Moving part fault diagnosis method based on orthogonal sound pressure
By using an orthogonal sound pressure-based fault diagnosis method, sound pressure signals from moving parts are collected and processed. Faults are then identified using signal reconstruction and a fault diagnosis network, solving the problem of high cost and achieving low-cost fault diagnosis.
Patent Information
- Application Number
- CN202511168324.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-08-20
- Publication Date
- 2025-11-18
AI Technical Summary
Existing methods for diagnosing faults in moving parts involve high costs for sensors, acquisition devices, and processing equipment. Furthermore, the transmission, storage, and processing of vibration signals involve a large amount of redundant data, which consumes hardware and communication resources.
A fault diagnosis method based on orthogonal sound pressure is adopted. The sound pressure analog signals of three mutually orthogonal channels are collected at the acquisition point, converted into square wave encoded signals for transmission, restored into pseudo sound pressure signals using a signal restoration network, and then used for fault identification through a fault diagnosis network.
It reduces the cost of sensors and data acquisition equipment, reduces data redundancy, and enables fault diagnosis of moving parts at extremely low cost.
Smart Images
Figure CN120971024A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of fault diagnosis technology, and in particular to a method for diagnosing faults in moving parts based on orthogonal sound pressure. Background Technology
[0002] The current mainstream method for fault diagnosis of rotating shafts and gears is to attach acceleration sensors to key parts of the supporting structure, collect vibration data through a professional data acquisition circuit, and finally process the signals through preprocessing and corresponding fault diagnosis algorithms to obtain the fault diagnosis results.
[0003] However, this process has certain drawbacks: Accelerometers require high dynamic performance and high linearity, capable of covering a frequency response range of 50Hz-3kHz, and these sensors are expensive; dedicated signal acquisition circuits often require high-precision, high-sampling-rate ADC components, with sampling frequencies typically above 25KSps and multiple sampling channels, making these acquisition circuits expensive; the transmission, storage, and processing of vibration signals often involve a large amount of redundant data, consuming hardware and communication resources. To reduce the cost of some fault diagnosis, some studies have used other indirect coupling methods such as acoustic-vibration coupling for fault diagnosis testing, such as using piezoelectric ceramic microphones to acquire sound data coupled with fault information, but the subsequent acquisition and processing costs remain high. Summary of the Invention
[0004] Based on the above analysis, the present invention aims to provide a method for diagnosing moving parts faults based on orthogonal sound pressure, in order to solve the problem of high cost of sensors, acquisition devices, and processing devices in existing moving parts fault diagnosis.
[0005] This invention provides a method for diagnosing faults in moving parts based on orthogonal sound pressure, comprising the following steps:
[0006] The signal acquisition process includes: acquiring the sound pressure analog signals of three mutually positively connected channels of the moving part to be diagnosed at at least one acquisition point; and processing the sound pressure analog signals of each channel at each acquisition point to obtain the square wave encoded signals of each channel at each acquisition point.
[0007] The signal processing procedure includes: obtaining the pseudo sound pressure signal of each channel at each acquisition point based on the square wave encoded signal of each channel at each acquisition point; and obtaining the fault diagnosis result of the moving part to be diagnosed based on the pseudo sound pressure signal of each channel at each acquisition point and the fault diagnosis network.
[0008] Furthermore, the square wave encoded signals for each channel at each acquisition point are generated separately using the following methods:
[0009] The analog sound pressure signal of each channel at the current acquisition point is filtered and amplified to obtain a bandpass amplified signal group for each channel at the current acquisition point; wherein, the bandpass amplified signal group includes multiple bandpass amplified signals with different passbands;
[0010] Square wave conditioning is performed on each bandpass amplified signal in each channel of the current acquisition point to obtain square wave signals for each channel of the current acquisition point.
[0011] Encode the square wave signals of each channel at the current acquisition point to obtain the square wave encoded signals of each channel at the current acquisition point.
[0012] Furthermore, the pseudo-sound pressure signal for each channel at each acquisition point is obtained in the following manner:
[0013] After decoding the square wave encoded signals of each channel at each acquisition point, the pseudo-analog signals of each channel at each acquisition point are obtained by passing them through a trained signal reconstruction network.
