A system for sampling and calculating the ripple signal of a seat motor

Abstract

This invention relates to the field of seat motor monitoring technology and discloses a seat motor ripple signal sampling and parameter calculation system. The system includes a ripple signal acquisition module for real-time acquisition and time-domain segmentation of the raw ripple signal; a noise filtering module for multi-scale decomposition and filtering of high-frequency noise; a feature extraction module for extracting frequency, amplitude, and periodic features to generate a ripple feature set; a dynamic mesh generation module for generating a multi-resolution analysis mesh based on signal resolution differences; a ripple load spectrum analysis module for obtaining the ripple load spectrum from the time-series spectrum of the feature distribution; a stress calculation module for calculating the stress distribution at key nodes based on finite element analysis of the load spectrum; a fatigue cycle identification module for identifying the number of fatigue cycles based on stress data; and a life prediction module for calculating the remaining life of key nodes by combining material fatigue curves and the number of cycles, thus obtaining the overall life prediction result of the motor. This system can accurately monitor the motor status and meet the reliability requirements of seat motors.

Description

Technical Field

[0001] This invention relates to the field of seat motor monitoring technology, specifically a seat motor ripple signal sampling and parameter calculation system. Background Technology

[0002] In modern automotive seat adjustment systems and various electric seat applications, the seat motor, as a core drive component, directly affects the overall performance and safety of the seat due to its operational stability and lifespan. As users increasingly demand higher precision in seat adjustments, faster response times, and greater long-term reliability, real-time monitoring of the seat motor's operating status and lifespan prediction are becoming key areas of focus in the industry.

[0003] For seat motor condition monitoring and lifespan assessment, the industry has developed various technical solutions. Some solutions directly detect basic electrical parameters such as motor voltage and current, and combine them with preset thresholds to determine whether the motor is in normal operating condition. However, these solutions can only provide simple fault alarms and cannot deeply analyze changes in the motor's internal operating state, let alone effectively predict the motor's remaining lifespan. Another solution uses vibration sensors to collect vibration signals generated during motor operation and analyzes the characteristics of the vibration signals to determine whether the motor has wear, abnormal noise, or other problems. However, vibration signals are easily affected by external environmental interference, such as seat frame vibration and external impacts, which can cause distortion of the vibration signals and thus affect the accuracy of the monitoring results. Furthermore, this type of solution has significant shortcomings in the ability to identify early, subtle faults in the motor and predict its lifespan.

[0004] In the field of motor life prediction, traditional methods rely heavily on the motor's rated operating parameters and cumulative operating time, using empirical formulas for life estimation. This approach fails to consider dynamic factors such as load fluctuations and changes in operating conditions during actual motor operation, leading to significant discrepancies between predicted life and actual conditions. For example, a seat motor experiences significantly different loads during different adjustment processes, such as seat height adjustment, forward and backward sliding, and backrest angle adjustment. Traditional methods cannot distinguish the impact of load changes at different operating stages on motor life, making accurate life prediction difficult. Furthermore, current technologies underutilize the ripple signals generated during motor operation. As an inherent signal during motor operation, the ripple signal directly reflects key information such as rotor position, speed, and electromagnetic characteristics, serving as a crucial carrier of the motor's operating status. However, a complete technical system is currently lacking for effectively acquiring and processing ripple signals, and for combining these signals to achieve motor condition analysis and life prediction.

[0005] In ripple signal processing, effectively filtering out noise interference, accurately extracting ripple features, and establishing a correlation between ripple features and motor stress distribution, fatigue cycles, and life prediction remain major challenges. Some existing ripple signal processing solutions employ single-scale filtering and feature extraction methods, which cannot adapt to the changing characteristics of ripple signals at different operating stages of the seat motor, resulting in insufficient feature extraction accuracy. Furthermore, when correlating ripple signals with motor life, the lack of effective mechanical analysis and fatigue calculation methods makes it difficult to achieve a precise mapping from ripple signals to motor life. These problems all restrict the further development and application of seat motor condition monitoring and life prediction technologies. Summary of the Invention

[0006] The purpose of this invention is to provide a system for sampling and calculating parameters of seat motor ripple signals to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides a system for sampling and calculating parameters of seat motor ripple signals, the system comprising:

[0008] The ripple signal acquisition module is used to acquire the raw ripple signal during the operation of the seat motor in real time, and to perform time-domain segmentation processing on the raw ripple signal to obtain a segmented ripple signal sequence.

[0009] The noise filtering module is used to perform multi-scale decomposition on the segmented ripple signal sequence, extract and filter out high-frequency noise components, and obtain a denoised ripple signal sequence.

[0010] The feature extraction module extracts ripple frequency features, ripple amplitude features, and ripple period features from the denoised ripple signal sequence to generate a ripple feature set.

[0011] The dynamic mesh generation module, based on the ripple feature set, generates analysis meshes of different fineness according to the signal resolution differences in different operating stages of the seat motor, thus obtaining a multi-resolution ripple feature distribution.

[0012] The ripple load spectrum analysis module is used to perform spectral analysis on the time series of multi-resolution ripple characteristic distributions to obtain the ripple load spectrum.

[0013] The stress calculation module, based on the ripple load spectrum, uses the finite element analysis method to calculate the stress distribution of key nodes of the seat motor and generate stress state data.

