A granular foreign object collision noise detection method and system
By combining multi-channel signal processing and finite element simulation with Bayesian iterative algorithms, spatial localization and material differentiation of multi-source particulate foreign objects were achieved, solving the problem of signal aliasing and decoupling in existing technologies, and improving the accuracy and reliability of fault diagnosis.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHENZHEN YESSYS TECH LTD
- Filing Date
- 2026-05-09
- Publication Date
- 2026-07-21
AI Technical Summary
Existing technologies cannot effectively separate and locate collision signals from multi-source particulate foreign objects, resulting in a lack of specificity and accuracy in fault diagnosis, and an inability to accurately identify particle material and collision force information.
By acquiring multi-channel vibration response signals, performing filtering and finite element dynamics simulation, collision template signals of different materials are obtained. Potential collision sources are determined using residual norm and Bayesian iterative formula, coarse localization and signal separation are performed, and finally, particle material and collision force are identified through temporal deconvolution reconstruction and time-frequency feature matching.
It enables spatial localization and material differentiation of multi-source particulate foreign objects, improves detection accuracy and fault diagnosis dimensions, breaks through the limitations of traditional methods, and can accurately identify the location and material type of particulate foreign objects.
Smart Images
Figure CN122430447A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of particle foreign object detection technology, and more specifically, to a method and system for detecting particle foreign object collision noise. Background Technology
[0002] Foreign object detection is one of the key technologies to ensure the safe operation of high-reliability sealed electronic devices, engines, and other precision equipment. During the manufacturing, assembly, and long-term service of devices, metal debris, non-metallic particles, and other foreign objects may remain or be generated inside the cavity. These particles may shift under vibration or impact, potentially causing fatal failures such as short circuits, wear, or blockages.
[0003] Particle collision noise detection is currently the most widely used non-destructive testing (NDT) technology. Its basic principle involves placing the test piece under standard impact excitation and using piezoelectric sensors attached to the outer wall of the cavity to collect transient acoustic emission signals generated by the collision of particles with the inner wall. The presence of foreign objects is then determined by setting an amplitude threshold. However, when multiple particles collide simultaneously or sequentially within the cavity, signals from different sources overlap. Current technologies cannot separate the independent collision components corresponding to each particle and cannot determine the specific location of the particles within the cavity. This leads to a lack of targeted troubleshooting or repair, and makes it difficult to assess the potential risk level. Furthermore, current technologies only perform qualitative judgments based on amplitude thresholds and cannot quantitatively invert the true temporal waveform of the collision force, limiting the depth of fault diagnosis. Therefore, how to achieve spatial localization and material differentiation of multi-source particle foreign objects to improve the detection accuracy and fault diagnosis dimensions remains a challenge for the industry. Summary of the Invention
[0004] This application provides a method and system for detecting collision noise of particulate foreign objects, which can realize the spatial positioning and material differentiation of multi-source particulate foreign objects, thereby improving the detection accuracy and fault diagnosis dimensions of particulate foreign objects.
[0005] In a first aspect, this application provides a method for detecting particle foreign object collision noise, the detection method comprising the following steps: Multi-channel vibration response signals are acquired, and filtering processing is performed on the multi-channel vibration response signals to obtain noise curves; Acquire collision template signals of different materials, determine the residual norm between the collision template signal and the noise curve, and determine the effective collision parameters of the potential collision source based on the residual norm; The coarse localization region is marked based on the effective collision parameters, and the failure point in the coarse localization region is determined according to the extreme value of the residual norm. The collision signal subspace is separated from the noise curve based on the prior accuracy and posterior covariance matrix of the failure point. Temporal deconvolution reconstruction and time-frequency feature matching and identification are performed on the collision signal subspace to obtain the peak collision force and foreign object material type at the failure point.
[0006] In this embodiment, a piezoelectric accelerometer array is used to acquire multi-channel vibration response signals.
[0007] In this embodiment, the filtering process performed on the multi-channel vibration response signal to obtain the noise curve specifically includes: The multi-channel vibration response signal is subjected to bandpass filtering to obtain a filtered signal; The filtered signal is normalized and energy aligned to obtain the noise curve.
[0008] In this embodiment, obtaining collision template signals of different materials specifically includes: Establish a finite element dynamic model for particle-foreign object collisions; Transient dynamic simulations of different materials were performed on the finite element dynamic model to extract collision template signals for different materials.