[0014] The pseudo-analog signals of each channel at each acquisition point are combined to obtain the pseudo sound pressure signal of each channel at each acquisition point.
[0015] Furthermore, the trained signal reconstruction network is obtained in the following way:
[0016] Based on a set time interval, the sound pressure simulation signals of three mutually positive communication channels at each sampling point of the moving part to be diagnosed are collected;
[0017] Based on the simulated sound pressure signals of the three mutually positively connected channels at each sampling point at each acquisition time, the square wave signals of each channel at each acquisition point after decoding at each acquisition time are obtained, and then a sample set is constructed; wherein, one sample is one square wave signal.
[0018] The signal reconstruction network is trained based on the sample set to obtain a trained signal reconstruction network.
[0019] Furthermore, the loss function S used when training the signal reconstruction network is expressed as:
[0020]
[0021] In the formula, V i V represents the voltage at the i-th edge of the pseudo-analog signal output by the signal reconstruction network. kLet represent the voltage at the k-th polarity error moment of the pseudo-analog signal output by the signal restoration network; F represents the hyperparameter matrix of the signal restoration network; α and β represent the first and second weighting coefficients, respectively; N represents the total number of edge moments of the pseudo-analog signal output by the signal restoration network; K represents the total number of polarity error moments of the pseudo-analog signal output by the signal restoration network; ||2 represents the 2-norm; and || represents the absolute value.
[0022] Furthermore, the signal restoration network employs an Encoder-Decoder CNN neural network.
[0023] Furthermore, the fault diagnosis results are obtained through the following methods:
[0024] Wavelet transform is performed on the pseudo sound pressure signal of each channel at each acquisition point to obtain the time-spectrum two-dimensional signal image of each channel at each acquisition point.
[0025] The time-spectrum two-dimensional signal image of each channel at each acquisition point is input into the trained fault diagnosis network to obtain the fault diagnosis result of the moving part to be diagnosed.
[0026] Furthermore, the analog sound pressure signal of each channel is input into multiple bandpass amplifier circuits with different passbands for filtering and amplification to obtain multiple bandpass amplified signals with different passbands; wherein, the range of each passband is determined according to the range of fault characteristic signals, and the fault characteristics are signal characteristics containing fault information.
[0027] Furthermore, each bandpass amplified signal is compared with the set reference voltage. If it is greater than the reference voltage, "1" is output; if it is less than the reference voltage, "0" is output, thus generating a square wave signal for each bandpass amplified signal.
[0028] Furthermore, the fault diagnosis network is trained in the following manner:
[0029] Construct a fault sample set, wherein each sample includes the pseudo sound pressure signal of each channel at each acquisition point, as well as the corresponding fault result and fault type;
[0030] Based on the fault sample set, the fault diagnosis network is trained to minimize the diagnosis loss function, resulting in a trained fault diagnosis network. The diagnosis loss function is set as the cross-entropy loss function.
[0031] Compared with the prior art, the present invention can achieve at least one of the following beneficial effects:
[0032] This invention provides a method for fault diagnosis of moving parts based on orthogonal sound pressure. In terms of signal acquisition, the method converts the acquired analog sound pressure signals from three mutually orthogonal channels into corresponding square wave encoded signals before transmission, replacing the traditional high-speed ADC for acquiring and encoding the acceleration analog signals. This saves ADC costs and reduces data redundancy, thus lowering acquisition costs at the signal acquisition level. In terms of signal processing, the received square wave signals are restored to analog signals through a signal restoration network to obtain pseudo-sound pressure signals. These orthogonal pseudo-sound pressure signals are then processed through a fault diagnosis network to identify faults, enabling fault diagnosis of moving parts at extremely low cost.
[0033] In this invention, the above-described technical solutions can be combined with each other to achieve more preferred combinations. Other features and advantages of this invention will be set forth in the following description, and some advantages may become apparent from the description or be learned by practicing the invention. The objects and other advantages of this invention can be realized and obtained from what is particularly pointed out in the description and drawings. Attached Figure Description
[0034] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts.
