Dynamic visual micro-vibration measurement method based on directional separation three-dimensional fast fourier transform

Abstract

A dynamic visual micro-vibration measurement method based on directional separation three-dimensional fast Fourier transform, utilizes an event camera to collect an asynchronous event stream of a four-beacon vibration measurement area on a measured object, performs geometric constraint filtering on an effective response area of a beacon ellipse, and establishes a target plane coordinate system according to centers of the four beacons; after mapping a target event into a target plane physical coordinate, spatial and temporal discretization is performed, polarity compensation weights are determined according to event polarity and event camera bias parameters, a weighted space-time body and a directional space-time array equivalent to a three-dimensional frequency spectrum directional slice are constructed; static texture and overall illumination drift are suppressed through double de-meaning, two-dimensional fast Fourier transform is performed on the directional space-time array, and non-zero spatial frequency energy is aggregated, so that micro-vibration spectra, equivalent acceleration spectra and envelope spectra of two reference directions are obtained; the method can retain the high time resolution advantage of the event camera under non-contact conditions, and output frequency domain representations with directional interpretability.

Description

Technical Field

[0001] This invention relates to the field of mechanical equipment micro-vibration measurement technology, and in particular to a dynamic visual micro-vibration measurement method based on direction-separated three-dimensional fast Fourier transform. Background Technology

[0002] Micro-vibrations during the operation of mechanical equipment contain health information about rotors, bearings, pumps, and transmission components. While contact-type accelerometers are mature technologies, they face challenges such as difficult placement, high maintenance costs, and limited measurement points in high-temperature environments, strong electromagnetic interference, confined installation spaces, near rotating parts, or in situations where wiring is impractical. Therefore, non-contact vibration measurement has become an important technical approach for equipment condition monitoring.

[0003] Traditional visual vibrometers typically rely on frame-based images. To capture minute displacements and high-frequency vibrations, frame-based cameras need to balance high frame rates, high resolution, and large data throughput; static backgrounds also generate a significant amount of redundant data. Event cameras, on the other hand, asynchronously output events triggered by changes in pixel brightness (Gallego, Guillermo, Tobi Delbrück, GarrickOrchard, Chiara Bartolozzi, Brian Taba, Andrea Censi, Stefan Leutenegger et al. "Event-based vision: A survey."). IEEE transactions on pattern analysis and machine intelligence The event camera (44, no. 1 (2020): 154-180.) features high temporal resolution, low redundancy, and sensitivity to minute edge movements, making it suitable for acquiring dynamic visual responses caused by equipment micro-vibrations. However, the event stream consists of discrete asynchronous events, lacking natural spatial correlation between pixel events. Simply accumulating the number of events within a region can easily lead to the loss of spatial distribution of events, making it difficult to distinguish vibrations in different directions. Constructing event frames before applying traditional template matching, corner tracking, or optical flow estimation weakens the advantages of high temporal resolution and low redundancy. Furthermore, background edges, reflective disturbances, static textures, overall brightness fluctuations, and asymmetric positive and negative polarity triggering all affect the stability of spectral peaks. Therefore, it is necessary to establish a dynamic visual micro-vibration measurement method based on standard vibration beacons, target plane coordinate mapping, polarity weighting, and direction-separated spectral analysis. Summary of the Invention

