Rotary machine blade fault early warning method and device based on self-encoder and medium

Abstract

The invention relates to the technical field of rotating machine vibration signal processing and fault early warning diagnosis, in particular to a rotating machine blade fault early warning method and device based on an auto-encoder and a medium. According to the method, the vibration acceleration sensor is installed on the equipment shell to collect signals, the sensor does not need to be directly installed on the blade, damage to the structure of the equipment shell due to punching is avoided, the integrity and stability of the overall structure of the rotating machine are guaranteed, and the accuracy and the reliability of the rotating machine are improved. Extra interference on equipment performance due to monitoring means is avoided, and the possibility that the blade fault risk is increased due to sensor installation is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of rotating machinery vibration signal processing and fault early warning diagnosis, and in particular to a rotating machinery blade fault early warning method, device and medium based on an autoencoder. Background Art

[0002] During the operation of rotating machinery, blades are core components, and their health is directly related to the performance and safety of the entire equipment. For example, blades in equipment such as gas turbines, smoke exhaust fans, and steam turbines are subject to complex operating conditions such as high temperatures, high pressures, high speeds, and alternating loads. These blades are prone to failures such as fouling, cracks, and even breakage. Blade failure can not only cause equipment downtime and impact production efficiency, but can also lead to serious safety incidents.

[0003] Currently, there are three main types of methods for rotating machinery blade fault detection.

[0004] (1) Direct blade measurement based on sensors: strain gauges, fiber optic sensors, etc. are directly installed on the blade surface to monitor the stress, strain and other parameters of the blade. 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. The sensor is easily damaged by the harsh environment, resulting in monitoring failure. The sampling tip timing method requires drilling holes in the shell to install the sensor, which will destroy the original structure of the shell. The blade tip timing monitoring system is difficult to implement long-term continuous online monitoring of field equipment; especially for high-speed rotating and complex blades, direct installation of sensors is extremely difficult, and even impossible in some cases.

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

[0006] (3) Based on traditional vibration analysis methods: Low-frequency vibration sensors installed on the bearing seat or casing are used to collect vibration signals, and blade faults are diagnosed through traditional methods such as spectrum analysis and time domain analysis. However, the blade fault characteristics contained in the low-frequency vibration signal are relatively weak, especially in the early stages of the fault. These weak characteristics can easily be overwhelmed by the vibration response of the bearing and rotor components or other noise signals. In addition, the traditional spectrum analysis method has limited ability to extract complex fault characteristics, making it difficult to achieve accurate fault warning.

[0007] In summary, existing methods for detecting rotor blade faults suffer from structural damage, low sensitivity, and difficulty providing early warnings. These drawbacks make them unable to meet the modern industrial demand for high-reliability and high-safety rotor blade operations. Therefore, a new method is urgently needed that can sensitively detect rotor blade faults early in the future, without damaging the original rotor casing structure, while incorporating information about blade failure mechanisms. Summary of the Invention

[0008] In view of this, the present invention proposes a rotating machinery blade fault warning method, device and medium based on an autoencoder, which can collect the vibration signal of the equipment shell by using a wide-band vibration acceleration sensor and perform sparse processing on its spectrum, and combine it with deep autoencoder reconstruction technology to achieve accurate warning of blade faults.

[0009] To achieve the above object, the technical solution of the present invention is: A rotating machinery blade fault early warning method based on an autoencoder comprises the following steps: S 1. Install a vibration acceleration sensor on the rotating equipment housing to collect vibration signals; S 2. Calculate the rotation frequency by improving the harmonic product spectrum; S 3. Based on the number of blades at each level, calculate the theoretical value of the blade passing frequency at each level and obtain the sparse spectrum; S 4. Construct a convolutional autoencoder to reconstruct the input sparse spectrum; S 5. Calculate the reconstruction error based on normal data to obtain the blade failure warning threshold; S 6. Reconstruct the new vibration data through the trained convolutional autoencoder, calculate the reconstruction error and issue an early warning.

[0010] Among them, the S 1, a vibration acceleration sensor is installed on the housing of the rotating equipment to collect vibration signals, including the following steps: Fix the broadband vibration accelerometer to the rotating equipment housing at one end of the blade through magnetic or threaded mounting. Then set the sampling rate to analyze the frequency of at least all blades at the highest speed. By doubling the frequency, ensure that the key frequency information related to the blade failure can be captured.

