Method, equipment, and medium for early warning of rotating machinery blade faults based on self-encoders

Abstract

This invention relates to the field of vibration signal processing and fault early warning diagnosis technology for rotating machinery, specifically to a method, device, and medium for early warning of blade faults in rotating machinery based on a self-encoder. It can acquire vibration signals from the equipment casing using a broadband vibration acceleration sensor, perform sparse processing on the spectrum, and combine this with deep self-encoder reconstruction technology to achieve accurate early warning of blade faults. This invention acquires signals by installing a vibration acceleration sensor on the equipment casing, eliminating the need to directly install sensors on the blades. This avoids damage to the equipment casing structure caused by drilling, ensuring the integrity and stability of the overall rotating machinery structure. It also prevents additional interference to equipment performance caused by monitoring methods and reduces the possibility of increased blade fault risk due to sensor installation.

Description

Technical Field

[0001] This invention relates to the field of vibration signal processing and fault early warning diagnosis technology for rotating machinery, specifically to a method, device, and medium for early warning of faults in rotating machinery blades based on a self-encoder. Background Technology

[0002] In the operation of rotating machinery, blades are core components, and their health directly affects the performance and safety of the entire equipment. For example, blades in equipment such as gas turbines, flue gas turbines, and steam turbines are subjected to complex operating conditions such as high temperature, high pressure, high speed, and alternating loads for extended periods, making them highly susceptible to failures such as scale buildup, cracks, and even breakage. Once a blade fails, it may not only cause equipment downtime and affect production efficiency, but may also lead to serious safety accidents.

[0003] Currently, there are three main categories of methods for detecting faults in rotating machinery blades.

[0004] (1) Direct measurement of blades based on sensors: Strain gauges, fiber optic sensors, etc. are directly installed on the blade surface to monitor parameters such as stress and strain. However, this method requires complex installation operations on the blade, which will change the original dynamic characteristics of the blade and affect its normal operation. Moreover, the sensors are easily damaged by harsh environments, leading to monitoring failure. The blade tip timing method requires drilling holes in the shell to install sensors, which will damage the original structure of the shell. The long-term continuous online monitoring of the blade tip timing monitoring system in the field is still difficult to implement; especially for high-speed rotating and complex blades, the direct installation of sensors is extremely difficult, and in some cases, it is even impossible.

[0005] (2) Aerodynamic parameter monitoring: The condition of the blades can be indirectly inferred by monitoring aerodynamic parameters such as airflow pressure, temperature, and flow rate at the inlet and outlet of rotating machinery. However, these parameters are affected by a variety of factors, such as changes in environmental conditions and fluctuations in the performance of other components. This makes the aerodynamic parameters less sensitive to blade failures, and it is difficult to accurately identify the blade failure status in the early stages of a failure. For example, when slight fouling or cracks appear on the blades, the changes in aerodynamic parameters may not be obvious, and it is impossible to issue an early warning in time.

[0006] (3) Based on traditional vibration analysis methods: Vibration signals are collected using low-frequency vibration sensors installed on the bearing housing or casing, and blade faults are diagnosed using traditional methods such as spectrum analysis and time-domain analysis. However, the blade fault characteristics contained in low-frequency vibration signals are relatively weak, especially in the early stages of the fault. These weak characteristics are easily drowned out by the vibration response of bearings, rotor components, or other noise signals. Moreover, traditional spectrum analysis methods have limited ability to extract complex fault features, making it difficult to achieve accurate fault early warning.

[0007] In summary, existing methods for detecting blade failures in rotating machinery suffer from drawbacks such as structural damage, low sensitivity, and difficulty in early warning, failing to meet the high reliability and safety requirements of modern industry for rotating machinery. Therefore, there is an urgent need for a new method that can sensitively provide early warnings of blade failures in rotating machinery without damaging the original machine casing structure, by incorporating information on blade failure mechanisms. Summary of the Invention

[0008] In view of this, the present invention proposes a method, device and medium for early warning of blade failure in rotating machinery based on autoencoder. It can collect vibration signals of the equipment housing by using a broadband vibration acceleration sensor, perform sparse processing on its spectrum, and combine it with deep autoencoder reconstruction technology to achieve accurate early warning of blade failure.