[0004] To overcome the shortcomings of the prior art, the present invention aims to provide a dynamic visual micro-vibration measurement method based on direction-separated three-dimensional fast Fourier transform, which can retain the high temporal resolution advantage of event cameras under non-contact conditions and output frequency domain characterization with directional interpretability, and is suitable for micro-vibration measurement, condition monitoring and fault characteristic frequency identification of mechanical equipment.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: A dynamic visual micro-vibration measurement method based on direction-separated three-dimensional fast Fourier transform includes: acquiring a dynamic visual event stream of a vibration beacon region on the object under test using an event camera; identifying the vibration beacon and filtering target events based on the geometric response region of the vibration beacon; establishing a target plane coordinate system based on the vibration beacon, mapping the target events to the target plane coordinate system, and discretizing them spatially and temporally; determining the polarity compensation weight of the target events based on the event polarity and the event camera bias parameters, and constructing a weighted three-dimensional spatiotemporal volume; accumulating the weighted three-dimensional spatiotemporal volume along non-interested spatial directions to obtain two directional spatiotemporal arrays equivalent to three-dimensional spectral directional slices; performing background term suppression and two-dimensional fast Fourier transform on the two directional spatiotemporal arrays, and aggregating non-zero spatial frequency components to obtain micro-vibration spectra in two reference directions; and outputting at least one of a velocity spectrum, an equivalent acceleration spectrum, and an envelope spectrum based on the micro-vibration spectra.

[0006] The vibration measurement beacon includes four circular machine vision beacons set on the surface of the object under test or on a carrier that vibrates synchronously with the object under test. The four circular machine vision beacons form a square or approximately square layout, and the two reference directions are determined by the line connecting the centers of the four circular machine vision beacons.

[0007] The selection of target events includes: performing ellipse fitting on the event accumulation region corresponding to each circular machine vision beacon to obtain the center, major semi-axis, minor semi-axis, and rotation angle of the circular machine vision beacon in the image plane; transforming the event coordinates to the local coordinate system of the corresponding ellipse; and retaining the event as the target event when it falls within the effective response region of the ellipse of any circular machine vision beacon.

[0008] Establishing the target plane coordinate system includes: using the centroids of the four circular machine vision beacons as coordinate reference points, forming the first reference direction and the second reference direction with the horizontal and vertical sides of the vibration beacons, respectively, and constructing a basis matrix from the vectors of the first reference direction and the vectors of the second reference direction, so as to map the image plane pixel offset to the target plane coordinates through oblique basis decomposition.

[0009] Spatial and temporal discretization includes: determining the spatial discretization interval based on the physical coordinate range of the target event in the first and second reference directions of the target plane; determining the temporal discretization unit based on the preset time sampling interval; and obtaining the first spatial discretization index, the second spatial discretization index, and the temporal discretization index for each target event.

[0010] The polarity compensation weight is applied to positive polarity events, and to negative polarity events, the weight is determined by the differential bias parameter, the positive event threshold bias parameter, the negative event threshold bias parameter, and the compensation coefficient. This is used to reduce the spectral amplitude bias caused by the asymmetry of positive and negative event triggering.

[0011] The two-directional spatiotemporal arrays include a first-directional spatiotemporal array and a second-directional spatiotemporal array; the first-directional spatiotemporal array is obtained by accumulating a weighted three-dimensional spatiotemporal volume along a second spatial direction and is used to extract vibration components in a first reference direction; the second-directional spatiotemporal array is obtained by accumulating a weighted three-dimensional spatiotemporal volume along a first spatial direction and is used to extract vibration components in a second reference direction.

[0012] Background suppression includes removing the temporal mean of each spatial location along the time dimension of the directional spatiotemporal array, and then removing the spatial mean of each time step along the spatial dimension, in order to reduce the impact of static texture, background illumination drift, and slow sensor response on spectral analysis.

[0013] Aggregating non-zero spatial frequency components includes: excluding the spatial DC component in the directional two-dimensional spectrum, aggregating the power spectrum of the non-zero spatial frequency components along the spatial frequency dimension, and using a one-sided spectral normalization coefficient to obtain a one-dimensional time vibration spectrum.

[0014] The envelope spectrum is obtained as follows: the dominant common vibration mode is extracted from the directional spatiotemporal array, the time derivative of the dominant common vibration mode is performed to obtain the acceleration-type time domain response, the acceleration-type time domain response is subjected to high-pass filtering and Hilbert envelope demodulation, and the envelope signal is subjected to frequency domain transformation to identify the rotation frequency, harmonics, fault characteristic frequency and its harmonic components of the mechanical equipment.

