A method for detecting the sealing state of an automobile water pump based on vibration data

Abstract

This invention relates to the field of automotive water pump testing technology, specifically a method for detecting the sealing status of automotive water pumps based on vibration data. The method includes: acquiring raw axial, radial, and tangential vibration signals of the water pump using a multi-dimensional vibration sensor; preprocessing and modal decomposition to obtain multi-scale intrinsic vibration modal components; and selecting sealing anomaly-related feature components for input into a diagnostic model. The model extracts time-frequency feature maps using a convolutional neural network, and after weighted focusing via an attention mechanism, outputs a sealing diagnosis result containing anomaly type and confidence level. The diagnostic result and feature components are fused to construct a state-enhanced vibration signal, which is then used to generate a three-dimensional vibration energy distribution cloud map through energy field simulation. Leakage energy channel features are extracted, and combined with a three-dimensional model of the water pump casing, suspected leak points and leakage levels are located in three-dimensional space. This method achieves refined diagnosis of automotive water pump sealing anomalies and spatial location of leak points, optimizing the accuracy and completeness of sealing status detection.

Description

Technical Field

[0001] This invention relates to the field of automotive water pump testing technology, and in particular to a method for detecting the sealing status of automotive water pumps based on vibration data. Background Technology

[0002] Current methods for detecting the vibration and sealing status of automotive water pumps primarily utilize single-dimensional vibration sensors to collect operational signals. These signals undergo only basic filtering preprocessing followed by simplistic mode decomposition, relying on conventional neural network models to extract single vibration features for sealing status determination. Some detection methods only assess the presence of seal anomalies based on vibration signal amplitude changes, without in-depth time-frequency feature mining. Existing detection processes only output a single judgment of normal or abnormal sealing, failing to incorporate attention mechanisms for weighted processing of feature information or generate refined diagnostic conclusions that include anomaly type and confidence scores. Furthermore, the automotive water pump sealing status detection process does not fuse diagnostic results with vibration feature components; it relies solely on the raw vibration signal for basic analysis, without conducting energy field simulation calculations, thus failing to generate a three-dimensional vibration energy distribution cloud map.

[0003] Single-dimensional vibration signal acquisition cannot fully capture vibration changes in the pump's sealing components. Conventional feature extraction methods easily overlook subtle features corresponding to sealing anomalies. The lack of a weighted focusing mechanism in diagnostic models leads to ambiguous anomaly judgments, making it impossible to distinguish specific anomaly types and their reliability. Existing detection technologies cannot reconstruct the energy distribution pattern corresponding to sealing failure, cannot extract energy channel features reflecting leakage paths, lack spatial positioning methods combined with a three-dimensional model of the pump casing, cannot pinpoint the leak point of sealing failure, and cannot classify the leakage level.

[0004] To address the issues of insufficient diagnostic accuracy and ambiguous anomaly type determination in automotive water pump sealing, as well as the inability to locate leak points and classify leak levels through three-dimensional energy distribution, it is necessary to build an appropriate diagnostic model and signal processing flow to achieve refined diagnosis of sealing status and three-dimensional spatial positioning of leak locations. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the existing technology and propose a method for detecting the sealing status of automotive water pumps based on vibration data.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a method for detecting the sealing status of an automotive water pump based on vibration data, comprising: The original vibration signals of the water pump during operation are collected by a multi-dimensional vibration sensor. The original vibration signals include axial vibration signals, radial vibration signals and tangential vibration signals. The original vibration signal is preprocessed to obtain an effective vibration signal that can be used for mode decomposition; Modal decomposition of effective vibration signals yields multi-scale intrinsic vibration modal components; Feature intrinsic vibration mode components related to sealing anomalies are selected from the multi-scale intrinsic vibration mode components. The feature intrinsic vibration mode components are input into a pre-constructed sealing anomaly diagnosis model. The sealing anomaly diagnosis model extracts the time-frequency feature map of the feature intrinsic vibration mode components through a convolutional neural network layer, and performs weighted focusing on the time-frequency feature map through an attention mechanism layer, outputting a sealing status diagnosis result that includes anomaly type and confidence score. The sealing state diagnosis results are fused with the characteristic intrinsic vibration mode components to construct a state-enhanced vibration signal. The energy field of the state-enhanced vibration signal is then simulated and calculated to generate a three-dimensional vibration energy distribution cloud map. Leakage energy channel features reflecting the sealing leakage path are extracted from the three-dimensional vibration energy distribution cloud map. Based on the leakage energy channel features and the three-dimensional model of the pump casing, the suspected leakage point of sealing failure and its leakage level are located in three-dimensional space.

[0007] As a further aspect of the present invention, the preprocessing of the original vibration signal to obtain an effective vibration signal that can be used for mode decomposition includes: Synchronous alignment and timestamp correction are performed on the raw vibration signals acquired by the multidimensional vibration sensor; The aligned original vibration signal is subjected to sliding mean filtering to eliminate low-frequency baseline drift; The filtered signal is subjected to wavelet threshold denoising to separate and suppress high-frequency background noise and impact interference, and outputs an effective vibration signal containing effective vibration information of the water pump operation.

[0008] As a further aspect of the present invention, the step of performing modal decomposition on the effective vibration signal to obtain multi-scale intrinsic vibration modal components includes: Empirical mode decomposition was performed on the effective vibration signals of axial vibration, radial vibration, and tangential vibration signals, respectively. Perform Hilbert transform on the eigenmode functions of each order generated by empirical mode decomposition, and calculate their instantaneous frequency and instantaneous amplitude; Based on the scale distribution of instantaneous frequency, intrinsic mode functions with similar frequency scales are classified and sorted from high to low frequency to form multi-scale intrinsic vibration mode components containing vibration characteristics at different time scales.

[0009] As a further aspect of the present invention, the characteristic intrinsic vibration mode components related to sealing anomalies are screened from the multi-scale intrinsic vibration mode components, including: Calculate the energy entropy and kurtosis index of multi-scale intrinsic vibration mode components; Multiscale intrinsic vibrational modal components with energy entropy higher than a preset energy entropy threshold and kurtosis index exceeding a preset kurtosis threshold are marked as candidate feature components; Principal component analysis was performed on the candidate feature components, and the top few principal components whose cumulative contribution rate exceeded the contribution rate threshold were selected and reconstructed into the feature intrinsic vibration mode components most related to the sealing anomaly.

