Signal processing method, monitoring method, device and equipment of rotating equipment and medium

Abstract

The application provides a signal processing method, a rotating equipment monitoring method, device, equipment and medium, wherein the signal processing method comprises: obtaining vibration data in a running state of a rotating equipment, the vibration data comprising a time domain signal and a revolution phase sequence; performing equiangular phase inverse mapping on the revolution phase sequence to obtain a target time sequence corresponding to a uniform angle; resampling the time domain signal according to the target time sequence by using a limited support band-limited reconstruction kernel to obtain an angular domain signal; performing angular domain spectrum estimation on the angular domain signal to obtain an angular domain spectrum; and extracting a target order amplitude based on the angular domain spectrum. Through equiangular phase inverse mapping, the requirement of angular domain uniform sampling is converted into sampling at specified target time points in the time domain. Under the premise of controlling the amount of calculation by using a limited support, resampling is performed by using a band-limited reconstruction kernel to obtain an angular domain signal with high fidelity in sampling structure and amplitude, thereby ensuring high precision and high robustness of order tracking results.

Description

Technical Field

[0001] This invention relates to the field of rotating equipment monitoring technology, specifically to a signal processing method, a method, apparatus, equipment, and medium for monitoring rotating equipment. Background Technology

[0002] Rotating equipment, such as motors, gearboxes, fans, and bearings, is widely used in industrial production, and the stability of its operating state directly affects production efficiency and safety. For monitoring the condition of rotating equipment under variable speed conditions, order tracking is often used to analyze its vibration signals. Order tracking extracts the order amplitude value proportional to the rotational speed from the vibration signal of the rotating equipment. This order amplitude value is then used to determine whether the equipment has faults such as rotor imbalance, shaft asymmetry, gear meshing wear, or abnormal electromagnetic harmonics. Furthermore, it can analyze the trend of order amplitude changes, enabling early identification of equipment faults and condition monitoring.

[0003] Existing order tracking techniques often rely on the Volk-Kalman tracking filter method or the computed order tracking (COT) method, which involves angular domain resampling and spectral estimation in the angular domain. However, these methods can lead to inaccurate extracted order amplitudes at high speeds or under conditions of rapid speed changes.

[0004] Therefore, how to improve the accuracy of order amplitude extraction in high-speed range or under conditions of rapid speed change has become an urgent technical problem to be solved. Summary of the Invention

[0005] In view of this, embodiments of the present invention provide a signal processing method, a monitoring method, apparatus, device, and medium for rotating equipment, to solve the technical problem of how to improve the accuracy of order tracking in extracting order amplitudes at high speeds or under conditions of rapid speed changes.

[0006] In a first aspect, embodiments of the present invention provide a signal processing method, comprising: acquiring vibration data of a rotating device in operation, the vibration data including a time-domain signal and a rotational phase sequence, wherein the rotational phase sequence is used to characterize a sequence of physical quantities representing the change of the angle of the rotating device over time; performing an isoangular phase inverse mapping on the rotational phase sequence to obtain a target time sequence corresponding to a uniform angle; resampling the time-domain signal according to the target time sequence using a finite-support band-limited reconstruction kernel to obtain an angular domain signal; performing angular domain spectrum estimation on the angular domain signal to obtain an angular domain spectrum; and extracting a target order amplitude based on the angular domain spectrum.

[0007] In one embodiment, before resampling the time-domain signal according to the target time series using a finite support band-limited reconstruction kernel, the method further includes: calculating the instantaneous rotational speed change rate based on the rotational speed phase sequence; and dynamically adjusting the support window length of the finite support band-limited reconstruction kernel based on the instantaneous rotational speed change rate, wherein the support window length is negatively correlated with the absolute value of the instantaneous rotational speed change rate.

[0008] In one embodiment, resampling the time-domain signal according to the target time series using a finite-support band-limited reconstruction kernel to obtain an angular domain signal includes: selecting a finite number of neighboring sampling points around each target time point in the target time series, wherein the neighboring sampling points are the original sampling points of the time-domain signal, and the target time point is used to characterize the discrete sampling time point in the time domain corresponding to the rotation phase in the rotation phase sequence when it reaches a set uniform angle; calculating the time difference between each neighboring sampling point and the corresponding target time point; determining the weight of the neighboring sampling points based on the finite-support band-limited reconstruction kernel and the time difference, wherein the weight is negatively correlated with the time difference; weighting and summing the signal value of the time-domain signal corresponding to each neighboring sampling point with the corresponding weight to obtain the angular domain signal value corresponding to the target time point; and using the angular domain signal values ​​corresponding to all the target time points as the angular domain signal.

[0009] In one embodiment, determining the weight of the neighborhood sampling point based on the finite support band-limited reconstruction kernel and the time difference includes: obtaining the weight by querying a pre-built weight query model through the time difference, wherein the weight query model is a correspondence model between different time differences and weights generated based on the finite support band-limited reconstruction kernel.

[0010] In one embodiment, the step of estimating the angular domain spectrum of the angular domain signal to obtain the angular domain spectrum includes: extracting the length features of the angular domain signal; and performing adaptive boundary angular domain spectrum estimation of the angular domain signal based on the length features to obtain the angular domain spectrum.

[0011] In one embodiment, the length feature includes the total length of the angular domain signal; the step of performing adaptive boundary angular domain spectrum estimation on the angular domain signal based on the length feature to obtain the angular domain spectrum includes: obtaining the total length of the angular domain signal; performing adaptive boundary processing on the angular domain signal based on the total length to obtain a processing result; and performing a short-time Fourier transform on the processing result to obtain the angular domain spectrum.

[0012] In one embodiment, the adaptive boundary processing of the angular domain signal based on the total length to obtain a processing result includes: determining whether the total length is less than a preset length, wherein the preset length is α times the window length of the short-time Fourier transform, and α is a constant greater than 1; in response to the total length being less than the preset length, performing a boundary extension operation on the angular domain signal; using the angular domain signal processed by the boundary extension operation as the processing result; and in response to the total length being greater than or equal to the preset length, using the angular domain signal as the processing result.

[0013] In one embodiment, the method for determining the preset length includes: obtaining the frame overlap rate of the short-time Fourier transform; determining the preset length based on the frame overlap rate, wherein the preset length is positively correlated with the frame overlap rate.

[0014] In one embodiment, performing an equal-angle phase inverse mapping on the revolution-phase sequence to obtain a target time series corresponding to a uniform angle includes: obtaining a target order range and determining an angular domain sampling rate based on the target order range; constructing a uniformly distributed angular domain sampling point sequence using the angular domain sampling rate; and performing a phase inverse mapping on the angular domain sampling point sequence based on the revolution-phase sequence to obtain the target time series.

[0015] In one embodiment, the step of extracting the target order amplitude based on the angular domain spectrum includes: obtaining a target order list; locating the index position corresponding to the target order list on the order axis of the angular domain spectrum; and extracting the amplitude data of each target order at all index positions to form an order amplitude matrix.

[0016] In one embodiment, before performing equal-angle phase inverse mapping on the rotation phase sequence, the method further includes: performing monotonicity correction on the rotation phase sequence to obtain a monotonic rotation phase sequence.

[0017] In one embodiment, the vibration data has multi-channel time-domain signals; resampling the time-domain signals according to the target time series using a finite-support band-limited reconstruction kernel to obtain angular domain signals includes: sharing the target time series to all channels as a reference target time series for all channels; and resampling the time-domain signal of each channel according to the reference target time series using a finite-support band-limited reconstruction kernel to obtain angular domain signals for each channel respectively.

[0018] According to a second aspect, embodiments of this application provide a method for monitoring a rotating device, comprising: acquiring a target order amplitude of the rotating device, the target order amplitude being obtained based on the signal processing method described in any one of the first aspects above; and monitoring the operating state of the rotating device based on the target order amplitude.

[0019] According to a third aspect, embodiments of this application provide a signal processing apparatus, comprising: an acquisition unit for acquiring vibration data of a rotating device in operation, the vibration data including a time-domain signal and a rotational phase sequence, wherein the rotational phase sequence is used to characterize a sequence of physical quantities representing the change of the angle of the rotating device over time; an inverse mapping unit for performing an equiangular phase inverse mapping on the rotational phase sequence to obtain a target time sequence corresponding to a uniform angle; a resampling unit for resampling the time-domain signal according to the target time sequence using a finite-support band-limited reconstruction kernel to obtain an angular domain signal; a spectrum estimation unit for performing angular domain spectrum estimation on the angular domain signal to obtain an angular domain spectrum; and an extraction unit for extracting a target order amplitude based on the angular domain spectrum.