[0014] The fatigue cycle identification module identifies the number of fatigue cycles at key nodes based on stress state data.

[0015] The life prediction module, by combining the preset material fatigue curve and fatigue cycle number, calculates the remaining life of key nodes and obtains the overall life prediction result of the seat motor based on the remaining life.

[0016] Preferably, the dynamic mesh partitioning module includes:

[0017] The area division sub-module divides the operation of the seat motor into multiple sub-stages;

[0018] The queue construction submodule constructs multiple computation queues, with each computation queue corresponding to a sub-stage.

[0019] The mesh parameter determination submodule determines the cell length, number of meshes, and mesh density of each sub-stage based on the ripple signal resolution of each sub-stage.

[0020] The mesh generation submodule is used to control multiple computing queues to generate the target mesh for each sub-stage according to the cell length, number of meshes and mesh density of each sub-stage.

[0021] The feature distribution generation submodule generates multi-resolution ripple feature distributions based on the ripple feature set and the target mesh of each sub-stage.

[0022] Preferably, the mesh generation submodule includes:

[0023] The surface information generation unit controls multiple computing queues to initially generate mesh surface information according to the mesh parameters of each sub-stage;

[0024] The volume mesh generation unit generates a 3D mesh for each sub-stage based on the mesh surface information and extracts the mesh volume information;

[0025] Numbering and allocation unit, used to number the vertices and volume elements of the 3D mesh and generate a unique string;

[0026] The synchronous communication unit controls the synchronous communication between multiple computing queues, enabling the volume units and unique strings of adjacent sub-stages to project onto each other and generate projection information.

[0027] The mesh optimization unit optimizes the 3D mesh of each sub-stage based on the projection information to obtain the target mesh.

[0028] Preferably, the noise filtering module includes:

[0029] The decomposition unit is used to perform multi-scale decomposition on the segmented ripple signal sequence to obtain sub-signals at different frequency scales.

[0030] A noise identification unit is used to extract high-frequency noise component features from sub-signals;

[0031] The filtering unit is used to adaptively adjust the filter parameters according to the characteristics of high-frequency noise components to filter the sub-signals;

[0032] The reconstruction unit is used to reconstruct the filtered sub-signals into a denoised ripple signal sequence.

[0033] Preferably, the feature extraction module includes:

[0034] The preliminary feature extraction unit is used to extract a preliminary feature set of ripple frequency, amplitude and period from the denoised ripple signal sequence;

[0035] The dimensionality reduction unit is used to reduce the dimensionality of the initial feature set when the dimensionality of the initial feature set exceeds a preset threshold.

[0036] The correlation enhancement unit is used to analyze the time series of the dimensionality-reduced data, enhance the dynamic correlation features between ripple frequency, amplitude and period, and generate a ripple feature set.

[0037] Preferably, the ripple load spectrum analysis module includes:

[0038] The spectrum conversion unit is used to convert the time series of multi-resolution ripple feature distributions into a frequency domain representation.

[0039] The energy calculation unit is used to calculate the energy distribution and peak amplitude of each frequency band.

[0040] The spectrum generation unit generates the ripple load spectrum based on the energy distribution and amplitude peak value.

[0041] Preferably, the stress calculation module includes:

[0042] Model building unit, used to build a three-dimensional finite element model of the seat motor;

[0043] The load application element generates external excitation based on the ripple load spectrum and applies it to the finite element model.

[0044] The stress analysis element is used to perform finite element calculations in conjunction with boundary conditions to obtain the stress distribution at key nodes.

[0045] Preferably, the fatigue cycle identification module includes:

[0046] The trend analysis unit is used to analyze the temporal and spatial trends of stress state data.

[0047] The cycle counting unit identifies the number of fatigue cycles based on the changing trend.

[0048] Preferably, the lifetime prediction module includes:

[0049] Curve matching unit, used to match the number of fatigue cycles with the material fatigue curve;

[0050] The remaining lifetime calculation unit is used to calculate the remaining lifetime of critical nodes;

[0051] The integration unit is used to integrate the remaining lifetime of each key node and generate an overall lifetime prediction result.

[0052] Preferably, the system further includes:

[0053] The dynamic adjustment module is used to monitor the changing trends of ripple signals and stress state data in real time, and dynamically update mesh generation parameters and load spectrum analysis parameters.

[0054] Compared with the prior art, the beneficial effects of the present invention are:

[0055] This seat motor ripple signal sampling and parameter calculation system provides a new technical approach for seat motor condition monitoring and life assessment by constructing a complete ripple signal acquisition, processing, analysis, and life prediction technology system. The system first uses a ripple signal acquisition module to acquire the raw ripple signal of the seat motor in real time during operation and performs time-domain segmentation processing. This accurately captures the ripple signal changes of the motor at different operating stages. Compared with traditional voltage and current detection schemes, it can more deeply reflect the internal operating state of the motor, providing richer and more accurate raw data for subsequent condition analysis and life prediction.

[0056] In the ripple signal processing stage, the noise filtering module employs a multi-scale decomposition method to extract and filter high-frequency noise components. Compared to traditional single-filtering methods, this approach better adapts to the noise distribution characteristics of ripple signals across different frequency bands, effectively preserving useful information within the ripple signal and avoiding feature extraction bias caused by noise interference, thereby improving the accuracy of subsequent feature analysis. The feature extraction module extracts ripple frequency features, ripple amplitude features, and ripple period features from the denoised ripple signal sequence. The generated ripple feature set comprehensively reflects key operating information such as motor speed changes and electromagnetic characteristic fluctuations, providing strong support for refined analysis of motor operating status and solving the problems of traditional vibration signals being easily interfered with and lacking sufficient feature information.