[0010] As a further aspect of the present invention, the step of inputting the characteristic intrinsic vibration modal components into the pre-constructed sealing anomaly diagnostic model includes: The characteristic intrinsic vibration modal components are converted into a two-dimensional time-frequency matrix; The two-dimensional time-frequency matrix is ​​input into the convolutional neural network layer of the sealing anomaly diagnosis model, and deep time-frequency feature maps are extracted through multi-layer convolution and pooling operations. The deep time-frequency feature map is input into the attention mechanism layer, the weight distribution of different regions in the feature map is calculated, and the feature map is weighted and fused to obtain the focused weighted time-frequency feature map. The weighted time-frequency feature map is input into the fully connected classification layer, and the output is the probability distribution corresponding to different sealing anomaly types. The anomaly type with the highest probability and its probability value are used as the sealing status diagnosis result.

[0011] As a further aspect of the present invention, the sealing condition diagnosis result is fused with the characteristic intrinsic vibration mode components to construct a state-enhanced vibration signal, including: The confidence score in the sealing condition diagnosis results is normalized into a weighting coefficient; The characteristic intrinsic vibration mode components are weighted and amplified using the aforementioned weighting coefficients; The weighted and amplified characteristic intrinsic vibration mode components are superimposed with the original vibration signal in the time domain to generate a state-enhanced vibration signal that highlights the sealing anomaly characteristics.

[0012] As a further aspect of the present invention, the step of performing energy field simulation calculations on the state-enhanced vibration signal to generate a three-dimensional vibration energy distribution cloud map includes: Obtain the meshed data of the 3D model of the car water pump housing, and map each monitoring position to the corresponding mesh node of the 3D model; The signal energy of the state-enhanced vibration signal at the monitoring location is used as the initial energy value of the grid node; Based on the three-dimensional diffusion model, the transmission and attenuation of the initial energy value between the grid nodes inside the three-dimensional model are calculated, the propagation process of vibration energy in the pump casing structure is simulated, and finally a three-dimensional vibration energy distribution cloud map covering the entire pump casing three-dimensional model is generated.

[0013] As a further aspect of the present invention, the extraction of leakage energy channel features reflecting the sealing leakage path from the three-dimensional vibration energy distribution cloud map includes: In the three-dimensional vibration energy distribution cloud map, an energy density threshold is set; A continuous three-dimensional spatial region with an energy density higher than the energy density threshold is identified as a high-energy region; By tracing the edge of the high-energy region, the path with the largest change in energy gradient is extracted. This path is the leakage energy channel characteristic that represents the transmission of vibration energy along the weak point of the seal.

[0014] As a further aspect of the present invention, the method of locating suspected leak points and their leakage levels in three-dimensional space based on the characteristics of the leakage energy channel and the three-dimensional model of the pump casing includes: The endpoints, inflection points, and points where the leakage energy channel intersects with known sealing structures are marked as key location points; Calculate the average energy density of the local area where each key location point is located in the three-dimensional vibration energy distribution cloud map; Based on the numerical range of average energy density, a corresponding leakage level is assigned to each critical location point, and the critical location points and their leakage levels together constitute the suspected leakage point information.

[0015] As a further aspect of the present invention, the method for constructing the three-dimensional model of the water pump casing includes: The initial computer-aided design model of the car water pump to be tested is obtained. The initial computer-aided design model is geometrically simplified and feature-preserving. Small chamfers, threaded holes and decorative structures that do not affect the transmission of vibration energy are removed, while the shell wall thickness, reinforcing rib layout and sealing surface structure are preserved to generate a simplified basic geometric model. The simplified basic geometric model is meshed using finite element methods. The mesh element type is set to tetrahedral or hexahedral. Mesh size control parameters are defined to make the mesh size of the sealing surface area and the expected energy propagation path area smaller than that of the non-critical area, thus generating a discretized mesh model with node coordinate information and element connection relationships. Material property parameters are assigned to the discretized mesh model. These material property parameters include at least density, elastic modulus, and Poisson's ratio. The material property parameters are set based on the actual manufacturing material of the car water pump housing. A mapping table between the unique identifier of each node in the discretized mesh model and its physical space coordinates was established, and the coordinates of the installation position of the multidimensional vibration sensor in physical space were mapped to the corresponding mesh nodes in the mapping table to complete the construction of the three-dimensional model of the water pump casing.

[0016] Compared with the prior art, the advantages and positive effects of the present invention are as follows: The system collects multi-dimensional original vibration signals from the water pump in the axial, radial, and tangential directions, performs preprocessing and mode decomposition, and filters out the intrinsic vibration mode components related to sealing anomalies. Then, it extracts the time-frequency feature map of the component through a convolutional neural network layer, and uses an attention mechanism layer to weight and focus the time-frequency feature map. The system outputs a sealing status diagnosis result that includes the anomaly type and confidence score. This system can accurately capture the time-frequency feature details corresponding to sealing anomalies, filter irrelevant vibration interference information during water pump operation, clearly distinguish different types of sealing anomaly states, quantify the credibility of sealing anomaly judgment, and refine the dimensions of the sealing status diagnosis result.

[0017] By fusing the sealing condition diagnosis results with the characteristic intrinsic vibration mode components, a state-enhanced vibration signal is constructed. Energy field simulation calculations are performed on this signal to generate a three-dimensional vibration energy distribution cloud map. Leakage energy channel features reflecting the sealing leakage path are extracted from the cloud map. Combined with the three-dimensional model of the pump casing, the suspected leakage point of sealing failure is located in three-dimensional space and the leakage level is determined. The diagnostic information and vibration characteristics can be integrated to form a signal carrier that fits the sealing failure state. The energy distribution pattern corresponding to the sealing leakage is restored through energy field simulation, the transmission path of leakage energy is clarified, and the spatial locking of the sealing failure location is achieved based on the three-dimensional model, intuitively classifying the level category corresponding to the leakage state. Attached Figure Description