[0015] Compared with the prior art, the present invention has at least the following beneficial effects: First, this invention establishes a correspondence between event data and the target plane direction defined by the vibration beacon, and can output a micro-vibration spectrum with a clear directional meaning; Second, the present invention can reduce the influence of background events, illumination drift, static textures and polarity asymmetry on the results by filtering the effective response region of the ellipse, polarity compensation weighting and double mean removal processing. Third, the present invention utilizes a directional spatiotemporal array to obtain three-dimensional spectrum directional slices, avoiding the storage and computational burden caused by directly calculating the complete three-dimensional spectrum; Fourth, the spectrum output of this invention can further form velocity spectrum, equivalent acceleration spectrum and envelope spectrum, providing non-contact measurement basis for frequency conversion, harmonic frequency, bearing fault characteristic frequency and its harmonic frequency identification. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the overall process of the method in an embodiment of the present invention.

[0017] Figure 2 This is a schematic diagram of the beacon target plane coordinate system in Embodiment 4 of the present invention.

[0018] Figure 3 This is a schematic diagram of event filtering based on the effective response region of the beacon ellipse in an embodiment of the present invention.

[0019] Figure 4 This is a schematic diagram of the directional separation spectrum and fault feature output in an embodiment of the present invention.

[0020] Figure 5 This is a schematic diagram of the results of the main frequency components in the dynamic visual micro-vibration experiment of this invention. Detailed Implementation

[0021] The present invention will be further described below with reference to embodiments and accompanying drawings. It should be understood that the embodiments are used to explain the technical solutions of the present invention and are not intended to limit the scope of protection of the present invention. Where there is no conflict, technical features in different embodiments can be combined with each other.

[0022] like Figure 1 As shown, a dynamic visual micro-vibration measurement method based on direction separation three-dimensional fast Fourier transform is proposed for equipment micro-vibration measurement tasks. First, the event camera is used to collect the four-beacon vibration measurement area on the surface of the object under test or on the carrier that vibrates synchronously with the object under test. The event flow after noise suppression and foundation quality optimization is expressed as formula (1). (1) in, This represents a set of dynamic visual event streams obtained from a single acquisition. Indicates the first The horizontal pixel coordinates of an event in the image plane; Indicates the first The vertical pixel coordinates of the event in the image plane; Indicates the first The timestamp of each event; Indicates the first The polarity of an event; Indicates the event sequence number; Indicates the total number of events; like Figure 2As shown, in this embodiment, four circular machine vision beacons are preferably used to form a square vibration beacon to enhance event response and provide a stable target plane coordinate reference; after identifying the centers of the four beacons, the center reference point of the vibration beacon is calculated as shown in formula (2); (2) in, The centroid of the four beacon centers represents the reference point for the target plane coordinates. , , and These represent the coordinate vectors of the four beacon centers in the image plane; To further eliminate non-beacon edge events that still exist after density screening, this embodiment performs ellipse fitting on the event response region of each beacon, and the ellipse parameters of the kth beacon are expressed as formula (3). (3) in, Indicates the first Ellipse fitting parameters for each beacon; Indicates the first One ellipse center; Indicates the first The major semi-axis of an ellipse; Indicates the first The minor semi-axis of an ellipse; Indicates the first The rotation angle of the ellipse's principal axis relative to the image's horizontal axis; Indicates the beacon serial number; For any event, rotate its displacement vector relative to the center of the kth beacon to the local coordinate system of the ellipse to obtain formula (4); (4) in, Indicates the first The event relative to the first The first coordinate component of an elliptical local coordinate system; Indicates the first The event relative to the first The second coordinate component of an elliptical local coordinate system; Indicates the rotation angle A defined two-dimensional rotation matrix; and They represent the first Image planar coordinates of each event; and They represent the first The horizontal and vertical coordinates of the beacon center; If an event falls within the elliptical valid response region of any beacon, it is retained as the target event; the joint retention criterion and the subset of target events are expressed as formula (5); like Figure 3 As shown, the elliptical effective response region is used to define the effective event range of the four circular machine vision beacons in the image plane. Events falling within any elliptical region are retained as target events, while background edge events, reflective disturbance events, and non-target motion events outside the ellipse are eliminated, thereby improving the input purity of subsequent directional spatiotemporal array construction.