[0011] Among them, the S 2, the rotation frequency is accurately calculated by improving the harmonic product spectrum, specifically: First, set the frequency range according to the normal working speed range of the equipment , select the multiple of the rotation frequency and the number of blades as N1 and N2; secondly, calculate the spectrum of the shell vibration signal; then, retain the three spectrum intervals 、N1* 、N2* , and the rest are set to zero to obtain the interval spectrum; further, the interval spectrum is upsampled by N1*N2 times, N2 times and N1 times respectively through linear interpolation to obtain three upsampled interval spectra; then the three upsampled interval spectra are point-multiplied to obtain the improved harmonic product spectrum; finally, the frequency corresponding to the maximum value in the spectrum is extracted, which is the accurately calculated conversion frequency .

[0012] Among them, the S In 3, based on the number of blades at each level, the theoretical value of the blade passing frequency at each level is calculated, and the sparse spectrum is obtained, specifically: First, record the number of blades at each level according to the structure or design information of the equipment. , m is the number of blades; then combined with the calculated rotation frequency With the known number of blades at each level, calculate the double and triple frequencies of the rotational frequency , calculate the passing frequency of blades at each level , and the blade passing frequency doubled , and obtain the set of all frequencies related to the blades mentioned above ; The spectrum of the original vibration signal is sparsely processed, and other frequency components within 3 frequency resolutions of the blade-related frequency are retained, that is, the retention interval [ , and obtain the sparse spectrum, where Indicates frequency resolution.

[0013] In S4, a convolutional autoencoder is constructed to perform reconstruction training on the input sparse spectrum, specifically: First, the convolutional autoencoder is constructed, including the encoder part and the decoder part. The encoder has two one-dimensional convolution layers. The convolution kernel size of each layer is set to 4, the stride is 2, and the padding is 1. The number of input channels of the first layer is 1, and the number of output channels is 16. The ReLU activation function is used to introduce nonlinear factors. This layer performs preliminary feature extraction on the input data; the second layer has 16 input channels and the number of output channels is increased to 32, and the ReLU activation function is used for nonlinear transformation; the decoder consists of two deconvolution layers and activation functions. The convolution kernel size of each layer is 4, the stride is 2, and the padding is 1, which 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. The ReLU activation function is used for nonlinear transformation. This layer upsamples the 32-channel features extracted by the encoder and gradually restores the dimension of the data; the second layer has 16 input channels and 1 output channel. The data in this layer is reconstructed into the same form as the original input data dimension.

[0014] Among them, the S In 5, the reconstruction error is calculated based on the training data to obtain the blade failure warning threshold, which is specifically: The sparse spectrum under normal working conditions is input into the trained convolutional autoencoder model to obtain the reconstructed spectrum, and the average error between the original sparse spectrum and the reconstructed spectrum is calculated. As the reconstruction error, the reconstruction error is statistically analyzed, and 1.5 times the 95% confidence interval of the reconstruction error is taken as a reasonable warning threshold, where L is the number of discrete frequency points of the sparse spectrum and the reconstructed spectrum, The original sparse spectrum i The frequency amplitude corresponding to the frequency point, To reconstruct the spectrum i The frequency amplitude corresponding to the frequency point, i The value range is 1 to L.

[0015] Among them, the S In step 6, the new vibration data is reconstructed through the 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 the same processing steps as the training data, the sparse spectrum of the newly collected data is obtained, which is input into the trained convolutional autoencoder for reconstruction. The reconstructed spectrum is output and the reconstruction error is calculated. Then, the reconstruction error is compared with the S If the reconstruction error exceeds the warning threshold, a blade failure warning signal is issued.

[0016] The present invention also provides an electronic device, which includes 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 rotating machinery blade fault warning method based on the autoencoder described in the present invention.

[0017] The present invention also provides a computer-readable storage medium, which stores a computer program. The computer program is used to execute the rotating machinery blade fault early warning method based on the autoencoder described in the present invention. Beneficial effects 1. The method of the present invention collects signals by installing a vibration acceleration sensor on the equipment shell, without the need to directly install the sensor on the blade, thus avoiding damage to the equipment shell structure caused by drilling, ensuring the integrity and stability of the overall structure of the rotating machinery, and preventing additional interference with equipment performance due to monitoring measures, thereby reducing the possibility of increased risk of blade failure due to the installation of sensors.