[0018] Figure 1 This is a flowchart of a method for detecting the sealing status of an automotive water pump based on vibration data, as described in this invention. Figure 2 A flowchart for obtaining multi-scale intrinsic vibration modal components from modal decomposition; Figure 3 This is a diagram of the original triaxial vibration signal of a car water pump. Figure 4 This is a three-dimensional vibration energy distribution cloud map; Figure 5 A comparison diagram showing the geometric simplification of CAD models of car water pumps. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0020] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0021] See Figure 1 This invention provides a method for detecting the sealing status of an automotive water pump based on vibration data. The specific method includes: By acquiring multidimensional vibration signals and performing a series of signal processing, feature extraction, and intelligent diagnostic steps, the sealing failure point is finally located in three-dimensional space. The original vibration signals of the water pump during operation are collected using multidimensional vibration sensors, including axial, radial, and tangential vibration signals. The original vibration signals are preprocessed to obtain effective vibration signals usable for mode decomposition. Mode decomposition is performed on the effective vibration signals to obtain multi-scale intrinsic vibration mode components. Feature intrinsic vibration mode components related to sealing anomalies are selected from these multi-scale intrinsic vibration mode components and input into a pre-constructed sealing anomaly diagnostic model. This model extracts the time-frequency feature maps of the feature intrinsic vibration mode components through a convolutional neural network layer and weights and focuses these feature maps through an attention mechanism layer, outputting a sealing state diagnostic result including anomaly type and confidence score. The sealing state diagnostic result is fused with the feature intrinsic vibration mode components to construct a state-enhanced vibration signal. Energy field simulation calculations are performed on the state-enhanced vibration signal to generate a three-dimensional vibration energy distribution cloud map. Leakage energy channel features reflecting the sealing leakage path are extracted from the three-dimensional vibration energy distribution cloud map. Based on the leakage energy channel features and the three-dimensional model of the pump casing, the suspected leakage point of sealing failure and its leakage level are located in three-dimensional space.

[0022] In one embodiment of the present invention, the raw vibration signals acquired by the multidimensional vibration sensor are synchronized and timestamped to ensure that the axial, radial, and tangential vibration signals are strictly synchronized on the time axis. The aligned raw vibration signals are then subjected to moving average filtering to eliminate low-frequency baseline drift. This operation removes slow variation trends in the signal caused by temperature changes or the sensor's inherent characteristics. The filtered signal is then subjected to wavelet threshold denoising to separate and suppress high-frequency background noise and impact interference, outputting a valid vibration signal containing effective vibration information of the water pump operation, providing a clean input for subsequent mode decomposition. After obtaining the valid vibration signals, empirical mode decomposition is performed on the valid vibration signals of the axial, radial, and tangential vibration signals, adaptively decomposing the signal in each direction into a series of intrinsic mode functions from high frequency to low frequency. Hilbert transforms are performed on the eigenmode functions generated by the empirical mode decomposition to calculate their instantaneous frequency and instantaneous amplitude, thereby obtaining the time-frequency characteristics of each component. Based on the scale distribution of instantaneous frequency, intrinsic mode functions with similar frequency scales are classified and sorted from high to low frequency to form multi-scale intrinsic vibration mode components containing vibration characteristics at different time scales. This set of components comprehensively characterizes the behavior of water pump vibration in different frequency bands.

[0023] In a specific implementation, taking the vibration signal acquisition of an automobile water pump under rated speed conditions as an example, the original vibration signal acquired by the multi-dimensional vibration sensor includes axial vibration signal, radial vibration signal, and tangential vibration signal. The amplitude range of the axial vibration signal is within ±5 grams, the amplitude range of the radial vibration signal is within ±3 grams, and the amplitude range of the tangential vibration signal is within ±2 grams. The sampling frequency is set to 10 kHz, and the original vibration signal is stored in time series form. In a specific implementation, the original vibration signal acquired by the multi-dimensional vibration sensor is synchronized and timestamped. The synchronization alignment operation uses a unified time base signal in the data acquisition system to interpolate and resample the vibration data streams of the three channels, so that each data point of the axial, radial, and tangential vibration signals corresponds to the same physical time. The timestamping operation assigns a precise time stamp to each frame of aligned data based on a GPS clock or a high-precision crystal oscillator clock. In some embodiments, the aligned original vibration signal is subjected to moving average filtering to eliminate low-frequency baseline drift. The mathematical expression for moving average filtering is: in: This represents the signal amplitude after moving average filtering at the nth discrete time point. This represents the amplitude of the original vibration signal that has been synchronized at the nth discrete time point. W represents the length of the sliding window, which is set according to the lowest frequency component of the effective vibration components to be retained in the signal. n and k are both discrete-time indices. In some embodiments, wavelet thresholding denoising is performed on the filtered signal. The wavelet thresholding denoising process selects the Symlets wavelet basis function to decompose the signal into 5 levels. The high-frequency detail coefficients obtained from the decomposition are shrunk using a soft thresholding function. The threshold of the soft threshold is adaptively determined according to the standard deviation of the high-frequency coefficients of the decomposition level. This separates and suppresses high-frequency background noise and impact interference, and outputs an effective vibration signal containing the effective vibration information of the water pump. It can be understood that after wavelet thresholding denoising, the high-frequency random noise component of the effective vibration signal is significantly suppressed, while the impact and periodic components reflecting the mechanical state of the water pump are retained.

[0024] In practical implementation, after obtaining a valid vibration signal, refer to... Figure 2 Empirical mode decomposition (EMD) is performed on the effective vibration signals of axial, radial, and tangential vibration signals. The EMD process involves identifying local maxima and minima, fitting the upper and lower envelopes using cubic spline interpolation, calculating the mean of the upper and lower envelopes, and continuously filtering out sub-modes from the original signal until the definition conditions of intrinsic mode functions (EMFs) are met. This decomposes the effective vibration signal in each direction into a series of EMFs. In practice, Hilbert transforms are applied to each order of EMFs generated by EMD to calculate the instantaneous frequency and amplitude. The Hilbert transform is achieved by constructing the analytic signal of the EMF; the instantaneous frequency is calculated from the derivative of the phase of the analytic signal, and the instantaneous amplitude is the magnitude of the analytic signal. Optionally, the Hilbert transform is applied to each order of EMF generated by EMD to construct the analytic signal form of the EMF. Based on the scale distribution of instantaneous frequency, intrinsic mode functions (EMFs) with similar frequency scales are grouped. This grouping is achieved by setting a frequency scale interval threshold. EMFs whose instantaneous frequency average values ​​fall within the same threshold interval are grouped together and sorted from high to low frequency, forming multi-scale intrinsic vibration mode components containing vibration characteristics at different time scales. Optionally, the frequency scale interval threshold is set based on the harmonic distribution of the pump's rotational frequency. It can be understood that the multi-scale intrinsic vibration mode components are arranged from high to low frequency; high-frequency components typically correspond to impact events, while low-frequency components typically correspond to slowly varying modulation or trend information.

