An accelerometer vibration rectification error test system and method

Abstract

The application relates to the technical field of data processing, in particular to an accelerometer vibration rectification error test system and method, which comprises an adaptive signal acquisition module, a bidirectional collaborative transmission module, a composite modal statistical module and a cross-validation error modeling module. The adaptive signal acquisition module generates a characteristic mark, the characteristic mark drives the bidirectional collaborative transmission module to adjust the buffer depth and inject a timestamp verification mark; the composite modal statistical module selects a wavelet packet time-frequency fusion algorithm based on a timestamp missing ratio, extracts time-frequency entropy mean statistics by aligning a physical cycle division window; the cross-validation error modeling module fuses the statistics and vibration parameters to generate a quantitative model, and a confidence evaluation factor distribution diagram triggers a double-path feedback: a stealth window optimizes and improves the data quality in a key period, and a frequency domain weight re-distribution optimizes a feature extraction strategy. The application solves the problems of insufficient dynamic data real-time processing efficiency and inaccurate transient feature extraction caused by the fragmentation of data acquisition and analysis functions in the prior art.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, and in particular to an accelerometer vibration rectification error test system and method. BACKGROUND

[0002] The accelerometer vibration rectification error is a distortion phenomenon in accelerometer measurement, which is essentially caused by the nonlinear physical characteristics of the sensing element. When the device is exposed to a symmetrical alternating acceleration environment, the average acceleration output should ideally be equal to zero. However, the nonlinear characteristics of the sensing element make the response amplitudes of positive and negative direction accelerations unequal, which directly leads to the asymmetrical bias of the output waveform. Next, the time average calculation acts on this asymmetric waveform, which will inevitably produce a non-zero DC offset component. The DC offset component is equivalent to an additional constant bias error, which is superimposed on the true measured acceleration value, ultimately causing the decline of the acceleration measurement accuracy.

[0003] The existing accelerometer vibration rectification error test has the following technical pain points, specifically: the existing test system uses discrete hardware and independent data processing units, and lacks deep integration between the data acquisition module and the backend analysis software. The test personnel need to manually switch different devices to complete signal acquisition, filtering and data recording, and then rely on third-party tools for statistical analysis, which leads to a fragmented operation process. Especially when observing the dynamic data changes in the vibration environment, because the key statistics cannot be extracted in real time, the original data must be repeatedly exported and the extreme value and mean value in a specific period are manually calculated. For example, when the test personnel analyze the response drift of the accelerometer in the vibration process, they need to intercept the vibration waveform data in different time periods and perform extreme value search and mean value calculation in segments, which greatly prolongs the analysis period; in addition, because the hardware sampling control and software processing are not optimized, the key feature points of high-frequency vibration signals are missed, which causes the rectification error calculation to deviate from the true working condition. SUMMARY

[0004] In view of the technical problems of the prior art, the present application provides an accelerometer vibration rectification error test system and method, which solves the technical problems of low real-time segmented statistical efficiency of dynamic data changes and insufficient key transient response feature extraction accuracy in the accelerometer vibration rectification error test process due to the fragmentation of the functions of the data acquisition hardware and the backend analysis software of the test system and the lack of deep collaborative optimization.

[0005] To solve the above technical problems, the specific technical solutions of the present application are as follows:

[0006] In a first aspect, the present application provides an accelerometer vibration rectification error test system, comprising:

[0007] The adaptive signal acquisition module acquires the current signal output by the accelerometer and performs analog-to-digital conversion, outputs a raw digital signal, extracts a frequency energy distribution parameter of the raw digital signal, generates a feature tag based on the frequency energy distribution parameter, appends the feature tag to a digital signal sequence, and outputs the digital signal sequence with the feature tag;

[0008] The bidirectional collaborative transmission module receives the digital signal sequence output by the adaptive signal acquisition module, analyzes the feature tag in the digital signal sequence, dynamically calculates a buffer depth adjustment amount according to the frequency energy distribution parameter value carried in the feature tag, sets the ring buffer depth to the buffer depth adjustment amount, synchronously detects a state jump edge of the feature tag, appends a timestamp verification identifier to the corresponding data frame header when the jump edge occurs, and outputs a data stream integrated with the timestamp verification identifier to the composite modal statistical module.

[0009] The composite modal statistical module receives the data stream output by the bidirectional collaborative transmission module, analyzes the timestamp verification identifier in the data stream, detects a missing proportion of the timestamp verification identifier, selects a wavelet packet time-frequency fusion algorithm from a preset algorithm library when the missing proportion exceeds a preset threshold, aligns a signal physical cycle to divide a statistical window, generates statistical data based on a signal analysis result, and the statistical data at least includes a time-frequency entropy mean value and a corresponding algorithm selection identifier.

[0010] The cross-validation error modeling module receives the statistical data output by the composite modal statistical module, extracts the algorithm selection identifier in the statistical data, adopts a range dynamic weighting calculation when the algorithm selection identifier indicates that the wavelet packet time-frequency fusion algorithm is adopted, adopts a fuzzy cognitive mapping calculation when the algorithm selection identifier indicates that a preset statistical algorithm of a non-wavelet packet time-frequency fusion algorithm is adopted, and fuses the statistical data to output a vibration rectification error quantization model and a confidence evaluation factor distribution map.

[0011] Further, the accelerometer vibration rectification error test system provided by the application has the advantages that the adaptive signal acquisition module includes a multi-core analog-to-digital conversion array.

[0012] The multi-core analog-to-digital conversion array inputs the current signal output by the accelerometer, performs analog-to-digital conversion, and outputs a raw digital signal.

[0013] The first 10ms time window of the raw digital signal is extracted, and the energy proportion of the 10-500Hz frequency band of the signal in the first 10ms time window is calculated.

[0014] When the energy proportion exceeds 20%, a sampling rate improvement instruction is output to the multi-core analog-to-digital conversion array, a feature tag is generated based on the energy proportion, and the feature tag is transmitted to the bidirectional collaborative transmission module.

[0015] Further, the bidirectional collaborative transmission module of the accelerometer vibration rectification error test system provided by the application is configured to:

[0016] receiving the digital signal sequence with the feature mark output by the adaptive signal acquisition module, and parsing the feature mark in the digital signal sequence;

[0017] calculating a buffer depth adjustment amount based on the energy proportion value in the feature mark and a preset scaling factor, the buffer depth adjustment amount being a product of the initial depth value and the energy proportion value and the preset scaling factor;

[0018] setting the annular buffer depth as the buffer depth adjustment amount;

[0019] detecting a state jump edge of the feature mark, and when the state jump edge occurs, attaching a timestamp check identifier to the head of a data frame, and outputting a data stream with the integrated timestamp check identifier to a composite modal statistical module.

[0020] Further, the accelerometer vibration rectification error test system, the composite modal statistical module is configured to:

[0021] receiving a data stream output by the bidirectional collaborative transmission module, and parsing a timestamp check identifier sequence from the data stream;

[0022] counting the number of losses per second of the timestamp check identifier sequence;

[0023] when the number of losses exceeds 5%, selecting a wavelet packet time-frequency fusion algorithm from a preset algorithm library;

[0024] aligning the received data stream with a signal physical cycle to divide a statistical window;

[0025] performing wavelet packet decomposition to the 5th layer, and calculating the energy entropy of all subbands obtained by the decomposition;

[0026] calculating the time-frequency entropy mean value based on the energy entropy of all subbands, generating statistical data including the time-frequency entropy mean value, marking a wavelet packet time-frequency fusion algorithm selection identifier, and outputting the statistical data marked with the wavelet packet time-frequency fusion algorithm selection identifier to a cross-validation error modeling module.

[0027] Further, the accelerometer vibration rectification error test system, the composite modal statistical module is further configured to:

[0028] based on the time-frequency entropy mean value, the signal digest hash stored by the composite modal statistical module, and the statistical window boundary coordinates, creating a statistical data object including a time-frequency entropy mean value field;

[0029] marking an algorithm selection identifier on the statistical data object, and outputting the statistical data object marked with the algorithm selection identifier to the cross-validation error modeling module.

[0030] Further, the accelerometer vibration rectification error test system disclosed by the application is characterized in that the cross-validation error modeling module is configured to:

[0031] receive the statistical data object labeled by the algorithm selection identifier output by the composite modal statistical module as a statistical data set;

[0032] extract the algorithm selection identifier from the statistical data set;

[0033] when the algorithm selection identifier indicates that the wavelet packet time-frequency fusion algorithm is used, input the statistical data set into the range dynamic weighting algorithm to perform calculation;

[0034] when the algorithm selection identifier indicates that the preset statistical algorithm of the non-wavelet packet time-frequency fusion algorithm is used, input the statistical data set into the fuzzy cognitive mapping algorithm to perform calculation;

[0035] receive the vibration frequency and amplitude parameters input from the outside;

[0036] fuse the calculation results of the range dynamic weighting algorithm or the fuzzy cognitive mapping algorithm with the vibration frequency and amplitude parameters to generate a vibration rectification error quantification model;

[0037] calculate a confidence evaluation factor based on the vibration rectification error quantification model and output a confidence evaluation factor distribution diagram.