[0035] Figure 1 A flowchart illustrating the moving part fault diagnosis method based on orthogonal sound pressure provided in an embodiment of the present invention;
[0036] Figure 2 This is a schematic diagram illustrating the specific processing flow of the moving part fault diagnosis method based on orthogonal sound pressure provided in an embodiment of the present invention. Detailed Implementation
[0037] Preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, which form part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not intended to limit the scope of the present invention.
[0038] A specific embodiment of the present invention discloses a method for diagnosing faults in moving parts based on orthogonal sound pressure, such as... Figure 1 As shown, it includes the following steps:
[0039] The signal acquisition process includes:
[0040] S11. Acquire the sound pressure analog signals of three mutually positively connected channels of the moving part to be diagnosed at at least one acquisition point; wherein, the moving part to be diagnosed may be a rotating shaft, gear, etc. in a mechanical transmission system;
[0041] S12. After processing the analog sound pressure signals of each channel at each acquisition point, the square wave encoded signals corresponding to the analog sound pressure signals of each channel at each acquisition point are obtained.
[0042] Signal processing includes:
[0043] S21. Based on the square wave encoded signals corresponding to the sound pressure analog signals of each channel at each acquisition point, obtain the pseudo sound pressure signals of each channel at each acquisition point.
[0044] S22. Based on the pseudo sound pressure signal of each channel at each acquisition point and the fault diagnosis network, the fault diagnosis result of the moving part to be diagnosed is obtained.
[0045] Compared with existing technologies, the fault diagnosis method provided in this embodiment converts the collected sound pressure analog signals from three mutually orthogonal channels into corresponding square wave coded signals before transmission, replacing the traditional high-speed ADC for collecting and encoding acceleration analog signals. This saves ADC costs and reduces data redundancy, thereby reducing acquisition costs from the signal acquisition level. In terms of signal processing, the received square wave signals are restored to analog signals through a signal restoration network to obtain pseudo sound pressure signals. The orthogonal pseudo sound pressure signals are then passed through a fault diagnosis network to identify faults, enabling fault diagnosis of moving parts at extremely low cost.
[0046] like Figure 2 The diagram shown is a schematic of the specific processing flow for fault diagnosis in this embodiment.
[0047] Specifically, in step S1, by setting three electret microphones with mutually orthogonal directions at the acquisition point to acquire the sound pressure simulation signals of three mutually orthogonal channels, it is possible to acquire the sound pressure generated by acoustic vibration coupling of moving parts at low cost.
[0048] Specifically, at least one acquisition point is set in the acquisition area of the moving part to be diagnosed, and the acquisition area is the area within the range where fault coupling signals can be received.
[0049] More specifically, the number of sampling points is set according to the specific situation, and the electret microphone uses a piezoelectric ceramic microphone.
[0050] Specifically, the length of the sound pressure analog signal is set according to actual needs.
[0051] It is understandable that this embodiment utilizes the acoustic-vibration coupling characteristics and orthogonally collects sound pressure through three low-cost electret microphones, achieving single-bit sampling of multi-frequency orthogonal acoustic-vibration signals. This replaces traditional vibration acceleration signal sampling with sound pressure sampling, significantly reducing the cost of the sensor. From the perspective of sensor type, it reduces the cost and deployment difficulty of collecting fault signals from moving parts.
[0052] During implementation, in step S12, the square wave encoded signals for each channel at each acquisition point are generated respectively in the following manner:
[0053] S121. Filter and amplify the analog sound pressure signal of each channel at the current acquisition point to obtain a bandpass amplified signal group for each channel at the current acquisition point; wherein, the bandpass amplified signal group includes multiple bandpass amplified signals with different passbands.
[0054] In practice, the sound pressure analog signal of each channel is input into multiple bandpass amplifier circuits with different passbands for filtering and amplification to obtain multiple bandpass amplified signals with different passbands. Among them, the range of each passband is determined according to the range of fault characteristic signals, and the fault characteristics are signal characteristics containing fault information.
[0055] Specifically, the bandpass amplifier circuit uses an LC filter circuit.
[0056] It should be noted that in this embodiment, the passband range is designed according to different fault characteristic signal ranges, so that multiple different characteristic frequencies in the sound pressure analog signal are used as the main frequencies of the sub-signals for separate output, so that the subsequent square wave conditioning can obtain its information and will not be overwhelmed by higher power signals; wherein, the fault characteristic is a signal characteristic containing fault information, mainly reflected in the amplitude change of a certain spectrum, and the fault characteristic signal range is set according to the actual working conditions.