[0025] In one embodiment of the invention, the energy entropy and kurtosis index of multi-scale intrinsic vibration mode components are calculated. The energy entropy is used to quantify the degree of disorder in the distribution of component energy, and the kurtosis index is used to characterize the intensity of the signal pulse impact. Multi-scale intrinsic vibration mode components with energy entropy higher than a preset energy entropy threshold and kurtosis index exceeding a preset kurtosis threshold are marked as candidate feature components. These components typically contain impact or modulation information related to sealing anomalies. Principal component analysis is performed on the candidate feature components, and the top few principal components with cumulative contribution rates exceeding a contribution rate threshold are selected and reconstructed into the feature intrinsic vibration mode components most relevant to sealing anomalies. This process achieves feature dimensionality reduction and focuses on the main variation directions.

[0026] After obtaining the characteristic intrinsic vibration mode components, they are converted into a two-dimensional time-frequency matrix. The rows of this matrix represent time points, the columns represent frequency points or scale indices, and the values ​​represent energy or amplitude. This two-dimensional time-frequency matrix is ​​input into the convolutional neural network layer of the sealing anomaly diagnosis model. Deep time-frequency feature maps are extracted through multiple convolutional and pooling operations. These feature maps can capture complex spatiotemporal patterns in the signal. The deep time-frequency feature maps are then input into an attention mechanism layer to calculate the weight distribution of different regions in the feature maps. The feature maps are then weighted and fused to obtain a focused weighted time-frequency feature map. This mechanism makes the model focus more on key regions related to the anomaly. Finally, the weighted time-frequency feature map is input into a fully connected classification layer, which outputs the probability distribution corresponding to different sealing anomaly types. The anomaly type with the highest probability and its probability value are used as the sealing condition diagnosis result.

[0027] In specific implementations, operations are performed based on the processed multi-scale intrinsic vibrational mode components, which include multiple components ordered from high frequency to low frequency, each component being a time series. In specific implementations, the energy entropy and kurtosis index of the multi-scale intrinsic vibrational mode components are calculated. The energy entropy is calculated for each intrinsic vibrational mode component, and its value reflects the uniformity and randomness of the energy distribution of that component. Similarly, the kurtosis index is calculated for each intrinsic vibrational mode component, and is used to quantify the sharpness and impulse characteristics of the signal distribution. In some embodiments, the energy entropy... The calculation formula is: in: The numerical value representing energy entropy. This represents the total number of equal-length intervals into which the time series of intrinsic vibration modal components is divided. The signal energy falls on the first The probability within a given interval is calculated as the ratio of the sum of squares of the signal amplitude within that interval to the total energy of the entire signal. This represents a logarithmic operation to the base 2. In some embodiments, the kurtosis index... The calculation is based on the ratio of the fourth central moment to the square of the second central moment of the intrinsic vibration modal component sequence. It can be understood that an intrinsic vibration modal component with obvious impact characteristics usually exhibits a high kurtosis index and a specific energy entropy value.

[0028] In practice, multi-scale intrinsic vibration mode components with energy entropy higher than a preset energy entropy threshold and kurtosis index exceeding a preset kurtosis threshold are marked as candidate feature components. The preset energy entropy threshold is set based on the 95th percentile of the statistical distribution of energy entropy of historical normal water pump samples, and the preset kurtosis threshold is set based on the 95th percentile of the statistical distribution of kurtosis index of historical normal water pump samples. For example, when the calculated energy entropy of a certain intrinsic vibration mode component is 0.7 and the preset energy entropy threshold is 0.5, and the calculated kurtosis index of this component is 5.0 and the preset kurtosis threshold is 4.0, this component is marked as a candidate feature component because the energy entropy of 0.7 is higher than the threshold of 0.5 and the kurtosis index of 5.0 exceeds the threshold of 4.0. In practice, principal component analysis is performed on the candidate feature components. The principal component analysis calculates the covariance and decomposes the matrix composed of all candidate feature components. The top few principal components whose cumulative contribution rate exceeds the contribution rate threshold are selected. The contribution rate threshold is set to 85%. The components are then reconstructed into the feature intrinsic vibration mode components most related to the sealing anomaly. The reconstruction operation is achieved by linearly combining the selected principal components with the corresponding eigenvectors.

[0029] In practical implementation, after obtaining the characteristic intrinsic vibration mode components, these components are converted into a two-dimensional time-frequency matrix. The two-dimensional time-frequency matrix is ​​generated through short-time Fourier transform or continuous wavelet transform. Rows in the matrix correspond to time frames, columns to frequency bins or scales, and matrix element values ​​are spectral amplitudes or energy. In practical implementation, the two-dimensional time-frequency matrix is ​​input into the convolutional neural network layer of the sealing anomaly diagnostic model. This convolutional neural network layer contains two convolutional layers and one max-pooling layer. Deep time-frequency feature maps are extracted through multiple convolutional and pooling operations. The first convolutional layer uses a 3x3 kernel and outputs 16 feature maps, and the second convolutional layer also uses a 3x3 kernel and outputs 32 feature maps. Optionally, a ReLU activation function is applied after the convolutional operation to introduce nonlinearity. In the specific implementation, the deep time-frequency feature map is input into the attention mechanism layer. The attention mechanism layer uses a channel attention module to calculate the weight distribution of different channels in the feature map and performs weighted fusion on the feature map to obtain a focused weighted time-frequency feature map. The weight distribution is generated by performing global average pooling on the feature map in the spatial dimension and then passing it through a fully connected layer and a sigmoid function. Optionally, the attention mechanism layer uses a spatial attention module to calculate the weights of different spatial locations in the feature map. In the specific implementation, the weighted time-frequency feature map is input into a fully connected classification layer. The fully connected classification layer contains a hidden layer with 64 neurons and an output layer with the number of output neurons equal to the total number of sealing anomaly types. It outputs the probability distribution corresponding to different sealing anomaly types, including "sealing ring wear", "sealing surface scratch", and "fastener loosening". The anomaly type with the highest probability and its probability value are used as the sealing status diagnosis result.