[0018] 2. This invention fully preserves the sensitive components and characteristics of blade faults, enhancing early warning capabilities. By employing a wideband vibration accelerometer and performing sparse spectrum processing based on the equipment's operating mechanism and the number of blades at each level, this method effectively preserves blade fault-related characteristics, removes irrelevant signal components, and highlights signal components sensitive to blade faults, thereby improving the ability to detect early blade faults. Compared to traditional methods, this method can capture more subtle changes in fault characteristics, providing more robust data support for early warning.

[0019] 3. In the method of the present invention, the constructed convolutional autoencoder model can effectively capture the sensitive features of the blades in the sparse spectrum and reconstruct them. By comparing the difference between the spectrum before and after reconstruction, it can be judged whether the blades are faulty. When the blades are normal, the autoencoder can reconstruct the spectrum well; when the blades are faulty, the reconstructed spectrum differs significantly from the original spectrum. This method can promptly detect signs of faults such as fouling and fine cracks on the blades, providing early warning, buying more time for equipment shutdown and maintenance, reducing equipment downtime and economic losses caused by blade failures, and improving the reliability and safety of rotating machinery operations.

[0020] 4. The method of the present invention has excellent versatility and scalability. This method can be applied to various types of rotating machinery, including gas turbines, smoke exhaust fans, steam turbines, and fans, as long as blade failures manifest through casing vibration. Furthermore, with the advancement of deep learning technology, deep autoencoder models can be further optimized and upgraded, adding more layers or adopting more advanced network structures to continuously improve fault warning performance and adapt to the future development of rotating machinery with higher parameters and more complex operating conditions.

[0021] 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 device shell to collect signals, there is no need to directly install the sensor on the blade, thus avoiding damage to the device shell structure caused by drilling, avoiding damage to the device shell structure, ensuring the integrity and stability of the overall structure of the rotating machinery, and not causing additional interference to the device performance due to monitoring means, thereby reducing the possibility of increased risk of blade failure due to the installation of sensors.

[0022] 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 device shell to collect signals, there is no need to directly install the sensor on the blade, thus avoiding damage to the device shell structure by drilling, avoiding damage to the device shell structure, ensuring the integrity and stability of the overall structure of the rotating machinery, and not causing additional interference to the device performance due to monitoring means, thereby reducing the possibility of increased risk of blade failure due to the installation of sensors. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 This is a flow chart of the rotating machinery blade fault warning method based on sparse convolutional autoencoder of the present invention.

[0024] Figure 2 This is a schematic diagram of the rotating machinery blade and casing structure and sensor installation position of the present invention, showing the structural relationship between the internal blades and casing of the rotating machinery, as well as the specific installation position of the broadband vibration acceleration sensor on the casing. The vibration transmission path is illustrated with dotted lines and annotations to make the location and source of signal collection clear.

[0025] Figure 3 This is a schematic diagram of the principle of improving the frequency conversion of harmonic product spectrum calculation in the present invention. It graphically shows the process of improving the frequency conversion of harmonic product spectrum calculation to help understand the principle and method of frequency conversion calculation.

[0026] 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 characteristics.

[0027] 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 displayed in the form of blocks and lines, which facilitates understanding of the model architecture and data flow.

[0028] Figure 6 This is a schematic diagram of the loss curve of the training process of the present invention. It plots the changes of training loss and validation loss with the number of training rounds during the training process, intuitively showing the training effect and convergence of the model, as well as the position of the optimal validation point and test loss on the curve, to assist in explaining the performance evaluation of model training.

[0029] Figure 7This is a schematic diagram of the comparison between the original sparse spectrum and the reconstructed spectrum of the present invention, which shows the comparison between the original sparse spectrum and the reconstructed spectrum under normal working conditions and fault conditions, respectively, to help understand how to judge blade faults through spectral differences.

[0030] Figure 8 This is a schematic diagram of the relationship between the reconstruction error and the warning threshold of the present invention. A curve of the reconstruction error changing with time or data points is drawn, and the position of the warning threshold is marked, which intuitively shows how to perform fault warning based on the comparison between the reconstruction error and the warning threshold.

[0031] Figure 9 A schematic structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0032] The present invention is described in detail below with reference to the accompanying drawings and embodiments.

[0033] This embodiment proposes a rotating machinery blade fault warning method based on a sparse convolutional autoencoder. This method can effectively utilize relevant features containing blade faults from the collected vibration signals without destroying the original structure of the equipment housing, and sensitively provide 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 housing to collect vibration signals; accurately calculating the rotational frequency by improving the harmonic product spectrum without an additional speed monitoring sensor; calculating the theoretical value of the blade passing frequency at each level based on the number of blades at each level 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.

