173]
[0174] The trained model was validated using validation sample data to obtain a fault identification model. Figure 20 This is a loss / accuracy graph of the predicted fault classification in the embodiments provided in this application. Figure 20 As shown, this is the "loss / accuracy curve" for classifying complex faults in a gearbox. Figure 21 This is the confusion matrix diagram provided in the embodiments of this application, such as... Figure 21 As shown in Table 5, the confusion matrix obtained by the self-attention network is tested, and the performance metrics are shown in Table 5. The results of the test set are output onto a two-dimensional plane. The final prediction accuracy of the network is 0.97, and the response time of the network is 26 minutes.
[0175] Table 5 shows the performance metrics obtained from testing the self-attention model.
[0176]
[0177] Step 7: Based on the operating data and fault identification model of the accessory gearbox bearing, diagnose the faults of the accessory gearbox bearing.
[0178] Through validation data testing, the final accuracy rate of the fault identification model for diagnosing and predicting complex fault data is 0.97, which can be used in practical applications to diagnose accessory gearbox faults.
[0179] This example provides a data processing method for accessory gearbox bearings. By collecting a composite fault dataset, cutting and transforming the composite fault dataset, adding one-hot labels, and training a model to obtain a fault identification model, the fault identification model obtained by this method can diagnose and predict composite faults occurring in the accessory gearbox with a prediction accuracy of 0.97.
[0180] Figure 22 A schematic diagram of the structure of the data processing device for the gearbox bearing provided in this application, as shown in Embodiment 1. Figure 22 As shown, the data processing device 300 includes:
[0181] Receiving module 311 is used to acquire the dataset of the accessory gearbox bearing, the dataset including bearing dataset or gearbox composite fault dataset;
[0182] The cutting module 312 is used to cut the dataset into multiple data sequences of a preset length according to the number of fault types; the data in each data sequence is data of different fault types;
[0183] Transformation module 313 is used to perform Fourier transform processing on the multiple data sequences to obtain multiple frequency domain data;
[0184] The tag module 314 is used to add a heat tag to each piece of frequency domain data according to the fault type corresponding to the frequency domain data, so as to obtain the corresponding sample data.
[0185] The sample module 315 is used to shuffle the sample data corresponding to all fault types and divide the shuffled sample data into training sample data and validation sample data according to a preset ratio.
[0186] The training module 316 is used to train the model based on the training sample data and to verify the trained model using the verification sample data to obtain a fault identification model.
[0187] Optionally, the receiving module 311 is further configured to:
[0188] Fault experiments were conducted on the accessory gearbox to obtain the bearing dataset, which includes: normal baseline data, 12kHz drive-end fault data, 48kHz drive-end fault data, and 12kHz fan-end fault data.
[0189] Optionally, the device further includes a diagnostic module 317, specifically used for:
[0190] Based on the operating data of the accessory gearbox bearing and the fault identification model, the faults of the accessory gearbox bearing are diagnosed.
[0191] Optionally, the receiving module 311 is further configured to:
[0192] A composite fault experiment was conducted on the accessory gearbox to obtain the gearbox composite fault dataset. The gearbox composite fault dataset includes: normal bearing data, bearing rolling element defect data, bearing inner ring defect data, bearing outer ring defect data, gear tooth breakage data, gear tooth breakage and bearing outer ring defect data, gear tooth breakage and bearing inner ring defect data, gear missing tooth defect data, gear missing tooth and bearing outer ring defect data, and gear missing tooth and bearing inner ring defect data.
[0193] Optionally, the device further includes a diagnostic module 317, which is specifically used for:
[0194] Based on the operating data of the accessory gearbox and the fault identification model, the faults of the accessory gearbox are diagnosed.
[0195] Optionally, the sample module 315 is further used for:
[0196] 90% of the shuffled sample data is determined as the training sample data, and the other 10% is determined as the validation sample data, with the preset ratio being 9:1.
[0197] Optionally, the tag module 314 further comprises:
[0198] The heat labels added to the frequency domain data corresponding to each fault type are different.
[0199] The data processing device provided in this embodiment is used to execute the technical solution of the data processing device side of the gearbox bearing in any of the foregoing method embodiments. Its implementation principle and technical effect are similar, and will not be described again here.
[0200] Figure 23 The schematic diagram of the electronic device structure provided in this application is as follows: Figure 23 As shown, the electronic device 400 includes:
[0201] Processor 411, memory 412, interface 413 for communicating with terminal devices;
[0202] The memory 412 stores computer-executed instructions;
[0203] The processor 411 executes the computer execution instructions stored in the memory 412, causing the processor 411 to execute the technical solution on the electronic device side in any of the foregoing method embodiments.
[0204] This application embodiment also provides a computer-readable storage medium, which may include various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a disk, or an optical disk. Specifically, the computer-readable storage medium stores program instructions, which are used in the methods described in the above embodiments.