[0009] In this embodiment, determining the residual norm between the collision template signal and the noise curve specifically includes: The noise curve is divided into different time segments according to a time window, wherein the length of each time segment is equal to the length of the collision template signal; The Euclidean distance between each time segment and the collision template signal of each material is calculated, thereby obtaining the residual norm between the collision template signal and the noise curve.
[0010] In this embodiment, determining the effective collision parameters of a potential collision source based on the residual norm specifically includes: Determine the initial prior accuracy and posterior covariance matrix of potential collision sources; The prior accuracy and posterior covariance matrix of the potential collision source are updated using the Bayesian iterative formula based on the residual norm, resulting in the updated prior accuracy and posterior covariance matrix. The effective collision parameters of potential collision sources are calculated based on the updated prior accuracy and posterior covariance matrix.
[0011] In this embodiment, marking the coarse localization region based on the effective collision parameters and determining the failure point in the coarse localization region based on the extreme value of the residual norm specifically includes: Spatial clustering is performed on the effective collision parameters, and continuous regions where the effective collision parameters exceed a preset background threshold are marked as coarse localization regions; Within the coarse localization region, local maxima are searched based on the variation curve of the residual norm. The failure points in the coarse localization region are obtained based on the local maxima filtering.
[0012] In this embodiment, separating the collision signal subspace from the noise curve based on the prior accuracy and posterior covariance matrix of the failure point specifically includes: The spatial filtering weights are determined based on the posterior covariance matrix of the failure points; The noise curve is fused using the spatial filtering weights to obtain the weighted timing signal of the failure point. A time window is extracted from the weighted time sequence signal and cross-interference separation is performed to obtain the collision signal subspace.
[0013] In this embodiment, temporal deconvolution reconstruction and time-frequency feature matching are performed on the collision signal subspace to obtain the peak collision force and foreign object material type at the failure point, specifically including: Perform temporal deconvolution reconstruction on the collision signal subspace to obtain the temporal waveform of the collision force at the failure point. The maximum amplitude value is extracted from the collision force time-series waveform as the collision force peak value; A time-frequency transformation is performed on the collision force time-series waveform, and the transformed time-frequency characteristics are matched and identified with the collision template signal to obtain the foreign object material type at the failure point.
[0014] Secondly, this application provides a particle collision noise detection system for performing a particle collision noise detection method, the detection system comprising: The vibration acquisition module is used to acquire multi-channel vibration response signals and perform filtering processing on the multi-channel vibration response signals to obtain noise curves. The template matching module is used to acquire collision template signals of different materials, determine the residual norm between the collision template signal and the noise curve, and determine the effective collision parameters of the potential collision source based on the residual norm. The positioning and separation module is used to mark the coarse positioning area based on the effective collision parameters, determine the failure point in the coarse positioning area according to the extreme value of the residual norm, and separate the collision signal subspace from the noise curve based on the prior accuracy and posterior covariance matrix of the failure point. The inversion identification module is used to perform temporal deconvolution reconstruction and time-frequency feature matching identification on the collision signal subspace to obtain the peak collision force and foreign object material type of the failure point.
[0015] The technical solutions provided by the embodiments disclosed in this application have the following beneficial effects: The technical solution provided in this application can improve the detection accuracy and fault diagnosis dimensions of particulate foreign objects. First, by acquiring multi-channel vibration response signals and performing filtering processing, a noise curve reflecting the global vibration state of the cavity is obtained, laying a wide-bandwidth, low-interference signal foundation for subsequent multi-source collision separation. Second, by acquiring collision template signals of different materials and matching them with the residual norm of the noise curve, the effective collision parameters of each potential collision source are determined, overcoming the limitation of traditional methods that cannot distinguish between metallic and non-metallic particles, and realizing the preliminary identification of foreign object materials and spatial quantification of collision energy distribution. Then, based on the effective collision parameters, coarse localization area marking is performed, and the failure point is determined using the residual norm extremum. Furthermore, based on the prior accuracy and posterior covariance of the failure point, the system can be further refined. The difference matrix separates the collision signal subspace corresponding to each failure point from the global noise, solving the problem of difficult decoupling of multi-particle signal aliasing. It realizes independent signal extraction and coarse spatial localization of multiple foreign objects under a single impact excitation. Finally, by performing temporal deconvolution reconstruction and time-frequency feature matching and recognition on each collision signal subspace, the peak waveform of the collision force at each failure point is obtained and the material type of the foreign object is accurately identified, further improving the quantitative dimension and reliability of fault diagnosis. The technical solution of this application can effectively solve the core defects of existing particle foreign object collision noise detection, such as the inability to distinguish, locate, identify materials, and lack of force information, through a progressive technical path from global response acquisition to local subspace separation and from material prediction to force waveform reconstruction.