[0030] In one embodiment of the present invention, the confidence score in the sealing condition diagnosis result is normalized to a weighting coefficient between 0 and 1. The intrinsic vibration modal components are weighted and amplified using this weighting coefficient; the higher the confidence level, the greater the amplification, thereby strengthening the abnormal features considered important by the diagnostic model. The weighted and amplified intrinsic vibration modal components are superimposed with the original vibration signal in the time domain to generate a state-enhanced vibration signal that highlights the sealing anomaly. This signal retains the background information of the original signal while amplifying the abnormal components. Mesh data of the three-dimensional model of the car water pump housing is acquired, and each monitoring location is mapped to a corresponding mesh node in the three-dimensional model. The signal energy of the state-enhanced vibration signal at the monitoring location is used as the initial energy value of the mesh node. The signal energy can be obtained by calculating the sum of squares or the root mean square value of the signal amplitude. Based on a three-dimensional diffusion model, the transmission and attenuation of the initial energy value between mesh nodes within the three-dimensional model are calculated to simulate the propagation process of vibration energy in the pump housing structure. Finally, a three-dimensional vibration energy distribution cloud map covering the entire three-dimensional model of the water pump housing is generated. The cloud map visually displays the spatial distribution of energy density using color or transparency.

[0031] In a specific implementation, taking the diagnostic result of the sealing anomaly diagnostic model outputting "sealing ring wear" with a confidence score of 0.92 as an example, the confidence score in the sealing condition diagnostic result is normalized into weight coefficients. The normalization operation is implemented through a linear mapping function, mapping confidence scores in the interval [0.5, 1.0] to weight coefficients in the interval [0.6, 1.0]. In some embodiments, the weight coefficients... The calculation formula is: in: This represents the final calculated weighting coefficient. This represents the confidence score in the sealing condition diagnosis results. When the confidence score... When the value is 0.92, the weighting coefficient is calculated by substituting it into the formula. The value is 0.936. Refer to Table 1 for the correspondence between different confidence scores and the calculated weight coefficients. It can be understood that the normalization process ensures that even with a low diagnostic confidence score, the constructed state-enhanced vibration signal still contains the original abnormal characteristic information.

[0032] Table 1: Correspondence between confidence scores and weighting coefficients In specific implementation, the characteristic intrinsic vibration modal components are weighted and amplified using the aforementioned weighting coefficients. The weighting amplification operation multiplies each data point in the characteristic intrinsic vibration modal component sequence by the weighting coefficient. In practice, the weighted and amplified intrinsic vibration modal components are superimposed on the original vibration signal in the time domain to generate a state-enhanced vibration signal that highlights the sealing anomaly characteristics. The superposition operation involves algebraically adding the data points of the weighted and amplified intrinsic vibration modal components to the data points of the original vibration signal at the same time. It can be understood that the state-enhanced vibration signal retains the overall background of the original vibration signal while significantly amplifying the amplitude of the vibration components identified as abnormal by the diagnostic model.

[0033] In specific implementation, the meshed data of the 3D model of the car water pump casing is acquired. The meshed data includes a list of node coordinates and a list of element connection relationships. Each monitoring position is mapped to the corresponding mesh node of the 3D model. The mapping operation is completed by calculating the Euclidean distance between the physical space coordinates of the monitoring point and the coordinates of all mesh nodes, and assigning the monitoring point to the nearest mesh node. In specific implementation, the signal energy of the state-enhanced vibration signal at the monitoring position is used as the initial energy value of the mesh node. The signal energy is obtained by calculating the sum of the squares of the amplitudes of all data points of the state-enhanced vibration signal within a complete analysis time window. In some embodiments, based on the 3D diffusion model, the transmission and attenuation of the initial energy value between mesh nodes within the 3D model are calculated. The 3D diffusion model is solved using the finite difference method. Its core is to simulate the diffusion process of energy from high-energy nodes to adjacent low-energy nodes, and to simulate the propagation process of vibration energy in the pump casing structure, ultimately generating a 3D vibration energy distribution cloud map covering the entire 3D model of the water pump casing. Optionally, the conduction coefficient of the 3D diffusion model is set by analogy with the thermal conductivity of the water pump casing material. Optionally, the energy attenuation coefficient is set based on empirical values ​​of structural damping.

[0034] See Figure 3 This is a raw triaxial vibration signal graph of an automotive water pump, showing the raw triaxial vibration time-domain signals collected by a multi-dimensional vibration sensor during the pump's operation. This is the first step in data acquisition for vibration sealing status detection. The axial vibration signal has the largest amplitude range (approximately -3g to 3g) and the most dramatic fluctuations, serving as a core indicator reflecting the pump's axial movement and the sealing surface's contact condition. The radial vibration signal has an amplitude range of approximately -2g to 2g, primarily reflecting impeller radial runout and bearing condition. The tangential vibration signal has an amplitude range of approximately -1g to 1.5g, directly related to the pump's rotational torque transmission. All three types of signals exhibit periodic fluctuations plus random noise, consistent with typical mechanical vibration characteristics. The noise originates from sensor background noise and internal fluid disturbances within the pump. The axial signal exhibits the largest amplitude fluctuation, indicating its highest sensitivity to sealing anomalies, making it a key focus for subsequent feature selection.

[0035] In one embodiment of the present invention, an energy density threshold is set in the three-dimensional vibration energy distribution cloud map. This threshold can be determined statistically based on the energy distribution of historical normal samples. Continuous three-dimensional spatial regions with energy densities higher than the energy density threshold are identified as high-energy regions, which are paths of abnormal concentration or conduction of vibration energy. The edges of the high-energy regions are traced, and the paths with the largest changes in energy gradient are extracted. These paths represent leakage energy channels characterized by the conduction of vibration energy along weak points in the seal. This characteristic manifests as one or more energy accumulation bands in three-dimensional space. The endpoints, inflection points, and points intersecting with known sealing structures of the leakage energy channel characteristics are marked as key location points. These points are often potential locations of stress concentration or seal failure. The average energy density of the local area where each key location point is located is calculated in the three-dimensional vibration energy distribution cloud map. The local area can be all grid nodes within a certain radius centered on the key point. Based on the numerical range of the average energy density, a corresponding leakage level is assigned to each key location point. For example, the energy density can be divided into three intervals: low, medium, and high, and different levels can be matched. The key location points and their leakage levels together constitute suspected leakage point information.