[0034] The following example illustrates a rotor blade failure. During normal operation, the device operates at a speed range of [8100, 9900] rpm and contains 17 stages of moving and stationary blades. The number of moving blades per stage is {16, 26, 26, 42, 45, 48, 54, 56, 64, 66, 66, 76, 76, 76, 76, 76, 76}. The following is a flowchart of the rotor blade failure warning method of this embodiment: Figure 1 The specific steps are as follows: S 1. Install a vibration acceleration sensor on the rotating equipment housing to collect vibration signals. In this embodiment, a wide-band vibration acceleration sensor with a frequency response range that meets the requirements is selected and fixed to the rotating equipment housing near the blades through threaded installation. The sensor installation diagram is as follows: Figure 2As shown in the figure, the maximum number of blades in the 17th stage is 76. At the maximum speed of 9900 rpm, the double blade pass frequency is 25080 Hz. Setting the sampling rate fs = 64000 Hz, the analysis frequency can reach 32000 Hz, exceeding the double blade pass frequency of each stage at the highest speed, ensuring that key frequency information related to blade failure can be captured. The casing vibration signal is continuously collected at the set sampling frequency, with each data segment lasting 0.256 seconds, and the number of sampling points N = 16384.

[0035] S 2. In the absence of an additional speed monitoring sensor, the speed frequency is accurately calculated by improving the harmonic product spectrum. In this embodiment, the speed frequency range is first set according to the normal operating speed range of the equipment. =[135Hz, 165Hz], select the 2-fold multiple of the rotation frequency and the number of second-stage blades as N1 and N2, that is, N1=2, N2=26; secondly, take a set of collected vibration signals as an example, calculate the spectrum of the shell vibration signal; then, retain the three spectrum intervals of the spectrum 、2* 、26* , and the rest are set to zero to obtain the interval spectrum; the interval spectrum is further upsampled by N1*N2 times, N2 times and N1 times respectively through linear interpolation to obtain three upsampled interval spectra; then the three upsampled interval spectra are point-multiplied to obtain the improved harmonic product spectrum, as shown Figure 3 Finally, for this section of the casing vibration signal, the frequency corresponding to the maximum value in the spectrum is extracted, which is the accurately calculated rotation frequency Similarly, the rotational frequency of other collected shell vibration signals can also be accurately calculated in this way.

[0036] S 3. Based on the number of blades at each level, calculate the theoretical value of the blade passing frequency at each level and obtain the sparse spectrum. In this embodiment, as mentioned above, the number of blades at each level of the device is {16, 26, 26, 42, 45, 48, 54, 56, 64, 66, 66, 76, 76, 76, 76, 76, 76}, and the number of blade stages m=17; then, considering the operating mechanism of the device, combined with the calculated rotation frequency With the known number of blades at each level, calculate the double and triple frequencies of the rotational frequency , calculate the passing frequency of blades at each level , and the blade passing frequency doubled , and obtain the set of all frequencies related to the blades mentioned above , a total of 37 frequency values; secondly, the spectrum of the original vibration signal is sparsely processed, and the frequency resolution , retain other frequency components within 3 frequency resolutions of the blade-related frequency, that is, retain the interval [ , and obtain the sparse spectrum, such as Figure 4 This processing method can effectively retain the relevant features of the blade fault, remove other interference components or redundant information such as noise, and improve the efficiency and accuracy of subsequent analysis.

[0037] S4. Construct a convolutional autoencoder to reconstruct the input sparse spectrum. Specifically: First, construct a convolutional autoencoder including the encoder part and the decoder part, such as Figure 5 As shown in Figure 2, the encoder consists of two convolutional layers. The first convolutional layer converts the input 1-channel data into 16-channel data with a kernel size of 4, a stride of 2, and a padding of 1. The second convolutional layer further converts the data into 32-channel data with similar kernel size and stride parameters. The decoder consists of two deconvolutional layers, which gradually restore the encoded 32-channel data to 1-channel reconstructed data. The constructed convolutional autoencoder structure can capture local features of the spectrum and achieve excellent reconstruction results. Next, sparse spectra are calculated from the vibration signals collected under normal operating conditions of the aforementioned equipment. These sparse spectra are used as training samples. Each sparse spectrum is 8192 points long. In this case, the training set consists of the first 200 data sets, and the remaining data sets serve as the test set. The model is built in the PyTorch framework, using the Adam optimizer, the MSE loss function, a learning rate of 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 to obtain the trained convolutional autoencoder model. During the training process, the training loss and validation loss of each epoch are recorded to observe the convergence of the model, such as Figure 6 shown.