[0057] For example, typical characteristic frequencies are as follows:
[0058] Rotational frequency refers to the frequency corresponding to one revolution of a rotating shaft. It is equal to the shaft's rotational speed divided by 60. For example, a shaft running at 3000 rpm has a rotational frequency of 50 Hz.
[0059] Bearing failure frequencies include inner ring failure frequencies, outer ring failure frequencies, rolling element failure frequencies, and cage failure frequencies. These frequencies depend on the bearing geometry and rotational speed. When bearing components are damaged, vibration peaks will appear at the corresponding failure frequencies.
[0060] Gear failure frequency involves the tooth pass frequency, which is the frequency at which each tooth passes through the meshing point, as well as the rotational frequency and harmonics of the gear shaft; gear failures usually produce obvious vibrations at the tooth pass frequency and its harmonics.
[0061] Motor electromagnetic frequency: The electric motor also has a vibration frequency caused by electromagnetic force, which is related to the power supply frequency (such as 50Hz or 60Hz), the number of pole pairs and the slip rate.
[0062] Coupling misalignment frequency: When the coupling is misaligned, specific fault frequencies will occur. These frequencies are usually multiples of the shaft rotation frequency.
[0063] For example, consider a bearing:
[0064] Low-frequency faults: Bearing low-frequency faults mainly refer to problems that occur when the speed is below 1000 rpm and the vibration frequency is in the range of 1 to 200 Hz. Common manifestations include humming sounds or low-frequency vibrations from the bearing, which leads to decreased machine efficiency, increased noise, or even machine shutdown. The causes of low-frequency faults include bearing looseness, external force vibration, bearing friction, poor lubrication, and other aspects. Solutions include checking the bearing suspension, reducing external interference, and timely replacement of lubricating oil.
[0065] Medium-frequency failure: Medium-frequency failure of bearings usually occurs when the speed is between 1000 and 10000 rpm and the vibration frequency is between 200 and 1000 Hz. It manifests as slight vibration, noise or temperature rise of the bearing, and may cause bearing damage in severe cases. The main causes of medium-frequency failure include insufficient bearing support stiffness, excessive inner and outer ring clearance, shaft misalignment, and imbalance. The solutions include optimizing bearing support, adjusting inner and outer ring clearance, reducing shaft misalignment, and balancing the mechanical structure.
[0066] High-frequency failures: High-frequency bearing failures occur at frequencies between 10,000 and 20,000 rpm, with vibration frequencies between 1,000 and 5,000 Hz. They are usually manifested as noticeable bearing vibration and noise. The main causes of high-frequency failures include excessive bearing wear, oil film rupture, excessive load, or overload. Solutions include replacing the bearing with a suitable one, controlling the load, and improving the quality of the oil.
[0067] For example, if three passbands are designed, the three channels of the sound pressure analog signal at one acquisition point will generate nine bandpass amplified signals.
[0068] S122. Perform square wave conditioning on each bandpass amplified signal in each channel of the current acquisition point to obtain square wave signals for each channel of the current acquisition point.
[0069] In practice, each bandpass amplified signal is compared with the set reference voltage. If it is greater than the reference voltage, "1" is output; if it is less than the reference voltage, "0" is output, thus generating a square wave signal for each bandpass amplified signal.
[0070] Specifically, the reference voltage is determined based on the specific fault characteristics of the moving part to be diagnosed.
[0071] Specifically, in this embodiment, square wave conditioning of each bandpass amplified signal is performed using multiple voltage comparators and a reference voltage source.
[0072] S213. Encode the square wave signals of each channel at the current acquisition point to obtain the square wave encoded signals of each channel at the current acquisition point.
[0073] In practice, the square wave encoded signal includes the time-encoded information of the square wave signal edges. The encoding of each square wave signal is performed using edge interrupts from the MCU's I / O ports in conjunction with an internal timer, including the following steps:
[0074] a1. Configure the MCU's I / O port to input mode and set it to edge-triggered interrupt; the edge includes rising edge and falling edge; initialize the internal timer, set it to continuous counting mode, and start the timer;
[0075] a2. When the edge of the square wave signal triggers an interrupt, the interrupt service function is entered, and the current internal timer value is recorded as T1.