[0020] According to a fourth aspect, this application provides a monitoring device for a rotating device, comprising: an acquisition module for acquiring a target order amplitude of the rotating device, the target order amplitude being obtained based on the signal processing method described in any one of the first aspects; and a monitoring module for monitoring the operating status of the rotating device based on the target order amplitude.

[0021] According to a fifth aspect, embodiments of this application provide a computer device, including: a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the signal processing method described in any one of the first aspects or the monitoring method of the rotating device described in the second aspect.

[0022] According to a sixth aspect, embodiments of this application provide a computer-readable storage medium storing computer instructions, the computer instructions being configured to cause a computer to perform the signal processing method described in any one of the first aspects or to perform the monitoring method for a rotating device described in the second aspect.

[0023] According to a seventh aspect, embodiments of this application provide a computer program product, the computer program product including a computer program, which, when executed by a processor, implements the signal processing method described in any of the first aspects or performs the monitoring method for a rotating device described in the second aspect.

[0024] This application has at least the following beneficial effects:

[0025] After acquiring vibration data of the rotating equipment under operating conditions, the rotational phase sequence in the vibration data is subjected to equiangular phase inverse mapping. By establishing a one-to-one correspondence between the uniform angular domain phase and the time domain time point, the requirement for uniform angular domain sampling is transformed into sampling at a specified target time point in the time domain. This ensures that all error sources are constrained within the same link, guaranteeing that the angular domain signal can be sampled at a uniform angle. After determining the target time sequence corresponding to the uniform angle, the time domain signal is resampled according to the target time sequence using a finite-support band-limited reconstruction kernel to obtain the angular domain signal. Based on an ideal band-limited reconstruction kernel, and under the premise of controlling the computational load through finite support, resampling is performed at the specified target time point corresponding to the uniform angular domain to obtain an angular domain signal with high fidelity in both sampling structure and amplitude information. This reduces the risk of systematic angular domain signal distortion at high speeds in variable speed conditions and ensures high accuracy and robustness of the order tracking results.

[0026] Specifically, when performing angular domain spectrum estimation, the boundary processing is adaptively applied to the angular domain signals of varying lengths and differences generated by the rotating equipment under different operating conditions. Under the premise of obtaining high-fidelity angular domain signals, the angular domain spectrum analysis results with consistent quality and stable format are output, thereby ensuring the high reliability and strong engineering adaptability of the order tracking results.

[0027] Therefore, by performing equal-angle phase inverse mapping on the rotation phase sequence, the requirement for uniform sampling in the angular domain is transformed into sampling at a specified target time point in the time domain. Under the premise of controlling the computational load with limited support, resampling is performed at the specified target time point corresponding to the uniform angular domain to obtain an angular domain signal with high fidelity in both sampling structure and amplitude information. At the same time, adaptive boundary processing is performed on the angular domain signals obtained under different working conditions to ensure high accuracy and high robustness of the order tracking results. Attached Figure Description

[0028] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0029] Figure 1 This is a schematic flowchart of a signal processing method according to some embodiments of the present invention;

[0030] Figure 2 This is a flowchart illustrating a resampling method in a signal processing method according to some embodiments of the present invention;

[0031] Figure 3This is a structural block diagram of a signal processing apparatus according to some embodiments of the present invention;

[0032] Figure 4 This is a structural block diagram of a monitoring device for a rotating device according to some embodiments of the present invention;

[0033] Figure 5 This is a schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. Detailed Implementation

[0034] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0035] According to embodiments of this application, a signal processing method, a method for monitoring rotating equipment, an apparatus, a device, and a medium are provided. It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowcharts, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0036] It should be noted that the information (including but not limited to user device information, user personal information, etc.), data (including but not limited to data used for analysis, data stored, data displayed, etc.) and signals involved in this application are all authorized by the user or fully authorized by all parties, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant regions.

[0037] For order tracking, existing technologies include the Vold-Kalman tracking filter method. Its core approach is to construct a frequency real-varying model based on the time-varying law of the preset order frequency, model the non-stationary vibration signal as a superposition of multiple time-varying sinusoidal components and noise, and then dynamically track the amplitude, frequency and phase of the order through state estimation.

[0038] However, in the high-speed range or rapidly changing conditions of rotating equipment, the speed change is often nonlinear. This can lead to a mismatch between the time-varying pattern of the preset order frequency and the time-varying pattern of the order frequency in the actual operating conditions, resulting in frequency tracking deviation. Furthermore, the frequency tracking accuracy of Volk-Kalman will decrease as the frequency increases, causing the frequency tracking deviation to increase further, which in turn leads to amplitude estimation deviation.

[0039] Existing technologies also employ the Coulomb-Operational (COT) method for order tracking. While the COT method can circumvent the limitation of the Vold-Kalman tracking filter method, which relies on a time-varying frequency model, when resampling using the COT method, if the target order is high or the rotational speed changes rapidly, the resampling interpolation error from the time domain to the angular domain is more likely to manifest as an underestimation of amplitude or phase distortion. This causes the tracking curve to deviate significantly from the reference result in the high-speed range, resulting in a systematic amplitude deviation of high-order orders in the high-speed range.

[0040] Based on this, embodiments of this application provide a signal processing method, apparatus, device, and medium to improve the accuracy of order tracking in extracting high-order amplitudes at high speeds or under conditions of rapid speed changes.

[0041] The following section will introduce the keywords or terms used in the embodiments of this application.

[0042] Time-domain sampling rate: refers to the number of sampling points per second, denoted by the symbol f in this paper. s Refers to a unit in Hz (Hertz).

[0043] Time-domain signal: When it is a time-domain sampled signal, it refers to a discrete sampled signal that collects the vibration physical quantity of rotating equipment during operation and shows how it changes over time using a time-domain sampling rate.

[0044] Rotational speed sequence: A discrete sequence of instantaneous rotational speed values ​​that change over time during the operation of rotating equipment. In this paper, these are represented by the symbol... Refers to units measured in RPM (revolutions per minute).

[0045] Rotational speed frequency: refers to the fundamental frequency of the rotating shaft, which is calculated from the rotational speed sequence. In this article, it is referred to as... Refers to the unit rev / s (revolutions per second).

[0046] Rotational phase: A physical quantity that characterizes the change of the rotational angle of rotating machinery over time, measured in rev (revolutions).

[0047] Order: This refers to the ratio of the vibration signal frequency of rotating machinery to the fundamental frequency of the rotating shaft, representing the number of vibration cycles per revolution of the rotating machinery. The order characteristic can identify the current operating state of the rotating machinery. Common order characteristics include, but are not limited to: integer orders (1X, 2X, 3X...), used to indicate direct mechanical responses such as rotor imbalance, misalignment, and looseness; fractional orders (e.g., 0.5X, 1.5X), used to indicate subharmonics or superharmonics caused by rolling bearing failure, belt slippage, combustion cycles, etc.; and high-order orders (e.g., 50X, 100X), used to indicate gear meshing frequencies (number of teeth x rotational speed), electromagnetic force harmonics, blade passing frequencies, etc. In this article, "rotating equipment" is used to refer to rotating machinery.

[0048] Angular domain sampling rate: the number of sampling points per unit revolution in the rotational angular domain, which is used in this paper. This refers to a unit of measurement: samples / rev (number of sampling points per revolution).

[0049] Angular domain signal: that is, the angular domain sampling signal, refers to the discrete sampling signal of the vibration physical quantity of rotating equipment during operation, which is collected using the angular domain sampling rate and is obtained by resampling the time domain signal.

[0050] Shannon sampling theorem, also known as Nyquist sampling theorem, states that when the sampling frequency is greater than twice the highest frequency of the original signal, the original signal can be recovered from the discrete sampled signal without distortion.

[0051] This embodiment provides a signal processing method. Figure 1 This is a flowchart of a signal processing method according to an embodiment of the present invention, such as... Figure 1 As shown, the process includes the following steps:

[0052] Step S101: Obtain vibration data of the rotating equipment under operating conditions. The vibration data includes time-domain signals and rotational phase sequences.

[0053] In this embodiment, the vibration data is acquired during the operation of the rotating equipment. The vibration data includes signals from the rotating equipment under variable speed conditions, which may include typical variable speed conditions such as acceleration, deceleration, and unsteady speed changes (e.g., speed fluctuations). The vibration data is used to analyze the operating status of the rotating equipment, including analyzing speed changes caused by faults. These faults may include, but are not limited to, at least one of rotor imbalance, shaft misalignment, gear meshing wear, and electromagnetic harmonic abnormalities.