[0016] In summary, the technical solution adopted in this application can realize the spatial positioning and material differentiation of multi-source particulate foreign objects, thereby improving the detection accuracy and fault diagnosis dimensions of particulate foreign objects. Attached Figure Description
[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0018] Figure 1 This is a flowchart of a particulate foreign object collision noise detection method provided in this application; Figure 2 This is a schematic diagram of the distribution of the piezoelectric accelerometer array provided in this application; Figure 3 This is an exemplary flowchart for determining the collision signal subspace provided in this application; Figure 4 This is a module structure diagram of a particle collision noise detection system provided in this application. Detailed Implementation
[0019] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0020] Example 1: To better understand the technical solution of this application, the above technical solution will be described in detail below with reference to the accompanying drawings and specific embodiments. Figure 1 As shown in the figure, this is a flowchart of a particulate foreign object collision noise detection method according to this embodiment of the present application. The detection method includes the following steps: In step S1, multi-channel vibration response signals are acquired, and filtering processing is performed on the multi-channel vibration response signals to obtain noise curves; In practical implementation, multi-channel vibration response signals are acquired using a piezoelectric accelerometer array. Preferably, such as... Figure 2 As shown, this figure is a schematic diagram of the distribution of a piezoelectric accelerometer array according to the present application. In this embodiment, the piezoelectric accelerometer array refers to a sensor set formed by fixing multiple piezoelectric accelerometers to the outer wall of the cavity of the test piece according to a preset geometric layout. Each sensor independently outputs a vibration response signal of one channel. The preset geometric layout is a uniform arrangement at equal angles along the circumference of the outer wall of the cavity of the test piece. The number of piezoelectric accelerometers in this application can be set to 9, which is beneficial for achieving sufficient spatial sampling of the vibration field in the circumference of the cavity. The configuration of 9 channels facilitates symmetry analysis and signal differential processing, and reduces common-mode noise interference. CH1-9 refer to the numbers of the piezoelectric accelerometers. The failure point mark indicates the number of points where particle collisions with foreign objects are detected, which can be indicated by flashing of the schematic diagram of the distribution of the piezoelectric accelerometer array. It should be noted that the multi-channel vibration response signal refers to a set of voltage sequences that change over time. The multi-channel vibration response signal is used to reflect the elastic wave propagation characteristics when particles collide with the wall.
[0021] In this embodiment, the noise curve is obtained by performing filtering processing on the multi-channel vibration response signal in the following manner: The multi-channel vibration response signal is subjected to bandpass filtering to obtain a filtered signal; The filtered signal is normalized and energy aligned to obtain the noise curve.
[0022] In practice, firstly, the multi-channel vibration response signals are input to a bandpass filter. This bandpass filter retains the effective frequency components generated by particle collisions and suppresses low-frequency environmental vibrations and high-frequency electrical noise. Based on the elastic wave propagation characteristics of the test component's cavity material and the expected frequency band of the collision signal, the lower and upper cutoff frequencies of the bandpass filter are set, ensuring that the filter outputs a filtered signal for each channel. Then, energy normalization is performed on the filtered signal for each channel. Energy normalization involves calculating the root mean square (RMS) value of the filtered signal for each channel on the time axis and dividing the filtered signal by its RMS value, giving each channel a uniform energy scale. All channel signals after energy normalization are stacked and arranged according to their sampling times to obtain a noise curve. This noise curve is a matrix-like data structure, where rows correspond to sampling times, columns correspond to sensor channel numbers, and each element in the matrix represents the vibration amplitude of that channel after filtering and energy normalization at that sampling time.
[0023] It should be noted that a bandpass filter is an electronic or digital filter that only allows signals within a specific frequency range to pass through while attenuating frequency components outside the range. In this embodiment, the bandpass filter is used to extract characteristic frequency bands related to particle collisions from the broadband vibration response signal to reduce the interference of irrelevant vibration components on subsequent residual norm calculations. Energy normalization refers to the operation of adjusting the signals of each channel to the same amplitude dimension, which is used to eliminate the problem of inconsistent amplitudes between channels caused by factors such as differences in sensor sensitivity and different installation coupling. The noise curve is a set representation of the multi-channel vibration response signals after filtering and normalization, and can be used as the benchmark data for residual norm calculation.