[0009] To achieve the above objectives, the technical solution of the present invention is as follows:

[0010] A method for early warning of faults in rotating machinery blades based on a self-encoder includes the following steps:

[0011] S 1. Install a vibration acceleration sensor on the housing of the rotating equipment to collect vibration signals;

[0012] S 2. Calculate the frequency transition by improving the harmonic product spectrum;

[0013] S 3. Based on the number of blades at each stage, calculate the theoretical values ​​of the passing frequency of each stage of blades and obtain the sparse spectrum;

[0014] S 4. Construct a convolutional autoencoder to reconstruct and train the sparse spectrum of the input;

[0015] S 5. Calculate the reconstruction error based on normal data to obtain the blade fault early warning threshold;

[0016] S 6. Reconstruct new vibration data using a trained convolutional autoencoder, calculate reconstruction error, and issue early warnings.

[0017] Among them, the S In step 1, a vibration acceleration sensor is installed on the housing of the rotating equipment to collect vibration signals, including the following steps:

[0018] The broadband vibration acceleration sensor is fixed to the rotating equipment housing at one end of the blade using magnetic or threaded mounting. Then, the sampling rate is set to analyze the frequency of each blade at least covering the highest speed. By using frequency doubling, key frequency information related to blade failure can be captured.

[0019] Among them, the S In section 2, the frequency shift is accurately calculated by improving the harmonic product spectrum, specifically as follows:

[0020] First, set the rotational frequency range according to the equipment's typical operating speed range. First, the frequency multiplier and the number of blades were selected as N1 and N2, respectively; second, the spectrum of the computer shell vibration signal was analyzed; then, three spectral intervals were retained. N1* N2* Set all others to zero to obtain the interval spectrum; further, upsample the interval spectrum by N1*N2, N2, and N1 times respectively using linear interpolation to obtain three upsampled interval spectra; then, multiply the three upsampled interval spectra by a dot product to obtain the improved harmonic product spectrum; finally, extract the frequency corresponding to the maximum value in this spectrum to obtain the accurately calculated rotational frequency. .

[0021] Among them, the S In step 3, based on the number of blades at each stage, the theoretical values ​​of the passing frequencies of each stage are calculated, and the sparse spectrum is obtained, specifically as follows:

[0022] First, based on the equipment's structure or design information, record the number of blades at each stage. m is the number of blade stages; then, combined with the calculated rotational frequency... Given the number of blades at each stage, calculate the second and third harmonics of the rotational frequency. The passing frequency of each stage of the blades was calculated. And the second harmonic of the blade's passing frequency This yields the set of all frequencies related to the blades mentioned above. The original vibration signal spectrum is sparsified, retaining other frequency components within three frequency resolutions of the blade-related frequency, i.e., the retention interval []. This yields a sparse spectrum, where... Indicates frequency resolution.

[0023] In step S4, a convolutional autoencoder is constructed to reconstruct and train the sparse spectrum of the input, specifically as follows:

[0024] First, the convolutional autoencoder is constructed, consisting of an encoder and a decoder. The encoder has two one-dimensional convolutional layers, each with a kernel size of 4, a stride of 2, and padding of 1. The first layer has 1 input channel and 16 output channels, using the ReLU activation function to introduce non-linearity. This layer performs preliminary feature extraction from the input data. The second layer has 16 input channels and increases the number of output channels to 32, using the ReLU activation function for non-linear transformation. The decoder consists of two deconvolutional layers and activation functions. Each layer has a kernel size of 4, a stride of 2, and padding of 1. It is used to reconstruct the features extracted by the encoder back to the original input data. The first layer has 32 input channels and 16 output channels, using the ReLU activation function for non-linear transformation. This layer upsamples the 32-channel features extracted by the encoder, gradually restoring the data dimensionality. The second layer has 16 input channels and 1 output channel. The data in this layer is reconstructed to have the same dimensionality as the original input data.

[0025] Among them, the S In step 5, the reconstruction error is calculated based on the training data to obtain the blade fault early warning threshold, specifically:

[0026] The sparse spectrum under normal operating conditions is input into a trained convolutional autoencoder model to obtain the reconstructed spectrum. The average error between the original sparse spectrum and the reconstructed spectrum is then calculated. As the reconstruction error, statistical analysis is performed on the reconstruction error, and a reasonable warning threshold is taken as 1.5 times the 95% confidence interval of the reconstruction error, where L is the number of discrete frequency points in the sparse spectrum and the reconstructed spectrum. In the original sparse spectrum, the first i The frequency amplitude corresponding to each frequency point To reconstruct the spectrum of the first i The frequency amplitude corresponding to each frequency point i The value ranges from 1 to L.