[0054] Among them, the time-domain signal represents the discrete sampling signal of the vibration physical quantity of the monitored part of the rotating equipment changing with time during operation. For example, it can include sampling signals of physical quantities such as acceleration, velocity, or displacement changing with time. Specifically, it can be obtained by using acceleration sensors, velocity sensors, displacement sensors, etc., deployed at the key vibration monitoring parts of the rotating equipment at a set time-domain sampling rate.

[0055] The rotational phase sequence represents a sequence of physical quantities showing how the rotational angle of rotating machinery changes over time; it can be the cumulative rotational rotation sequence of the rotor of a rotating device over time. Specifically, there is a fixed mapping relationship between the time index corresponding to the rotational phase sequence in the time domain and the sampling time point of the time-domain signal. For example, there can be a one-to-one correspondence between the time index corresponding to the rotational phase sequence in the time domain and the sampling time point of the time-domain signal.

[0056] In this embodiment, the rotational speed phase sequence can be obtained by integrating the rotational speed sequence of the rotating equipment. The rotational speed sequence represents a discrete sequence of instantaneous rotational speed changes over time during the operation of the rotating equipment. The rotational speed sequence can be obtained by directly sampling the instantaneous rotational speed of the rotating equipment using a speed sensor. Alternatively, the rotational speed sequence can be determined using pulse signals or encoder signals associated with the rotating equipment. For example, a phase-locked loop can be used to estimate the frequency of the pulse signals or encoder signals, and the frequency estimation result can be converted into instantaneous rotational speed values ​​to form the rotational speed sequence. Alternatively, rotational speed features can be extracted from vibration, current, or other signals of the rotating equipment to estimate the rotational speed sequence.

[0057] After obtaining the rotational speed sequence, the rotational speed sequence is converted into a rotational speed frequency, and the rotational speed frequency is integrated to obtain the rotational speed phase sequence. Specifically, the following formula can be used to convert the rotational speed sequence into a rotational speed phase sequence:

[0058] ;

[0059] ;

[0060] in, For time The rotational speed at that location, For time Rotational speed and frequency values; The phase value of the number of revolutions at time t; For integration variables; It is a time infinitesimal.

[0061] The above formula is used to sequentially convert time-continuous rotational speed values ​​into time-continuous rotational speed phase values, forming a rotational speed phase sequence. Theoretically, the rotational speed phase sequence is strictly monotonically increasing. However, in engineering data, rotational speed sequences often contain noise, sampling jitter, or local outliers, which can lead to local non-monotonic segments in the rotational speed phase sequence. Directly performing an equal-angle phase inverse mapping on non-monotonic data can result in abnormal interpolation such as multi-value / backlash, leading to disordered angular domain resampling time points, and consequently, local phase distortion and spectral artifacts in the angular domain signal.

[0062] Therefore, in one embodiment, after obtaining the revolution-phase sequence, the revolution-phase sequence can be checked and corrected for monotonicity to obtain a monotonic revolution-phase sequence, thereby ensuring the stability of the data during isotropic phase inverse mapping and resampling. For example, monotonicity modification can be performed by retaining sampling points with positive phase increments and removing outliers that cause non-monotonicity, thus ensuring the monotonicity and numerical stability of the revolution-phase sequence.

[0063] In one embodiment, a monotonic fitting correction method can also be used to test and correct the monotonicity of the revolution-phase sequence. Specifically, when a non-monotonic segment is detected in a local portion of the revolution-phase sequence, a local window containing the non-monotonic segment is selected. A monotonic-constrained least-squares fitting or equivalent monotonic regression is performed on the revolution-phase sequence within the local window to obtain a locally monotonic phase sequence. This locally monotonic phase sequence then replaces the original phase sequence in the local window. This approach can suppress non-monotonicity caused by sampling jitter while avoiding the impact of time axis holes on the density and stability of resampling points.

[0064] For example, non-monotonic corrections can be performed within a local window in the following form:

[0065] ;

[0066] For the first One corrected revolution phase value; For the first One original rotational phase value; The preset condition is: the next corrected rotational phase value within the local window must be greater than or equal to the previous corrected rotational phase value.

[0067] In another embodiment, monotonicity correction can also be performed through local smoothing and interpolation. When non-monotonic segments or densely populated anomalous points are detected in the local area of ​​the revolution phase sequence, local smoothing can be performed on the revolution phase sequence or its difference sequence, for example, by using sliding window smoothing to suppress jitter. The revolution phases that are determined to be abnormal are then filled by interpolation, and the monotonicity test is performed on the corrected sequence, thereby improving the coverage of different engineering implementation calibers.

[0068] Step S102. Perform equal-angle phase inverse mapping on the rotation phase sequence to obtain the target time sequence corresponding to the uniform angle.

[0069] In this embodiment, the target time series is used to characterize the sequence of discrete sampling time points in the time domain when the revolutions phase reaches a set uniform angle. A one-to-one correspondence between the uniform angle (phase) sampling time points in the time domain is established through isoangular phase inverse mapping. This transforms the isoangular sampling constraint into a non-uniform time sampling position in the time domain.

[0070] When performing isoangular phase inverse mapping on the revolution phase sequence, an isoangular target phase grid uniformly distributed in the angular domain can be predefined. Each target grid in the isoangular target phase grid corresponds to an angular domain phase value. By performing phase inverse mapping on the revolution phase sequence, the target time point corresponding to each isoangular target phase grid is determined. After traversing all the target time points corresponding to the isoangular target phase grids, the target time sequence is obtained.

[0071] Based on this, regardless of the variable speed conditions under which the original time-domain signal is acquired, as long as an equiangular target phase grid is set, and the rotational phase sequence is inversely mapped using equiangular phase, the target time point corresponding to the uniform angle can be obtained. Subsequent sampling of the time-domain signal into an angular-domain signal according to the target time point can avoid the influence of time-domain signal non-uniformity, resulting in a high-precision angular-domain phase with the same angle length and resolution. This ensures the consistency of the angular-domain signal or its order amplitude under different measurements and operating conditions. Therefore, transforming the requirement for uniform angular-domain sampling into sampling at a specified target time point in the time domain can constrain all error sources within the same link, providing a unified foundation for high-fidelity angular-domain signal construction and reducing high-order amplitude deviations.

[0072] For example, the expression for the isoangular phase inverse mapping is:

[0073]

[0074] in, The phase grid for the isoangular target is k grids, each grid corresponding to an angular domain phase value. k starts from 0 and takes integer values ​​sequentially. Different integer values ​​represent the index of the grid or angular domain phase value. For the k-th angular domain phase value, the target time point in the time domain is... (⋅) is the rotational phase inverse mapping function.

[0075] In an optional embodiment, when performing equiangular phase inverse mapping on the revolution phase sequence, constructing an equiangular target phase grid is the benchmark for the order tracking resampling process, which can determine the sampling density, range and accuracy of the angular domain signal. Therefore, it is necessary to construct a uniform and reliable phase sequence in the angular domain as the target position for resampling.

[0076] For example, the phase mesh of an isoangular target can be constructed in the following way:

[0077] Obtain the target order range, which is the interval range of all orders that need to be extracted or tracked from the vibration data.

[0078] The angular sampling rate is determined based on the target order range, wherein the angular sampling rate is determined based on the highest order within the target order range, and the higher the highest order, the greater the angular sampling rate.

[0079] Furthermore, in order to avoid spectral aliasing and recover the highest-order components without distortion, the angular domain sampling rate can be set to more than twice the highest order. The higher the angular domain sampling rate, the more finely the details of the high-order signal can be described.

[0080] For example, the highest order is extracted within the target order range. To avoid angular domain spectral aliasing, the angular domain sampling rate is... Must meet: ,in, It is the highest order.

[0081] The angular domain sampling rate is the number of sampling points per revolution. Based on the number of sampling points per revolution, the angular domain sampling point sequence is determined as the isoangular target phase grid. That is, with the number of sampling points per revolution as the basic interval unit, an angular domain phase value sequence that is uniformly distributed in the rotation angle is generated, and the uniformly distributed angular domain phase value sequence is used as the isoangular target phase grid.

[0082] For example, the phase mesh expression for an isoangular target is:

[0083]

[0084] in, The phase grid for an isoangular target consists of k grids, each corresponding to an angular phase value. k starts from 0 and sequentially takes integer values, with different integer values ​​representing the index of the grid or angular phase value. The initial angular domain phase value, is the angular sampling rate.

[0085] Step S103. Resample the time-domain signal according to the target time series using a finite-support band-limited reconstruction kernel to obtain the angular domain signal.

[0086] In this embodiment, the finite-support band-limited reconstruction kernel is used to reconstruct and resample non-uniform time-domain sampled signals onto an equiangular target phase grid with high fidelity.