[0023] (5) in, Indicates the first The event relative to the first The result of determining the interior of the ellipse of a beacon; This represents an indicator function, which takes the value 1 if the condition within the parentheses is true, and 0 otherwise. Indicates the first The joint retention flag for each event; This represents the subset of target events obtained by filtering the effective response regions of the four beacons; Indicates the first One event; and Indicates the first The event in the 1st Two coordinate components in an elliptical local coordinate system; and Indicates the first The major and minor axes of an ellipse; Indicates the beacon serial number; After the target event screening is completed, the vectors of the first reference direction and the second reference direction are defined by the horizontal and vertical sides of the vibration beacon to form a basis matrix. Considering that the two directions may not be strictly orthogonal in the image plane under perspective projection, this embodiment uses oblique basis decomposition to map the pixel offset to the target plane, as shown in formula (6). (6) in, Indicates the first Normalized coordinates of an event in the first reference direction; Indicates the first Normalized coordinates of an event in the second reference direction; This represents the basis matrix consisting of two non-collinear reference direction vectors; The inverse of the basis matrix; and Represents the planar coordinates of the event image; and This represents the two coordinate components of the central reference point in the image plane.

[0024] Further combining the pixel dimensions in the first reference direction and the second reference direction, the event coordinates are converted into the physical coordinates of the target plane; assuming that the physical coordinate range of the first direction is a closed interval and the physical coordinate range of the second direction is a closed interval, and both directions are divided into spatial units, the spatial resolution and discrete index are expressed as formula (7) and formula (8) respectively. (7) in, Indicates the spatial discrete interval in the first direction; Indicates the spatial discrete interval in the second direction; and These represent the maximum and minimum values ​​of the physical coordinates in the first direction, respectively. and These represent the maximum and minimum values ​​of the physical coordinates in the second direction, respectively. Indicates the number of divisions in a single spatial direction; (8) in, Indicates the first Discrete indices of events in the first direction; Indicates the first Discrete indices of events in the second direction; Indicates the first Discrete indexes of events in the time direction; and Indicates the first The two physical coordinates of an event in the target plane; and Represents the minimum physical coordinates in two directions; and Represents the spatial discrete interval in two directions; Indicates the first The timestamp of each event; Indicates the start time of data collection; Indicates the time sampling interval; This indicates the floor function.

[0025] To mitigate the amplitude bias caused by the asymmetry in the triggering sensitivity of positive and negative polarity events, this embodiment determines the polarity compensation weight based on the event polarity and bias parameters, as shown in formula (9). (9) in, Indicates the first Polarity compensation weights for individual events; Indicates the first The polarity of an event; This represents the compensation coefficient, and its value ranges from greater than 0 to no greater than 1. Indicates the event camera differential bias parameters; This represents the negative event threshold bias parameter for the event camera; This represents the positive event threshold bias parameter for the event camera.

[0026] The weighted target events are accumulated according to the discrete index to obtain the complete three-dimensional spatiotemporal volume, as shown in formula (10); (10) in, Represents a weighted three-dimensional spacetime volume at discrete positions The value at that location; , and These represent discrete index variables in the first spatial direction, the second spatial direction, and the time direction, respectively. Indicates the event sequence number; Indicates the total number of events; to These represent the polarity compensation weights for each event; to Represent each event with respect to discrete positions. The attribution function; when the three discrete indices of the corresponding event equal to If the value is 1, then the value is 0; otherwise, the value is 0.