[0076] a3. When the next square wave edge is triggered, enter the interrupt service function and record the current internal timer value as T2.
[0077] a4. Calculate the time difference T2-T1 between the two edges, take the high 15 bits of its binary representation as the code Q, and record the current level state to store whether this edge is a rising edge or a falling edge.
[0078] a5. Repeat steps a2 to a5 until the encoding of the current square wave signal is completed.
[0079] Specifically, the Q encoding can reduce the number of bits for high-frequency signals and reduce the number of bits for low-frequency signals by decreasing the timer clock, thus achieving the effect of data compression.
[0080] Understandably, by passing the three-channel orthogonal acoustic signals from each acquisition point through multiple bandpass filters, amplifications, and voltage comparators to generate a square wave that can be represented by a single bit, and continuously acquiring the timing information of the square wave edges using I / O ports, the acquisition cost is reduced at the signal acquisition level. The encoded data has low redundancy, allowing for transmission and storage while reducing resource consumption. It should be noted that the acquired signals need to be transmitted to high-computing devices such as servers and computers for further processing; encoding them can reduce the data size.
[0081] It should be noted that the simulated sound pressure signals from the three interconnected positive communication channels of the moving part to be diagnosed are collected and processed, and then transmitted to high-computing-power devices such as servers or computers as the data basis for subsequent fault diagnosis.
[0082] During implementation, in step S21, the pseudo sound pressure signal of each channel at each acquisition point is obtained in the following manner:
[0083] S211. After decoding the square wave encoded signals of each channel of each acquisition point, obtain the pseudo-analog signals of each channel of each acquisition point through the trained signal restoration network.
[0084] During implementation, the square wave encoded signals of each channel at each acquisition point are decoded to obtain the square wave signals of each channel at each acquisition point; the square wave signals of each channel at each acquisition point are then passed through a trained signal restoration network to obtain the corresponding pseudo-analog signals of each channel at each acquisition point.
[0085] In practice, the signal restoration network uses an Encoder-Decoder CNN neural network.
[0086] Specifically, the trained signal reconstruction network is obtained in the following way:
[0087] Based on a set time interval, the sound pressure simulation signals of three mutually positive communication channels at each sampling point of the moving part to be diagnosed are collected;
[0088] Based on the simulated sound pressure signals of the three mutually positively connected channels at each sampling point at each acquisition time, the square wave signals of each channel at each acquisition point after decoding at each acquisition time are obtained, and then a sample set is constructed; wherein, one sample is one square wave signal.
[0089] The signal reconstruction network is trained based on the sample set to obtain a trained signal reconstruction network.
[0090] Specifically, the loss function used when training the signal reconstruction network is expressed as:
[0091]
[0092] In the formula, V i V represents the voltage at the i-th edge of the pseudo-analog signal output by the signal reconstruction network. k Let represent the voltage at the k-th polarity error moment of the pseudo-analog signal output by the signal restoration network; F represents the hyperparameter matrix of the signal restoration network; α and β represent the first and second weighting coefficients, respectively; N represents the total number of edge moments of the pseudo-analog signal output by the signal restoration network; K represents the total number of polarity error moments of the pseudo-analog signal output by the signal restoration network; ||2 represents the 2-norm; and || represents the absolute value.
[0093] Specifically, the edge time of the pseudo-analog signal is the zero-crossing point of the pseudo-analog signal voltage minus the reference voltage; when the polarity error is represented by a square wave, the restored pseudo-analog signal that should be high level is low level or should be low level is high level, because the square wave can be 1 or 0. If the restored voltage is not greater than the reference voltage when it is 1, or if the restored voltage is not less than the reference voltage when it is 0, then it is a polarity error.