[0038] Further, the accelerometer vibration rectification error test system disclosed by the application is characterized in that the composite modal statistical module is further configured to:

[0039] receive a user box selection time interval instruction, analyze the user box selection time interval instruction, and obtain the time interval start point and end point coordinates;

[0040] cut the data segment between the start point and the end point from the data stream output by the bidirectional collaborative transmission module;

[0041] use the frequency domain energy distribution parameter extraction method of the adaptive signal acquisition module to calculate the frequency domain main frequency component of the data at the time interval start point of the cut data segment;

[0042] calculate the physical extension length based on the frequency domain main frequency component, and the physical extension length is equal to an integer N multiplied by 1 divided by the main frequency component, N being an integer;

[0043] set an optimized statistical window boundary, and the start point of the optimized statistical window boundary is the time interval start point, and the end point is the time interval start point plus the physical extension length;

[0044] re-analyze the timestamp verification identifier sequence within the optimized statistical window boundary to detect the missing proportion of the timestamp verification identifier;

[0045] When the missing proportion exceeds the preset threshold, a wavelet packet time-frequency fusion algorithm is selected from a preset algorithm library, a physical cycle division of data streams in an optimized statistical window is aligned, wavelet packet decomposition is performed to the 5th layer, and energy entropy of all subbands obtained by the decomposition is calculated;

[0046] A time-frequency entropy mean value is calculated based on the energy entropy of all subbands, statistical data including the time-frequency entropy mean value is generated, and an algorithm selection identification is marked.

[0047] Further, the accelerometer vibration rectification error test system provided by the application further comprises:

[0048] After the cross-validation error modeling module generates the confidence evaluation factor distribution map, the average confidence evaluation factor of the distribution map is calculated, and the average confidence evaluation factor feedback signal is output to the composite modal statistical module and the adaptive signal acquisition module;

[0049] When the feedback signal value received by the composite modal statistical module is lower than 0.9, the invisible window optimization process is executed:

[0050] The user's box selection instruction is received and the time interval coordinates are analyzed, the data segment is intercepted and the main frequency component is calculated, the optimization window is set and the data is reprocessed, and the regenerated statistical data is output;

[0051] When the feedback signal value received by the adaptive signal acquisition module is lower than 0.9, the confidence evaluation factor distribution map is analyzed, the frequency interval with a confidence less than 0.8 is identified, and the weight allocation proportion of the frequency interval in the frequency energy distribution parameter is reduced.

[0052] Further, the accelerometer vibration rectification error test system provided by the application further comprises:

[0053] The input of the bidirectional collaborative transmission module is the digital signal sequence with feature markers output by the adaptive signal acquisition module;

[0054] The input of the composite modal statistical module is the data stream with time stamp verification marks output by the bidirectional collaborative transmission module;

[0055] The input of the cross-validation error modeling module is the statistical data with algorithm selection identification output by the composite modal statistical module.

[0056] Secondly, referring to Figure 1 The application provides an accelerometer vibration rectification error test method, which is applied to the accelerometer vibration rectification error test system and comprises the following steps:

[0057] Step 1: Acquire the current signal output by the accelerometer and perform analog-to-digital conversion to output the original digital signal. Extract the frequency domain energy distribution parameters of the original digital signal, generate feature tags based on the frequency domain energy distribution parameters, attach the feature tags to the digital signal sequence, and output the digital signal sequence with feature tags.

[0058] Step 2: Receive the digital signal sequence output in Step 1, parse the feature markers in the digital signal sequence, dynamically calculate the buffer depth adjustment amount based on the frequency domain energy distribution parameter value carried in the feature markers, set the ring buffer depth as the buffer depth adjustment amount, synchronously detect the state transition edge of the feature markers, when the transition edge appears, add a timestamp verification mark to the header of the corresponding data frame, and output the data stream with integrated timestamp verification mark to Step 3.

[0059] Step 3: Receive the data stream output from Step 2, parse the timestamp verification identifier in the data stream, detect the missing ratio of timestamp verification identifiers, and when the missing ratio exceeds the preset threshold, select the wavelet packet time-frequency fusion algorithm from the preset algorithm library, align the signal physical period segmentation statistical window, generate statistical data based on the signal analysis results, and the statistical data shall include at least the average time-frequency entropy and be marked with the corresponding algorithm selection identifier.

[0060] Step 4: Receive the statistical data output in Step 3, extract the algorithm selection identifier from the statistical data, and use range dynamic weighted calculation when the algorithm selection identifier indicates that the wavelet packet time-frequency fusion algorithm is used. When the algorithm selection identifier indicates that the preset statistical algorithm of the non-wavelet packet time-frequency fusion algorithm is used, fuzzy cognitive mapping calculation is used. The fused statistical data outputs the vibration rectification error quantification model and confidence evaluation factor distribution map.

[0061] Beneficial effects of this invention;

[0062] This invention addresses the disconnect between data acquisition and analysis by dynamically adjusting the buffer depth and injecting timestamp verification identifiers through a feature-marked bidirectional collaborative transmission module. This overcomes the limitations of fixed sampling modes and avoids the omission of high-frequency transient features. The composite modal statistics module selects a wavelet packet time-frequency fusion algorithm based on the proportion of missing timestamps, aligns the physical period segmentation window to extract the mean time-frequency entropy statistics, and improves the efficiency of dynamic data segmentation statistics and the accuracy of transient feature extraction. The cross-validation error modeling module fuses statistics and vibration parameters to generate a vibration rectification error quantification model. The confidence assessment factor distribution map triggers dual-path feedback, including hidden window optimization to improve data quality in key periods and frequency domain weight redistribution to optimize feature extraction strategies. This forms a closed-loop evolutionary mechanism for hardware acquisition and software analysis. Standardized data stream encapsulation maintains the integrity of feature transmission and eliminates information attenuation from multiple tool transfers, ultimately improving the efficiency of real-time dynamic data processing and enhancing the accuracy of key transient response feature extraction. Attached Figure Description

[0063] In order to more clearly illustrate the technical solutions of the present application, the following will briefly introduce the drawings needed to be used in the embodiments. Obviously, for those skilled in the art, other drawings can also be obtained from the drawings without any creative labor.

[0064] Figure 1 A flowchart of an accelerometer vibration rectification error test method provided by the embodiment of the present application. DETAILED DESCRIPTION

[0065] In order to make the purpose, technical solutions and advantages of the present application more clear, the following will combine the specific embodiments of the present application and the corresponding drawings to clearly and completely describe the technical solutions of the present application. Obviously, the described embodiments are only one modular embodiment of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without any creative labor are within the protection scope of the present application. The following will combine the drawings to specifically describe the technical solutions provided by the embodiments of the present application. In order to better understand the purpose of the present application, the following will further describe the present application in detail.

[0066] In the first aspect, the present application provides an accelerometer vibration rectification error test system, comprising:

[0067] The adaptive signal acquisition module acquires the current signal output by the accelerometer and performs analog-to-digital conversion, outputs the original digital signal, extracts the frequency energy distribution parameter of the original digital signal, generates a feature tag based on the frequency energy distribution parameter, appends the feature tag to the digital signal sequence, and outputs the digital signal sequence with the feature tag;

[0068] The bidirectional cooperative transmission module receives the digital signal sequence output by the adaptive signal acquisition module, analyzes the feature tag in the digital signal sequence, dynamically calculates the buffer depth adjustment amount according to the frequency energy distribution parameter value carried in the feature tag, sets the ring buffer depth to the buffer depth adjustment amount, synchronously detects the state jump edge of the feature tag, appends a timestamp verification mark to the corresponding data frame header when the jump edge appears, and outputs the data stream integrated with the timestamp verification mark to the composite modal statistical module;

[0069] The composite modal statistical module receives the data stream output by the bidirectional cooperative transmission module, analyzes the timestamp verification mark in the data stream, detects the missing proportion of the timestamp verification mark, selects a wavelet packet time-frequency fusion algorithm from a preset algorithm library when the missing proportion exceeds a preset threshold, aligns the signal physical period to divide the statistical window, generates statistical data based on the signal analysis result, and the statistical data at least includes the time-frequency entropy mean value and the corresponding algorithm selection mark is labeled;

[0070] The cross-validation error modeling module receives the statistical data output by the composite modal statistical module, extracts the algorithm selection identifier in the statistical data, adopts a range dynamic weighting calculation when the algorithm selection identifier indicates the wavelet packet time-frequency fusion algorithm, adopts a fuzzy cognitive mapping calculation when the algorithm selection identifier indicates the preset statistical algorithm of the non-wavelet packet time-frequency fusion algorithm, and fuses the statistical data to output a vibration rectification error quantization model and a confidence evaluation factor distribution map.

[0071] The adaptive signal acquisition module first acquires the current signal output by the accelerometer, and performs an analog-to-digital conversion operation to convert the analog current signal into a raw digital signal. The conversion process uses a multi-core parallel processing architecture to improve signal acquisition efficiency. After conversion, the module performs frequency domain energy analysis on the raw digital signal, and calculates the energy distribution parameters of the signal in a specific frequency band through fast Fourier transform. The frequency domain energy distribution parameters reflect the core frequency spectrum characteristics of the signal. Based on the frequency domain energy distribution parameters, a digital feature marker is generated, and the feature marker code carries the abstract information of the signal frequency spectrum characteristics. The feature marker is attached to the tail of the digital signal sequence to form a digital signal sequence output with a feature marker. This step realizes the preliminary integration of the hardware acquisition layer and feature analysis.