[0057] The dynamic mesh generation module, based on the ripple feature set, generates analysis meshes of varying fineness according to the signal resolution differences at different operating stages of the seat motor, resulting in a multi-resolution ripple feature distribution. This dynamic mesh generation method can adopt an analysis precision that matches the changing characteristics of the ripple signal at different operating stages of the motor. It ensures the accuracy of analysis during periods of rapid signal change while avoiding unnecessary waste of computational resources during periods of stable signal, achieving a balance between analysis accuracy and computational efficiency. Compared to traditional fixed-scale analysis methods, it is more adaptable to the complex and variable operating conditions of seat motors.

[0058] The ripple load spectrum analysis module performs spectral analysis on the time series of multi-resolution ripple characteristic distributions. The resulting ripple load spectrum clearly reflects the load variation patterns of the motor over different time periods, establishing a correlation between ripple characteristics and motor load. Based on the ripple load spectrum, the stress calculation module uses finite element analysis to calculate the stress distribution of key nodes in the seat motor. This accurately captures the stress state changes of key internal components under different loads. Compared to traditional stress estimation methods that rely on empirical formulas, this module more accurately reflects the mechanical state of the motor, providing a reliable stress data foundation for subsequent fatigue analysis.

[0059] The fatigue cycle identification module identifies the number of fatigue cycles at key nodes based on stress state data. It accurately counts the number of load cycles experienced by key nodes during actual motor operation, considering the impact of load fluctuations at different operating stages on fatigue damage, thus solving the problem of traditional life prediction methods ignoring changes in operating conditions. The life prediction module, combining preset material fatigue curves and fatigue cycle counts, calculates the remaining life of key nodes and obtains the overall motor life prediction result. This achieves a closed-loop technology process from ripple signal acquisition to life prediction, providing precise guidance for the maintenance of seat motors, avoiding cost waste due to over-maintenance or safety hazards caused by insufficient maintenance. It also provides important practical operating data references for seat motor design optimization, helping to improve the overall design level and reliability of the motor. Attached Figure Description

[0060] Figure 1 This is a timing diagram of the seat motor ripple signal sampling and parameter calculation system described in this invention;

[0061] Figure 2 A schematic diagram illustrating the working principle of a dynamic mesh partitioning module;

[0062] Figure 3 This is a schematic diagram illustrating the working principle of the noise filtering module. Detailed Implementation

[0063] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0064] Please see Figure 1The present invention provides a system for sampling and calculating parameters of ripple signals of a seat motor. The system includes: a ripple signal acquisition module, a noise filtering module, a feature extraction module, a dynamic mesh generation module, a ripple load spectrum analysis module, a stress calculation module, a fatigue cycle identification module, and a life prediction module.

[0065] The ripple signal acquisition module acquires the raw ripple signal during the operation of the seat motor in real time and performs time-domain segmentation processing on the raw ripple signal to generate a segmented ripple signal sequence. The noise filtering module performs multi-scale decomposition on the segmented ripple signal sequence, extracts and filters out high-frequency noise components, and outputs a denoised ripple signal sequence. The feature extraction module extracts ripple frequency features, ripple amplitude features, and ripple period features from the denoised ripple signal sequence to form a ripple feature set. The dynamic mesh generation module generates analysis meshes of different fineness based on the ripple feature set and the signal resolution differences at different operating stages of the seat motor, obtaining a multi-resolution ripple feature distribution. The ripple load spectrum analysis module performs spectral analysis on the time series of the multi-resolution ripple feature distribution to generate a ripple load spectrum. The stress calculation module calculates the stress distribution of key nodes of the seat motor using the finite element analysis method based on the ripple load spectrum, generating stress state data. The fatigue cycle identification module identifies the number of fatigue cycles of key nodes based on the stress state data. The life prediction module calculates the remaining life of key nodes by combining the preset material fatigue curve and the number of fatigue cycles, and obtains the overall life prediction result of the seat motor based on the remaining life.

[0066] Example 1: See Figure 2The implementation of the dynamic mesh partitioning module involves the collaborative work of multiple components. The region partitioning submodule divides the entire operation process into multiple sub-stages based on the characteristics of the motor control signal changes. These sub-stages include different states such as start-up acceleration, steady-state operation, and deceleration and stopping. The partitioning of each sub-stage is based on the inflection point and rate of change of the motor current and speed curves, thereby ensuring that the partitioning results can reflect the physical process changes in actual operation. The queue construction submodule creates an independent computing queue for each sub-stage. The computing queues use a first-in-first-out data structure to manage data processing tasks. Each queue is allocated independent memory space and computing resources to achieve a parallel processing architecture to improve efficiency. Data exchange between queues is completed through a shared cache mechanism to avoid resource conflicts and maintain timing consistency. The mesh parameter determination submodule analyzes the ripple signal resolution characteristics of each sub-stage. The resolution is dynamically evaluated by the signal sampling frequency and signal-to-noise ratio. The high-resolution stage corresponds to a smaller cell length and a higher mesh density, while the low-resolution stage uses a larger cell length and a lower number of meshes. The specific value of the cell length is calculated by the ratio of the signal wavelength to the sampling interval. The number of meshes is determined based on the quotient of the signal duration and the cell length. The mesh density is further calculated from the cell length and the number of meshes, thus forming the mesh parameter set for each sub-stage.