[0094] It should be noted that the signal reconstruction network used in this embodiment has the ability to understand and reconstruct frequency signals. It employs three physically meaningful loss functions for unsupervised learning of the pseudo-analog signal reconstruction, achieving the reconstruction solution of the pseudo-signal under underdetermined conditions. This provides a data foundation for subsequent fault diagnosis. Therefore, in this embodiment, the sample set of the moving part to be diagnosed is first obtained. After training the signal reconstruction network, fault diagnosis is performed according to the diagnostic method described in this embodiment. It is understandable that after the moving part to be diagnosed acquires signals, the signals need to be transmitted to high-computing devices such as servers or computers. Traditional methods using dedicated signal acquisition circuits for acquisition and encoding often require high-precision, high-sampling-rate ADC components and numerous sampling channels, resulting in high acquisition circuit costs. In this embodiment, by converting the acquired sound pressure signal into a square wave before encoding and transmission, the acquisition cost and signal data redundancy are reduced. After the signal processing device receives the encoded square wave signal, it decodes the corresponding square wave signal for subsequent processing. Encoding and decoding reduce the data size during transmission, lowering costs and improving efficiency.
[0095] It should be noted that the signal restoration network training in this embodiment is based on the following theory:
[0096] First, a random signal of the same dimension as the signal to be decomposed is input. Then, the signal output from the ED neural network is multiplied by the sampling matrix of the edge points of the square wave. The result should be a zero matrix (where the sampling matrix is a tensor matrix with 1s only at the edge positions and 0s elsewhere). In addition, the frequency domain of the signal should satisfy the bandwidth dilution characteristic. Therefore, the loss function of the ED network is constructed based on these two prior knowledge.
[0097] It is understandable that the CNN network of Encoder-Decoder in this embodiment has a natural ability to understand and reconstruct frequency signals. It uses three loss functions with physical meaning to perform unsupervised learning on the reconstruction of pseudo signals, so as to realize the reconstruction solution of pseudo signals under underdetermined conditions and provide a data foundation for subsequent fault diagnosis.
[0098] S212. Combine the pseudo-analog signals of each channel at each acquisition point to obtain the pseudo sound pressure signal of each channel at each acquisition point.
[0099] Specifically, the pseudo-analog signals from each channel at each acquisition point are superimposed and merged to obtain the corresponding pseudo-sound pressure signal.
[0100] During implementation, in step S22, the fault diagnosis result is obtained in the following manner:
[0101] Wavelet transform is performed on the pseudo sound pressure signal of each channel at each acquisition point to obtain the time-spectrum two-dimensional signal image of each channel at each acquisition point.
[0102] The time-spectrum two-dimensional signal image of each channel at each acquisition point is input into the trained fault diagnosis network to obtain the fault diagnosis result of the moving part to be diagnosed.
[0103] Specifically, the fault diagnosis results include whether there is a fault, and if so, the fault type.
[0104] Specifically, the fault diagnosis network is trained in the following manner:
[0105] Construct a fault sample set, wherein each sample includes the pseudo sound pressure signal of each channel at each acquisition point, as well as the corresponding fault result and fault type;
[0106] Based on the fault sample set, the fault diagnosis network is trained to minimize the diagnosis loss function, resulting in a trained fault diagnosis network. The diagnosis loss function is set as the cross-entropy loss function.
[0107] More specifically, different moving parts have different fault types. For example, when the moving part to be diagnosed is a bearing, the fault type can be inner ring fault, outer ring fault, rolling element fault, or cage fault.
[0108] Specifically, the fault diagnosis network uses a CNN neural network, which includes an encoding network and a fully connected network.
[0109] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware, and the program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.
[0110] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.
Claims
1. A method for diagnosing faults in moving parts based on orthogonal sound pressure, characterized in that, Includes the following steps: The signal acquisition process includes: acquiring the sound pressure analog signals of three mutually positively connected channels of the moving part to be diagnosed at at least one acquisition point; and processing the sound pressure analog signals of each channel at each acquisition point to obtain the square wave encoded signals of each channel at each acquisition point. The signal processing procedure includes: obtaining the pseudo sound pressure signal of each channel at each acquisition point based on the square wave encoded signal of each channel at each acquisition point; and obtaining the fault diagnosis result of the moving part to be diagnosed based on the pseudo sound pressure signal of each channel at each acquisition point and the fault diagnosis network.