[0024] In step S2, collision template signals of different materials are acquired, the residual norm between the collision template signal and the noise curve is determined, and the effective collision parameters of the potential collision source are determined based on the residual norm.
[0025] In this embodiment, the collision template signals of different materials can be obtained in the following ways: Establish a finite element dynamic model for particle-foreign object collisions; Transient dynamic simulations of different materials were performed on the finite element dynamic model to extract collision template signals for different materials.
[0026] In practice, firstly, a three-dimensional solid model can be established based on the actual geometric dimensions and cavity structure of the tested part. The three-dimensional solid model is then imported into finite element analysis software for tetrahedral mesh generation to obtain a discretized finite element dynamic model. In this finite element dynamic model, the elastic modulus, Poisson's ratio, and density of the tested part material can be set as structural property parameters. The contact type between the foreign particles and the inner wall of the cavity is set as face-to-face frictional contact, and the friction coefficient is set to simulate energy dissipation during the collision process. Then, the finite element dynamics model is assigned the material properties of metal, non-metal, and mixed materials as the material parameters of the particulate foreign object. The same standard impact excitation as in the actual measurement is applied to the outside of the cavity of the finite element dynamics model as the boundary condition. Explicit dynamic transient solution is performed to calculate the acceleration response generated when the particulate foreign object collides with the inner wall of the cavity at each time step, and the acceleration time series waveform corresponding to each material is obtained. A fixed-length time window before and after the collision occurs is extracted from the acceleration time series waveform of each material, and the extracted waveforms are stored according to the material category. Each material category corresponds to a set of acceleration time series waveforms, so that the set of all acceleration time series waveforms is used as the collision template signal.
[0027] It should be noted that the finite element dynamics model is a numerical calculation model based on finite element technology used to simulate the mechanical behavior of physical systems under dynamic loads. In this embodiment, it is used to predict the vibration response generated when foreign particles of different materials collide with the inner wall of the cavity in a computer environment. Face-to-face frictional contact refers to the contact algorithm that simulates the frictional force generated when two surfaces come into contact with each other and slide relative to each other, and is used to reflect the energy loss during the collision process between the foreign particles and the inner wall of the cavity. Explicit dynamic transient solution refers to the calculation process of solving the differential equation of motion step by step using the time integration method, which is suitable for the simulation analysis of short-term transient events such as collisions and impacts. The collision template signal is a set of acceleration time-series waveforms extracted from the simulation results, which can be used as a reference benchmark for comparison with the measured noise curve in the calculation of residual norm.
[0028] In this embodiment, the residual norm between the collision template signal and the noise curve can be determined in the following way: The noise curve is divided into different time segments according to a time window, wherein the length of each time segment is equal to the length of the collision template signal; The Euclidean distance between each time segment and the collision template signal of each material is calculated, thereby obtaining the residual norm between the collision template signal and the noise curve.
[0029] In practical implementation, firstly, the total duration of the noise curve can be taken as the sampling time length, and the length of the collision template signal can be taken as the time window length. A sliding time window is used to divide the noise curve into multiple continuous time segments, where the length of each time segment is equal to the length of the collision template signal. The sliding step size between two adjacent time segments is set to half the time window length, so that there is partial overlap between adjacent time segments. Then, for each time segment, the Euclidean distance between the time segment and the collision template signal of each material is calculated. The smaller the Euclidean distance, the higher the similarity between the time segment and the material template. The Euclidean distances calculated for all time segments and all material templates are arranged in chronological order into a three-dimensional array. The first dimension of this three-dimensional array corresponds to the time segment number, the second dimension corresponds to the material type, and the third dimension stores the Euclidean distance value. This three-dimensional array can be used as the residual norm between the collision template signal and the noise curve.
[0030] It should be noted that the residual norm is a multidimensional data structure containing multiple Euclidean distance values. It is used to quantify the degree of difference between templates of different materials and at different times and the measured noise curve. The smaller value in the residual norm indicates the presence of a collision event with a high degree of matching with the template material in the vicinity at that time, which can provide a basic similarity measure for determining effective collision parameters. In this embodiment, the sliding time window is a means of dividing a long sequence signal into multiple subsequences of equal length. By setting overlap, it is possible to avoid collision events falling exactly at the window boundary, which would lead to missed detection.