[0067] The mesh generation submodule drives multiple computation queues to generate the target mesh based on the determined mesh parameters. The surface information generation unit generates preliminary mesh surface information in each queue based on the cell length and number of meshes. This information includes vertex coordinate sets, edge connectivity matrices, and patch topology data. Vertex coordinates are calculated based on signal time series and spatial mapping relationships. Edge connectivity is constructed using the Delaunay triangulation method, and patch topology records geometric shapes and adjacency relationships. The volume mesh generation unit constructs a three-dimensional mesh structure based on the mesh surface information. The three-dimensional mesh uses hexahedral cells to fill the three-dimensional space. The type, volume, and spatial location of the volume cells are generated using patch expansion and voxelization methods. Volume cell attributes include material type and physical parameters, thus forming complete mesh volume information. The numbering and allocation unit uniquely identifies the vertices and volume cells of the three-dimensional mesh. Vertex numbers are generated as unique strings using a hash function, combining spatial coordinates and timestamps. Volume cell numbers are generated using a hybrid encoding method based on their spatial location and topological relationships. These unique strings ensure global uniqueness and traceability throughout the entire mesh system. The synchronous communication unit controls data synchronization and exchange between computation queues. Volume elements and unique strings from adjacent sub-stages are associated through projection mapping. The projection process uses nearest neighbor interpolation and coordinate transformation to generate projection information containing overlapping region matching degrees and consistency verification data. This projection information ensures the continuity and smooth transition of mesh data between sub-stages. The mesh optimization unit optimizes the 3D mesh of each sub-stage based on the projection information. Optimization operations include mesh smoothing, element merging, and adaptive resolution adjustment. Mesh smoothing eliminates local distortion using the Laplacian operator, element merging reduces redundant meshes in overlapping regions, and resolution adjustment dynamically refines or coarsens the mesh based on signal characteristic changes, ultimately outputting the target mesh structure for each sub-stage.

[0068] The feature distribution generation submodule maps the ripple feature set onto the target grid of each sub-stage. The ripple feature set includes multi-dimensional data such as frequency, amplitude, and period. The mapping process is based on the correspondence between the spatiotemporal coordinates of the grid vertices and the feature time series. A bilinear interpolation method and a weight allocation algorithm are used to distribute the feature values ​​to the grid nodes, ultimately forming a multi-resolution ripple feature distribution. This distribution is stored in a grid data structure, containing node values, gradient information, and a time evolution sequence. Standardized interfaces are used for data transfer between components to ensure the system's scalability and stability. Simultaneously, the dynamic grid partitioning mechanism effectively adapts to changes in signal resolution during motor operation, improving the accuracy and reliability of subsequent analysis.

[0069] Example 2: See Figure 3The noise filtering module implements signal purification through multi-scale decomposition and adaptive filtering. The decomposition unit receives the segmented ripple signal sequence from the previous module and performs multi-scale decomposition on each segment using wavelet transform. A five-level decomposition is performed using the Db4 wavelet as the basis function, resulting in a set of sub-signals containing high-frequency detail coefficients and low-frequency approximation coefficients. Each sub-signal corresponds to a different frequency bandwidth, covering all components from high-frequency noise to the low-frequency fundamental wave. The noise identification unit extracts noise component features from the decomposed high-frequency sub-signals. It distinguishes noise from valid signals by calculating the statistical characteristics of each level of detail coefficients, including amplitude distribution, energy proportion, and zero-crossing rate. A dynamic threshold is set to identify significant noise segments; this threshold is adaptively adjusted based on the overall signal energy. The high-frequency noise component features are quantized into feature vectors to guide the filtering process. The filtering unit adaptively adjusts the filter parameters based on the noise feature vectors, employing different filtering strategies for different frequency bands. Median filtering is used for impulse noise, while frequency-response-based IIR filters are used to suppress broadband noise. The filter cutoff frequency and order are calculated in real-time based on the noise bandwidth, achieving targeted noise suppression. The reconstruction unit reconstructs the filtered sub-signals of each layer, uses inverse wavelet transform to merge the processed detail coefficients and approximation coefficients, and reconstructs the complete signal sequence after denoising. The reconstruction process maintains the phase consistency of the signal and performs special processing on edge effects to avoid distortion. Finally, it outputs a smooth ripple signal sequence that retains key features.