[0087] In this context, "band-limited" refers to the fact that the current reconstruction kernel has an ideal, flat passband in the frequency domain (order domain) and can completely suppress frequency components outside the passband. This ensures that during resampling, only the time-domain signal within the predefined target order range is preserved and reconstructed without distortion, while components outside the target order range are filtered out. "Finite support" means that the current reconstruction kernel does not extend infinitely in the time domain, but only has non-zero values ​​within a finite local interval, rapidly decaying to zero outside this finite local interval. This is used to reduce computational complexity by relying on only a finite number of original data points near the target time point in each interpolation calculation.

[0088] After obtaining the target time point for each uniform angle in the time domain sampling, since the target time point usually does not fall on the discrete sampling points of the original time domain signal, it is necessary to determine the signal value corresponding to the target time point from the original time domain signal by interpolation or reconstruction. However, common interpolation algorithms in existing technologies may lead to high-order amplitude deviations.

[0089] For example, existing linear interpolation or cubic spline interpolation methods are essentially based on polynomial fitting for geometric connections. Therefore, at high speeds and when the frequencies corresponding to higher-order components are close to the Nyquist frequency, the amplitude-frequency response of linear or cubic spline interpolation will show significant attenuation, leading to a systematic underestimation of higher-order amplitudes. Specifically, the equivalent amplitude-frequency response of linear interpolation is significantly attenuated at high frequencies, easily resulting in a decrease in higher-order amplitudes; while cubic spline interpolation may introduce ringing or boundary overshoot under certain operating conditions, causing spurious peaks or increased noise floor in the angular domain spectrum.

[0090] To reduce amplitude-frequency response attenuation, existing technologies often employ methods such as high-frequency gain compensation and adaptive piecewise interpolation to improve linear or cubic spline interpolation. However, the essence of linear and cubic spline interpolation is to use geometric fitting methods to fit the interpolation results to geometric lines. Based on this fundamental method, whether high-frequency gain compensation or adaptive piecewise interpolation is used to improve linear or cubic spline interpolation, it is impossible to guarantee that the amplitude of a specific frequency component is distortion-free.

[0091] Therefore, in this embodiment, the geometric approach based on polynomial fitting is abandoned, and a finite-support band-limited reconstruction kernel is used to perform precise non-uniform resampling of the time domain signal, thereby realizing the conversion from non-uniform time point sampling in the time domain to uniform angular domain phase value, achieving attenuation-free amplitude-frequency response within the target order range, and suppressing aliasing and artifacts through band-limiting.

[0092] Based on Shannon's sampling theorem, a band-limited signal can be completely and distortion-freely recovered using ideal band-limited reconstruction through its discrete sampling points. However, the support of the ideal band-limited reconstruction kernel is infinitely long, making practical computation impossible. Therefore, in this embodiment, a finite-support band-limited reconstruction kernel is used to resample the time-domain signal according to the target time series. This approach, based on the ideal band-limited reconstruction kernel, better preserves the amplitude of high-frequency / high-order components while controlling the computational load through finite support, reducing the risk of systematic amplitude underestimation of high-order components at high speeds, and obtaining a high-fidelity angular domain signal.

[0093] Step S104. Perform angular domain spectrum estimation on the angular domain signal to obtain the angular domain spectrum.

[0094] Specifically, the time-domain signal is resampled according to the target time series using a finite-support band-limited reconstruction kernel, converting time-domain signals with unequal time intervals into angular-domain signals with equal angular intervals.

[0095] Angular domain spectrum estimation is used to perform spectral analysis on angular domain signals to obtain the angular domain spectrum, where the horizontal axis of the angular domain spectrum represents the order.

[0096] For example, angular domain spectrum estimation can be performed by processing the angular domain signal using Fourier transform. In order to achieve order tracking, that is, to extract order amplitudes proportional to the rotational speed from the vibration data under varying rotational speed conditions, in this embodiment, short-time Fourier transform can be used to analyze the obtained stationary angular domain signal to achieve accurate tracking of the evolution of each order amplitude with the angle.

[0097] Typically, the length characteristics of angular domain signals are inconsistent. This is because angular domain signals are obtained by resampling the time domain signal according to the target time series using a finite-support band-limited reconstruction kernel. In actual testing, the acquisition of time domain signals usually does not begin when the rotating shaft is exactly at 0° phase value, nor does it end at an integer number of revolutions. Furthermore, the acquisition of time domain signals may begin or stop based on actual operating conditions. For example, it may start and terminate at any time due to the start and stop of rotating equipment, fault triggering, or timed saving or storage space limitations of time domain signals. As a result, the total length of time domain signal segments is often random, and the corresponding total length of angular domain signals is also random.

[0098] In addition, the signals collected before or after the rotating equipment is started may be noise or meaningless vibration data; the middle of the collected signal may contain invalid segments due to interference, which causes the effective signal to be divided into multiple segments, resulting in the effective length of the effective signal in the time domain signal often being random, and the corresponding effective length of the angular domain signal is also random.

[0099] The length characteristic of a corner domain signal can include the total length of the corner domain signal or the effective length of the effective signal within the corner domain signal.

[0100] Given the inconsistency in the length characteristics of angular domain signals, if the same processing strategy is applied to angular domain signals with different length characteristics during angular domain spectrum estimation, the number of output columns becomes unstable, the angular domain spectrum estimation bias increases, and boundary artifacts become significant, resulting in inconsistent output apertures for angular domain signals with different length characteristics. In this embodiment, we will illustrate this by taking an example where the length characteristic includes the total length of the angular domain signal, and angular domain spectrum estimation is performed using short-time Fourier transform:

[0101] When performing short-time Fourier transform, there is a contradiction when performing boundary processing for angular domain signals of different total lengths: for angular domain signals with shorter total lengths, a certain boundary extension strategy is needed to improve the stability of the frame sequence and output aperture; however, for the boundaries of angular domain signals with longer total lengths, extension may copy / mirror the endpoint structure into the window, resulting in unnecessary artifacts or energy leakage, which affects the purity of the spectrum.

[0102] Based on this, in this embodiment, when performing angular domain spectrum estimation on the angular domain signal, boundary processing is adaptively performed on the angular domain signals with inconsistent total lengths generated by the rotating equipment under different operating conditions, producing analysis results with consistent quality and stable format, ensuring that the output aperture of all types of angular domain signals is consistent, thereby ensuring the high reliability and strong engineering adaptability of the order tracking results.

[0103] Step S105. Extract the target order amplitude based on the angular domain spectrum.

[0104] The target order amplitude is obtained by extracting the corresponding amplitude value from the angular domain spectrum according to the target order list.

[0105] The target order list is a set of predefined orders within the target order range. For example, the predefined orders may include at least one of the following: the theoretical characteristic order corresponding to the mechanical structure in the rotating equipment, the fault characteristic order corresponding to a specific fault mode, and the custom order that needs to be tracked or analyzed.

[0106] For example, the target order amplitude can be converted into commonly used engineering indicators, which include one or more of the following: RMS value, peak value, and power.

[0107] In this embodiment, after acquiring vibration data of the rotating equipment under operating conditions, an equal-angle phase inverse mapping is performed on the rotational phase sequence in the vibration data. By establishing a one-to-one correspondence between uniform angular domain phase values ​​and time points in the time domain, the requirement for uniform angular domain sampling is transformed into sampling at a specified target time point in the time domain. This ensures that all error sources are constrained within the same link. Under variable speed conditions, it can guarantee that the angular domain signal can be sampled at a uniform angle. After determining the target time sequence corresponding to the uniform angle, a finite-support band-limited reconstruction kernel is used to resample the time domain signal according to the target time sequence to obtain the angular domain signal, which can be rationally... Based on a band-limited reconstruction kernel, and under the premise of controlling computational load with limited support, resampling is performed at a specified target time point corresponding to the uniform angular domain. This results in angular domain signals with high fidelity in both sampling structure and amplitude information, reducing the risk of systematic angular domain signal distortion at high speeds in variable-speed conditions. Adaptive boundary angular domain spectrum estimation is performed on the angular domain signals to adaptively address the significantly different transient or steady-state angular domain signals generated by rotating equipment under different operating conditions. While obtaining high-fidelity angular domain signals, consistent quality and stable format analysis results are output, ensuring high reliability and strong engineering adaptability of the order tracking results. Therefore, by performing equiangular phase inverse mapping on the revolution phase sequence, the requirement for uniform angular domain sampling is transformed into sampling at a specified target time point in the time domain. Under the premise of controlling computational load with limited support, resampling is performed at the specified target time point corresponding to the uniform angular domain to obtain angular domain signals with high fidelity in both sampling structure and amplitude information. Simultaneously, adaptive boundary processing is performed on the angular domain signals obtained under different operating conditions to ensure high accuracy and robustness of the order tracking results.