[0031] In this embodiment, the effective collision parameters of a potential collision source can be determined based on the residual norm in the following manner: Determine the initial prior accuracy and posterior covariance matrix of potential collision sources; The prior accuracy and posterior covariance matrix of the potential collision source are updated using the Bayesian iterative formula based on the residual norm, resulting in the updated prior accuracy and posterior covariance matrix. The effective collision parameters of potential collision sources are calculated based on the updated prior accuracy and posterior covariance matrix.
[0032] In practice, the interior of the cavity of the test part is first discretized into three-dimensional spatial grid points. Each of the three-dimensional spatial grid points is used as a potential collision source. An initial prior accuracy value and an initial posterior covariance matrix are assigned to each potential collision source. The initial prior accuracy value can be set to any small positive number, and the initial posterior covariance matrix is set to a unit diagonal matrix with diagonal elements equal to the reciprocal of the initial prior accuracy value. Then, each Euclidean distance value in the residual norm is obtained, and the Euclidean distance value is used as the observation residual of the potential collision source under the corresponding time segment and the corresponding material type. The prior accuracy and posterior covariance matrix of the potential collision source are updated according to the Bayesian iterative formula. The Bayesian iterative formula includes the posterior covariance matrix update step and the prior accuracy update. The posterior covariance matrix update is obtained by adding the inverse of the current posterior covariance matrix to the product of the transpose of the observation matrix and the observation matrix and the noise accuracy, and then taking the inverse. The prior accuracy update is obtained by adding the square of the observation residual to the current prior accuracy and the ratio of the trace of the posterior covariance matrix to the trace of the posterior covariance matrix. The above update process is repeated until the rate of change of the prior accuracy in two adjacent iterations converges. The prior accuracy and posterior covariance matrix at the time of stopping iteration can be used as the updated prior accuracy and updated posterior covariance matrix of the potential collision source, respectively. Finally, the product of the updated posterior covariance matrix and the updated prior accuracy can be used as the effective collision parameter of the potential collision source.
[0033] It should be noted that prior accuracy refers to the confidence parameter in Bayesian inference, which represents the degree of initial uncertainty about a potential collision source. The larger the prior accuracy value, the more reliable the initial estimate. The posterior covariance matrix is a matrix used to describe the uncertainty of the state estimate of a potential collision source after fusing observation data. The trace of the posterior covariance matrix reflects the total uncertainty of the estimate. The effective collision parameter is a scalar value that combines the prior accuracy and the posterior covariance matrix. It is used to quantify the probability of the presence of particulate foreign objects at each potential collision source. The larger the effective collision parameter, the more likely there are foreign objects at that location.
[0034] In step S3, coarse localization regions are marked based on effective collision parameters, and failure points in the coarse localization regions are determined based on the extreme values of the residual norm. The collision signal subspace is then separated from the noise curve based on the prior accuracy and posterior covariance matrix of the failure points.
[0035] In this embodiment, the coarse localization region is marked based on the effective collision parameters, and the failure point in the coarse localization region is determined according to the extreme value of the residual norm. Specifically, the following method can be used: Spatial clustering is performed on the effective collision parameters, and continuous regions where the effective collision parameters exceed a preset background threshold are marked as coarse localization regions; Within the coarse localization region, local maxima are searched based on the variation curve of the residual norm. The failure points in the coarse localization region are obtained based on the local maxima filtering.
[0036] In practice, firstly, the mean and standard deviation of the effective collision parameters of all potential collision sources are calculated. The sum of twice the mean and standard deviation is used as a background threshold. Potential collision sources whose effective collision parameters exceed the background threshold are marked as high-response points. Then, spatial clustering is performed on these high-response points. Spatial clustering refers to grouping spatially adjacent high-response points into the same cluster. Specifically, a connected component labeling algorithm based on Euclidean distance is used to treat high-response points as adjacent points and group them into the same cluster. The continuous spatial region covered by each cluster is used as the coarse localization region. Then, within each coarse localization region, the curve of the residual norm changing with spatial location is visualized using a visualization library. This curve is obtained by arranging the effective collision parameter values of each potential collision source in the region according to spatial coordinates. Local maxima are searched on this curve. A local maximum is a spatial location where the residual norm value of a point is greater than the residual norm values of all its neighboring spatial points. Finally, the spatial locations corresponding to all local maxima in each coarse localization region are used as failure points in that region. Multiple local maxima within the same coarse localization region correspond to different failure points.