[0036] In practical implementation, the three-dimensional vibration energy distribution cloud map uses the color depth of grid nodes to represent energy density values, with the energy density values ​​ranging from 0 to 1.0. In this implementation, an energy density threshold of 0.8 is set in the three-dimensional vibration energy distribution cloud map. In practice, continuous three-dimensional spatial regions with energy densities higher than the energy density threshold are identified as high-energy regions. This identification operation involves traversing all grid nodes, marking nodes with energy density values ​​greater than 0.8, and aggregating spatially adjacent marked nodes into continuous regions based on a three-dimensional connected component algorithm. These aggregated regions are the high-energy regions. For example, a spatial region containing 150 adjacent grid nodes, where each node has an energy density greater than 0.8, is identified as a high-energy region.

[0037] In specific implementation, the edge of the high-energy region is traced, and the path with the largest change in energy gradient is extracted. This path represents the leakage energy channel characteristic of vibration energy transmission along the weak point of the seal. The tracing operation starts from the boundary node of the high-energy region, and the energy gradient of each boundary node along all its adjacent directions is calculated. The energy gradient is calculated using the energy density difference between adjacent nodes. The direction with the most dramatic gradient decrease is preferentially selected for path extension until the path extends to a region where the energy density is below the energy density threshold. In some embodiments, the energy gradient... The calculation formula is: in: This represents the absolute value of the energy gradient from the current node to one of its neighboring nodes. This represents the energy density value of the current node. This represents the energy density value of a neighboring node of the current node. This represents the spatial distance between the current node and its neighboring nodes. This indicates that the maximum value is taken after calculating the gradients of all adjacent directions of the current node. Optionally, during the path extension process, the coordinate sequence of all traversed nodes is recorded simultaneously, and this coordinate sequence defines a leakage energy channel feature.

[0038] In practical implementation, the endpoints, inflection points, and intersections with known sealing structures of the leakage energy channel characteristics are marked as key location points. Endpoints refer to the starting and ending coordinates of the leakage energy channel characteristic coordinate sequence. Inflection points are points where the direction of the leakage energy channel characteristic path changes significantly; they are determined by calculating the cosine of the angle between adjacent line segments of the path, and a cosine value less than a set threshold is considered an inflection point. Points intersecting with known sealing structures are obtained by comparing the leakage energy channel characteristic coordinates with the predefined three-dimensional coordinate region of the pump sealing surface; if the channel characteristic coordinates fall within the sealing surface region, they are marked as intersection points. See Table 2 for examples of key location points.

[0039] Table 2: Correspondence between Energy Density and Leakage Level at Critical Locations In specific implementation, the average energy density of the local area where each key location point is located is calculated in the three-dimensional vibration energy distribution cloud map. The local area is defined as all grid nodes within a spherical space with a radius of 2 mm centered on the coordinates of the key location point. The average energy density is the arithmetic mean of the energy density values ​​of all nodes within this spherical space. In specific implementation, a corresponding leakage level is assigned to each key location point according to the numerical range of the average energy density. The average energy density numerical range is divided into three levels: 0.8 to 1.0 corresponds to a "high" leakage level, 0.6 to 0.8 corresponds to a "medium" leakage level, and below 0.6 corresponds to a "low" leakage level. The key location points and their leakage levels together constitute the suspected leakage point information, as shown in Table 2. In some embodiments, the numerical range of the leakage level classification can be adjusted based on historical detection data of different models of water pumps. It can be understood that the higher the average energy density of a key location point, the higher the probability of a seal leakage.

[0040] See Figure 4This is a 3D vibration energy distribution cloud map, a core visualization result in the energy field simulation and leak location stage of the automotive water pump vibration sealing condition detection process. It visually presents the spatial distribution of vibration energy across the water pump casing cross-section. The main high-energy zone is located at X=15~33mm and Y=15~35mm, with an energy density close to 1.0 at its center (approximately X=25mm, Y=25mm), representing the core accumulation point of vibration energy. The secondary high-energy zone is located at X=32~44mm and Y=30~45mm, partially overlapping with the main high-energy zone, with an energy density of approximately 0.6~0.9, representing leakage channels formed by the diffusion of vibration energy along structural weak points. The remaining areas have an energy density of 0, representing non-critical structural areas where no vibration energy propagates. The connection path between the two high-energy zones (X=30~33mm, Y=30~35mm) is the leakage energy channel, corresponding to the water pump sealing surface or weak points in the casing wall. Vibration energy is conducted outward along this path, indicating a risk of seal failure. The center of the main high-energy zone (X=25mm, Y=25mm) is the suspected leak point, where the energy density is the highest and the leak level is the highest; the secondary high-energy zone is the next lower level of leak risk point.

[0041] In one embodiment of the present invention, an initial computer-aided design model of the automotive water pump to be tested is obtained. The initial computer-aided design model is then geometrically simplified and feature-preserving, removing minor chamfers, threaded holes, and decorative structures that do not affect vibration energy transmission, while retaining the shell wall thickness, reinforcing rib layout, and sealing surface structure, generating a simplified basic geometric model. The simplified basic geometric model is then meshed using finite element methods, setting the mesh element type to tetrahedral or hexahedral elements, and defining mesh size control parameters so that the mesh size of the sealing surface region and the expected energy propagation path region is smaller than that of non-critical regions, generating a discretized mesh model with node coordinate information and element connection relationships. Material property parameters are assigned to the discretized mesh model, including at least density, elastic modulus, and Poisson's ratio. These material property parameters are set according to the actual manufacturing material of the automotive water pump shell, such as aluminum alloy or engineering plastics. A mapping table between the unique identifier of each node in the discretized mesh model and its physical space coordinates was established. The coordinates of the installation position of the multidimensional vibration sensor in physical space were then mapped to the corresponding mesh nodes in the mapping table, thus completing the construction of the three-dimensional model of the water pump casing. This model provides the geometric and topological basis for subsequent energy field simulation and spatial positioning.