[0027] A complete three-dimensional spectrum can describe the coupling relationship of the target event response in two spatial and temporal directions; however, for equipment micro-vibration measurement, the focus is mainly on the temporal vibration frequencies in two reference directions. Based on the linearity of the Fourier transform, the two-direction spatiotemporal arrays can be obtained by directly summing along the non-interested spatial directions during the spatiotemporal volume construction stage, as shown in formula (11). (11) in, This represents a first-direction spatiotemporal array used to extract vibration components in the first reference direction; This indicates a second-direction spatiotemporal array used to extract vibration components in the second reference direction; Represents the discrete index variable in the first spatial direction; Represents the discrete index variable in the second spatial direction; Represents a time-discrete index variable; Indicates the number of divisions in a single spatial direction; Represents a weighted three-dimensional spacetime volume.

[0028] To suppress low-frequency background terms caused by static texture, background lighting drift, and slow sensor response, this embodiment performs double mean-reduction processing on the directional spatiotemporal array, as shown in formula (12). (12) in, Indicates direction The directional spatiotemporal array after background term suppression; Indicates direction The original orientation spacetime array on; This indicates the operation of averaging along the time dimension; This indicates the operation of calculating the mean along the spatial dimension; Indicates direction; and These represent the first reference direction and the second reference direction, respectively.

[0029] After double mean removal, two-dimensional fast Fourier transform is performed on the two-directional spatiotemporal arrays respectively, as shown in formula (13); (13) in, Represents the two-dimensional time-space spectrum in the first direction; Represents the two-dimensional time-space spectrum in the second direction; Indicates the spatial frequency index in the first direction; Indicates the spatial frequency index in the second direction; Represents physical frequency in the time dimension; Represents the two-dimensional Fast Fourier Transform operator; and These represent the directional spatiotemporal arrays after background suppression in two directions, respectively.

[0030] After excluding the spatial DC component, the power spectrum of the non-zero spatial frequency component is aggregated along the spatial frequency dimension to obtain a one-dimensional time vibration spectrum, as shown in formula (14). (14) in, This represents the power aggregation of the non-zero spatial frequency components in the first direction; This represents the power aggregation of the non-zero spatial frequency components in the second direction; This represents the one-dimensional time vibration spectrum in the first reference direction; This represents the one-dimensional time vibration spectrum in the second reference direction; Represents physical frequency in the time dimension; Indicates the number of divisions in a single spatial direction; This represents the number of discrete sampling points on the time axis; This represents the number of non-zero spatial frequencies participating in aggregation in the first direction; This indicates the number of non-zero spatial frequencies participating in aggregation in the second direction; to Indicates the non-zero spatial frequency indices of the first direction participating in the aggregation; to Indicates the non-zero spatial frequency indices of the aggregations in the second direction; and These represent two-dimensional spectra in two directions, respectively. This represents the operator for complex number magnitude.

[0031] When the target texture and lighting conditions are relatively stable, the cumulative polarity value in the spatial-temporal field can approximate the local brightness change rate, and the directional vibration spectrum can be regarded as a velocity spectrum. According to the frequency domain differential relationship, the velocity spectrum is converted into an equivalent acceleration spectrum, as shown in formula (15); (15) in, Indicates direction The equivalent acceleration spectrum on; Indicates the vibration frequency; Indicates direction The velocity-type vibration spectrum on the surface; Indicates direction; and These represent two reference directions; This represents the constant value of pi.

[0032] For impact-type or modulation-type faults such as bearings, this embodiment can also extract the dominant common vibration mode from the directional spatiotemporal array. Let the spatial-time matrix of any direction be P, then the covariance matrix and the principal component time-domain signal are expressed as formula (16); (16) in, This represents a covariance matrix constructed from the directional space-time matrix; Represents the space-time matrix in any direction; Representation matrix transpose; This represents the time-domain signal corresponding to the dominant common vibration mode; Representation matrix The eigenvector corresponding to the largest eigenvalue; Represents a time variable; This indicates a matrix or vector transpose operation.