[0037] It should be noted that the coarse localization region refers to a continuous spatial range delineated from high-response points through spatial clustering. This is used to roughly delineate local areas within the cavity where particulate matter may exist, reducing the spatial search range for subsequent fine-grained calculations. Local maxima are peak points in the residual norm variation curve, describing the location where the matching degree between the collision template and the noise curve reaches a local maximum, indicating that an actual particle collision event is most likely to occur at that location. Failure points are the finally confirmed locations of particulate matter collisions. Each failure point corresponds to an independent particulate matter collision source and can be used for signal separation and material identification. The same coarse localization region may have multiple local maxima points, therefore the number of failure points is uncertain, which helps avoid missing particulate matter collision detection.
[0038] Preferably, in this embodiment, reference Figure 3 As shown, this figure is an exemplary flowchart for determining the collision signal subspace according to the present application. In this embodiment, the collision signal subspace is separated from the noise curve based on the prior accuracy and posterior covariance matrix of the failure point, which can be achieved by the following steps: First, in step S31, the spatial filtering weights are determined based on the posterior covariance matrix of the failure points; Then, in step S32, the noise curve is fused by channel weighting using the spatial filtering weights to obtain the weighted timing signal of the failure point; Finally, in step S33, a time window is extracted from the weighted timing signal and cross-interference separation is performed to obtain the collision signal subspace.
[0039] In practice, firstly, all elements on the diagonal of the updated posterior covariance matrix are extracted to form the posterior variance vector. The reciprocal of each element in the posterior variance vector is divided by the sum of the reciprocals of all elements to obtain the spatial filtering weight vector. Each element in the spatial filtering weight vector corresponds to a sensor channel, and the sum of all elements is 1. Then, the noise curve matrix is multiplied by the spatial filtering weight vector. Multiplication involves performing a dot product operation between each row vector of the noise curve matrix and the spatial filtering weight vector to obtain a new time series. This time series is used as the weighted time sequence signal for the failure point. Finally, the time point at which the failure point reaches a local maximum in the residual norm determines the center position of the time window. The time window is then extended forward and backward by half the preset length from the center position. Signal segments within this time window are extracted from the weighted time series signal. Principal component analysis is then performed on the set of signal segments extracted from all failure points. Principal component analysis involves constructing a matrix from multiple signal segments and performing singular value decomposition. The principal components with the highest contribution rates are retained as the independent signal components of each failure point, and the independent signal components corresponding to the failure point are used as the collision signal subspace of that failure point.
[0040] It should be noted that the spatial filtering weight vector is a set of coefficients calculated based on the posterior covariance matrix. It is used to spatially enhance the vibration energy from the direction of the failure point and suppress interference from other directions, which is beneficial for extracting signals from specific failure points from multi-channel mixed signals. The weighted time-series signal is a one-dimensional time series obtained by linearly fusing multi-channel signals according to the spatial filtering weights. It is used to describe the main vibration components generated by particle collisions at the failure point. The collision signal subspace is the independent signal components further separated from the weighted time-series signal, and the pure collision signal obtained after removing cross-interference between different failure points. It can be used for temporal deconvolution reconstruction and material identification.
[0041] In step S4, temporal deconvolution reconstruction and time-frequency feature matching and identification are performed on the collision signal subspace to obtain the peak collision force and foreign object material type of the failure point.
[0042] In this embodiment, temporal deconvolution reconstruction and time-frequency feature matching are performed on the collision signal subspace to obtain the peak collision force and foreign object material type at the failure point. Specifically, this can be achieved in the following manner: Perform temporal deconvolution reconstruction on the collision signal subspace to obtain the temporal waveform of the collision force at the failure point. The maximum amplitude value is extracted from the collision force time-series waveform as the collision force peak value; A time-frequency transformation is performed on the collision force time-series waveform, and the transformed time-frequency characteristics are matched and identified with the collision template signal to obtain the foreign object material type at the failure point.