[0070] The feature extraction module extracts key parameters from the denoised signal. The preliminary feature extraction unit performs time-domain and frequency-domain analysis on the signal, calculates the instantaneous frequency through zero-crossing detection, extracts amplitude modulation information through envelope analysis, and determines periodic characteristics through autocorrelation function, forming a preliminary feature set containing the original frequency, amplitude, and period parameters. This feature set has high dimensionality and contains redundant information. The dimensionality reduction unit starts processing when the feature set dimension exceeds a preset threshold. It uses principal component analysis to transform the preliminary features, calculates the covariance matrix of the feature vectors, and solves for eigenvalues ​​and eigenvectors. It retains principal component components with a cumulative contribution rate exceeding a set value, mapping the original high-dimensional features to a low-dimensional space while preserving most of the original information. The correlation enhancement unit analyzes the time-series data of the dimensionality-reduced features, uses a time-delay embedding method to construct a dynamic relationship model between features, calculates the mutual information and cross-correlation between frequency, amplitude, and period parameters, and enhances the dynamic correlation features between parameters through state space reconstruction. Finally, it generates a compact and information-rich ripple feature set. This feature set fully captures the dynamic behavior of the signal and has a reasonable dimensionality, providing effective input for subsequent processing. The noise filtering module dynamically adjusts processing parameters based on signal characteristics to avoid signal feature loss due to over-filtering. The feature extraction module balances information integrity and computational complexity through dimensionality reduction and enhancement processing. Standardized data interfaces are used between units to ensure collaborative work between modules. The system achieves real-time processing capabilities while maintaining high-precision feature extraction performance.

[0071] Taking the operation of a car seat position adjustment motor as an example, the motor generates a raw ripple signal containing a mixture of electromagnetic noise and mechanical vibration during the adjustment process. The signal sampling frequency is set to 10kHz, and the operating data is continuously recorded for 2 seconds. The noise filtering module first receives a segmented signal sequence from the acquisition module, with each segment having a length of 500 sampling points, corresponding to a 50-millisecond time window. The decomposition unit uses the Db4 wavelet basis function to perform a five-level decomposition on each signal segment, obtaining five detail components D1-D5 and one approximate component A5. Among them, the D1 and D2 components mainly contain high-frequency noise components, which are characterized by signal fluctuations with large amplitudes and strong randomness. The noise identification unit analyzes the statistical characteristics of each detail component, calculates the peak-to-peak value, root mean square value, and zero-crossing rate of each component within the time window, and finds that the peak-to-peak value of the D1 component reaches 35% of the basic signal amplitude, and its zero-crossing rate is significantly higher than that of other components. Based on this, the D1 and D2 components are determined to be the main noise components. The filtering unit adaptively adjusts the filter parameters based on the recognition results. A fourth-order Butterworth high-pass filter with a cutoff frequency of 800Hz is used for components D1 and D2, while a low-pass filter with a cutoff frequency of 200Hz is used for components D3-D5, preserving the main signal features. The reconstruction unit resynthesizes the processed components using inverse wavelet transform, outputting a smoothed signal segment. During reconstruction, special attention is paid to the connections between segments, employing an overlap-preservation method to avoid edge distortion.

[0072] The feature extraction module processes the denoised signal sequence. The preliminary feature extraction unit calculates the time and frequency domain parameters of each signal segment, including the instantaneous frequency value obtained through zero-crossing detection, the signal envelope amplitude extracted through Hilbert transform, and the period length obtained through peak localization of the autocorrelation function, forming a preliminary feature set containing 15 dimensions. When the dimensionality reduction unit detects that the feature dimension exceeds a preset threshold of 10 dimensions, it initiates principal component analysis, calculates the covariance matrix of the feature vectors and solves for the eigenvalues, retaining the top 8 principal component components with a cumulative contribution rate of 95%, reducing the original feature dimension from 15 to 8. The correlation enhancement unit analyzes the time series of the dimensionality-reduced features and constructs an 8-dimensional feature vector state-space model. By calculating the delay mutual information, a significant phase lag relationship is found between the frequency features and the amplitude features. The dynamic time warping algorithm is used to enhance this time-varying correlation characteristic, and finally a ripple feature set containing 10 enhanced features is generated. This set retains the essential information of individual features while enhancing the dynamic correlation between features. The processing time of each signal segment is controlled within 5 milliseconds, meeting the real-time requirements. The output ripple feature set accurately captures the state change characteristics of the motor during operation, including the frequency gradual change process in the start-up phase, the amplitude modulation characteristics in the stable operation phase, and the oscillation decay mode in the stopping phase, providing high-quality feature data for subsequent analysis.

[0073] Example 3: The implementation of the ripple load spectrum analysis module completes the spectral feature extraction through the collaborative work of three core units. The spectrum conversion unit receives the time series data of multi-resolution ripple feature distribution. This data contains information on the changes of spatial grid node feature values ​​over time at different operating stages of the motor. The unit first preprocesses the feature time series of each grid node, including removing linear trends and applying window functions to reduce spectral leakage. The Hamming window function is used to window each segment of 4096 data points. Then, the fast Fourier transform is applied to convert the time domain signal into a frequency domain representation, generating a frequency domain complex sequence containing real and imaginary parts with a frequency resolution of 0.244Hz. During the conversion process, the frequency axis range is automatically adjusted according to the signal sampling rate. The energy calculation unit processes the complex frequency data after spectrum conversion. This unit defines the analysis frequency band range from the DC component to the Nyquist frequency. It uses a critical band division method to divide the frequency axis into 24 non-uniform sub-bands. The low-frequency region has a narrower bandwidth and the high-frequency region has a wider bandwidth. The signal energy in each sub-band is calculated by integrating the square of the power spectrum modulus of all frequency points in the band. At the same time, the peak amplitude and its corresponding precise frequency position in each sub-band are detected. The peak detection uses parabolic interpolation to improve the frequency estimation accuracy, and the peak amplitude is recorded in decibels.