[0205] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this application are indicated by the following claims.
[0206] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.
Claims
1. A data processing method for accessory gearbox bearings, characterized in that, include: Obtain the dataset of the accessory gearbox bearings, wherein the dataset includes a bearing dataset or a gearbox composite fault dataset; Based on the number of fault types, the dataset is divided into multiple data sequences of a preset length; each data sequence contains data of different fault types. The multiple data sequences are subjected to Fourier transform processing to obtain multiple frequency domain data. For each piece of frequency domain data, a tag represented by one-hot encoding is added to the frequency domain data according to the fault type corresponding to the frequency domain data to obtain the corresponding sample data; The sample data corresponding to all fault types are shuffled, and the shuffled sample data is divided into training sample data and validation sample data according to a preset ratio; the preset ratio of the training sample data to the validation sample data is 9:1, 8:2, 7:3 or 6:
4. The model is trained using the training sample data, and the trained model is validated using the validation sample data to obtain a fault identification model; the model training uses a self-attention model. If the dataset is a bearing dataset, then obtaining the dataset of the accessory gearbox bearing includes: conducting a fault experiment on the accessory gearbox to obtain the bearing dataset, which includes: normal baseline data, 12kHz drive-end fault data, 48kHz drive-end fault data and 12kHz fan-end fault data. The dataset is a gearbox composite fault dataset. Therefore, obtaining the dataset for the accessory gearbox bearing includes: conducting composite fault experiments on the accessory gearbox to obtain the gearbox composite fault dataset. The gearbox composite fault dataset includes: normal bearing data, bearing rolling element defect data, bearing inner ring defect data, bearing outer ring defect data, gear tooth breakage data, gear tooth breakage and bearing outer ring defect data, gear tooth breakage and bearing inner ring defect data, gear missing tooth defect data, gear missing tooth and bearing outer ring defect data, and gear missing tooth and bearing inner ring defect data.
2. The method according to claim 1, characterized in that, The method further includes: Based on the operating data of the accessory gearbox bearing and the fault identification model, the faults of the accessory gearbox bearing are diagnosed.
3. The method according to claim 1, characterized in that, The method further includes: Based on the operating data of the accessory gearbox and the fault identification model, the faults of the accessory gearbox are diagnosed.
4. The method according to any one of claims 1 to 3, characterized in that, The step of dividing the shuffled sample data into training sample data and validation sample data according to a preset ratio includes: 90% of the shuffled sample data is determined as the training sample data, and the other 10% is determined as the validation sample data, with the preset ratio being 9:
1.
5. The method according to any one of claims 1 to 3, characterized in that, The heat labels added to the frequency domain data corresponding to each fault type are different.
6. A data processing device for an accessory gearbox bearing, characterized in that, include: The receiving module is used to acquire the dataset of the accessory gearbox bearing, the dataset including the bearing dataset or the gearbox composite fault dataset; The segmentation module is used to segment the dataset into multiple data sequences of a preset length based on the number of fault types; each data sequence contains data of different fault types. The transformation module is used to perform Fourier transform processing on the multiple data sequences to obtain multiple frequency domain data. The tagging module is used to add a tag represented by one-hot encoding to each piece of frequency domain data according to the fault type corresponding to the frequency domain data, so as to obtain the corresponding sample data. The sample module is used to shuffle the sample data corresponding to all fault types and divide the shuffled sample data into training sample data and validation sample data according to a preset ratio. The preset ratio of the training sample data to the validation sample data is 9:1, 8:2, 7:3 or 6:4; The training module is used to train the model based on the training sample data and to validate the trained model using the validation sample data to obtain a fault identification model; the model training adopts a self-attention model. If the dataset is a bearing dataset, then the receiving module is specifically used to conduct a fault experiment on the accessory gearbox to obtain the bearing dataset. The bearing dataset includes: normal baseline data, 12kHz drive end fault data, 48kHz drive end fault data, and 12kHz fan end fault data. The dataset is a gearbox composite fault dataset. The receiving module is specifically used to conduct composite fault experiments on the accessory gearbox to obtain the gearbox composite fault dataset. The gearbox composite fault dataset includes: normal bearing data, bearing rolling element defect data, bearing inner ring defect data, bearing outer ring defect data, gear tooth breakage data, gear tooth breakage and bearing outer ring defect data, gear tooth breakage and bearing inner ring defect data, gear missing tooth defect data, gear missing tooth and bearing outer ring defect data, and gear missing tooth and bearing inner ring defect data.
7. An electronic device, characterized in that, include: The processor, the memory communicatively connected to the processor, and the communication interface for interacting with other devices; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory to implement the data processing method for the accessory gearbox bearing as described in any one of claims 1 to 5.
8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the data processing method for the accessory gearbox bearing as described in any one of claims 1 to 5.