[0043] In specific implementation, firstly, the transfer function matrix of the test object from the particle collision point to each sensor channel is obtained. This transfer function matrix is a matrix pre-calibrated through pulse excitation experiments on the test object. The number of rows in this transfer function matrix equals the total number of sensor channels, and the number of columns equals the length of the collision signal subspace. Then, temporal deconvolution reconstruction is performed between the transfer function matrix and the collision signal subspace. Temporal deconvolution reconstruction refers to using a least-squares iterative algorithm to solve for a collision force time-series waveform, such that the convolution result of this collision force time-series waveform and the transfer function matrix most closely approximates the collision signal. In the actual implementation, a zero vector is first initialized as the initial estimate of the collision force time-series waveform. Then, the convolution result of the current estimate and the transfer function matrix is calculated. The convolution result is subtracted from the collision signal subspace to obtain the error vector. Based on the error vector, the estimate of the collision force time-series waveform is iteratively updated using gradient descent. In each iteration, the current estimate is multiplied by a step size factor and the error vector to obtain the updated estimate. This iterative update process is repeated until the L2 norm of the error vector converges. The estimated value at the point where iteration stops is taken as the collision force time-series waveform at each failure point. Finally, the maximum amplitude in the collision force time-series waveform is taken as the peak collision force at each failure point. Finally, a time-frequency transformation operation is performed on the collision force time-series waveform at each failure point. The time-frequency transformation refers to converting the collision force time-series waveform in the time domain into a time-frequency spectrum through a short-time Fourier transform. The horizontal axis of the time-frequency spectrum corresponds to the time sampling point, and the vertical axis corresponds to the frequency component. The value at each coordinate position represents the energy intensity at that time and frequency. The energy intensity sequence in the time-frequency spectrum can be matched with the energy intensity sequence of the time-frequency spectrum corresponding to each material in the collision template signal in two dimensions. The two-dimensional cross-correlation matching refers to multiplying the energy intensity matrices of the two time-frequency spectra element by element, summing them, and then dividing by the root mean square of the energy of the two matrices to obtain the correlation coefficient. The material type with the largest correlation coefficient is then selected as the foreign object material type at each failure point.
[0044] It should be noted that temporal deconvolution reconstruction is a numerical calculation process that deduces the system input excitation from the known system output response and system transfer characteristics. In this embodiment, it can be used to recover the complete waveform of the collision force applied to the inner wall of the cavity during particle collision from the collision signal subspace, which helps to quantitatively assess the collision energy. The collision force temporal waveform is a complete record of the force change over time during particle collision, including features such as peak size, rise time, and pulse width. The time spectrum is a representation of the energy distribution of the signal on the time-frequency plane. The time spectrum generated by the collision of particles of different materials has different characteristic patterns. Material classification of metal particles and non-metal particles can be achieved through two-dimensional cross-correlation matching.
[0045] In summary, the technical solution adopted in this application can realize the spatial positioning and material differentiation of multi-source particulate foreign objects, thereby improving the detection accuracy and fault diagnosis dimensions of particulate foreign objects.
[0046] Example 2: This application provides a particle collision noise detection system, referring to... Figure 4 As shown, this figure is a modular structure diagram of a particle collision noise detection system according to this application. The detection system includes: The vibration acquisition module 100 is used to acquire multi-channel vibration response signals and perform filtering processing on the multi-channel vibration response signals to obtain noise curves. Template matching module 200 is used to acquire collision template signals of different materials, determine the residual norm between the collision template signal and the noise curve, and determine the effective collision parameters of the potential collision source based on the residual norm. The positioning and separation module 300 is used to mark the coarse positioning area based on the effective collision parameters, determine the failure point in the coarse positioning area according to the extreme value of the residual norm, and separate the collision signal subspace from the noise curve based on the prior accuracy and posterior covariance matrix of the failure point. The inversion identification module 400 is used to perform temporal deconvolution reconstruction and time-frequency feature matching identification on the collision signal subspace to obtain the peak collision force and foreign object material type of the failure point.
[0047] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart. Figure 1One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0048] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, including read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically-Erasable Programmable Read-Only Memory (EEPROM), compactdisc read-only memory (CD-ROM) or other optical disc storage, disk storage, magnetic tape storage, or any other computer-readable medium capable of carrying or storing data.
[0049] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.
Claims
1. A method for detecting particle collision noise, characterized in that, The detection method includes the following steps: Multi-channel vibration response signals are acquired, and filtering processing is performed on the multi-channel vibration response signals to obtain noise curves; Acquire collision template signals of different materials, determine the residual norm between the collision template signal and the noise curve, and determine the effective collision parameters of the potential collision source based on the residual norm; The coarse localization region is marked based on the effective collision parameters, and the failure point in the coarse localization region is determined according to the extreme value of the residual norm. The collision signal subspace is separated from the noise curve based on the prior accuracy and posterior covariance matrix of the failure point. Temporal deconvolution reconstruction and time-frequency feature matching and identification are performed on the collision signal subspace to obtain the peak collision force and foreign object material type at the failure point.