[0038] S 5. Calculate the reconstruction error based on the normal data to obtain the blade fault warning threshold. In this embodiment, the sparse spectrum under normal working conditions is input into the trained convolutional autoencoder model to obtain the reconstructed spectrum, such as Figure 7 As shown; calculate the average error between the original sparse spectrum and the reconstructed spectrum As the reconstruction error, the reconstruction error is statistically analyzed, and 1.5 times the 95% quantile of the reconstruction error of 200 sets of training data is taken as a reasonable warning threshold. In this embodiment, the calculated warning threshold is 0.000628. Where L is the number of discrete frequency points of the sparse spectrum and the reconstructed spectrum, The original sparse spectrum i The frequency amplitude corresponding to the frequency point, To reconstruct the spectrum i The frequency amplitude corresponding to the frequency point, i The value range is 1 to L.

[0039] S6. Reconstruct the new vibration data through the trained convolutional autoencoder, calculate the reconstruction error and issue an early warning. In this embodiment, first, during the actual operation of the rotating equipment, the vibration data is collected in real time, and the rotation frequency calculation and sparse spectrum acquisition are performed according to the same steps as above. The newly obtained sparse spectrum data is input into the trained convolutional autoencoder for reconstruction, and the reconstruction error is calculated; then, the reconstruction error is compared with the S 5. If the reconstruction error exceeds the warning threshold, a blade failure warning signal is issued. In this case, the test set is followed by 200 groups, which are used as new data to input the autoencoder for reconstruction. The reconstruction error trend is as follows: Figure 8 At point 317, the reconstruction error is 0.000704, exceeding the warning threshold and issuing an alarm.

[0040] It can be seen from this embodiment that the rotating machinery blade fault warning method based on the sparse convolutional autoencoder of the present invention can effectively realize early warning of blade failure in practical applications, and provide strong guarantee for the safe and stable operation of rotating machinery such as gas turbines and smoke turbines.

[0041] The embodiment of the present application also provides an electronic device, Figure 9The structure of an electronic device provided by an embodiment of the present invention is shown. For example, the electronic device 90 may include a processor 91, a memory 92, and a transmission device 93. The processor is configured to execute the autoencoder-based rotating machinery blade fault warning method described in the above embodiment. The processor and memory may be connected via a bus or other means, with bus connection being used as an example. The transmission device may be connected to the processor and memory via a wired or wireless connection. The memory, as a non-transitory computer-readable storage medium, may be used to store non-transitory software programs, non-transitory computer executable programs, and modules, such as the program instructions / modules corresponding to the autoencoder-based rotating machinery blade fault warning method described in the embodiment of the present application. The processor executes the non-transitory software programs, instructions, and modules stored in the memory to execute various processor functions and data processing, thereby implementing the autoencoder-based rotating machinery blade fault warning method described in the above method embodiment. The memory may include a program storage area and a data storage area. The program storage area may store an operating system and at least one application required for the function; the data storage area may store data created by the processor. Furthermore, 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 a memory remote from the processor, which may be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof. The one or more modules are stored in the memory and, when executed by the processor, perform the autoencoder-based rotating machinery blade fault early warning method of the embodiment.

[0042] As another aspect, the present application also provides a computer-readable storage medium, which may be the computer-readable storage medium included in the device described in the above embodiment; or it may be a computer-readable storage medium that exists independently and is not assembled into the device. The computer-readable storage medium may be a tangible storage medium, such as a random access memory (RAM), internal 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 technical field. The computer-readable storage medium stores one or more programs, and the programs are used by one or more processors to execute the rotating machinery blade fault warning method based on the autoencoder described in the present application.

[0043] In summary, the above are only preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A rotating machinery blade fault early warning method based on an autoencoder, characterized in that: The following steps are included: S 1. Install a vibration acceleration sensor on the rotating equipment housing to collect vibration signals; S 2. Calculate the rotation frequency by improving the harmonic product spectrum; S 3. Based on the number of blades at each level, calculate the theoretical value of the blade passing frequency at each level and obtain the sparse spectrum; S 4. Construct a convolutional autoencoder to reconstruct the input sparse spectrum; S 5. Calculate the reconstruction error based on normal data to obtain the blade failure warning threshold; S 6. Reconstruct the new vibration data through the trained convolutional autoencoder, calculate the reconstruction error and issue an early warning.