[0042] In practice, an initial computer-aided design (CAD) model of the vehicle water pump to be tested is obtained. This initial CACAD model is a STEP or IGES format file containing the complete geometric features of the pump casing. The initial CACAD model undergoes geometric simplification and feature preservation processing. Minor chamfers, threaded holes, and decorative structures that do not affect vibration energy transfer are removed. The size thresholds for feature removal are set to chamfers with a radius less than 0.5 mm and threaded holes with a diameter less than 2 mm. Simultaneously, the casing wall thickness, reinforcing rib layout, and sealing surface structure are preserved, generating a simplified basic geometric model. It can be understood that geometric simplification reduces the complexity and computational load of subsequent finite element mesh generation, while the preservation of key structural features ensures the physical realism of the energy simulation.

[0043] In practical implementation, the simplified basic geometric model is meshed using finite element methods, with the mesh element type set to tetrahedral or hexahedral elements. In some embodiments, tetrahedral elements are preferentially used for automatic meshing of the complex-shaped pump casing. Mesh size control parameters are defined to ensure that the mesh size of the sealing surface area and the expected energy propagation path area is smaller than that of non-critical areas. For example, the global maximum mesh size of the sealing surface area and the expected energy propagation path areas such as bearing housings and flange connection surfaces is set to 0.3 mm, while the global maximum mesh size of other main areas of the pump casing is set to 1.0 mm, generating a discretized mesh model with node coordinate information and element connection relationships. In some embodiments, the total number of nodes in the discretized mesh model is between 100,000 and 500,000, and the total number of elements is between 300,000 and 1,500,000. Optionally, a mesh quality check is required after meshing to ensure that the element Jacobian ratio is greater than 0.7.

[0044] In practical implementation, material property parameters are assigned to the discretized mesh model. These parameters include at least density, elastic modulus, and Poisson's ratio, and are set based on the actual manufacturing material of the automotive water pump housing. For example, for a water pump housing made of die-cast aluminum alloy A380, the material property parameters are set as follows: density... elastic modulus Poisson's ratio It is understandable that accurately assigning material property parameters is fundamental to simulating the propagation of vibration energy within a structure. In practical implementation, a mapping table is established between the unique identifiers of each node in the discretized mesh model and their physical space coordinates. This mapping is achieved through a function... definition: in: Indicates the mapping relationship. The unique identifier representing a node is a consecutive integer starting from 1. This represents the Cartesian coordinates of the node corresponding to the identifier in three-dimensional space, with the coordinate unit being millimeters. In some embodiments, the coordinates of the installation position of the multi-dimensional vibration sensor in physical space are mapped to the corresponding grid nodes in the mapping table. The mapping operation calculates the Euclidean distance between the sensor's physical coordinates and the coordinates of all nodes in the discretized grid model, finds the grid node with the smallest distance, and binds the unique identifier of this node to the sensor channel number, thus completing the construction of the three-dimensional model of the pump casing. For example, an axial vibration sensor located at physical coordinates (10.2, 5.1, 0.0) millimeters is found to be closest to the node with identifier #45021 in the discretized grid model (distance 0.05 millimeters), so a correspondence is established between sensor channel A1 and node #45021. Optionally, for multiple sensors, the above nearest neighbor search and binding operations are performed sequentially.

[0045] See Figure 5 This is a comparison chart of the geometric simplification of a car water pump CAD model, visually demonstrating the differences in the two-dimensional contours of the original CAD model and the simplified model. The original CAD model contour contains numerous high-frequency jagged edges and minute fluctuations, reflecting details such as small chamfers, threaded holes, and decorative structures in the design. The simplified model contour is smooth and continuous, removing high-frequency noise while retaining the overall shell contour, wall thickness variations, and geometric features of key structures (such as sealing surfaces and reinforcing ribs). The overall trend and peak / valley positions of the two curves are completely consistent, indicating that the simplification operation did not destroy the core geometry and key features for mechanical analysis. Removing minor features that do not affect vibration energy transfer significantly reduces the number of model meshes, which can significantly reduce the computation time for finite element analysis and energy field simulation. Key structures such as shell wall thickness, reinforcing rib layout, and sealing surfaces are retained, ensuring that the analysis accuracy of vibration energy transfer paths and mechanical responses is not affected. The simplified model is more suitable for subsequent finite element mesh generation, material property assignment, and sensor position mapping, serving as the basic geometric carrier for generating three-dimensional vibration energy distribution cloud maps.

[0046] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for detecting the sealing status of an automotive water pump based on vibration data, characterized in that, include: The original vibration signals of the water pump during operation are collected by a multi-dimensional vibration sensor, which includes axial vibration signals, radial vibration signals and tangential vibration signals. The original vibration signal is preprocessed to obtain an effective vibration signal that can be used for mode decomposition; Modal decomposition of effective vibration signals yields multi-scale intrinsic vibration modal components; Feature intrinsic vibration mode components related to sealing anomalies are selected from the multi-scale intrinsic vibration mode components. The feature intrinsic vibration mode components are input into a pre-constructed sealing anomaly diagnosis model. The sealing anomaly diagnosis model extracts the time-frequency feature map of the feature intrinsic vibration mode components through a convolutional neural network layer, and performs weighted focusing on the time-frequency feature map through an attention mechanism layer, outputting a sealing status diagnosis result that includes anomaly type and confidence score. The sealing state diagnosis results are fused with the characteristic intrinsic vibration mode components to construct a state-enhanced vibration signal. The energy field of the state-enhanced vibration signal is then simulated and calculated to generate a three-dimensional vibration energy distribution cloud map. Leakage energy channel features reflecting the sealing leakage path are extracted from the three-dimensional vibration energy distribution cloud map. Based on the leakage energy channel features and the three-dimensional model of the pump casing, the suspected leakage point of sealing failure and its leakage level are located in three-dimensional space.

2. The method for detecting the sealing status of an automotive water pump based on vibration data according to claim 1, characterized in that, The preprocessing of the original vibration signal to obtain an effective vibration signal that can be used for mode decomposition includes: Synchronous alignment and timestamp correction are performed on the raw vibration signals acquired by the multidimensional vibration sensor; The aligned original vibration signal is subjected to sliding mean filtering to eliminate low-frequency baseline drift; The filtered signal is subjected to wavelet threshold denoising to separate and suppress high-frequency background noise and impact interference, and outputs an effective vibration signal containing effective vibration information of the water pump operation.