[0108] In one embodiment, the finite-support bandgap reconstruction kernel includes windowing. Band-limited reconstruction kernel.

[0109] In this embodiment, with ideal Function-based finite-support windowing The band-limited reconstruction kernel possesses band-limited reconstruction characteristics that match the preset band limit. Within the effective band limit, the amplitude-frequency response is basically flat, which can completely preserve the true amplitude of each order component, especially avoiding the systematic amplitude underestimation of high-order components in the high-speed range. Outside the band limit, the amplitude-frequency response decays rapidly, which can suppress irrelevant noise interference. At the same time, the computational load is controllable by relying on the finite support design. Under the premise of controllable computational load, the systematic amplitude underestimation of high-order components / high-speed range can be reduced, and the risk of noise flooring of interpolation error in the angular domain spectrum can be reduced.

[0110] In an alternative embodiment, the finite-supported bandgap reconstruction core can also employ a non- Windowed band-limited reconstruction kernels, such as windowed compactly supported wavelet kernels, are based on compactly supported wavelet functions and optimize frequency domain characteristics through window functions. They are suitable for coarse feature extraction of non-stationary signals, such as impact location of vibration signals and precise non-amplitude measurement.

[0111] To address the requirement of distortion-free high-order amplitudes, this embodiment employs windowing. The band-limited reconstruction kernel is used for resampling to extract key features such as gear meshing and bearing failure.

[0112] Specifically, a window function with main lobe suppression and side lobe attenuation characteristics is used for finite-supported... Modify the class-band-limited reconstruction kernel to obtain a windowed form. Band-limited reconstruction kernel. Among them, ideal... The function is defined as:

[0113]

[0114] in, for In this embodiment, the input variable of the function can represent the time difference between the neighboring sampling points and the target time point. The sine function The parameters are used for association. The location of the zeros of the function.

[0115] The Fourier transform of the function is a rectangular window. Within the Nyquist frequency range, the amplitude-frequency response is completely flat with no amplitude attenuation. Beyond this range, the amplitude attenuates to zero. Therefore, for band-limited signals whose highest frequency does not exceed the Nyquist frequency, such as the vibration data and order signals in the above embodiments, the Fourier transform of the function is... The function interpolates discrete sampling points and can recover the original continuous signal without distortion.

[0116] Based on this, in this embodiment, the ideal sinc function is used as a basis, and a window function is used to modify it to obtain the kernel function of the reconstruction kernel, which is expressed as follows:

[0117]

[0118] in, It is a window function with main lobe suppression and side lobe attenuation characteristics; The time difference between a neighboring sampling point and the target time point can be expressed as follows: ,

[0119] in, For the target time point, The original sampling time of the time-domain signal. The original sampling interval of the time-domain signal. The sampling rate is in the time domain. Here, is the kernel half-width parameter, which refers to the distance from the center of the kernel function to the edge of the finite-supported band-limited reconstructed kernel. The kernel half-width parameter is calculated using the following formula:

[0120]

[0121] in, ) indicates the length of the support window for the finite support banded reconstruction core. Indicates the sequence number of the discrete sampling point.

[0122] In practice, the instantaneous rotational speed change rate can be calculated based on the rotational speed phase sequence; the support window length of the finite support band-limited reconstruction core can be dynamically adjusted based on the instantaneous rotational speed change rate, wherein the support window length is negatively correlated with the absolute value of the instantaneous rotational speed change rate.

[0123] Specifically, based on the rotational phase value at any moment in the rotational phase sequence (e.g., the phase change data corresponding to each rotation), the instantaneous rotational speed change rate at any moment can be calculated by numerical differentiation or difference methods.

[0124] Specifically, the instantaneous speed change rate can be calculated using the following formula. :

[0125]

[0126]

[0127] Among them, the rotational phase value is The corresponding time point is , This represents the instantaneous rotational speed. The rate of change of the instantaneous rotational speed can be obtained by further differentiating the instantaneous rotational speed with respect to time. The instantaneous rate of change of rotational speed reflects the degree of fluctuation in the rotational speed of a rotating system at the current moment. Indicates the sequence number of the discrete sampling point.

[0128] Obtain the instantaneous speed change rate Then, the system will base its decision on its absolute value. Dynamically adjustable finite support bandgap reconstruction kernel (e.g., windowed) Support window length (e.g., band-limited reconstruction kernel, Kaiser window) Specifically, there is a negative correlation between the length of the support window and the absolute value of the instantaneous rate of change of rotational speed, that is: when When the value is large, it indicates a rapid change in rotational speed. In this case, shortening the support window length improves the time-positioning capability of the reconstruction kernel and reduces errors caused by non-stationary rotational speed. When the value is relatively small, it indicates that the rotation speed is stabilizing. At this point, the length of the support window should be increased to preserve the frequency resolution characteristics of the band-limited reconstruction core and improve the signal reconstruction accuracy.

[0129] This negative correlation can be achieved using linear functions, piecewise functions, or mapping tables, for example: Where k is a preset proportional coefficient, The maximum window length; or a nonlinear function.

[0130] .

[0131] in It is a proportionality coefficient, also known as a sensitivity coefficient or adjustment constant, and its core function is to control the length of the support window. absolute value of instantaneous speed change Sensitivity to change.

[0132] The above method ensures that the window length is always within a reasonable range and decreases monotonically as the instantaneous rotational speed changes.

[0133] The ideal sinc function is modified using a window function, and optimizations are applied to the ideal sinc function including a finite support stage, main lobe suppression, and side lobe attenuation. The finite support range of the kernel function is defined by the kernel half-width parameter, i.e., the finite support range is [- , ].

[0134] When resampling at the target time point, a finite support range is constructed with the target time point as the center, and all original sampling points within the finite support range are selected to participate in the calculation. This reduces the systematic amplitude underestimation in the high-order / high-speed range and reduces the risk of noise flooring of resampling error in the angular domain spectrum while keeping the computational load under control.

[0135] For example, taking the kernel function mentioned above, we can select all original sampling points within the finite support range defined by the kernel half-width parameter, centered on the target time point, where the time difference between each original sampling point and the target time point satisfies |d| < The weights of all original sampling points are calculated using a kernel function, and then weighted summation is performed to obtain the interpolation result for the target time point. Detailed implementation methods are described in steps A1-A4 below, and will not be repeated here.

[0136] By leveraging the smooth truncation property of window functions, the ideal... The function transitions from infinite support to finite support. When the time difference exceeds the kernel half-width parameter, the window function automatically reduces the amplitude to zero, truncating oscillations and controlling computational load while ensuring resampling accuracy. Therefore, utilizing a window function with main lobe suppression and side lobe attenuation characteristics for ideal... Functions are modified to construct windows. Band-limited reconstruction kernel, utilizing windowing The band-limited reconstruction kernel resamples the time-domain signal according to the target time series. Under the premise of controlling the amount of computation, it can retain the amplitude of high-order / high-frequency components without distortion, reduce the risk of systematic underestimation of the amplitude of high-order components in the high-speed range, and provide accurate order amplitude data for fault diagnosis and condition monitoring of rotating equipment.

[0137] In one embodiment, windowing Band-limited reconstruction kernels can use Lanczos kernels, i.e., in... At that time, As a window function, the function itself can achieve a good balance between the main lobe width and the side lobe attenuation, thus achieving high amplitude fidelity with controllable computational complexity.

[0138] In optional embodiments, the window function can also employ other window functions with sidelobe attenuation capabilities, such as Blackman and Kaiser window functions, to expand the coverage of the resampling kernel and improve robustness to equivalent evasion. In one embodiment, such as Figure 2 As shown, the step of resampling the time-domain signal according to the target time series using a finite-support band-limited reconstruction kernel to obtain the angular domain signal includes:

[0139] Step A1: Select a finite number of neighborhood sampling points around each target time point in the target time series, wherein the neighborhood sampling points are the original sampling points of the time domain signal.

[0140] In this embodiment, after obtaining the target time point corresponding to each uniform angle in the time domain, a finite number of neighborhood sampling points around each target time point are selected. For example, the original sampling points corresponding to N times the kernel half-width parameter range on both sides can be selected as neighborhood sampling points with the target time point as the center, where N is a positive integer greater than or equal to 1.

[0141] The larger the kernel half-width parameter, the more data needs to be calculated, the greater the computational overhead, and the more computational resources required. In addition, the higher the order of the target analysis, the more accurate the signal value may be required, which may require more data to participate in the calculation.