2. The method for diagnosing moving parts based on orthogonal sound pressure as described in claim 1, characterized in that, The square wave encoded signals for each channel at each acquisition point are generated using the following methods: The analog sound pressure signal of each channel at the current acquisition point is filtered and amplified to obtain a bandpass amplified signal group for each channel at the current acquisition point; wherein, the bandpass amplified signal group includes multiple bandpass amplified signals with different passbands; Square wave conditioning is performed on each bandpass amplified signal in each channel of the current acquisition point to obtain square wave signals for each channel of the current acquisition point. Encode the square wave signals of each channel at the current acquisition point to obtain the square wave encoded signals of each channel at the current acquisition point.
3. The method for diagnosing moving parts based on orthogonal sound pressure as described in claim 1, characterized in that, The pseudo sound pressure signal for each channel at each acquisition point was obtained in the following manner: After decoding the square wave encoded signals of each channel at each acquisition point, the pseudo-analog signals of each channel at each acquisition point are obtained by passing them through a trained signal reconstruction network. The pseudo-analog signals of each channel at each acquisition point are combined to obtain the pseudo sound pressure signal of each channel at each acquisition point.
4. The method for diagnosing moving parts based on orthogonal sound pressure as described in claim 3, characterized in that, The trained signal reconstruction network is obtained through the following method: Based on a set time interval, the sound pressure simulation signals of three mutually positive communication channels at each sampling point of the moving part to be diagnosed are collected; Based on the simulated sound pressure signals of the three mutually positively connected channels at each sampling point at each acquisition time, the square wave signals of each channel at each acquisition point after decoding at each acquisition time are obtained, and then a sample set is constructed; wherein, one sample is one square wave signal. The signal reconstruction network is trained based on the sample set to obtain a trained signal reconstruction network.
5. The method for diagnosing moving parts based on orthogonal sound pressure according to claim 4, characterized in that, The loss function S used when training the signal reconstruction network is expressed as: In the formula, V i V represents the voltage at the i-th edge of the pseudo-analog signal output by the signal reconstruction network. k Let represent the voltage at the k-th polarity error moment of the pseudo-analog signal output by the signal restoration network, F represent the hyperparameter matrix of the signal restoration network, α and β represent the first and second weighting coefficients respectively, N represent the total number of edge moments of the pseudo-analog signal output by the signal restoration network, K represent the total number of polarity error moments of the pseudo-analog signal output by the signal restoration network, ||||2 represents the 2-norm, and || represents the absolute value.
6. The method for diagnosing moving parts based on orthogonal sound pressure according to claim 4, characterized in that, The signal restoration network uses an Encoder-Decoder CNN neural network.
7. The method for diagnosing moving parts based on orthogonal sound pressure as described in claim 4, characterized in that, The fault diagnosis results are obtained through the following methods: Wavelet transform is performed on the pseudo sound pressure signal of each channel at each acquisition point to obtain the time-spectrum two-dimensional signal image of each channel at each acquisition point. The time-spectrum two-dimensional signal image of each channel at each acquisition point is input into the trained fault diagnosis network to obtain the fault diagnosis result of the moving part to be diagnosed.
8. The method for diagnosing moving parts based on orthogonal sound pressure according to claim 2, characterized in that, The analog sound pressure signal of each channel is input into multiple bandpass amplifier circuits with different passbands for filtering and amplification to obtain multiple bandpass amplified signals with different passbands; among them, the range of each passband is determined according to the range of fault characteristic signals, and the fault characteristics are signal characteristics containing fault information.
9. The method for diagnosing moving parts based on orthogonal sound pressure according to claim 2, characterized in that, Each bandpass amplified signal is compared with the set reference voltage. If it is greater than the reference voltage, "1" is output; if it is less than the reference voltage, "0" is output, thus generating a square wave signal for each bandpass amplified signal.
10. The method for diagnosing moving parts based on orthogonal sound pressure according to claim 7, characterized in that, The fault diagnosis network is trained using the following method: Construct a fault sample set, wherein each sample includes the pseudo sound pressure signal of each channel at each acquisition point, as well as the corresponding fault result and fault type; Based on the fault sample set, the fault diagnosis network is trained to minimize the diagnosis loss function, resulting in a trained fault diagnosis network. The diagnosis loss function is set as the cross-entropy loss function.