2. The method according to claim 1, wherein described S 1, a vibration acceleration sensor is installed on the housing of the rotating equipment to collect vibration signals, including the following steps: Fix the broadband vibration accelerometer to the rotating equipment housing at one end of the blade through magnetic or threaded mounting. Then set the sampling rate to analyze the frequency of at least all blades at the highest speed. By doubling the frequency, ensure that the key frequency information related to the blade failure can be captured.

3. The method according to claim 1, wherein described S 2, the rotation frequency is accurately calculated by improving the harmonic product spectrum, specifically: First, set the frequency range according to the normal working speed range of the equipment , select the multiple of the rotation frequency and the number of blades as N1 and N2; secondly, calculate the spectrum of the shell vibration signal; then, retain the three spectrum intervals 、N1* 、N2* , and the rest are set to zero to obtain the interval spectrum; further, the interval spectrum is upsampled by N1*N2 times, N2 times and N1 times respectively through linear interpolation to obtain three upsampled interval spectra; then the three upsampled interval spectra are point-multiplied to obtain the improved harmonic product spectrum; finally, the frequency corresponding to the maximum value in the spectrum is extracted, which is the accurately calculated conversion frequency .

4. The method according to claim 1, wherein described S In 3, based on the number of blades at each level, the theoretical value of the blade passing frequency at each level is calculated, and the sparse spectrum is obtained, specifically: First, record the number of blades at each level according to the structure or design information of the equipment. , m is the number of blades; then combined with the calculated rotation frequency With the known number of blades at each level, calculate the double and triple frequencies of the rotational frequency , calculate the passing frequency of blades at each level , and the blade passing frequency doubled , and obtain the set of all frequencies related to the blades mentioned above ; The spectrum of the original vibration signal is sparsely processed, and other frequency components within 3 frequency resolutions of the blade-related frequency are retained, that is, the retention interval [ , and obtain the sparse spectrum, where Indicates frequency resolution.

5. The method according to claim 1, wherein In S4, a convolutional autoencoder is constructed to perform reconstruction training on the input sparse spectrum, specifically: First, we construct a convolutional autoencoder, which includes an encoder and a decoder. The encoder has two one-dimensional convolutional layers. The convolution kernel size of each layer is set to 4, the stride is 2, and the padding is 1. The first layer has 1 input channel and 16 output channels. The ReLU activation function is used to introduce nonlinear factors. This layer performs preliminary feature extraction on the input data. The second layer has 16 input channels and the number of output channels is increased to 32. The ReLU activation function is used for nonlinear transformation. The decoder consists of two deconvolution layers and activation functions. The convolution kernel size of each layer is 4, the stride is 2, and the padding is 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. The ReLU activation function is used for nonlinear transformation. This layer upsamples the 32-channel features extracted by the encoder and gradually restores the dimension of the data; the second layer has 16 input channels and 1 output channel. The data in this layer is reconstructed into the same form as the original input data dimension.

6. The method according to any one of claims 1 to 5, wherein: described S In 5, the reconstruction error is calculated based on the training data to obtain the blade failure warning threshold, which is specifically: The sparse spectrum under normal working conditions is input into the trained convolutional autoencoder model to obtain the reconstructed spectrum, and the average error between the original sparse spectrum and the reconstructed spectrum is calculated. As the reconstruction error, the reconstruction error is statistically analyzed, and 1.5 times the 95% confidence interval of the reconstruction error is taken as a reasonable warning threshold, where L is the number of discrete frequency points of the sparse spectrum and the reconstructed spectrum, The original sparse spectrum i The frequency amplitude corresponding to the frequency point, To reconstruct the spectrum i The frequency amplitude corresponding to the frequency point, i The value range is 1 to L.

7. The method according to any one of claims 1 to 5, wherein: described S In step 6, the new vibration data is reconstructed through the 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 the same processing steps as the training data, the sparse spectrum of the newly collected data is obtained, which is input into the trained convolutional autoencoder for reconstruction. The reconstructed spectrum is output and the reconstruction error is calculated. Then, the reconstruction error is compared with the S If the reconstruction error exceeds the warning threshold, a blade failure warning signal is issued.

8. 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 used to read the executable instructions from the memory and execute the instructions to implement the rotating machinery blade fault warning method based on the autoencoder as described in any one of claims 1 to 7.

9. A computer-readable storage medium, characterized in that The storage medium stores a computer program, and the computer program is used to execute the rotating machinery blade fault early warning method based on the autoencoder according to any one of claims 1 to 7.

Images