[0072] The bidirectional collaborative transmission module receives the digital signal sequence with the feature marker, and analyzes the content of the feature marker embedded in the digital signal sequence. The frequency domain energy distribution parameter values in the feature marker are used to calculate the buffer depth adjustment amount. The calculation process is based on the product relationship of the energy proportion value and the preset scaling factor added to the initial depth value to dynamically determine the storage capacity of the ring buffer. After setting the depth of the ring buffer, the module detects the state jump event of the feature marker. The state jump event triggers the time stamp check identifier addition operation, and the time stamp check identifier is embedded in the corresponding data frame header. The time stamp check identifier provides a microsecond-level timing reference. Finally, the data stream integrated with the time stamp check identifier is output to the downstream module. This process establishes a feature-driven transmission mechanism to ensure the integrity of high-frequency transient signals.

[0073] The composite modal statistical module receives the data stream with the time stamp check identifier, and analyzes the sequence of time stamp check identifiers in the data stream. The number of lost time stamp check identifiers per second is counted, which reflects the data transmission reliability. When the number of lost time stamp check identifiers exceeds the set threshold, the wavelet packet time-frequency fusion algorithm is called from the preset algorithm library. The algorithm selection is associated with the data transmission quality. The module aligns the signal physical period to divide the statistical window, and the window boundary matches the inherent period of the signal. Wavelet packet decomposition is performed to a fixed number of layers, and the decomposition generates multi-level sub-band frequency domain components. The energy entropy of each sub-band is calculated and aggregated to output the time-frequency entropy mean value. The time-frequency entropy mean value is taken as the core statistical quantity and included in the statistical data. The statistical data are output after being labeled with the algorithm selection identifier. This step solves the problem of dynamic signal segmentation and statistical efficiency.

[0074] The cross-validation error modeling module receives statistical data labeled with algorithm selection identifiers. The algorithm selection identifiers in the statistical data are extracted, and the algorithm selection identifiers declare the upstream processing logic. When the algorithm selection identifiers indicate a wavelet packet time-frequency fusion algorithm, a dynamic range weighting calculation is performed. When the algorithm selection identifiers indicate other statistical algorithms, a fuzzy cognitive mapping calculation is performed. The module receives external vibration frequency and amplitude parameters. The fusion algorithm calculation results and the vibration parameters generate a vibration rectification error quantization model. The quantization model represents the mapping relationship between the error and the vibration parameters. Based on the quantization model, a confidence evaluation factor is calculated, and a two-dimensional confidence evaluation factor distribution map is generated. The distribution map identifies the spatial distribution of model reliability.

[0075] The invisible window optimization technology is explained. The composite modal statistical module receives user box selection time interval instructions, and analyzes the instructions to obtain time interval start point and end point coordinates. The corresponding time interval data segment is intercepted from the data stream. The main frequency component at the start point is extracted, and the component calculation method is consistent with the adaptive signal acquisition module. The physical extension length is calculated based on the main frequency component, and the extension length adapts to the signal period characteristics. The optimization statistical window boundary is set, and the boundary range covers an integer multiple of the signal period. The time stamp sequence is reanalyzed, the missing proportion is detected, the statistical algorithm is selected, and the wavelet packet decomposition is performed in the optimization window. The statistical data are calculated and labeled with algorithm identifiers. This mechanism improves the analysis accuracy of the key period.

[0076] The double-path feedback control technology is explained. The cross-validation error modeling module calculates the spatial mean of the confidence evaluation factor distribution map and outputs the average confidence evaluation factor feedback signal. The feedback signal is transmitted to the composite modal statistical module and the adaptive signal acquisition module. When the feedback signal value received by the composite modal statistical module is lower than the threshold value, the invisible window optimization process is triggered to regenerate the statistical data. When the feedback signal value received by the adaptive signal acquisition module is lower than the threshold value, the confidence evaluation factor distribution map is analyzed, and the low confidence frequency interval is identified. The weight allocation proportion of the low confidence frequency interval in the frequency energy distribution parameter is reduced. This closed-loop system realizes the co-evolution of hardware sampling and software analysis.

[0077] The system-level data stream technology is explained. The adaptive signal acquisition module outputs a digital signal sequence with feature markers to the bidirectional collaborative transmission module. The bidirectional collaborative transmission module outputs a data stream with timestamp verification identifiers to the composite modal statistical module. The composite modal statistical module outputs statistical data labeled with algorithm selection identifiers to the cross-validation error modeling module. The data stream adopts a three-level standardized packaging format: the feature marker carries the signal spectrum characteristics, the timestamp identifier ensures time synchronization, and the algorithm identifier declares the processing method. The structured data stream maintains the integrity of feature transmission and eliminates information attenuation during cross-module storage.

[0078] The application performs signal acquisition and feature label generation by step 1; step 2 realizes feature-driven transmission control; step 3 completes dynamic signal statistics and algorithm selection; step 4 carries out error modeling and confidence evaluation. The feature-labeled digital signal sequence output by step 1 is input to step 2, the time-stamped data stream output by step 2 is input to step 3, and the labeled identification statistical data output by step 3 is input to step 4. The four-level processing forms a "acquisition, transmission, statistics, modeling" technical chain, and the input and output of each step is strictly closed loop.

[0079] Specifically, the accelerometer vibration rectification error test system disclosed by the application comprises a multi-core analog-to-digital conversion array in the adaptive signal acquisition module;

[0080] The multi-core analog-to-digital conversion array inputs the current signal output by the accelerometer and outputs the original digital signal after analog-to-digital conversion;

[0081] The first 10ms time window of the original digital signal is extracted, and the energy proportion of the signal in the 10-500Hz frequency band in the first 10ms time window is calculated;

[0082] When the energy proportion exceeds 20%, a sampling rate improvement instruction is output to the multi-core analog-to-digital conversion array, a feature label is generated based on the energy proportion, and the feature label is transmitted to the bidirectional collaborative transmission module.

[0083] The adaptive signal acquisition module processes the current signal output by the accelerometer through the multi-core analog-to-digital conversion array. The multi-core analog-to-digital conversion array adopts a parallel conversion channel architecture, and each conversion core independently performs analog-to-digital conversion. The conversion process applies oversampling technology to suppress quantization noise and outputs high-fidelity original digital signals. The multi-core in the array works collaboratively to improve the signal acquisition throughput efficiency.

[0084] The module intercepts the first fixed time length data segment of the original digital signal as an analysis window. The intercepted data segment is subjected to frequency domain transformation calculation, and the energy distribution parameters of the signal in the target frequency band are extracted by using the fast Fourier transform algorithm. The energy distribution parameters quantitatively reflect the signal energy concentration degree in the target frequency band and represent the core frequency spectrum characteristics of the current vibration environment.

[0085] The dynamic sampling strategy is triggered based on the analysis result of the energy distribution parameters. When the energy proportion in the target frequency band exceeds the set threshold, a sampling rate improvement instruction is generated and sent to the multi-core analog-to-digital conversion array. After receiving the instruction, the array switches the sampling frequency configuration to enhance the high-frequency signal component capture capability. The dynamic sampling mechanism avoids the transient feature omission problem under the fixed sampling mode.

[0086] The frequency domain energy distribution parameter is digitally coded to generate a feature marker, and the feature marker data structure includes energy proportion numerical value and frequency band identification information.

[0087] The multi-core analog-to-digital conversion array converts the analog current signal into a digital signal to provide an input basis for frequency domain analysis.

[0088] The fast Fourier transform processing converts the time domain signal into a frequency domain energy distribution, and the target frequency band energy proportion calculation focuses on the vibration sensitive frequency band and filters out the interference of non-key noise frequency band.

[0089] The energy proportion threshold comparison result directly drives the sampling rate switching.

[0090] The energy distribution parameter coding generates a feature marker to convert the physical feature into a machine processable digital identifier.

[0091] Specifically, the accelerometer vibration rectification error test system disclosed by the application is characterized in that the bidirectional collaborative transmission module is configured to:

[0092] Receive the digital signal sequence with the feature marker output by the adaptive signal acquisition module, and analyze the feature marker in the digital signal sequence;

[0093] Based on the energy proportion numerical value in the feature marker and the preset scaling factor, the buffer depth adjustment amount is calculated, and the buffer depth adjustment amount is the initial depth value plus the product of the energy proportion numerical value and the preset scaling factor;

[0094] Set the ring buffer depth to the buffer depth adjustment amount;

[0095] Detect the state jump edge of the feature marker, when the state jump edge appears, add a timestamp verification identifier to the head of the data frame, and output the data stream integrated with the timestamp verification identifier to the composite modal statistical module.

[0096] The bidirectional cooperative transmission module receives the digital signal sequence with feature markers output by the adaptive signal acquisition module. The module analyzes the content of the feature markers embedded in the digital signal sequence, and the feature markers include frequency energy distribution parameter values. The analysis process extracts the energy proportion value in the feature markers, which reflects the energy concentration degree of the signal in the target frequency band.

[0097] Based on the extracted energy proportion value and the preset scaling factor, the module calculates the buffer depth adjustment amount. The calculation process uses a product superposition mechanism, and the buffer depth adjustment amount is the product of the initial depth value and the product of the energy proportion value and the preset scaling factor. The calculation logic associates signal feature strength with storage resource demand, and dynamically adapts to high-frequency transient signal processing demand.