[0033] Subsequently, the principal component time-domain signal is differentiated by time to obtain the acceleration-type time-domain response, and the low-frequency trend is removed by high-pass filtering; the analytic signal is constructed using Hilbert transform, and its envelope is expressed as formula (17). (17) in, The envelope signal representing the acceleration-type time-domain response; Represents a time variable; Represents the Hilbert transform operator; This represents the acceleration-type time-domain response after differentiating and filtering the principal component time-domain signal; This represents the operator for complex number magnitude.

[0034] The acceleration envelope spectrum can be obtained by performing a Fourier transform on the envelope signal. This spectrum can highlight the amplitude modulation frequency components caused by impulse faults and is suitable for identifying the characteristic frequencies of frequency conversion, harmonics, and bearing faults, as well as their harmonic components.

[0035] like Figure 4 As shown, in this embodiment, the first-direction spatiotemporal array and the second-direction spatiotemporal array are respectively input into the two-dimensional fast Fourier transform and spatial frequency aggregation module to form velocity-type spectra and equivalent acceleration spectra in two reference directions, and further merged into the envelope spectrum analysis module to output the fault characteristic frequency identification results. This figure corresponds to the direction separation spectrum processing process of formulas (11) to (17).

[0036] In experimental verification, this method can correspond to the envelope spectrum of a contact accelerometer and can decompose the main vibration frequency components in the first and second reference directions of the vibration beacon. Compared with methods that only count the number of events, this method retains directional information. Compared with methods based on event centroid tracking, this method does not rely on a single overall centroid trajectory and can improve the identifiability of the dominant frequency and harmonic components through the aggregation of non-zero spatial frequency energy. Therefore, this method can provide non-contact dynamic visual measurement results with directional interpretability for mechanical equipment fault diagnosis.

[0037] Based on experimental data from a pump-machine cycle fault simulation test bench, this embodiment can output the main frequency components corresponding to the envelope spectrum of the contact accelerometer under various operating conditions, including inner ring fault, outer ring fault, and normal operation. For example... Figure 5As shown, under the inner ring fault condition of 10 Hz, this embodiment can identify the 10 Hz and 20 Hz frequency shift and harmonics, and highlight the fault characteristic frequencies of 52.3 Hz and 104.7 Hz and their harmonics; under the inner ring fault condition of 6 Hz, it can identify the 6 Hz and 12 Hz frequency shift and harmonics, and highlight the fault characteristic frequencies of 30.7 Hz and 61.4 Hz and their harmonics; under the outer ring fault condition of 10 Hz, it can identify the fault characteristic frequencies of 37.2 Hz, 74.4 Hz, and 111.6 Hz and their harmonics; under the normal 10 Hz condition, it can identify the 10 Hz, 20 Hz, and 30 Hz frequency shift and harmonics, and present other prominent frequency components such as 62.4 Hz and 92.2 Hz.

[0038] The above experimental results show that this embodiment can not only obtain the main frequency components corresponding to the contact sensor under non-contact conditions, but also output the spectrum results in the first reference direction and the second reference direction respectively. Compared with the method that only counts the total number of regional events, this embodiment retains the spatial distribution and direction information of events. Compared with the method based on event centroid tracking, this embodiment enhances the identifiability of the main frequency, harmonics and fault characteristic frequencies through non-zero spatial frequency aggregation. Therefore, it has the advantages of accurate frequency identification, clear direction decomposition and strong resistance to background disturbance.

Claims

1. A dynamic visual micro-vibration measurement method based on direction-separated three-dimensional fast Fourier transform, characterized in that, include: The dynamic visual event stream of the vibration beacon area on the object under test is captured using an event camera; Identify vibration beacons and filter target events based on the geometric response regions of the vibration beacons; A target plane coordinate system is established based on the vibration beacon. The target events are mapped to the target plane coordinate system and discretized in space and time. The polarity compensation weight of the target events is determined according to the event polarity and the event camera bias parameters, and a weighted three-dimensional spatiotemporal volume is constructed. The weighted three-dimensional spatiotemporal volume is accumulated along the non-interested spatial direction to obtain two spatiotemporal arrays equivalent to three-dimensional spectrum directional slices. Background term suppression and two-dimensional fast Fourier transform are performed on the two spatiotemporal arrays, and the non-zero spatial frequency components are aggregated to obtain the micro-vibration spectrum in two reference directions. The output is based on at least one of the following: a velocity spectrum, an equivalent acceleration spectrum, and an envelope spectrum, derived from the micro-vibration spectrum.