[0027] Among them, the S In step 6, the new vibration data is reconstructed using a trained convolutional autoencoder, the reconstruction error is calculated, and an early warning is issued. Specifically:

[0028] First, during the actual operation of the rotating equipment, vibration data is collected in real time. After undergoing the same processing steps as the training data, the sparse spectrum of the newly collected data is obtained. This spectrum is then input into a trained convolutional autoencoder for reconstruction, outputting the reconstructed spectrum and calculating the reconstruction error. Finally, the reconstruction error is compared with... S The warning thresholds calculated in step 5 are compared. If the reconstruction error exceeds the warning threshold, a blade fault warning signal is issued.

[0029] The present invention also provides an electronic device, the electronic device including a processor and a memory for storing executable instructions of the processor; the processor is used to read the executable instructions from the memory and execute the instructions to implement the self-encoder-based rotary machinery blade fault early warning method of the present invention.

[0030] The present invention also provides a computer-readable storage medium storing a computer program for executing the self-encoder-based rotary machinery blade fault early warning method described in the present invention.

[0031] Beneficial effects

[0032] 1. The method of the present invention collects signals by installing a vibration acceleration sensor on the equipment housing, without the need to install the sensor directly on the blade. This avoids damage to the equipment housing structure caused by drilling, ensures the integrity and stability of the overall structure of the rotating machinery, and does not cause additional interference to the equipment performance due to the monitoring method. It also reduces the possibility of increased blade failure risk due to the installation of sensors.

[0033] 2. This invention fully preserves the sensitive components and characteristics of blade faults, enhancing early warning capabilities. By employing a broadband vibration acceleration sensor and performing sparse processing of the spectrum based on equipment operating mechanisms and the number of blades at each stage, it effectively retains blade fault-related characteristics, removes irrelevant signal components, and highlights the fault-sensitive components in the signal, thus improving the detection capability for early blade faults. Compared to traditional methods, it can capture more subtle changes in fault characteristics, providing stronger data support for early warning.

[0034] 3. In the method of this invention, the constructed convolutional autoencoder model can effectively capture and reconstruct the blade's sensitive features in the sparse spectrum. By comparing the differences in the spectrum before and after reconstruction, it can be determined whether the blade has failed. When the blade is normal, the autoencoder can reconstruct the spectrum well; when the blade fails, the reconstructed spectrum differs significantly from the original spectrum. This method can detect early signs of failure in the blade, such as fouling and microcracks, enabling early warning and allowing more time for equipment downtime and maintenance. This reduces downtime and economic losses caused by blade failures, and improves the reliability and safety of rotating machinery operation.

[0035] 4. The method of this invention has good versatility and scalability. Whether it is a gas turbine, a flue gas fan, a steam turbine, or a fan, as long as blade faults are manifested through casing vibration, this method can be applied for fault early warning. Furthermore, with the development of deep learning technology, the deep autoencoder model can be further optimized and upgraded, adding more layers or adopting more advanced network structures to continuously improve fault early warning performance and adapt to the future needs of rotating machinery developing towards higher parameters and more complex operating conditions.

[0036] 5. The device of the present invention is used to implement the method of the present invention. By installing a vibration acceleration sensor on the equipment shell to collect signals, there is no need to install the sensor directly on the blade, which avoids the damage to the equipment shell structure caused by drilling. This ensures the integrity and stability of the overall structure of the rotating machinery, and will not cause additional interference to the equipment performance due to the monitoring method. It also reduces the possibility of increased blade failure risk due to the installation of sensors.

[0037] 6. The device of the present invention is used to implement the method of the present invention. By installing a vibration acceleration sensor on the equipment shell to collect signals, there is no need to install the sensor directly on the blade, which avoids damage to the equipment shell structure caused by drilling. This ensures the integrity and stability of the overall structure of the rotating machinery, and will not cause additional interference to the equipment performance due to the monitoring method. It also reduces the possibility of increased blade failure risk due to the installation of sensors. Attached Figure Description

[0038] Figure 1 This is a flowchart of the rotating machinery blade fault early warning method based on sparse convolutional autoencoder of the present invention.