[0098] The module applies the calculated buffer depth adjustment amount to the ring buffer depth configuration. The ring buffer uses a head-to-tail storage structure, and the depth adjustment amount determines the number of data frames that the buffer can accommodate. The configuration process updates the buffer storage space in real time, so that the buffer depth matches the strength of the signal feature change. The buffer depth adjustment eliminates the resource redundancy or overflow risk of fixed buffer configuration.

[0099] The module continuously monitors the state change of the feature markers, and identifies the step mutation of the feature marker value through the jump edge detection circuit. When the feature marker state has a jump event, the module embeds a timestamp verification identifier in the header of the corresponding data frame. The timestamp verification identifier includes accurate clock information, which is used to mark the absolute time point of the signal feature mutation. The timestamp addition position is located in the data frame header to ensure the traceability of the identifier.

[0100] After completing the addition of the timestamp verification identifier, the module integrates the original digital signal sequence, the feature markers, and the timestamp verification identifier into a structured data stream. The structured data stream is output to the composite modal statistical module through a high-speed transmission channel. The output process maintains the signal timing integrity and the feature marker correlation, and establishes a timing alignment basis for downstream statistical analysis.

[0101] The feature marker analysis provides an energy proportion value input, which is a core parameter for subsequent calculation. The analysis action establishes a direct association between upstream features and transmission control.

[0102] The product of the energy proportion value and the preset scaling factor determines the adjustment amount. The product relationship reflects the influence degree of signal feature strength on resource demand. Superimposing the initial depth value realizes gradual adjustment of resource configuration.

[0103] The ring buffer depth setting directly applies the calculation result of the previous step. The head-to-tail characteristic of the ring structure guarantees data continuity. The depth adjustment mechanism breaks through the limitations of fixed buffer configuration.

[0104] State jump along detects the moment of feature mutation. Timestamp identification binds absolute time reference, solving multi-device timing synchronization problem. Adding position selection data frame header to achieve fast positioning.

[0105] Original signal, feature mark and timestamp identification three elements are integrated. Structured packaging maintains feature transmission integrity. High-speed transmission channel minimizes transmission delay.

[0106] Specifically, the accelerometer vibration rectification error test system is configured with a composite modal statistical module.

[0107] The data stream output by the bidirectional collaborative transmission module is received, and the timestamp verification identification sequence is parsed from the data stream;

[0108] The number of lost per second of the timestamp verification identification sequence is counted;

[0109] When the number of lost exceeds 5%, a wavelet packet time-frequency fusion algorithm is selected from a preset algorithm library;

[0110] The received data stream is aligned with the signal physical cycle to divide a statistical window;

[0111] Wavelet packet decomposition is performed to the 5th layer, and the energy entropy of all subbands obtained by decomposition is calculated;

[0112] The time-frequency entropy mean value is calculated based on the energy entropy of all subbands, and statistical data including the time-frequency entropy mean value is generated; a wavelet packet time-frequency fusion algorithm selection identification is marked, and the statistical data marked with the wavelet packet time-frequency fusion algorithm selection identification is output to a cross-validation error modeling module.

[0113] The composite modal statistical module receives the data stream output by the bidirectional collaborative transmission module, and the data stream includes vibration signal data with timestamp verification identification. The module first parses the timestamp verification identification sequence from the data stream, and the timestamp verification identification carries accurate time information of signal acquisition. The parsing process identifies and extracts the content of the timestamp verification identification, and establishes a timestamp sequence database for subsequent integrity analysis.

[0114] The module performs integrity detection on the timestamp verification identification sequence, and counts the number of lost per second of the timestamp verification identification. The number of lost reflects the loss degree of timing synchronization signal in the data transmission process. The counting adopts a sliding time window counting mechanism to quantify the data transmission reliability index. This step generates a data transmission quality evaluation parameter.

[0115] Based on the statistical result of the number of lost, the module performs statistical algorithm dynamic selection. When the number of lost exceeds a preset threshold, a wavelet packet time-frequency fusion algorithm is called from a preset algorithm library. The algorithm selection mechanism associates data transmission quality with analysis method, and enables a high-robustness algorithm in a high-loss-rate scenario to improve analysis stability. The preset algorithm library includes various noise suppression algorithms.

[0116] The module performs a physical cycle segmentation window alignment operation on the received data stream. The alignment process identifies the signal inherent cycle boundary through zero-crossing detection, adjusts the start and end positions of the statistical window, and matches the window length to an integer multiple of the signal physical cycle. The synchronization operation eliminates cycle truncation errors and ensures the accuracy of subsequent feature calculation. After window alignment, the data is processed within a complete cycle.

[0117] Wavelet packet decomposition is performed to a fixed decomposition level, and the decomposition process converts the time-domain signal into multi-level sub-band frequency-domain components. Wavelet packet decomposition uses a tree structure for frequency band division, and each sub-band corresponds to a specific frequency range signal component. Fixed decomposition depth ensures consistency in feature extraction and balances the complexity of calculation and resolution requirements.

[0118] The energy entropy feature of each sub-band component is calculated, and the energy entropy represents the complexity of the energy distribution of the sub-band signal. Energy entropy calculation is based on the energy probability distribution of the sub-band signal, which quantifies the uncertainty of the frequency domain component. The energy entropy calculation results of all sub-bands are aggregated to output the time-frequency entropy mean value representing the global time-frequency feature. The time-frequency entropy mean value integrates the joint characteristics of the signal time-frequency domain.

[0119] A statistical data set including the time-frequency entropy mean value is generated, and the data set integrates the time-frequency entropy mean value and the original signal summary information. The statistical data is labeled with a wavelet packet time-frequency fusion algorithm selection identifier, and the identifier clearly records the type of algorithm used. The labeled structured data is output to the downstream modeling module through a high-speed interface.

[0120] Specifically, the accelerometer vibration rectification error test system described in the application is further configured as:

[0121] Based on the time-frequency entropy mean value, the signal summary hash stored in the composite modal statistical module, and the statistical window boundary coordinates, a statistical data object including a time-frequency entropy mean value field is created;

[0122] An algorithm selection identifier is labeled on the statistical data object, and the statistical data object with the labeled algorithm selection identifier is output to the cross-validation error modeling module.

[0123] The composite modal statistical module creates a structured statistical data object based on the time-frequency entropy mean value data, the signal summary hash value stored in the module, and the statistical window boundary coordinates. The time-frequency entropy mean value data is derived from the aggregation result of wavelet packet decomposition calculation and represents the global time-frequency characteristics of the signal. The signal summary hash value is generated by performing a hash algorithm on the original digital signal and provides a basis for data integrity verification. The statistical window boundary coordinates record the start and end time points of the analysis window and declare the time range of feature calculation. The three elements integrate to build a core data structure including a time-frequency entropy mean value field.

[0124] After the statistical data object is created, the module marks the algorithm selection identifier in the object metadata area. The algorithm selection identifier declares the specific statistical algorithm type adopted by the data processing process, and clearly records the application state of the wavelet packet time-frequency fusion algorithm or other preset algorithms. The marking operation permanently binds the algorithm selection identifier with the statistical data object, providing processing method traceability basis for downstream modules.

[0125] After the marking is completed, the module outputs the statistical data object of the marked algorithm selection identifier to the cross-validation error modeling module. The output process uses a structured data transmission protocol, and the protocol header includes a data type identifier and a check code. The data structure maintains the correlation of the time-frequency entropy mean value, the signal digest hash value, the window boundary coordinates, and the algorithm selection identifier, avoiding information attenuation in the data transfer process.

[0126] Specifically, the accelerometer vibration rectification error test system according to the present application is configured as follows:

[0127] Receiving the statistical data object of the marked algorithm selection identifier output by the composite modal statistical module as a statistical data set;

[0128] Extracting the algorithm selection identifier from the statistical data set;

[0129] When the algorithm selection identifier indicates the use of the wavelet packet time-frequency fusion algorithm, inputting the statistical data set into the range dynamic weighting algorithm for calculation;

[0130] When the algorithm selection identifier indicates the use of a preset statistical algorithm other than the wavelet packet time-frequency fusion algorithm, inputting the statistical data set into the fuzzy cognitive mapping algorithm for calculation;

[0131] Receiving the vibration frequency and amplitude parameters input from the outside;

[0132] Fusing the calculation results of the range dynamic weighting algorithm or the fuzzy cognitive mapping algorithm with the vibration frequency and amplitude parameters to generate a vibration rectification error quantification model;

[0133] Calculating a confidence evaluation factor based on the vibration rectification error quantification model and outputting a confidence evaluation factor distribution graph.

[0134] The cross-validation error modeling module receives the statistical data object of the marked algorithm selection identifier output by the composite modal statistical module, which is input as a statistical data set. The module first parses the algorithm selection identifier content in the statistical data object, and the algorithm selection identifier clearly indicates the specific algorithm type adopted by the upstream statistical processing. The identification parsing process distinguishes between the application scenarios of the wavelet packet time-frequency fusion algorithm and other preset statistical algorithms.

[0135] When the algorithm selection identifier indicates that the wavelet packet time-frequency fusion algorithm is used, the module inputs the statistical data set into the range dynamic weighting algorithm to perform calculation. The range dynamic weighting algorithm calculates the range of data based on the time-frequency entropy mean feature, and dynamically allocates a weight coefficient to enhance the contribution degree of high-frequency transient features. The calculation process focuses on the discrete degree analysis of the time-frequency characteristics of the signal.