2. The method for detecting particle collision noise as described in claim 1, characterized in that, Multi-channel vibration response signals are acquired using a piezoelectric accelerometer array.
3. The method for detecting particle collision noise as described in claim 1, characterized in that, The noise curve obtained by performing filtering processing on the multi-channel vibration response signal specifically includes: The multi-channel vibration response signal is subjected to bandpass filtering to obtain a filtered signal; The filtered signal is normalized and energy aligned to obtain the noise curve.
4. The method for detecting particle collision noise as described in claim 1, characterized in that, Obtaining collision template signals for different materials specifically includes: Establish a finite element dynamic model for particle-foreign object collisions; Transient dynamic simulations of different materials were performed on the finite element dynamic model to extract collision template signals for different materials.
5. The method for detecting particle collision noise as described in claim 1, characterized in that, Determining the residual norm between the collision template signal and the noise curve specifically includes: The noise curve is divided into different time segments according to a time window, wherein the length of each time segment is equal to the length of the collision template signal; The Euclidean distance between each time segment and the collision template signal of each material is calculated, thereby obtaining the residual norm between the collision template signal and the noise curve.
6. The method for detecting particle collision noise as described in claim 1, characterized in that, Determining the effective collision parameters of a potential collision source based on the residual norm specifically includes: Determine the initial prior accuracy and posterior covariance matrix of potential collision sources; The prior accuracy and posterior covariance matrix of the potential collision source are updated using the Bayesian iterative formula based on the residual norm, resulting in the updated prior accuracy and posterior covariance matrix. The effective collision parameters of potential collision sources are calculated based on the updated prior accuracy and posterior covariance matrix.
7. The method for detecting particle collision noise as described in claim 1, characterized in that, The process of marking coarse localization regions based on effective collision parameters and determining failure points within these regions based on the extreme values of the residual norm specifically includes: Spatial clustering is performed on the effective collision parameters, and continuous regions where the effective collision parameters exceed a preset background threshold are marked as coarse localization regions; Within the coarse localization region, local maxima are searched based on the variation curve of the residual norm. The failure points in the coarse localization region are obtained based on the local maxima filtering.
8. The method for detecting particle collision noise as described in claim 1, characterized in that, Separating the collision signal subspace from the noise curve based on the prior accuracy and posterior covariance matrix of the failure point specifically includes: The spatial filtering weights are determined based on the posterior covariance matrix of the failure points; The noise curve is fused using the spatial filtering weights to obtain the weighted timing signal of the failure point. A time window is extracted from the weighted time sequence signal and cross-interference separation is performed to obtain the collision signal subspace.
9. The method for detecting particle collision noise as described in claim 1, characterized in that, Performing temporal deconvolution reconstruction and time-frequency feature matching on the collision signal subspace yields the peak collision force and foreign object material type at the failure point, specifically including: Perform temporal deconvolution reconstruction on the collision signal subspace to obtain the temporal waveform of the collision force at the failure point. The maximum amplitude value is extracted from the collision force time-series waveform as the collision force peak value; A time-frequency transformation is performed on the collision force time-series waveform, and the transformed time-frequency characteristics are matched and identified with the collision template signal to obtain the foreign object material type at the failure point.
10. A particle collision noise detection system, characterized in that, For performing a particulate foreign object collision noise detection method as described in any one of claims 1 to 9, the detection system comprises: The vibration acquisition module is used to acquire multi-channel vibration response signals and perform filtering processing on the multi-channel vibration response signals to obtain noise curves. The template matching module is used to acquire collision template signals of different materials, determine the residual norm between the collision template signal and the noise curve, and determine the effective collision parameters of the potential collision source based on the residual norm. The positioning and separation module is used to mark the coarse positioning area based on the effective collision parameters, determine the failure point in the coarse positioning area according to the extreme value of the residual norm, and separate the collision signal subspace from the noise curve based on the prior accuracy and posterior covariance matrix of the failure point. The inversion identification module is used to perform temporal deconvolution reconstruction and time-frequency feature matching identification on the collision signal subspace to obtain the peak collision force and foreign object material type of the failure point.