[0142] Therefore, in an optional embodiment, the value of the kernel half-width parameter can be based on the target highest analysis order and computing resources. The kernel half-width parameter is positively correlated with the target highest analysis order and negatively correlated with computing resources. If the computing resources are sufficient, the kernel half-width parameter can be selected based on the target highest analysis order.

[0143] For example, when the highest analysis order of the target is less than 50, the kernel half-width parameter can be 2 to 3; when the highest analysis order of the target is greater than 50 and less than 100, the kernel half-width parameter can be 3 to 4; and when the highest analysis order of the target is greater than 100, the kernel half-width parameter can be 4 to 6. It should be noted that the above data are only for the convenience of the specification's value logic and do not represent all embodiments; other values ​​are within the protection scope of this embodiment.

[0144] Step A2: Calculate the time difference between each of the neighboring sampling points and the corresponding target time point.

[0145] Once the kernel half-width parameter is determined, the finite support range of the kernel function can be determined. , ], taking the target time point as the center, select [- , ] Calculate the time difference between each selected neighboring sampling point and the target time point, taking all neighboring sampling points within the range of ] .

[0146] In this embodiment, the time difference can be normalized and converted into a dimensionless time difference. For example, for each neighboring sampling point, its corresponding original sampling time point is first determined, and the time difference between the original sampling time point and the target time point is calculated. For instance, the time difference can be calculated using the following formula:

[0147]

[0148] in, The sampling rate is in the time domain. For the first The time difference between each neighboring sampling point and the target time point; For the target time point; For the first neighborhood sampling points The corresponding original sampling time point.

[0149] Step A3: Determine the weight of the neighborhood sampling points based on the finite support band-limited reconstruction kernel and the time difference.

[0150] In this embodiment, the finite-support bandgap reconstruction kernel is windowed. Taking the band-limited reconstruction kernel as an example, after obtaining the time difference between each neighboring sampling point and the target time point, windowing is used. The class-band-limited reconstruction kernel calculates the weight of each neighborhood sampling point.

[0151] Where the time difference is less than or equal to the kernel half-width parameter, i.e. The window function is non-zero, and the weights are calculated using both the window function and the sinc function; when the time difference is greater than the kernel half-width parameter, i.e. When the window function is zero, the corresponding neighborhood sampling points are outside the support range, and their weights are zero.

[0152] When the time difference is less than or equal to the kernel half-width parameter, i.e., when the neighborhood sampling points are within a finite support range, the corresponding time difference can be substituted into the windowing. Band-limited reconstruction kernel, and The functions jointly calculate their weights. Window functions such as Lanczos and Blackman can be selected to adjust the basic weights of the corresponding neighborhood sampling points based on the relative proportion of the time difference and the kernel half-width parameter, thus suppressing spectral sidelobes within a limited computational range. The weights characterize the contribution of the neighborhood sampling points to the angular domain signal value corresponding to the target time point; the closer the neighborhood sampling point is to the target time point, the greater its corresponding weight.

[0153] Lanczos Taking function windowing as an example, windowing The kernel function of the function is:

[0154] ;

[0155] Based on this, the weights of the neighborhood sampling points corresponding to the target time point are:

[0156] ;

[0157] in, For the first The time difference between each neighboring sampling point and the target time point For the first The weights of each neighboring sampling point.

[0158] Step A4: The signal value of the time domain signal corresponding to each of the neighboring sampling points is weighted and summed with the corresponding weight to obtain the angular domain signal value corresponding to the target time point.

[0159] Multiply the values ​​of all original sampling points within the neighborhood by the corresponding weights calculated in the previous step, and sum them to obtain the angular domain signal value corresponding to the target time point. Specifically, refer to the following formula:

[0160]

[0161] in, For the first The angular domain signal values ​​corresponding to each target time point Let i be the signal value of the i-th neighboring sampling point. The weights corresponding to the i-th neighborhood sampling point are: The kernel half-width parameter.

[0162] Step A5: Use the angular domain signal values ​​corresponding to all the target time points as the angular domain signals.

[0163] For each target time point, repeat steps A1-A4 above to obtain a angular domain signal sequence that is uniformly sampled in terms of angle.

[0164] When resampling, it is necessary to calculate the weight of the neighboring sampling points corresponding to each target time point in real time, which will result in a large computational cost. In order to save computational cost, in one embodiment, a weight query model can be established in advance. After obtaining the time difference between the neighboring sampling points and the target time point, the weight of each neighboring sampling point can be obtained by querying the weight query model.

[0165] In this embodiment, a model of the correspondence between different time differences and weights can be generated in advance based on the limited support band limit reconstruction kernel.

[0166] Specifically, the weights corresponding to different time differences are calculated in advance based on the kernel half-width parameter and the window function type, and the calculation results are stored as a weight query model. During the resampling process, the corresponding weights are obtained by querying the weight query model through the time difference, so as to reduce the amount of real-time computation.

[0167] In a specific embodiment, the weights stored in the weight query model can be represented by a relational expression. After obtaining the time difference, the weights are substituted into the relational expression to calculate the corresponding weights.

[0168] In another alternative embodiment, the weights stored in the weight query model can also be stored in a discrete form. When querying by time difference, it may not be possible to accurately match the specific weight value. The accurate weight corresponding to any time difference can be quickly calculated by linear interpolation of adjacent discrete weight values.

[0169] As described in the above embodiments, due to the support of multiple factors such as engineering acquisition, signal preprocessing, and equipment operating conditions, the total length of the acquired time-domain signal and the effective length of the effective signal are inconsistent. Consequently, when the signal is transmitted to the angular domain, the length characteristics such as the total length or effective length are different. Traditional angular domain spectrum estimation often uses fixed boundaries, which cannot adapt to the differences in the length characteristics of the angular domain signal. This results in insufficient frames for angular domain signals with short total length or effective length, and artifacts appearing in angular domain signals with long total length or effective length.

[0170] Therefore, in one embodiment, adaptive boundary angular domain spectrum estimation is performed on the angular domain signal for different length characteristics. The boundary processing method is automatically selected according to the length characteristics of the angular domain signal, which takes into account both the result stability of shorter angular domain signals and the spectral purity of longer angular domain signals.

[0171] In one embodiment, the length feature may include the total length of the angular domain signal, where the total length is the actual length of the angular domain signal obtained after resampling the time-domain signal. The length feature may also include the effective length of multiple effective signal segments in the angular domain signal, where the effective length may be based on the length of the effective signal segments extracted from the angular domain signal, or it may be the length of the segment with the most core effective feature in the angular domain signal.

[0172] In this embodiment, the total length of the angular domain signal is taken as an example for illustration:

[0173] Angular domain spectral estimation is essentially a Fourier transform of the angular domain signal, requiring operations such as framing, windowing, and performing Fourier transforms on the angular domain. After resampling to obtain the angular domain signal sequence at a uniform angle, its total length is obtained. When the total length is short, the number of analysis frames available for short-time Fourier transform is small, resulting in too few data points for the output order amplitude curve. Furthermore, the endpoint effect causes unstable output length and large spectral estimation deviation. Therefore, to improve the stability of the frame sequence and output aperture, a certain degree of boundary extension is required. When the total length is long, boundary extension may mirror the endpoint structure of the angular domain signal into the window, introducing the discontinuities at both ends of the angular domain signal into the window calculation, leading to unnecessary artifacts or energy leakage and affecting the purity of the spectrum.

[0174] The adaptive boundary spectral estimation of the angular domain signal specifically includes: after obtaining the total length of the angular domain signal, performing adaptive boundary processing on the angular domain signal based on the total length to obtain the processing result. The adaptive boundary processing may include performing boundary extension and skipping boundary extension.

[0175] In this embodiment, a preset length can be set in advance, and it can be determined whether the total length is less than the preset length. When the total length is less than the preset length, a boundary extension operation is performed on the corner domain signal; when the total length is greater than or equal to the preset length, the boundary extension operation is skipped, and the corner domain signal is directly used as the processing result.

[0176] In one embodiment, the preset length can be α times the window length of the short-time Fourier transform, where α is a constant greater than 1.

[0177] The frame-segmentation logic based on short-time Fourier transform determines whether the current signal needs boundary extension, as follows:

[0178] Obtain the total length of the angular domain signal, the window length of the short-time Fourier transform, and the frame overlap rate;

[0179] The frame shift of the short-time Fourier transform is determined based on the window length and the frame overlap rate.

[0180] For example, the frame shift can be calculated using the following formula:

[0181] ,in, For window length, For frame overlap rate, For frame shifting;

[0182] The number of available frames is calculated based on the total length of the angular domain signal and the frame shift.