[0136] When the algorithm selection identifier indicates that the preset statistical algorithm of the non-wavelet packet time-frequency fusion algorithm is used, the module inputs the statistical data set into the fuzzy cognitive mapping algorithm to perform calculation. The fuzzy cognitive mapping algorithm constructs a nonlinear relationship network between the input features and the output results, and processes data uncertainty through fuzzy rule reasoning. The calculation process is suitable for low-quality data scenarios.

[0137] The module receives the vibration frequency and amplitude parameters input by the external test equipment. The vibration frequency parameter describes the base frequency characteristics of the excitation signal, and the amplitude parameter quantifies the vibration environment intensity. The parameter input interface supports real-time environment monitoring data access, realizing the matching of the model and the physical working condition.

[0138] The fusion process multi-dimensionally couples the algorithm calculation results and the vibration frequency and amplitude parameters. The coupling operation maps heterogeneous data to a unified feature space through matrix transformation, and establishes an association matrix of signal features and vibration parameters. The fusion output generates a vibration rectification error quantization model, which represents the mapping relationship between the accelerometer output error and the vibration parameters in the form of a function.

[0139] Based on the vibration rectification error quantization model, the module calculates a confidence evaluation factor. The factor value reflects the reliability level of the model in different feature regions, and the calculation process combines residual analysis and feature coverage density evaluation. The output confidence evaluation factor distribution map uses a heat map to mark the boundaries of high-confidence and low-confidence regions, providing a visual criterion for the reliability of the test results.

[0140] The statistical data object marked with the algorithm selection identifier is taken as the core input, and the algorithm selection identifier determines the subsequent calculation path. The content of the identifier directly determines the selection of the statistical algorithm. The range dynamic weighting algorithm enhances the contribution of high-frequency features, and the fuzzy cognitive mapping algorithm handles data uncertainty. The algorithm branches match the upstream data processing methods. The external vibration parameters provide a physical working condition benchmark, the frequency parameters define the excitation characteristics, and the amplitude parameters quantify the environment intensity. The parameter input maintains the synchronization of the model and the actual vibration environment. The algorithm output results and the vibration parameters are coupled through matrix transformation to construct a unified feature space. The fusion process eliminates the dimensional differences to generate a quantitative mathematical model. Based on the model residual and feature coverage, a confidence factor is calculated, and a spatial distribution heat map is used to identify the reliability regions of the model. The visual output supports the decision-making of the test conclusion.

[0141] Specifically, the accelerometer vibration rectification error test system described in the present application is further configured to:

[0142] receiving a user box selection time interval instruction, parsing the user box selection time interval instruction, obtaining time interval start point and end point coordinates;

[0143] cutting the data segment between the start point and the end point from the data stream output by the bidirectional collaborative transmission module;

[0144] using the frequency energy distribution parameter extraction method of the adaptive signal acquisition module to calculate the frequency domain main frequency component of the data at the time interval start point of the cut data segment;

[0145] calculating the physical extension length based on the frequency domain main frequency component, the physical extension length being equal to an integer N multiplied by 1 divided by the main frequency component, N being an integer;

[0146] setting an optimized statistical window boundary, the start point of the optimized statistical window boundary being the time interval start point, and the end point being the time interval start point plus the physical extension length;

[0147] reanalyzing the timestamp verification mark sequence within the optimized statistical window boundary, and detecting the missing proportion of the timestamp verification mark;

[0148] when the missing proportion exceeds a preset threshold, selecting a wavelet packet time-frequency fusion algorithm from a preset algorithm library, aligning the physical period division of the data stream within the optimized statistical window, performing wavelet packet decomposition to the 5th layer, and calculating the energy entropy of all subbands obtained by decomposition;

[0149] calculating the time-frequency entropy mean value based on the energy entropy of all subbands, generating statistical data including the time-frequency entropy mean value, and marking the algorithm selection identification.

[0150] The composite modal statistical module receives the box selection time interval instruction input by the user through the graphical interface. The instruction analysis process maps the user's box selection screen coordinates to the start point and end point timestamps in the system time coordinate system. The coordinate conversion is based on the data stream time axis calibration parameters to realize the accurate correspondence of the user's intention to the signal time dimension. The time interval start point and end point coordinates provide the positioning reference for the key period.

[0151] The module cuts the data segment between the start point and the end point from the data stream output by the bidirectional collaborative transmission module. The cutting operation positions the data stream according to the timestamp coordinates to extract the complete signal data block within the target time interval. The data segment includes the original vibration signal and its associated timestamp verification mark sequence, maintaining the integrity of the original data structure. The cutting process inherits the time sequence characteristics and identification association of the data stream.

[0152] The frequency domain main frequency component calculation is performed on the neighborhood signal at the starting point of the time interval of the intercepted data segment. The calculation method reuses the core algorithm process of the adaptive signal acquisition module: the fast Fourier transform is used to analyze the frequency spectrum of the neighborhood signal at the starting point, and the frequency component with the highest energy concentration degree is identified as the main frequency component. The feature extraction method reuse ensures the consistency of the frequency domain analysis of the whole system.

[0153] The physical extension length is calculated based on the frequency domain main frequency component. The physical extension length is composed of an integer multiple of the signal period length, and the calculation formula reflects the signal physical period characteristics: the physical extension length is equal to the integer N multiplied by 1 divided by the main frequency component. The calculation process adapts to the signal fundamental frequency characteristics, so that the extension range matches the signal physical oscillation law.

[0154] The optimization statistical window boundary is set, the starting point of the boundary remains unchanged as the starting point of the user's selected time interval, and the ending point extends to the starting point plus the physical extension length. The window boundary reconstruction eliminates the problem of incomplete period caused by artificial truncation. The optimized window covers the original area of interest and the physical extension area, enhancing the physical rationality of feature calculation.

[0155] The timestamp verification identification sequence is reanalyzed within the optimization statistical window boundary. The analysis operation detects the continuity and integrity of the timestamp identification within the window, and establishes a timestamp sequence database. The timestamp sequence analysis provides input basis for subsequent integrity detection.

[0156] The missing proportion of timestamp verification identification is detected, and the percentage of missing identification quantity in the total expected identification quantity is calculated. The missing proportion calculation uses a sliding time window counting mechanism to quantify the data transmission reliability index within the optimization window. This parameter reflects the data quality status in the key period.

[0157] When the missing proportion exceeds the preset threshold, the wavelet packet time-frequency fusion algorithm is selected from the preset algorithm library. The algorithm selection mechanism preferentially calls high-robustness processing methods to adapt to possible noise interference in the key period. The threshold comparison result directly determines the calculation algorithm type.

[0158] The physical period segmentation of the data stream within the optimization statistical window is aligned. The segmentation operation identifies the inherent period boundary of the signal through zero-crossing detection, and adjusts the data segment boundary within the window to match the complete signal period. The period alignment eliminates truncation errors and ensures the accuracy of subsequent feature calculation.

[0159] The wavelet packet decomposition is performed to the fifth layer, and the decomposition process converts the time domain signal within the optimization window into thirty-two sub-band frequency domain components. Fixed layer decomposition balances feature resolution and calculation efficiency, and sub-band division covers the full frequency spectrum range of the target frequency band.

[0160] The energy entropy features of all subbands obtained by calculation are decomposed, and the energy entropy quantifies the complexity of the energy distribution of each subband signal.

[0161] The time-frequency entropy mean value is calculated based on the energy entropy feature vectors of all subbands, and the mean value operation aggregates the subband features into global statistics. The time-frequency entropy mean value represents the time-frequency joint distribution characteristics of the signal within the optimization window.

[0162] Statistical data including the time-frequency entropy mean value is generated, and the data structure integrates the time-frequency entropy mean value, the signal digest hash value, and the optimization window boundary coordinates. The statistical data is labeled with the wavelet packet time-frequency fusion algorithm selection identifier, and the feature generation method is declared.

[0163] Specifically, the accelerometer vibration rectification error test system also includes:

[0164] After the cross-validation error modeling module generates the confidence evaluation factor distribution map, the average confidence evaluation factor of the distribution map is calculated, and the average confidence evaluation factor feedback signal is output to the composite modal statistical module and the adaptive signal acquisition module;

[0165] When the feedback signal value received by the composite modal statistical module is less than 0.9, the invisible window optimization process is executed:

[0166] Receive the user's box selection instruction and parse the time interval coordinates, intercept the data segment and calculate the main frequency component, set the optimization window and reprocess the data, and output the regenerated statistical data;

[0167] When the feedback signal value received by the adaptive signal acquisition module is less than 0.9, the confidence evaluation factor distribution map is analyzed, the frequency interval with a confidence less than 0.8 is identified, and the weight allocation proportion of the frequency interval in the frequency energy distribution parameter is reduced.

[0168] After the cross-validation error modeling module generates the confidence evaluation factor distribution map, the spatial mean of the distribution map is calculated to generate the average confidence evaluation factor. The average confidence evaluation factor is transmitted synchronously to the composite modal statistical module and the adaptive signal acquisition module as a feedback signal through the internal bus of the system. The feedback signal value quantitatively represents the global reliability level of the vibration rectification error quantification model.

[0169] The composite modal statistical module monitors the received feedback signal value in real time. When the feedback signal value is lower than the set threshold, the invisible window optimization process is triggered. The trigger mechanism automatically activates the optimization program based on the numerical comparison result. The invisible window optimization process includes four consecutive operations: receiving the user's box selection instruction and parsing the time interval coordinates, intercepting the target data segment and calculating the main frequency component, setting the physical expansion optimization window, and re-executing the complete statistical processing flow. The process outputs the regenerated statistical data, enhancing the quality of the key period features.