[0039] Figure 2 This is a schematic diagram of the rotating machinery blade and housing structure and sensor installation position of the present invention. It shows the structural relationship between the blade and housing inside the rotating machinery, as well as the specific installation position of the broadband vibration acceleration sensor on the housing. The vibration transmission path is indicated by dashed lines and markings to make the location and source of signal acquisition clear.

[0040] Figure 3 This is a schematic diagram illustrating the principle of the improved harmonic product spectrum for calculating frequency conversion in this invention. It graphically demonstrates the process of calculating frequency conversion using the improved harmonic product spectrum, helping to understand the principle and method of frequency conversion calculation.

[0041] Figure 4 This is a schematic diagram of the sparse spectrum generation process of the present invention, which describes in detail the generation process from the original spectrum to the sparse spectrum, making it easier to understand the principle of sparse spectrum preserving fault features.

[0042] Figure 5 This is a schematic diagram of the convolutional autoencoder model structure of the present invention. The structure of each layer of the encoder and decoder of the convolutional autoencoder is shown in the form of blocks and lines, which facilitates understanding of the model architecture and data flow.

[0043] Figure 6 This diagram illustrates the loss curves during the training process of this invention. It shows the changes in training loss and validation loss with each training round, visually demonstrating the training effect and convergence of the model, as well as the positions of the optimal validation point and test loss on the curves, thus aiding in the performance evaluation of model training.

[0044] Figure 7 This is a schematic diagram comparing the original sparse spectrum and the reconstructed spectrum of the present invention, showing the comparison between the original sparse spectrum and the reconstructed spectrum under normal operating conditions and fault conditions, respectively, to help understand how to judge blade faults through spectral differences.

[0045] Figure 8 This diagram illustrates the relationship between reconstruction error and warning threshold in this invention. It plots the curve of reconstruction error changing over time or data points and marks the position of the warning threshold, visually demonstrating how to provide fault warnings based on the comparison between reconstruction error and warning threshold.

[0046] Figure 9 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0047] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0048] This embodiment proposes a method for early warning of rotating machinery blade faults based on a sparse convolutional autoencoder. This method effectively utilizes relevant features containing blade fault information from collected vibration signals without damaging the original structure of the equipment casing, and sensitively provides early warnings for faults such as fouling, cracks, and fractures in rotating machinery blades. The method includes the following steps: installing a vibration acceleration sensor on the rotating equipment casing to collect vibration signals; accurately calculating the rotational frequency by improving the harmonic product spectrum without additional rotational speed monitoring sensors; calculating the theoretical passing frequency of each blade stage based on the number of blades at each stage and the accurately extracted rotational frequency, and obtaining a sparse spectrum; constructing a convolutional autoencoder to reconstruct and train the input sparse spectrum; calculating the reconstruction error based on normal operating data to obtain a blade fault warning threshold; reconstructing new vibration data using the trained convolutional autoencoder, calculating the reconstruction error, and issuing a warning.

[0049] The following is an illustration of a blade failure case in a certain type of rotating machinery. The equipment operates at a speed range of [8100, 9900] rpm and includes 17 stages of moving / stationary blades. The number of moving blades in each stage is {16, 26, 26, 42, 45, 48, 54, 56, 64, 66, 66, 76, 76, 76, 76, 76}. The flow of the rotating machinery blade failure early warning method in this embodiment is as follows: Figure 1 As shown, the specific steps are as follows:

[0050] S 1. Install a vibration acceleration sensor on the rotating equipment housing to collect vibration signals. In this embodiment, a wideband vibration acceleration sensor with a sufficient frequency response range is selected. The wideband vibration acceleration sensor is fixed to the rotating equipment housing near the blades using a threaded mounting method. The sensor installation diagram is shown below. Figure 2 As shown; the 17th stage has a maximum of 76 blades. At a maximum speed of 9900 rpm, the passing frequency of twice the blades is 25080 Hz. With a sampling rate of fs=64000 Hz, the analysis frequency can reach 32000 Hz, exceeding the passing frequency of each stage of blades at the highest speed, ensuring the capture of key frequency information related to blade faults. The casing vibration signal is continuously collected according to the set sampling frequency, with each data segment lasting 0.256 seconds, and the number of sampling points N=16384.