[0074] The spectrum generation unit constructs a complete ripple load spectrum based on energy distribution and peak amplitude data, and calculates the normalized energy characteristics of each frequency band using the following formula:

[0075] ,

[0076] in: Indicates in frequency band The relative peak energy percentage within, Represents frequency variables. and These represent the lower and upper frequency limits of the bandwidth, respectively. Indicates the signal at frequency Fourier transform coefficients at the point, This represents the total number of frequency points within the frequency band. It is the first Each frequency point, This refers to frequency resolution. The formula quantifies the importance of the maximum peak energy within each frequency band relative to the total energy.

[0077] The load spectrum stores frequency, energy, and amplitude information in matrix form. The frequency axis uses a logarithmic scale to annotate key characteristic frequency points, and energy values ​​are stored in decibels for easy representation of a large dynamic range. Amplitude peak values ​​are associated with their corresponding frequency positions and occurrence times. Simultaneously, the cumulative energy distribution curves for each frequency band are recorded. The resulting ripple load spectrum provides a complete characteristic description of the energy in the frequency domain. Spectrum calculations use an overlap-preservation method to improve frequency resolution, and energy calculations employ parallel processing to accelerate frequency band analysis. Load spectrum generation supports a real-time update mechanism, and all parameters are adaptively adjusted based on the input signal characteristics. This includes window function type, frequency band division scheme, and peak detection threshold, all dynamically optimized based on signal statistical characteristics to ensure applicability to motor ripple signal analysis under different operating conditions.

[0078] Taking the ripple signal analysis of a certain type of automotive seat adjustment motor under typical operating conditions as an example, this motor generates a composite ripple signal containing multiple frequency components during position adjustment. The ripple load spectrum analysis module receives multi-resolution ripple characteristic distribution data from the preceding module. This data includes characteristic time series of three stages: motor start-up, constant speed operation, and deceleration / stop. Each stage has a sampling duration of 2 seconds, a sampling rate of 2000Hz, and a total of 4000 data points. The spectrum conversion unit first preprocesses the characteristic time series of each stage, using a Hanning window function to window the data to reduce spectral leakage. The window function length is consistent with the signal segment length, 4000 points, and the overlap rate is set to 50%. The fast Fourier transform algorithm is applied to convert the time-domain signal into a frequency-domain representation. During the transformation, the frequency resolution reaches 0.5Hz, generating a frequency-domain dataset containing amplitude and phase spectra. The amplitude spectrum values ​​are stored in decibels for easy subsequent analysis. The energy calculation unit processes the converted frequency-domain data, defining the analysis frequency band as 0-1000Hz, covering the main harmonic components of motor operation. The frequency band division adopts a non-uniform approach, with a 10Hz subdivision interval within the 0-200Hz fundamental frequency band and a 50Hz interval within the 200-1000Hz frequency band. The signal energy value within each frequency band is calculated by integrating the power spectral density of all frequency points within that band. Simultaneously, the peak amplitude and its corresponding precise frequency position within each frequency band are detected. Peak detection employs a local maximum comparison algorithm combined with an adaptive threshold judgment, with the threshold set at three times the standard deviation of the average amplitude of that frequency band.

[0079] The spectrum generation unit constructs a ripple load spectrum based on energy distribution and peak amplitude data. The load spectrum is stored using a three-dimensional matrix data structure. The first dimension represents the frequency axis, marked with 128 characteristic frequency points using a non-uniform scale. The second dimension stores the energy value of the corresponding frequency point, expressed in logarithmic form. The third dimension records the peak amplitude information and its occurrence time. During load spectrum generation, the cumulative energy percentage of each frequency band is also calculated, reflecting the energy distribution characteristics within the frequency range. The final generated ripple load spectrum contains complete frequency-energy-amplitude relationship information. The entire analysis process is completed on an embedded processing platform, with computation time controlled within 200 milliseconds, meeting real-time processing requirements. The output ripple load spectrum accurately reflects the dynamic load characteristics of the motor during operation, including the dominance of low-frequency components during startup, the coexistence of multiple harmonics during constant speed operation, and the attenuation of high-frequency components during deceleration. During processing, all parameters are adaptively adjusted according to signal characteristics. The window function type and length are dynamically selected based on signal stationarity, and the frequency band division range is adaptively determined based on the distribution of major harmonic components, ensuring the analysis results have broad applicability and accuracy.

[0080] Example 4: The implementation process of the stress calculation module is carried out using a specific analysis of a certain type of seat motor as an example. The model building unit first obtains the three-dimensional geometric model data of the motor, including detailed dimensional parameters of the stator assembly, rotor assembly, bearing housing and housing structure. The finite element preprocessing software is used to generate a mesh model suitable for calculation. The mesh type adopts ten-node tetrahedral elements. The mesh is refined in stress concentration areas such as tooth groove and bearing contact area. The number of elements is controlled within a reasonable range to ensure a balance between calculation accuracy and efficiency. The final finite element model includes all key structural components and their connection relationships.

[0081] The load application unit constructs an external excitation based on the ripple load spectrum data generated by the aforementioned module. Taking a load spectrum obtained from a certain actual sampling as an example, the load spectrum contains 12 significant frequency components and their corresponding amplitudes. These frequency domain data are converted into time domain force functions through inverse Fourier transform. The time length of the force function is 2 seconds, the number of sampling points is 2000, and the maximum amplitude is controlled within the range of 50N. According to the actual installation of the motor, the force function is applied to 6 key positions such as the motor base fixing point and the output shaft connection point. The force direction at each position is determined according to the actual force conditions.