[0170] The adaptive signal acquisition module receives the average confidence evaluation factor feedback signal in parallel. The module analyzes the feedback signal value and extracts the confidence evaluation factor distribution graph data structure. The distribution graph space data is analyzed to identify frequency intervals with confidence below a certain standard. For the identified low-confidence frequency intervals, the weight allocation proportion in the frequency energy distribution parameter is reduced. The weight adjustment process dynamically optimizes the feature extraction strategy, focusing on high-confidence frequency band signal features.

[0171] The confidence evaluation factor distribution graph space mean value calculation provides a comprehensive evaluation of model reliability. Feedback signal values below the threshold indicate insufficient system reliability, triggering the optimization mechanism. Bus transmission ensures synchronized signal distribution.

[0172] The low-confidence state activates the invisible window optimization process, and the user's marquee instruction guides the analysis focus. The physical expansion window enhances the physical rationality of the features, and the data reprocessing improves the input quality. The confidence evaluation factor distribution graph analysis identifies low-confidence frequency intervals, and the weight allocation proportion adjustment reduces the influence of interference frequency bands. Dynamic optimization of the feature extraction strategy makes subsequent acquisition more focused on key features.

[0173] The statistical module optimizes the data quality problem, and the acquisition module optimizes the feature generation accuracy problem. The dual-path cooperation forms a closed loop to enhance the system, and improves the subsequent modeling confidence level.

[0174] Specifically, the accelerometer vibration rectification error test system described in the application further comprises:

[0175] The input of the bidirectional cooperative transmission module is the digital signal sequence with feature markers output by the adaptive signal acquisition module;

[0176] The input of the composite modal statistical module is the data stream with timestamp verification marks output by the bidirectional cooperative transmission module;

[0177] The input of the cross-validation error modeling module is the statistical data with algorithm selection marks output by the composite modal statistical module.

[0178] The adaptive signal acquisition module outputs the digital signal sequence with feature markers to the bidirectional cooperative transmission module. The digital signal sequence with feature markers includes original digital signal blocks and feature marker blocks, and the feature marker blocks encode frequency energy distribution parameters. The output sequence uses differential signal transmission protocol, and the protocol frame header includes data type identification and length verification code to ensure signal transmission anti-interference. This output serves as the input basis for the bidirectional cooperative transmission module.

[0179] The bidirectional cooperative transmission module receives the input, performs feature marker analysis and timestamp injection operation, and outputs a data stream with timestamp verification mark to the composite modal statistical module. The data stream with timestamp verification mark adopts a packaging format binding timestamp mark and signal frame. The timestamp mark is located in the data frame header and accurately records the system clock count value when the feature marker jumps. The data stream is transmitted through a direct memory access channel, maintaining a millisecond-level transmission delay. This output constitutes the input source of the composite modal statistical module.

[0180] The composite modal statistical module receives the data stream with timestamp verification mark, performs timestamp integrity detection and statistical analysis process, and outputs statistical data with algorithm selection mark to the cross-validation error modeling module. The statistical data with algorithm selection mark adopts a triple structure: time-frequency entropy mean core feature, signal digest hash value, and algorithm selection mark. The triple data is transmitted through a high-speed serial bus after protocol packaging. The packaging header includes version mark and cyclic redundancy check code. This output serves as the input basis for the cross-validation error modeling module.

[0181] In a second aspect, referring to Figure 1 The accelerometer vibration rectification error test method of the application is applied to the accelerometer vibration rectification error test system and includes the following steps.

[0182] Step 1: Obtain the current signal output by the accelerometer and perform analog-to-digital conversion to output a raw digital signal. Extract the frequency energy distribution parameter of the raw digital signal, generate a feature marker based on the frequency energy distribution parameter, attach the feature marker to the digital signal sequence, and output the digital signal sequence with the feature marker.

[0183] Step 2: Receive the digital signal sequence output by step 1, analyze the feature marker in the digital signal sequence, dynamically calculate the buffer depth adjustment amount according to the frequency energy distribution parameter value carried in the feature marker, set the ring buffer depth to the buffer depth adjustment amount, and simultaneously detect the state jump edge of the feature marker. When the jump edge appears, a timestamp verification mark is attached to the corresponding data frame header. The data stream with the integrated timestamp verification mark is output to step 3.

[0184] Step 3: Receive the data stream output by step 2, analyze the timestamp verification mark in the data stream, detect the missing proportion of the timestamp verification mark, and when the missing proportion exceeds a preset threshold, select a wavelet packet time-frequency fusion algorithm from a preset algorithm library, align the signal physical period, divide the statistical window, generate statistical data based on the signal analysis result, and the statistical data at least includes time-frequency entropy mean and the corresponding algorithm selection mark.

[0185] Step 4, receive the statistical data output in step 3, extract the algorithm selection identifier in the statistical data, when the algorithm selection identifier indicates the wavelet packet time-frequency fusion algorithm, use the range dynamic weighting calculation, when the algorithm selection identifier indicates the preset statistical algorithm of the non-wavelet packet time-frequency fusion algorithm, use the fuzzy cognitive mapping calculation, fuse the statistical data to output the vibration rectification error quantization model and the confidence evaluation factor distribution map.

[0186] Step 1, obtain the current signal output by the accelerometer through the multi-core analog-to-digital conversion array, perform parallel analog-to-digital conversion and output the original digital signal. A fixed time length analysis window is cut off from the original digital signal, and the energy distribution parameters of the signal in the target frequency band within the window are calculated by using the fast Fourier transform. The energy distribution parameters represent the core spectral characteristics of the signal. Based on the energy distribution parameters, a digital feature marker is generated, and the feature marker code contains energy proportion values and frequency band identification information. The feature marker is attached to the tail of the digital signal sequence to form a digital signal sequence with feature markers output. This step realizes the conversion of physical signals to digital features, providing a feature-driven basis for downstream collaborative transmission.

[0187] Step 2, receive the digital signal sequence with feature markers, and analyze the feature marker content in the sequence. According to the energy proportion value in the feature marker and the preset scaling factor, calculate the buffer depth adjustment amount. The depth adjustment amount is obtained by superimposing the initial depth value, the product of the energy proportion value and the scaling factor. Set the ring buffer depth to the calculated adjustment amount, dynamically match the signal feature intensity. Synchronously detect the feature marker state jump edge, and when the jump edge appears, embed a timestamp verification identifier in the corresponding data frame header. The timestamp identifier records the microsecond-level system clock count value of the feature mutation. Output the data stream integrated with the timestamp identifier to the downstream statistical module to establish the transmission timing reference.

[0188] Step 3, receive the data stream with timestamp verification identifiers, and analyze the timestamp sequence in the data stream. Count the number of lost timestamp identifiers per second, which reflects the data transmission integrity. When the number of lost identifiers exceeds the set threshold, call the wavelet packet time-frequency fusion algorithm from the preset algorithm library. Perform physical cycle segmentation window alignment operation on the data stream, and match the inherent cycle boundary of the signal through zero-crossing detection. Perform wavelet packet decomposition to the fifth layer to generate multi-level sub-band frequency domain components. Calculate the energy entropy features of each sub-band, aggregate and output the time-frequency entropy mean global statistics. Generate statistical data containing time-frequency entropy mean, and label the used algorithm selection identifier.

[0189] Step 4, receive statistical data marked by algorithm selection identifier. When the algorithm identifier indicates wavelet packet time-frequency fusion algorithm, use range dynamic weighting calculation to process data; when indicating other preset algorithms, use fuzzy cognitive mapping calculation. Receive externally input vibration frequency and amplitude parameters, vibration frequency defines excitation fundamental frequency characteristics, and amplitude quantifies environmental strength. Fusion algorithm calculation result and vibration parameters generate vibration rectification error quantization model, and model establishes function mapping relationship of error and vibration parameters. Based on model residual error analysis, calculate confidence evaluation factor, output confidence evaluation factor distribution map, and distribution map marks model reliability space distribution in the form of heat map.

[0190] The adaptive signal acquisition module acquires the accelerometer current signal through the multi-core analog-to-digital conversion array, extracts the frequency energy distribution parameters of the front time window signal in the target frequency band, generates a feature marker based on the frequency energy distribution parameters, and the feature marker carries a signal spectrum feature abstract. The feature marker is attached to the digital signal sequence output, realizing real-time transmission of the hardware acquisition layer feature to the transmission layer. The feature marker drives the bidirectional collaborative transmission module to dynamically calculate the buffer depth adjustment amount, and the ring buffer depth dynamically changes with the frequency energy distribution parameter value. The feature marker state jump triggers the timestamp verification identifier injection, and the microsecond-level clock identifier is embedded in the data frame header. The hardware feature marker directly controls the transmission layer resource configuration and timing synchronization, breaking through the limitation of the fixed sampling transmission mode.