[0183] For example, the number of available frames can be calculated using the following formula:

[0184] ;

[0185] in, For the total length, This represents the number of available frames.

[0186] It can be seen that when and At this time, a maximum of 2 frames can be formed, and both frames are close to the endpoints, making the boundary effect more pronounced; when and At this time, at least 3 frames can be formed, so there is at least 1 frame of internal overlapping frames that do not depend on endpoint extension, which is more conducive to stabilizing the output aperture between short and long sequences.

[0187] Therefore, in this embodiment, the preset length is a fixed twice the window length of the short-time Fourier transform.

[0188] In another embodiment, the frame overlap rate used when performing short-time Fourier transform on angular domain signals of different total lengths may be different. For example, when the total length of the angular domain signal is short, a higher frame overlap rate can be selected in order to generate a more stable number of output frames and have more internal overlapping frames participating in the calculation, so as to reduce the boundary effect.

[0189] When the total length of the angular domain signal is long, there are enough internal overlapping frames. Therefore, the number of output frames is more stable and the boundary influence is lower. However, due to the large amount of data, the corresponding computational overhead is large, requiring a lower overlap rate.

[0190] In this embodiment, to address the issue of different angular domain signals having different frame overlap rates, the generation of no less than a preset number of internal overlapping frames is used as a constraint. The preset length value is dynamically adjusted based on the frame overlap rate. The larger the frame overlap rate, the larger the preset length, so that the results under different configurations are consistent.

[0191] In one embodiment, when extending the boundary, extension methods such as even-symmetric extension, periodic extension, and zero-filling can be used. In this embodiment, the specific extension method can be determined based on the physical characteristics of the angular domain signal.

[0192] For example, feature analysis can be performed on angular domain signals, such as periodic intensity detection, endpoint smoothness evaluation, and spectral sparsity analysis. If the angular domain signal is determined to have strong periodicity, periodic extension can be used; if the endpoints are determined to have a symmetrical trend of peaks and / or valleys, even-symmetric extension can be used; if the number of significant spectral lines is determined to be small and the spectral sparsity is high, zero filling can be used.

[0193] In one embodiment, the effective boundary of the angular domain signal is taken as an example, using the length feature as the illustration:

[0194] For angular domain signals with a total length less than the preset length, boundary extension processing can be performed directly. For angular domain signals with a total length greater than the preset length, in order to save computational overhead and maximize the use of effective information of the angular domain signal, effective phase segments can be identified first, and then the actual effective length can be obtained based on the effective phase segments for adaptive boundary processing.

[0195] For example, the effective phase segment can be dynamically determined based on the effective phase range, energy distribution, signal-to-noise ratio, rotational speed variation characteristics, and target order characteristics of the angular domain signal. For angular domain signals of varying lengths, the optimal analysis boundary can be dynamically determined without the need for a uniform fixed value, maximizing the utilization of the effective phase segment and avoiding spectral artifacts such as unstable length, large spectral estimation deviation, or truncation.

[0196] For example, the following method can be used when performing adaptive boundary processing based on the effective length:

[0197] The core features of the angular domain signal are extracted, including: energy distribution features, signal-to-noise ratio features, rotational speed variation features, and phase segment features that are significant for the target order.

[0198] For example, regarding energy distribution characteristics, the angular domain signal can be divided into fixed phase segments to obtain multiple phase segments, and the energy value of each phase segment can be calculated to obtain the energy distribution characteristics. Regarding signal-to-noise ratio (SNR) characteristics, the SNR of each phase segment can be calculated to obtain the SNR distribution characteristics. Regarding rotational speed variation characteristics, the rotational speed variation rate of each phase segment of the corresponding angular domain signal can be inferred from the rotational speed phase sequence, where the rotational speed variation rate of the rotational speed phase sequence can be calculated based on the obtained rotational speed sequence according to the time point corresponding to the time domain signal synchronized with it. Regarding the phase segment characteristics of the target order, a fast Fourier transform can be performed on the angular domain signal to pre-estimate the energy distribution of the target order and determine the phase segments of the target order.

[0199] Using the initial phase range of the angular domain signal as the initial boundary, invalid phase segments are determined within the initial boundary based on energy distribution characteristics, signal-to-noise ratio characteristics, and rotational speed change characteristics. A phase segment can be determined as invalid if one of the following conditions is met: the energy is less than the preset energy corresponding to the phase segment; the signal-to-noise ratio is less than the preset signal-to-noise ratio corresponding to the phase segment; or the rotational speed change rate is higher than the preset change rate corresponding to the phase segment. The preset energy, preset signal-to-noise ratio, and preset change rate can be set specifically based on actual conditions.

[0200] If the remaining phase segment is greater than the minimum effective length and less than or equal to the maximum effective length, the complete remaining phase segment is used as the effective length for subsequent short-time Fourier transforms. If the remaining phase segment is less than the minimum effective length, boundary extension is performed according to the boundary extension strategy described in the above embodiments. If the remaining phase segment is greater than the maximum effective length, the phase segment with significant target order is selected as the effective length for subsequent short-time Fourier transforms. The minimum effective length phase segment can be twice the window length of the short-time Fourier transform, and the maximum effective phase segment can be determined based on current computing resources.

[0201] In one embodiment, when the vibration data includes multi-channel time-domain signals, the target time series can be shared to all channels as a reference target time series for all channels.

[0202] When resampling multi-channel time-series signals, for each channel's time-domain signal, the same finite-support band-limited reconstruction kernel is used to resample according to the aforementioned benchmark target time series, obtaining the angular domain signal for each channel. After obtaining the angular domain signal for each channel, independent angular domain spectrum estimation is performed for each channel's angular domain signal. This ensures that the phases of the vibration data from different channels are strictly aligned after conversion to the angular domain, allowing for direct cross-channel comparison, superposition, or joint analysis, avoiding additional phase differences introduced by inconsistencies in the resampling process.

[0203] This application also provides a method for monitoring rotating equipment, including:

[0204] The target order amplitude of the rotating device is obtained based on the signal processing method described in any of the above embodiments; the operating status of the rotating device is monitored based on the target order amplitude.

[0205] In this embodiment, after acquiring the vibration data of the rotating equipment, the vibration data is processed using the signal processing method described in the above embodiment to obtain the target order amplitude. The target order amplitude is a key characteristic indicator characterizing the operating state of the rotating equipment.

[0206] After obtaining the target order amplitude, fault analysis of rotating equipment can be performed based on the target order amplitude. For example, by analyzing the order amplitude, faults such as rotor imbalance, shaft misalignment, and gear wear in rotating equipment can be identified.

[0207] In another embodiment, the trend of target order amplitude change can be tracked. By tracking its change trend over time or mileage, faults can be predicted in advance, and fault warnings can be issued.

[0208] This embodiment provides a signal processing device, such as... Figure 3 As shown, it includes:

[0209] The acquisition unit 301 is used to acquire vibration data of the rotating equipment under operating conditions, the vibration data including time-domain signals and rotational phase sequences;

[0210] The inverse mapping unit 302 is used to perform equiangular phase inverse mapping on the rotation phase sequence to obtain the target time sequence corresponding to the uniform angle.

[0211] The resampling unit 303 is used to resample the time-domain signal according to the target time series using a finite-support band-limited reconstruction kernel to obtain the angular domain signal.

[0212] The spectrum estimation unit 304 is used to perform adaptive boundary corner domain spectrum estimation on the corner domain signal to obtain the corner domain spectrum;

[0213] Extraction unit 305 is used to extract the target order amplitude based on the angular domain spectrum.

[0214] This embodiment provides a monitoring device for rotating equipment, such as... Figure 4 As shown, it includes:

[0215] The acquisition module 401 is used to acquire the target order amplitude of the rotating device, wherein the target order amplitude is obtained based on the signal processing method described in any one of the above embodiments;

[0216] Monitoring module 402 is used to monitor the operating status of the rotating equipment based on the target order amplitude. Please refer to [link / reference]. Figure 5 , Figure 5 This is a schematic diagram of the structure of a computer device provided in an optional embodiment of the present invention, such as... Figure 5 As shown, the computer device includes one or more processors 10, memory 20, and interfaces for connecting the components, including high-speed interfaces and low-speed interfaces. The components communicate with each other via different buses and can be mounted on a common motherboard or otherwise installed as needed. The processors can process instructions executed within the computer device, including instructions stored in or on memory to display graphical information of a GUI on external input / output devices (such as display devices coupled to the interfaces). In some alternative implementations, multiple processors and / or multiple buses can be used with multiple memories and multiple memory modules, if desired. Similarly, multiple computer devices can be connected, each providing some of the necessary operations (e.g., as a server array, a group of blade servers, or a multiprocessor system).