[0051] S 2. In the absence of an additional speed monitoring sensor, the rotational frequency is accurately calculated by improving the harmonic product spectrum. In this embodiment, the rotational frequency range is first set according to the typical operating speed range of the equipment. =[135Hz, 165Hz], select multiples of the rotational frequency and the number of blades in the second stage as N1 and N2, i.e., N1=2, N2=26; next, take a set of collected vibration signals as an example, the spectrum of the computer shell vibration signal; then, retain the three spectral intervals of the spectrum. 2* 26* Set the rest to zero to obtain the interval spectrum; further, upsample the interval spectrum by N1*N2 times, N2 times, and N1 times respectively using linear interpolation to obtain three upsampled interval spectra; then, multiply the three upsampled interval spectra by a dot product to obtain the improved harmonic product spectrum, as shown below. Figure 3 Finally, the frequency corresponding to the maximum value in the spectrum of the vibration signal of the casing is extracted, which is the accurately calculated rotational frequency. Similarly, the rotational frequency of other acquired shell vibration signals can also be accurately calculated using this method.

[0052] S 3. Based on the number of blades at each stage, calculate the theoretical values ​​of the passing frequencies of each blade stage and obtain the sparse spectrum. In this embodiment, as mentioned above, the number of blades at each stage of the device is {16, 26, 26, 42, 45, 48, 54, 56, 64, 66, 66, 76, 76, 76, 76, 76}, and the number of blade stages m=17; then, considering the device's operating mechanism, and combining the calculated frequency... Given the number of blades at each stage, calculate the second and third harmonics of the rotational frequency. The passing frequency of each stage of the blades was calculated. And the second harmonic of the blade's passing frequency This yields the set of all frequencies related to the blades mentioned above. There are a total of 37 frequency values; secondly, the spectrum of the original vibration signal is sparsified, and the frequency resolution is improved. Other frequency components within three frequency resolutions of the frequency related to the blade are retained, i.e., the retention interval [ This yields a sparse spectrum, such as Figure 4 This processing method can effectively preserve blade fault-related features, remove other interfering components or redundant information such as noise, and improve the efficiency and accuracy of subsequent analysis.

[0053] S4. Construct a convolutional autoencoder to reconstruct and train the sparse spectrum of the input. Specifically: First, construct the convolutional autoencoder, which includes an encoder part and a decoder part, such as... Figure 5 As shown, the encoder consists of two convolutional layers. The first convolutional layer converts the input 1-channel data into 16 channels with a kernel size of 4, a stride of 2, and padding of 1. The second convolutional layer further converts the data into 32 channels, with similar parameters as the first layer in terms of kernel size and stride. The decoder consists of two deconvolutional layers that gradually restore the encoded 32-channel data to the reconstructed 1-channel data. The constructed convolutional autoencoder structure can capture local features of the spectrum and has a good reconstruction effect. Secondly, the sparse spectrum of the vibration signal collected under normal operating conditions of the aforementioned equipment is calculated and used as the training set samples. Each sparse spectrum has a length of 8192 points. In this case, the training set consists of the first 200 sets of data, and the rest are the test set. The model is built on the PyTorch framework, using the Adam optimizer, the MSE loss function, a learning rate of lr = 0.001, and a batch size of 20. The sparse spectrum of the training set is input into the constructed convolutional autoencoder model for reconstruction training, resulting in the trained convolutional autoencoder model. During training, the training loss and validation loss for each epoch are recorded, and the convergence of the model is observed. Figure 6 As shown.

[0054] S 5. Calculate the reconstruction error based on normal data to obtain the blade fault early warning threshold. In this embodiment, the sparse spectrum under normal operating conditions is input into the trained convolutional autoencoder model to obtain the reconstructed spectrum, such as... Figure 7 As shown; calculate the average error of the original sparse spectrum and the reconstructed spectrum. As the reconstruction error, statistical analysis is performed on the reconstruction error. A reasonable warning threshold is determined by taking 1.5 times the 95th percentile of the reconstruction error from 200 sets of training data. In this embodiment, the calculated warning threshold is 0.000628. Where L represents the number of discrete frequency points in the sparse spectrum and the reconstructed spectrum. In the original sparse spectrum, the first i The frequency amplitude corresponding to each frequency point To reconstruct the spectrum of the first i The frequency amplitude corresponding to each frequency point i The value ranges from 1 to L.