[0191] The bidirectional collaborative transmission module outputs the data stream with the timestamp verification identifier to the composite modal statistical module. The module analyzes the timestamp verification identifier sequence, and calculates the identification loss ratio per second. When the loss ratio exceeds the threshold, the wavelet packet time-frequency fusion algorithm is called from the preset algorithm library. The algorithm selection is associated with data transmission quality: high loss rate scene automatically enables high robustness algorithm, and low loss rate scene enables high precision algorithm. Align the signal physical period to divide the statistical window, and match the period boundary through zero-crossing detection. Perform wavelet packet decomposition to the fifth layer to calculate the sub-band energy entropy, and output the time-frequency entropy mean statistical quantity. The statistical data marked by the algorithm selection identifier declares the processing method, forming a dynamic adaptation mechanism of analysis strategy and data quality.

[0192] The statistical data marked by the algorithm selection identifier is input into the cross-validation error modeling module. The module selects range dynamic weighting or fuzzy cognitive mapping calculation according to the algorithm identifier, generates a vibration rectification error quantization model by fusing external vibration frequency and amplitude parameters. After outputting the confidence evaluation factor distribution map, calculate the average confidence evaluation factor as a feedback signal. When the value is lower than the threshold, trigger the double-path optimization: the composite modal statistical module performs the invisible window optimization, reconstructs the physical period window based on the user's selected interval to generate statistical data again; the adaptive signal acquisition module reduces the weight allocation proportion of the low-confidence frequency band, and optimizes the feature extraction strategy. The feedback closed loop realizes the co-evolution of hardware sampling and software analysis.

[0193] The system maintains feature integrity through standardized data flow: the adaptive signal acquisition module outputs a digital signal sequence with feature markers (original signal block + feature marker block), the bidirectional collaborative transmission module outputs a data stream with timestamp verification markers (signal frame bound with timestamp marker), and the composite modal statistical module outputs statistical data with algorithm selection markers (time-frequency entropy mean + algorithm marker). The encapsulation format eliminates information attenuation caused by multi-tool transfer, and ensures the cross-module association and transmission of feature markers, timing information and processing methods.

[0194] The application constructs a three-level feature transmission chain of hardware feature markers, transmission timing markers and algorithm selection markers, and realizes the depth collaboration of acquisition and analysis through feature-driven transmission configuration, timing quality adaptation analysis and double-path feedback optimization, so as to finally improve the real-time segmented statistical efficiency and strengthen the transient feature extraction accuracy.

[0195] The embodiment of the application is based on the accelerometer vibration rectification error test scene, and aims at the problem of low dynamic data statistical efficiency and inaccurate transient feature extraction caused by the functional fragmentation of data acquisition hardware and backend analysis software. The technical problem is solved through the following technical contents:

[0196] The adaptive signal acquisition module acquires the accelerometer current signal in real time through a multi-core analog-to-digital conversion array. The array synchronously processes multiple signals through parallel conversion channels, and outputs the original digital signal and then intercepts the front time window data segment. The data segment is subjected to fast Fourier transform to calculate the energy distribution parameters of the target frequency band, and a sampling rate improvement instruction is triggered when the energy proportion exceeds the set threshold. Based on the energy distribution parameters, a feature marker is generated and attached to the digital signal sequence. This process realizes real-time extraction of hardware layer features, and solves the problem of missing high-frequency transient features under the fixed sampling mode. The feature marker carries a spectral feature abstract to drive downstream collaborative optimization.

[0197] The bidirectional collaborative transmission module receives the digital signal sequence with feature markers, and analyzes the energy proportion value in the feature marker. Based on the value and a preset scaling factor, the buffer depth adjustment amount is dynamically calculated, and the ring buffer depth is set in real time. The feature marker state jump edge is detected synchronously, and when the jump occurs, a microsecond timestamp verification marker is embedded in the corresponding data frame header. The structured data stream is output to the composite modal statistical module, solving the timing misalignment problem caused by discrete devices. The transmission layer drives resource allocation through feature markers, avoiding data overflow in strong vibration scenes.

[0198] The composite modal statistical module analyzes the timestamp verification identification sequence in the data stream, and counts the identification loss ratio per second. When the loss ratio exceeds the threshold, the wavelet packet time-frequency fusion algorithm is called. The physical cycle segmentation window alignment is performed on the data stream, and the signal inherent cycle boundary is matched through zero-crossing detection. The five-layer wavelet packet decomposition is performed to calculate the sub-band energy entropy, and the time-frequency entropy mean statistical quantity is output. The statistical data labeling algorithm selects the identification declaration processing method. When the user selects the concerned period, the invisible window optimization is triggered: the starting point main frequency component is extracted, the physical window is expanded to an integer multiple of the period based on the formula N*1 / main frequency component, and the statistical process is re-executed in the window. This mechanism eliminates the artificial truncation error and improves the accuracy of the impact response analysis.

[0199] The cross-validation error modeling module selects the range dynamic weighting or fuzzy cognitive mapping algorithm according to the algorithm identification. The time-frequency entropy mean statistical quantity and the external vibration frequency and amplitude parameters are fused to generate a vibration rectification error quantitative model. After the confidence evaluation factor distribution map is output, the average confidence evaluation factor is fed back to the statistical and acquisition module: when the threshold is lower than the threshold, the statistical module performs invisible window optimization to regenerate data; the acquisition module analyzes the distribution map to identify the low-confidence frequency band, and reduces the weight distribution proportion of the corresponding frequency band. The double-path feedback forms a co-evolution closed loop of hardware sampling and software analysis.

[0200] The system transmits the spectral features through the digital signal sequence with feature markers, guarantees the time sequence synchronization through the data stream with timestamp verification identification, and selects the statistical data declaration processing method through the labeling algorithm identification. The structured packaging eliminates the information attenuation of multi-tool transfer, and realizes the end-to-end optimization from the current signal input to the error model output.

[0201] The technical features involved in the present application are explained as follows:

[0202] The frequency domain energy distribution parameter extraction method is that the adaptive signal acquisition module performs fast Fourier transform processing on the original digital signal, and calculates the energy concentration degree of the signal in a specific frequency band. The target frequency band range covers the core frequency spectrum interval of the accelerometer vibration response, and the energy distribution parameter represents the energy distribution characteristics of the signal in the frequency domain space. The frequency domain energy distribution parameter is generated by digital coding to generate a feature marker, and the feature marker is used as a machine-readable abstract of the signal spectrum feature.

[0203] The dynamic buffer depth adjustment strategy is that the bidirectional collaborative transmission module calculates the ring buffer depth adjustment amount based on the energy proportion value in the feature marker and the preset scaling factor. The buffer depth adjustment amount is the initial depth value plus the product of the energy proportion value and the scaling factor. The ring buffer depth is set to the calculated adjustment amount, so that the storage capacity can match the change of the signal feature intensity in real time.

[0204] Wavelet packet time-frequency fusion algorithm, composite modal statistical module calls the wavelet packet time-frequency fusion algorithm when the proportion of timestamp missing exceeds the threshold. The algorithm performs five-layer wavelet packet decomposition to decompose the time-domain signal into multi-level sub-band frequency components. Calculate the energy entropy feature of each sub-band, and the energy entropy quantifies the complexity of the sub-band signal energy distribution. Aggregate all sub-band energy entropy output time-frequency entropy mean, and the time-frequency entropy mean represents the global time-frequency characteristics of the signal.

[0205] Range dynamic weighting algorithm, cross-validation error modeling module performs range dynamic weighting calculation when the algorithm selection identifier indicates the wavelet packet time-frequency fusion algorithm. The algorithm calculates the data range based on the time-frequency entropy mean feature, and dynamically allocates weight coefficients to enhance the contribution of high-frequency transient features in the model. The weight allocation is positively correlated with the feature dispersion degree.

[0206] Fuzzy cognitive mapping algorithm, when the algorithm selection identifier indicates a non-wavelet packet time-frequency fusion algorithm, the modeling module performs fuzzy cognitive mapping calculation. The algorithm constructs a nonlinear relationship network between input features and output results, and processes data uncertainty through fuzzy rule-based reasoning. The cognitive mapping network adapts to the modeling needs of low-quality data scenarios.

[0207] Invisible window optimization strategy, the composite modal statistical module receives user box selection time interval instructions, extracts the frequency domain main frequency component of the data at the start point of the time interval. Based on the main frequency component, calculate the physical extension length, which is an integer multiple of the signal period length. Set the optimization statistical window boundary to cover the original interval and the physical extension area, and re-execute data analysis and feature calculation in the window.

[0208] Dual-path feedback control strategy, the average confidence evaluation factor of the confidence evaluation factor distribution map is used as the feedback signal distribution. When the composite modal statistical module receives a feedback signal below the threshold, it triggers the invisible window optimization to regenerate statistical data. The adaptive signal acquisition module analyzes the distribution map to identify low-confidence frequency intervals and reduces the weight allocation proportion of corresponding intervals in the frequency energy distribution parameters. The dual-path cooperative optimization improves the accuracy of hardware sampling and software analysis.