[0217] Processor 10 may be a central processing unit, a network processor, or a combination thereof. Processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The programmable logic device may be a complex programmable logic device (CAMP), a field-programmable gate array (FPGA), a general-purpose array logic (GDA), or any combination thereof.

[0218] The memory 20 stores instructions executable by at least one processor 10 to cause the at least one processor 10 to perform the method shown in the above embodiments.

[0219] The memory 20 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the computer device as shown by a landing page for an app. Furthermore, the memory 20 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 alternative embodiments, the memory 20 may optionally include memory remotely located relative to the processor 10, which can be connected to the computer device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0220] The memory 20 may include volatile memory, such as random access memory; the memory may also include non-volatile memory, such as flash memory, hard disk or solid-state drive; the memory 20 may also include a combination of the above types of memory.

[0221] The computer device also includes a communication interface 30 for communicating with other devices or communication networks.

[0222] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code, which, when accessed and executed by the computer, processor, or hardware, implements the methods shown in the above embodiments.

[0223] This invention also provides a computer program product, which includes a computer program that, when executed by a processor, implements the above-described signal processing method or the monitoring method for rotating equipment.

[0224] In some embodiments, the computer program product involved in this disclosure may be deployed and executed on a computer device, or on multiple computer devices located in one location, or on multiple computer devices distributed in multiple locations and interconnected through a communication network. Multiple computer devices distributed in multiple locations and interconnected through a communication network may form a blockchain system.

[0225] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A signal processing method, characterized in that, The method includes: Vibration data of a rotating device under operating conditions is acquired. The vibration data includes a time-domain signal and a rotational phase sequence, wherein the rotational phase sequence is used to characterize a sequence of physical quantities representing the change of the angle of the rotating device over time. Perform an equal-angle phase inverse mapping on the revolution phase sequence to obtain the target time sequence corresponding to the uniform angle; The time-domain signal is resampled according to the target time series using a finite-support band-limited reconstruction kernel to obtain the angular domain signal; Angular domain spectrum estimation is performed on the angular domain signal to obtain the angular domain spectrum; The target order amplitude is extracted based on the angular domain spectrum.

2. The signal processing method as described in claim 1, characterized in that, Before resampling the time-domain signal according to the target time series using a finite-support band-limited reconstruction kernel, the method further includes: Calculate the instantaneous speed change rate based on the aforementioned rotational phase sequence; The length of the support window of the finite support band-limited reconstruction core is dynamically adjusted according to the instantaneous rotational speed change rate, wherein the length of the support window is negatively correlated with the absolute value of the instantaneous rotational speed change rate.

3. The signal processing method as described in claim 2, characterized in that, The resampling of the time-domain signal using a finite-support band-limited reconstruction kernel according to the target time series to obtain the angular domain signal includes: A finite number of neighborhood sampling points are selected around each target time point in the target time series. The neighborhood sampling points are the original sampling points of the time-domain signal. The target time point is used to characterize the discrete sampling time point in the time domain when the rotation phase in the rotation phase sequence reaches a set uniform angle. Calculate the time difference between each of the neighboring sampling points and the corresponding target time point; The weights of the neighborhood sampling points are determined based on the finite support band-limited reconstruction kernel and the time difference, wherein the weights are negatively correlated with the time difference; The signal value of the time domain signal corresponding to each of the neighboring sampling points is weighted and summed with the corresponding weight to obtain the angular domain signal value corresponding to the target time point. The angular domain signal values ​​corresponding to all the target time points are taken as the angular domain signals.

4. The signal processing method as described in claim 3, characterized in that, The determination of the weights of the neighborhood sampling points based on the finite support band-limited reconstruction kernel and the time difference includes: The weight is obtained by querying the pre-constructed weight query model through the time difference query. The weight query model is a correspondence model between different time differences and weights generated based on the finite support band-limited reconstruction kernel.

5. The signal processing method as described in claim 1, characterized in that, The angle domain spectrum estimation of the angle domain signal to obtain the angle domain spectrum includes: Extract the length features of the angular domain signal; Based on the length feature, adaptive boundary angular domain spectrum estimation is performed on the angular domain signal to obtain the angular domain spectrum.

6. The signal processing method as described in claim 5, characterized in that, The length feature includes the total length of the angular domain signal; The adaptive boundary corner domain spectrum estimation based on the length feature to obtain the corner domain spectrum includes: Obtain the total length of the angular domain signal; Based on the total length, adaptive boundary processing is performed on the angular domain signal to obtain the processing result; The processing result is subjected to a short-time Fourier transform to obtain the angular domain spectrum.

7. The signal processing method as described in claim 6, characterized in that, The adaptive boundary processing of the angular domain signal based on the total length yields the following results: Determine whether the total length is less than a preset length, where the preset length is α times the window length of the short-time Fourier transform, and α is a constant greater than 1; In response to the total length being less than a preset length, a boundary extension operation is performed on the corner domain signal; the corner domain signal after the boundary extension operation is used as the processing result. In response to the total length being greater than or equal to the preset length, the angular domain signal is used as the processing result.

8. The signal processing method as described in claim 7, characterized in that, The methods for determining the preset length include: Obtain the frame overlap rate of the short-time Fourier transform; The preset length is determined based on the frame overlap rate, wherein the preset length is positively correlated with the frame overlap rate.

9. The signal processing method as described in claim 1, characterized in that, Performing an equal-angle phase inverse mapping on the revolution-phase sequence to obtain the target time sequence corresponding to the uniform angle includes: Obtain the target order range, and determine the angular domain sampling rate based on the target order range; A uniformly distributed sequence of corner sampling points is constructed using the aforementioned corner sampling rate; Based on the revolution phase sequence, the angular domain sampling point sequence is subjected to phase inverse mapping to obtain the target time sequence.

10. The signal processing method as described in claim 1, characterized in that, The extraction of the target order magnitude based on the angular domain spectrum includes: Get the list of target orders; Locate the index position corresponding to the target order list on the order axis of the angular domain spectrum; Extract the amplitude data of each target order at all index positions to form an order amplitude matrix.

11. The signal processing method as described in claim 1, characterized in that, Before performing the isoangular phase inverse mapping on the revolution phase sequence, the method further includes: The rotational phase sequence is modified to be monotonic to obtain a monotonic rotational phase sequence.

12. The signal processing method as described in claim 1, characterized in that, The vibration data has multi-channel time-domain signals; Using a finite-support band-limited reconstruction kernel, the time-domain signal is resampled according to the target time series to obtain the angular domain signal, including: The target time series is shared to all channels as the benchmark target time series for all channels; For the time-domain signal of each channel, the finite-support band-limited reconstruction kernel is used to resample according to the reference target time series to obtain the angular domain signal of each channel.

13. A method for monitoring rotating equipment, characterized in that, include: The target order amplitude of the rotating device is obtained based on the signal processing method according to any one of claims 1-12; The operating status of the rotating equipment is monitored based on the target order amplitude.

14. A signal processing apparatus, characterized in that, include: The acquisition unit is used to acquire vibration data of the rotating equipment under operating conditions. The vibration data includes a time-domain signal and a rotational phase sequence, wherein the rotational phase sequence is used to characterize the physical quantity sequence of the angle of the rotating equipment changing with time. The inverse mapping unit is used to perform an equal-angle phase inverse mapping on the rotation phase sequence to obtain the target time sequence corresponding to the uniform angle. A resampling unit is used to resample the time-domain signal according to the target time series using a finite-support band-limited reconstruction kernel to obtain an angular domain signal; A spectrum estimation unit is used to perform angular domain spectrum estimation on the angular domain signal to obtain the angular domain spectrum; The extraction unit is used to extract the target order amplitude based on the angular domain spectrum.

15. A monitoring device for rotating equipment, characterized in that, include: An acquisition module is used to acquire the target order amplitude of the rotating device, wherein the target order amplitude is obtained based on the signal processing method according to any one of claims 1-12; The monitoring module is used to monitor the operating status of the rotating equipment based on the target order amplitude.

16. A computer device, characterized in that, include: A memory and a processor are communicatively connected, the memory stores computer instructions, and the processor executes the computer instructions to perform the signal processing method of any one of claims 1 to 12 or the monitoring method of the rotating device of claim 13.

17. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to perform the signal processing method of any one of claims 1 to 12 and / or the monitoring method of the rotating device of claim 13.

18. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the signal processing method of any one of claims 1 to 12 and / or performs the monitoring method of the rotating device of claim 13.

Images