[0082] The stress analysis unit performs finite element calculations after applying loads. Boundary conditions are set, including full constraints on the base mounting surface, radial constraints on the bearing area, and contact relationships between components. The calculation adopts the transient dynamic analysis method with a time step of 0.001 seconds, and a total of 2000 time steps are calculated. The stress distribution results for each time step are output. The nonlinear characteristics of materials are considered during the calculation, including the anisotropic properties of the stator silicon steel sheet and the plastic deformation characteristics of the bearing steel.

[0083] The fatigue cycle identification module processes the stress analysis results data, and the trend analysis unit extracts the stress time history data of five key nodes (including the tooth root, bearing raceway contact point, and shaft shoulder transition area). After eliminating high-frequency fluctuation components through moving average filtering, the stress change trend is identified. The spatial change trend is evaluated by comparing the stress difference and gradient distribution of adjacent nodes, and it is found that the maximum stress gradient occurs in the transition area from the tooth root to the tooth tip.

[0084] The cycle counting unit uses the rainflow counting method to process stress time history data. Taking the tooth root node as an example, its stress time series contains obvious cyclic characteristics. All stress cycles are identified by peak and valley detection. After filtering out small cycles with amplitudes less than 5 MPa, significant cycle events are retained. The influence of average stress is considered and corrected during the counting process, and finally, the fatigue cycle number of each key node is obtained. The stress distribution and cycle counting data of key nodes are shown in Table 1.

[0085] Table 1: Stress and Cyclic Data of Key Nodes in a Certain Model of Seat Motor

[0086]

[0087] The model construction fully considers actual manufacturing tolerances and assembly clearances, the load application accurately reflects actual working conditions, the stress analysis adopts a validated constitutive model, and the fatigue identification is based on an industry-recognized counting method to ensure that the analysis results have engineering reference value. All calculation parameters and boundary conditions are recorded to ensure repeatability, and the final output stress state data and fatigue cycle count provide reliable input for life prediction.

[0088] Example 5: The curve matching unit receives the number of cycles and their corresponding stress amplitude and mean from the fatigue cycle identification module. This unit has a built-in material database storing fatigue characteristic curves of various commonly used motor materials, including SN curves and ε-N curves. Taking stator silicon steel sheet material as an example, its SN curve contains 256 data points, covering 10... 3 Up to 10 7 The range of the number of cycles is determined by the bilinear interpolation algorithm in logarithmic coordinates during the matching process. The corresponding permissible number of cycles is found based on the actual stress level. The influence of average stress is considered and adjusted using the Goodman correction method. The matching results generate damage parameters for each stress level.

[0089] The remaining life calculation unit uses Miner's linear cumulative damage theory based on damage parameters. This unit first calculates the damage degree caused by each stress cycle, defined as the ratio of the actual number of cycles to the permissible number of cycles. Then, it accumulates the damage degree of all cycles to obtain the total damage amount. When the total damage amount reaches a critical value, node failure is determined. The remaining life is calculated by multiplying the reciprocal of the total damage amount by the operating time. The calculation considers the load sequence effect and adjusts it with correction coefficients. Finally, it outputs the predicted remaining life value for each critical node. The integration unit comprehensively processes the remaining life data of all critical nodes. This unit uses the weakest link theory to determine the overall system life, identifies the minimum remaining life value among all nodes as the benchmark for the overall motor life, and considers the correlation effect between nodes. The final prediction result is adjusted through a correlation coefficient matrix. For node groups with functional redundancy, a parallel system model is used to calculate system reliability, outputting a comprehensive report including overall life prediction and the life distribution of critical nodes.

[0090] The dynamic adjustment module monitors signal changes and stress states in real time during operation. Monitoring data includes ripple signal amplitude, frequency characteristics, and stress values ​​at key nodes. A sliding time window statistical method is used to analyze data trends, with the window length set to 10 operating cycles. The dynamic update mechanism adjusts mesh generation parameters and load spectrum analysis parameters according to the changing trends. Mesh parameter updates include element length coefficients and mesh density factors, while load spectrum parameter updates involve frequency band division ranges and amplitude thresholds. The update frequency is synchronized with the monitoring window to ensure timely parameter adjustments in response to changes in operating status. A complete parameter update log is established, recording the parameter values ​​before each adjustment, the reason for the adjustment, and the adjusted values, forming a parameter change history sequence for easy traceability and analysis. All calculations employ a dual verification mechanism to ensure data accuracy. The lifetime prediction results are output in probability distribution form and provide confidence interval assessments. Finally, a complete lifetime prediction report is generated, including remaining lifetime values, reliability indicators, and analysis of key influencing factors.

[0091] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, 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.