Claims

1. An accelerometer vibration rectification error testing system, characterized in that, include; The adaptive signal acquisition module acquires the current signal output by the accelerometer and performs analog-to-digital conversion, outputs the raw digital signal, extracts the frequency domain energy distribution parameters of the raw digital signal, generates feature tags based on the frequency domain energy distribution parameters, appends the feature tags to the digital signal sequence, and outputs the digital signal sequence with feature tags. The bidirectional collaborative transmission module receives the digital signal sequence output by the adaptive signal acquisition module, parses the feature markers in the digital signal sequence, dynamically calculates the buffer depth adjustment amount based on the frequency domain energy distribution parameter values ​​carried in the feature markers, sets the ring buffer depth as the buffer depth adjustment amount, synchronously detects the state transition edge of the feature markers, and appends a timestamp verification identifier to the header of the corresponding data frame when the transition edge occurs, and outputs the data stream integrating the timestamp verification identifier to the composite modality statistics module. The composite modal statistics module receives the data stream output by the bidirectional collaborative transmission module, parses the timestamp verification identifier in the data stream, detects the missing ratio of timestamp verification identifiers, and when the missing ratio exceeds a preset threshold, selects a wavelet packet time-frequency fusion algorithm from a preset algorithm library, aligns the signal physical period segmentation statistical window, generates statistical data based on the signal analysis results, and the statistical data includes at least the mean time-frequency entropy and is labeled with the corresponding algorithm selection identifier. The cross-validation error modeling module receives statistical data output from the composite modal statistics module, extracts the algorithm selection identifier from the statistical data, and uses dynamic weighted range calculation when the algorithm selection identifier indicates that the wavelet packet time-frequency fusion algorithm is used. When the algorithm selection identifier indicates that the preset statistical algorithm of the non-wavelet packet time-frequency fusion algorithm is used, fuzzy cognitive mapping calculation is used. The fused statistical data outputs a vibration rectification error quantification model and a confidence evaluation factor distribution map.

2. The accelerometer vibration rectification error testing system according to claim 1, characterized in that, The adaptive signal acquisition module includes a multi-core analog-to-digital converter array; The multi-core analog-to-digital converter array takes the current signal output from the accelerometer as input, performs analog-to-digital conversion, and outputs the original digital signal. Extract the first 10ms time window of the original digital signal and calculate the energy proportion of the signal in the 10-500Hz frequency band during the first 10ms time window. When the energy percentage exceeds 20%, the output sampling rate increase instruction is sent to the multi-core analog-to-digital converter array, a feature tag is generated based on the energy percentage, and the feature tag is transmitted to the bidirectional collaborative transmission module.

3. The accelerometer vibration rectification error testing system according to claim 2, characterized in that, The bidirectional collaborative transmission module is configured as follows: Receive the digital signal sequence with feature markers output by the adaptive signal acquisition module, and parse the feature markers in the digital signal sequence; Based on the energy percentage value in the feature marker and the preset scaling factor, the buffer depth adjustment amount is calculated. The buffer depth adjustment amount is the product of the initial depth value, the energy percentage value, and the preset scaling factor. Set the ring buffer depth to the buffer depth adjustment amount; The system detects state transition edges of feature markers. When a state transition edge occurs, a timestamp verification flag is appended to the header of the data frame, and the data stream integrating the timestamp verification flag is output to the composite modality statistics module.

4. The accelerometer vibration rectification error testing system according to claim 3, characterized in that, The composite modal statistics module is configured as follows: Receive the data stream output by the bidirectional collaborative transmission module and parse the timestamp check identifier sequence from the data stream; Count the number of timestamp verification identifier sequences lost per second; When the number of lost packets exceeds 5%, the wavelet packet time-frequency fusion algorithm is selected from the preset algorithm library; The received data stream is aligned with the physical periodic segmentation statistical window of the signal. Perform wavelet packet decomposition up to level 5, and calculate the energy entropy of all subbands obtained from the decomposition; Calculate the mean time-frequency entropy based on the energy entropy of all sub-bands, generate statistical data including the mean time-frequency entropy; label the wavelet packet time-frequency fusion algorithm selection identifier, and output the statistical data labeled with the wavelet packet time-frequency fusion algorithm selection identifier to the cross-validation error modeling module.

5. The accelerometer vibration rectification error testing system according to claim 4, characterized in that, The composite modal statistics module is also configured as follows: Based on the mean time-frequency entropy, the signal digest hash stored in the composite modality statistics module, and the coordinates of the statistical window boundary, create a statistical data object including the mean time-frequency entropy field; Label the statistical data objects with algorithm selection identifiers, and output the statistical data objects labeled with algorithm selection identifiers to the cross-validation error modeling module.

6. The accelerometer vibration rectification error testing system according to claim 5, characterized in that, The cross-validation error modeling module is configured as follows: The annotation algorithm output from the composite modality statistics module selects the identified statistical data objects as the statistical data set. Identifiers are selected from statistical datasets using algorithms. When the algorithm selection indicator indicates that the wavelet packet time-frequency fusion algorithm is used, the statistical data set is input into the range dynamic weighting algorithm to perform the calculation; When the algorithm selection indicator indicates the use of the preset statistical algorithm of the non-wavelet packet time-frequency fusion algorithm, the statistical data set is input into the fuzzy cognitive mapping algorithm to perform calculations; Receives external input vibration frequency and amplitude parameters; By integrating the calculation results of the range dynamic weighting algorithm or the fuzzy cognitive mapping algorithm with the vibration frequency and amplitude parameters, a vibration rectification error quantification model is generated. The confidence assessment factor is calculated based on the vibration rectification error quantification model, and the confidence assessment factor distribution map is output.

7. The accelerometer vibration rectification error testing system according to claim 6, characterized in that, The composite modal statistics module is also configured as follows: Receive the user's command to select a time interval, parse the user's command to select a time interval, and obtain the coordinates of the start and end points of the time interval; Extract the data segment from the start point to the end point from the data stream output by the bidirectional collaborative transmission module; For the data at the beginning of the time interval of the intercepted data segment, the frequency domain energy distribution parameter extraction method of the adaptive signal acquisition module is used to calculate the frequency domain dominant frequency component; The physical extension length is calculated based on the frequency domain main frequency component. The physical extension length is equal to an integer N multiplied by 1 and divided by the main frequency component, where N is an integer. Set the boundaries of the optimized statistics window. The starting point of the optimized statistics window boundary is the starting point of the time interval, and the ending point is the starting point of the time interval plus the physical extension length. Within the optimized statistical window boundary, the timestamp verification identifier sequence is re-parsed to detect the proportion of missing timestamp verification identifiers; When the missing proportion exceeds the preset threshold, the wavelet packet time-frequency fusion algorithm is selected from the preset algorithm library, the physical period segmentation of the data stream within the optimization statistical window is aligned, the wavelet packet decomposition is performed up to the 5th level, and the energy entropy of all sub-bands obtained by the decomposition is calculated. The mean time-frequency entropy is calculated based on the energy entropy of all sub-bands, and statistical data including the mean time-frequency entropy is generated, with the labeling algorithm selection identifier.

8. The accelerometer vibration rectification error testing system according to claim 7, characterized in that, Also includes: After the cross-validation error modeling module generates a confidence assessment factor distribution map, it calculates the average confidence assessment factor of the distribution map and outputs the average confidence assessment factor feedback signal to the composite modal statistics module and the adaptive signal acquisition module. When the feedback signal value received by the composite modal statistics module is lower than 0.9, the hidden window optimization process is executed: Receive user selection instructions and parse time interval coordinates, extract data segments and calculate main frequency components, set optimization windows and reprocess data, and output regenerated statistical data; When the feedback signal value received by the adaptive signal acquisition module is lower than 0.9, the confidence evaluation factor distribution map is analyzed to identify the frequency range with a confidence level of less than 0.8, and the weight allocation ratio of the frequency range in the frequency domain energy distribution parameter is reduced.

9. The accelerometer vibration rectification error testing system according to claim 8, characterized in that, Also includes: The input to the bidirectional cooperative transmission module is a sequence of digital signals with feature tags output by the adaptive signal acquisition module; The input to the composite modal statistics module is a data stream with timestamp verification marks output from the bidirectional collaborative transmission module; The input to the cross-validation error modeling module is the statistical data of the labeling algorithm selected by the composite modality statistics module.

10. A method for testing the vibration rectification error of an accelerometer, applied to the accelerometer vibration rectification error testing system as described in any one of claims 1 to 9, characterized in that, include: Step 1: Acquire the current signal output by the accelerometer and perform analog-to-digital conversion to output the original digital signal. Extract the frequency domain energy distribution parameters of the original digital signal, generate feature tags based on the frequency domain energy distribution parameters, attach the feature tags to the digital signal sequence, and output the digital signal sequence with feature tags. Step 2: Analyze the feature markers in the digital signal sequence, dynamically calculate the buffer depth adjustment amount based on the frequency domain energy distribution parameter values ​​carried in the feature markers, set the ring buffer depth as the buffer depth adjustment amount, synchronously detect the state transition edge of the feature markers, and when the transition edge appears, add a timestamp verification mark to the header of the corresponding data frame, and output the data stream with integrated timestamp verification mark to Step 3. Step 3: Parse the timestamp verification identifier in the data stream, detect the missing ratio of timestamp verification identifiers, and when the missing ratio exceeds the preset threshold, select the wavelet packet time-frequency fusion algorithm from the preset algorithm library, align the signal physical period segmentation statistical window, generate statistical data based on the signal analysis results, the statistical data shall include at least the mean time-frequency entropy, and mark the corresponding algorithm selection identifier. Step 4: Extract the algorithm selection identifier from the statistical data. When the algorithm selection identifier indicates that the wavelet packet time-frequency fusion algorithm is used, the range dynamic weighted calculation is used. When the algorithm selection identifier indicates that the preset statistical algorithm of the non-wavelet packet time-frequency fusion algorithm is used, the fuzzy cognitive mapping calculation is used. The fused statistical data outputs the vibration rectification error quantification model and the confidence evaluation factor distribution map.

Images