[0055] S 6. Reconstruct the new vibration data using a trained convolutional autoencoder, calculate the reconstruction error, and issue an early warning. In this embodiment, firstly, vibration data is collected in real time during the actual operation of the rotating equipment, and the same steps as described above are followed for frequency calculation and sparse spectrum acquisition. The newly obtained sparse spectrum data is then input into the trained convolutional autoencoder for reconstruction, and the reconstruction error is calculated; then, the reconstruction error is compared with... S The calculated warning thresholds in step 5 are compared. If the reconstruction error exceeds the warning threshold, a blade fault warning signal is issued. In this case, the set after 200 sets is the test set, which is used as new data input to the autoencoder for reconstruction. The resulting reconstruction error trend is as follows. Figure 8 As shown. At point 317, the reconstruction error is 0.000704, exceeding the warning threshold, and an alarm is issued.

[0056] As can be seen from this embodiment, the rotating machinery blade fault early warning method based on sparse convolutional autoencoder of the present invention can effectively realize early warning of blade faults in practical applications, and provide strong protection for the safe and stable operation of rotating machinery such as gas turbines and flue gas turbines.

[0057] This application also provides an electronic device. Figure 9The structure of an electronic device provided in an embodiment of the present invention is illustrated. For example, the electronic device 90 may include a processor 91, a memory 92, and a transmission device 93. The processor is used to execute the self-encoder-based rotating machinery blade fault early warning method mentioned in the above embodiments. The processor and the memory can be connected via a bus or other means, taking a bus connection as an example. The transmission device can be connected to the processor and the memory via wired or wireless means. The memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs, non-transitory computer-executable programs, and modules, such as the program instructions / modules corresponding to the self-encoder-based rotating machinery blade fault early warning method in the embodiments of this application. The processor executes various functional applications and data processing by running the non-transitory software programs, instructions, and modules stored in the memory, thereby realizing the self-encoder-based rotating machinery blade fault early warning method in the above method embodiments. The memory may include a program storage area and a data storage area. The program storage area may store the operating system and at least one application program required for a function; the data storage area may store data created by the processor, etc. In addition, the memory may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, the memory may optionally include memory remotely located relative to the processor, and this remote memory may be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof. The one or more modules are stored in the memory and, when executed by the processor, perform the self-encoder-based rotating machinery blade fault early warning method described in the embodiments.

[0058] In another aspect, this application also provides a computer-readable storage medium, which may be the computer-readable storage medium included in the apparatus described in the above embodiments; or it may be a standalone computer-readable storage medium not assembled into the device. The computer-readable storage medium may be a tangible storage medium, such as random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, register, floppy disk, hard disk, removable storage disk, CD-ROM, or any other form of storage medium known in the art. The computer-readable storage medium stores one or more programs, which are used by one or more processors to execute the self-encoder-based rotary machinery blade fault early warning method described in this application.

[0059] In summary, the above are merely preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for early warning of faults in rotating machinery blades based on a self-encoder, characterized in that, It includes the following steps: S 1. Install a vibration acceleration sensor on the housing of the rotating equipment to collect vibration signals; S 2. Calculate the frequency transition by improving the harmonic product spectrum; S 3. Based on the number of blades at each stage, calculate the theoretical values ​​of the passing frequency of each stage of blades and obtain the sparse spectrum; S 4. Construct a convolutional autoencoder to reconstruct and train the sparse spectrum of the input; S 5. Calculate the reconstruction error based on normal data to obtain the blade fault early warning threshold; S 6. Reconstruct new vibration data using a trained convolutional autoencoder, calculate reconstruction error, and issue early warnings; The S In section 2, the frequency shift is accurately calculated by improving the harmonic product spectrum, specifically as follows: First, set the rotational frequency range according to the operating speed range of the equipment. First, the frequency multiplier and the number of blades were selected as N1 and N2, respectively; second, the spectrum of the computer shell vibration signal was analyzed; then, three spectral intervals were retained. N1* N2* Set all others to zero to obtain the interval spectrum; further, upsample the interval spectrum by N1*N2, N2, and N1 times respectively using linear interpolation to obtain three upsampled interval spectra; then, multiply the three upsampled interval spectra by a dot product to obtain the improved harmonic product spectrum; finally, extract the frequency corresponding to the maximum value in the improved harmonic product spectrum, which is the accurately calculated rotational frequency. ; The S In step 3, based on the number of blades at each stage, the theoretical values ​​of the passing frequencies of each stage are calculated, and the sparse spectrum is obtained, specifically as follows: First, based on the equipment's structure or design information, record the number of blades at each stage. m is the number of blade stages; then, combined with the calculated rotational frequency... Given the number of blades at each stage, calculate the second and third harmonics of the rotational frequency. The passing frequency of each stage of the blades was calculated. And the second harmonic of the blade's passing frequency This yields the set of all frequencies related to the blades mentioned above. The original vibration signal spectrum is sparsified, retaining other frequency components within three frequency resolutions of the blade-related frequency, i.e., the retention interval []. This yields a sparse spectrum, where... Indicates frequency resolution.