3. The method for detecting the sealing status of an automotive water pump based on vibration data according to claim 2, characterized in that, The modal decomposition of the effective vibration signal to obtain multi-scale intrinsic vibration modal components includes: Empirical mode decomposition was performed on the effective vibration signals of axial vibration, radial vibration, and tangential vibration signals, respectively. Perform Hilbert transform on the eigenmode functions of each order generated by empirical mode decomposition, and calculate their instantaneous frequency and instantaneous amplitude; Based on the scale distribution of instantaneous frequency, intrinsic mode functions with similar frequency scales are classified and sorted from high to low frequency to form multi-scale intrinsic vibration mode components containing vibration characteristics at different time scales.

4. The method for detecting the sealing status of an automotive water pump based on vibration data according to claim 3, characterized in that, Selecting characteristic intrinsic vibration mode components related to sealing anomalies from the multi-scale intrinsic vibration mode components includes: Calculate the energy entropy and kurtosis index of multi-scale intrinsic vibration mode components; Multiscale intrinsic vibrational modal components with energy entropy higher than a preset energy entropy threshold and kurtosis index exceeding a preset kurtosis threshold are marked as candidate feature components; Principal component analysis was performed on the candidate feature components, and the top few principal components whose cumulative contribution rate exceeded the contribution rate threshold were selected and reconstructed into the feature intrinsic vibration mode components most related to the sealing anomaly.

5. The method for detecting the sealing status of an automotive water pump based on vibration data according to claim 4, characterized in that, The step of inputting the characteristic intrinsic vibration mode components into the pre-constructed sealing anomaly diagnostic model includes: The characteristic intrinsic vibration modal components are converted into a two-dimensional time-frequency matrix; The two-dimensional time-frequency matrix is ​​input into the convolutional neural network layer of the sealing anomaly diagnosis model, and deep time-frequency feature maps are extracted through multi-layer convolution and pooling operations. The deep time-frequency feature map is input into the attention mechanism layer, the weight distribution of different regions in the feature map is calculated, and the feature map is weighted and fused to obtain the focused weighted time-frequency feature map. The weighted time-frequency feature map is input into the fully connected classification layer, and the output is the probability distribution corresponding to different sealing anomaly types. The anomaly type with the highest probability and its probability value are used as the sealing status diagnosis result.

6. The method for detecting the sealing status of an automotive water pump based on vibration data according to claim 5, characterized in that, The sealing condition diagnosis results are fused with the characteristic intrinsic vibration mode components to construct a state-enhanced vibration signal, including: The confidence score in the sealing condition diagnosis results is normalized into a weighting coefficient; The characteristic intrinsic vibration mode components are weighted and amplified using the aforementioned weighting coefficients; The weighted and amplified characteristic intrinsic vibration mode components are superimposed with the original vibration signal in the time domain to generate a state-enhanced vibration signal that highlights the sealing anomaly characteristics.

7. The method for detecting the sealing status of an automotive water pump based on vibration data according to claim 6, characterized in that, The step of performing energy field simulation calculations on the enhanced vibration signal to generate a three-dimensional vibration energy distribution cloud map includes: Obtain the meshed data of the 3D model of the car water pump casing, and map each monitoring position to the corresponding mesh node of the 3D model; The signal energy of the state-enhanced vibration signal at the monitoring location is used as the initial energy value of the grid node; Based on the three-dimensional diffusion model, the transmission and attenuation of the initial energy value between the grid nodes inside the three-dimensional model are calculated, the propagation process of vibration energy in the pump casing structure is simulated, and finally a three-dimensional vibration energy distribution cloud map covering the entire pump casing three-dimensional model is generated.

8. The method for detecting the sealing status of an automotive water pump based on vibration data according to claim 7, characterized in that, The extraction of leakage energy channel features reflecting the sealing leakage path from the three-dimensional vibration energy distribution cloud map includes: In the three-dimensional vibration energy distribution cloud map, an energy density threshold is set; A continuous three-dimensional spatial region with an energy density higher than the energy density threshold is identified as a high-energy region; By tracing the edge of the high-energy region, the path with the largest change in energy gradient is extracted. This path is the leakage energy channel characteristic that represents the transmission of vibration energy along the weak point of the seal.

9. The method for detecting the sealing status of an automotive water pump based on vibration data according to claim 8, characterized in that, The method of locating suspected leak points and their leakage levels in three-dimensional space based on the characteristics of the leakage energy channel and the three-dimensional model of the pump casing includes: The endpoints, inflection points, and points where the leakage energy channel intersects with known sealing structures are marked as key location points; Calculate the average energy density of the local area where each key location point is located in the three-dimensional vibration energy distribution cloud map; Based on the numerical range of average energy density, a corresponding leakage level is assigned to each critical location point, and the critical location points and their leakage levels together constitute the suspected leakage point information.

10. The method for detecting the sealing status of an automotive water pump based on vibration data according to claim 9, characterized in that, The method for constructing the three-dimensional model of the water pump casing includes: The initial computer-aided design model of the car water pump to be tested is obtained. The initial computer-aided design model is geometrically simplified and feature-preserving. Small chamfers, threaded holes and decorative structures that do not affect the transmission of vibration energy are removed, while the shell wall thickness, reinforcing rib layout and sealing surface structure are preserved to generate a simplified basic geometric model. The simplified basic geometric model is meshed using finite element methods. The mesh element type is set to tetrahedral or hexahedral. Mesh size control parameters are defined to make the mesh size of the sealing surface area and the expected energy propagation path area smaller than that of the non-critical area, thus generating a discretized mesh model with node coordinate information and element connection relationships. Material property parameters are assigned to the discretized mesh model. These material property parameters include at least density, elastic modulus, and Poisson's ratio. The material property parameters are set based on the actual manufacturing material of the car water pump housing. A mapping table between the unique identifier of each node in the discretized mesh model and its physical space coordinates was established, and the coordinates of the installation position of the multidimensional vibration sensor in physical space were mapped to the corresponding mesh nodes in the mapping table to complete the construction of the three-dimensional model of the water pump casing.

Images