2. The method according to claim 1, characterized in that, The vibration measurement beacon includes four circular machine vision beacons set on the surface of the object under test or on a carrier that vibrates synchronously with the object under test. The four circular machine vision beacons form a square or approximately square layout, and the two reference directions are determined by the line connecting the centers of the four circular machine vision beacons.

3. The method according to claim 2, characterized in that, The selection of target events includes: performing ellipse fitting on the event accumulation region corresponding to each circular machine vision beacon to obtain the center, major semi-axis, minor semi-axis, and rotation angle of the circular machine vision beacon in the image plane; transforming the event coordinates to the local coordinate system of the corresponding ellipse; and retaining the event as the target event when it falls within the effective response region of the ellipse of any circular machine vision beacon.

4. The method according to claim 2, characterized in that, Establishing the target plane coordinate system includes: using the centroids of the four circular machine vision beacons as coordinate reference points, forming the first reference direction and the second reference direction with the horizontal and vertical sides of the vibration beacons, respectively, and constructing a basis matrix from the vectors of the first reference direction and the vectors of the second reference direction, so as to map the image plane pixel offset to the target plane coordinates through oblique basis decomposition.

5. The method according to claim 4, characterized in that, Spatial and temporal discretization includes: determining the spatial discretization interval based on the physical coordinate range of the target event in the first and second reference directions of the target plane; determining the temporal discretization unit based on the preset time sampling interval; and obtaining the first spatial discretization index, the second spatial discretization index, and the temporal discretization index for each target event.

6. The method according to claim 1, characterized in that, The polarity compensation weight is applied to positive polarity events, and to negative polarity events, the weight is determined by the differential bias parameter, the positive event threshold bias parameter, the negative event threshold bias parameter, and the compensation coefficient. This is used to reduce the spectral amplitude bias caused by the asymmetry of positive and negative event triggering.

7. The method according to claim 4, characterized in that, The two-directional spatiotemporal arrays include a first-directional spatiotemporal array and a second-directional spatiotemporal array; the first-directional spatiotemporal array is obtained by accumulating a weighted three-dimensional spatiotemporal volume along a second spatial direction and is used to extract vibration components in a first reference direction; the second-directional spatiotemporal array is obtained by accumulating a weighted three-dimensional spatiotemporal volume along a first spatial direction and is used to extract vibration components in a second reference direction.

8. The method according to claim 1, characterized in that, Background suppression includes removing the temporal mean of each spatial location along the time dimension of the directional spatiotemporal array, and then removing the spatial mean of each time step along the spatial dimension, in order to reduce the impact of static texture, background illumination drift, and slow sensor response on spectral analysis.

9. The method according to claim 1, characterized in that, Aggregating non-zero spatial frequency components includes: excluding the spatial DC component in the directional two-dimensional spectrum, aggregating the power spectrum of the non-zero spatial frequency components along the spatial frequency dimension, and using a one-sided spectral normalization coefficient to obtain a one-dimensional time vibration spectrum.

10. The method according to claim 1, characterized in that, The envelope spectrum is obtained as follows: the dominant common vibration mode is extracted from the directional spatiotemporal array, the time derivative of the dominant common vibration mode is performed to obtain the acceleration-type time domain response, the acceleration-type time domain response is subjected to high-pass filtering and Hilbert envelope demodulation, and the envelope signal is subjected to frequency domain transformation to identify the rotation frequency, harmonics, fault characteristic frequency and its harmonic components of the mechanical equipment.

Images