2. The method as described in claim 1, characterized in that, The S In step 1, a vibration acceleration sensor is installed on the housing of the rotating equipment to collect vibration signals, including the following steps: The broadband vibration acceleration sensor is fixed to the rotating equipment housing at one end of the blade using magnetic or threaded mounting. Then, the sampling rate is set to analyze the frequency of each blade at least covering the highest speed. By using frequency doubling, key frequency information related to blade failure can be captured.

3. The method as described in claim 1, characterized in that, In step S4, a convolutional autoencoder is constructed to reconstruct and train the sparse spectrum of the input, specifically as follows: First, a convolutional autoencoder is constructed, consisting of an encoder and a decoder. The encoder has two one-dimensional convolutional layers, each with a kernel size of 4, a stride of 2, and padding of 1. The first layer has 1 input channel and 16 output channels, and uses the ReLU activation function to introduce non-linearity. This layer performs preliminary feature extraction on the input data. The second layer has 16 input channels and increases the number of output channels to 32, and uses the ReLU activation function for non-linear transformation. The decoder consists of two deconvolutional layers and activation functions. Each layer has a kernel size of 4, a stride of 2, and padding of 1. It is used to reconstruct the features extracted by the encoder into the original input data. The first layer has 32 input channels and 16 output channels. It uses the ReLU activation function for non-linear transformation. This layer upsamples the 32 channels of features extracted by the encoder, gradually restoring the dimensionality of the data. The second layer has 16 input channels and 1 output channel. The data in this layer is reconstructed into a form with the same dimensionality as the original input data.

4. The method according to any one of claims 1-3, characterized in that, The S In step 5, the reconstruction error is calculated based on the training data to obtain the blade fault early warning threshold, specifically: The sparse spectrum under normal operating conditions is input into a trained convolutional autoencoder model to obtain the reconstructed spectrum. The average error between the original sparse spectrum and the reconstructed spectrum is then calculated. As the reconstruction error, statistical analysis is performed on the reconstruction error, and a reasonable warning threshold is taken as 1.5 times the 95% confidence interval of the reconstruction error, where L is the number of discrete frequency points in the sparse spectrum and the reconstructed spectrum. In the original sparse spectrum, the first i The frequency amplitude corresponding to each frequency point To reconstruct the spectrum of the first i The frequency amplitude corresponding to each frequency point i The value ranges from 1 to L.

5. The method as described in claim 1, characterized in that, The S In step 6, the new vibration data is reconstructed using a trained convolutional autoencoder, the reconstruction error is calculated, and an early warning is issued. Specifically: First, during the actual operation of the rotating equipment, vibration data is collected in real time. After undergoing the same processing steps as the training data, the sparse spectrum of the newly collected data is obtained. This spectrum is then input into a trained convolutional autoencoder for reconstruction, outputting the reconstructed spectrum and calculating the reconstruction error. Finally, the reconstruction error is compared with... S The warning thresholds calculated in step 5 are compared. If the reconstruction error exceeds the warning threshold, a blade fault warning signal is issued.

6. An electronic device, characterized in that, The electronic device includes a processor and a memory for storing executable instructions of the processor; the processor is configured to read the executable instructions from the memory and execute the instructions to implement the self-encoder-based rotary machinery blade fault early warning method according to any one of claims 1-5.

7. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which is used to execute the self-encoder-based rotary machinery blade fault early warning method according to any one of claims 1-5.

Images