[0092] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A system for sampling and calculating parameters of a seat motor ripple signal, characterized in that, include: The ripple signal acquisition module is used to acquire the raw ripple signal during the operation of the seat motor in real time, and to perform time-domain segmentation processing on the raw ripple signal to obtain a segmented ripple signal sequence. The noise filtering module is used to perform multi-scale decomposition on the segmented ripple signal sequence, extract and filter out high-frequency noise components, and obtain a denoised ripple signal sequence. The feature extraction module extracts ripple frequency features, ripple amplitude features, and ripple period features from the denoised ripple signal sequence to generate a ripple feature set. The dynamic mesh generation module, based on the ripple feature set, generates analysis meshes of different fineness according to the signal resolution differences in different operating stages of the seat motor, thus obtaining a multi-resolution ripple feature distribution. The ripple load spectrum analysis module is used to perform spectral analysis on the time series of multi-resolution ripple characteristic distributions to obtain the ripple load spectrum. The stress calculation module, based on the ripple load spectrum, uses the finite element analysis method to calculate the stress distribution of key nodes of the seat motor and generate stress state data. The fatigue cycle identification module identifies the number of fatigue cycles at key nodes based on stress state data. The life prediction module, by combining the preset material fatigue curve and fatigue cycle number, calculates the remaining life of key nodes and obtains the overall life prediction result of the seat motor based on the remaining life.

2. The seat motor ripple signal sampling and parameter calculation system as described in claim 1, characterized in that, The dynamic mesh generation module includes: The area division sub-module divides the operation of the seat motor into multiple sub-stages; The queue construction submodule constructs multiple computation queues, with each computation queue corresponding to a sub-stage. The mesh parameter determination submodule determines the cell length, number of meshes, and mesh density of each sub-stage based on the ripple signal resolution of each sub-stage. The mesh generation submodule is used to control multiple computing queues to generate the target mesh for each sub-stage according to the cell length, number of meshes and mesh density of each sub-stage. The feature distribution generation submodule generates multi-resolution ripple feature distributions based on the ripple feature set and the target mesh of each sub-stage.

3. The seat motor ripple signal sampling and parameter calculation system as described in claim 2, characterized in that, The mesh generation submodule includes: The surface information generation unit controls multiple computing queues to initially generate mesh surface information according to the mesh parameters of each sub-stage; The volume mesh generation unit generates a 3D mesh for each sub-stage based on the mesh surface information and extracts the mesh volume information; Numbering and allocation unit, used to number the vertices and volume elements of the 3D mesh and generate a unique string; The synchronous communication unit controls the synchronous communication between multiple computing queues, enabling the volume units and unique strings of adjacent sub-stages to project onto each other and generate projection information. The mesh optimization unit optimizes the 3D mesh of each sub-stage based on the projection information to obtain the target mesh.

4. The seat motor ripple signal sampling and parameter calculation system as described in claim 1, characterized in that, The noise filtering module includes: The decomposition unit is used to perform multi-scale decomposition on the segmented ripple signal sequence to obtain sub-signals at different frequency scales. A noise identification unit is used to extract high-frequency noise component features from sub-signals; The filtering unit is used to adaptively adjust the filter parameters according to the characteristics of high-frequency noise components to filter the sub-signals; The reconstruction unit is used to reconstruct the filtered sub-signals into a denoised ripple signal sequence.

5. The system for sampling and calculating parameters of a seat motor ripple signal as described in claim 1, characterized in that, The feature extraction module includes: The preliminary feature extraction unit is used to extract a preliminary feature set of ripple frequency, amplitude and period from the denoised ripple signal sequence; The dimensionality reduction unit is used to reduce the dimensionality of the initial feature set when the dimensionality of the initial feature set exceeds a preset threshold. The correlation enhancement unit is used to analyze the time series of the dimensionality-reduced data, enhance the dynamic correlation features between ripple frequency, amplitude and period, and generate a ripple feature set.

6. The system for sampling and calculating parameters of a seat motor ripple signal as described in claim 1, characterized in that, The ripple load spectrum analysis module includes: The spectrum conversion unit is used to convert the time series of multi-resolution ripple feature distributions into a frequency domain representation. The energy calculation unit is used to calculate the energy distribution and peak amplitude of each frequency band. The spectrum generation unit generates the ripple load spectrum based on the energy distribution and amplitude peak value.

7. The system for sampling and calculating parameters of a seat motor ripple signal as described in claim 1, characterized in that, The stress calculation module includes: Model building unit, used to build a three-dimensional finite element model of the seat motor; The load application element generates external excitation based on the ripple load spectrum and applies it to the finite element model. The stress analysis element is used to perform finite element calculations in conjunction with boundary conditions to obtain the stress distribution at key nodes.

8. The system for sampling and calculating parameters of a seat motor ripple signal as described in claim 1, characterized in that, The fatigue cycle identification module includes: The trend analysis unit is used to analyze the temporal and spatial trends of stress state data. The cycle counting unit identifies the number of fatigue cycles based on the changing trend.

9. The system for sampling and calculating parameters of a seat motor ripple signal as described in claim 1, characterized in that, The lifetime prediction module includes: Curve matching unit, used to match the number of fatigue cycles with the material fatigue curve; The remaining lifetime calculation unit is used to calculate the remaining lifetime of critical nodes; The integration unit is used to integrate the remaining lifetime of each key node and generate an overall lifetime prediction result.

10. The system for sampling and calculating parameters of a seat motor ripple signal as described in claim 1, characterized in that, Also includes: The dynamic adjustment module is used to monitor the changing trends of ripple signals and stress state data in real time, and dynamically update mesh generation parameters and load spectrum analysis parameters.

Images