A data acquisition and verification system for a miniature data acquisition instrument

Abstract

This invention relates to the field of electronic digital data processing technology, specifically to a data acquisition and verification system for a miniature data acquisition instrument. It solves the technical problem that traditional verification methods are unable to adapt to dynamic changes in data characteristics, resulting in insufficient accuracy in data verification. The system includes: a dataset construction unit, a data verification unit, a dataset update unit, and a data correction unit; the dataset construction unit is used to determine a first dataset based on first vibration data; the data verification unit is used to determine abnormal data in the second vibration data based on the matching degree between the second vibration data and the first dataset; the dataset update unit is used to update the first dataset based on non-abnormal data in the second vibration data; the data correction unit is used to correct the matching degree judgment result between the third vibration data and the first dataset based on a system error compensation coefficient; this invention is used for data acquisition and verification scenarios in helicopter mechanical environment testing.

Description

Technical Field

[0001] This invention relates to the field of electronic digital data processing technology, and more specifically to a data acquisition and verification system for a miniature data acquisition instrument. Background Technology

[0002] With the rapid development of the Internet of Things (IoT) and intelligent sensing technologies, miniature data acquisition devices have been widely used in industrial equipment condition monitoring and structural health diagnosis due to their advantages of small size, low power consumption, and ease of integration and deployment. Particularly in helicopter mechanical testing, real-time and accurate monitoring of blade vibration is crucial for ensuring flight safety and extending equipment lifespan. This requires the supporting data acquisition system to operate stably for extended periods under space-constrained and harsh airborne conditions. Currently, in such scenarios, verification methods based on fixed thresholds or static historical data models are commonly used to identify anomalies in the acquired vibration data. However, the highly integrated sensors and signal conditioning circuits within miniature data acquisition devices are constantly subjected to vibration, temperature changes, and complex electromagnetic interference during actual helicopter flight. This causes the baseline of the acquired data to drift slowly over time, making traditional verification methods relying on fixed baselines unable to adapt to the dynamic changes in data characteristics, ultimately resulting in insufficient accuracy and long-term reliability of the data verification. Summary of the Invention

[0003] To address the technical problem that traditional verification methods relying on fixed benchmarks are ill-suited to adapt to dynamic changes in data characteristics, ultimately resulting in insufficient accuracy and long-term reliability in data verification, this invention aims to provide a data acquisition and verification system for a miniature data acquisition instrument. The specific technical solution adopted is as follows:

[0004] In a first aspect, the present invention provides a data acquisition and verification system for a miniature data acquisition instrument. The system includes: a dataset construction unit for determining a first dataset based on first vibration data; the first vibration data being all vibration data of a target blade within a preset time period; the preset time period being any historical time period; the first dataset including at least one first vibration data point; a data verification unit for determining abnormal data in the second vibration data based on the degree of matching between the second vibration data and the first dataset; the degree of matching between the abnormal data and the first dataset being less than a first preset threshold; the second vibration data being real-time vibration data of the target blade; a dataset update unit for updating the first dataset based on non-abnormal data in the second vibration data to obtain the second dataset; and a data correction unit for correcting the degree of matching between the third vibration data and the first dataset based on a system error compensation coefficient; the system error compensation coefficient being determined based on the second dataset.

[0005] In conjunction with the first aspect above, in one possible implementation, the first vibration data includes multiple data segments; the dataset construction unit includes: a signal decomposition subunit for performing signal decomposition on the first vibration data to determine multiple component signals; a reference signal determination subunit for determining the reference signal among the multiple component signals that has the highest similarity to the first vibration data; and a data filtering subunit for determining at least one target data segment from the first vibration data based on the reference signal; the at least one target data segment is used to constitute the first dataset.

[0006] In conjunction with the first aspect above, in one possible implementation, the reference signal determination subunit is specifically used to: determine the absolute value of the frequency difference between each component signal and the first vibration data among the multiple component signals; determine the similarity value between the multiple component signals and the first vibration data based on the absolute value of the frequency difference between each component signal and the first vibration data; and determine the component signal with the largest similarity value among the multiple component signals as the reference signal.

[0007] In conjunction with the first aspect mentioned above, in one possible implementation, the data filtering subunit is specifically used for: determining the degree of difference between each data segment and the time period data of the reference signal based on the fluctuation amplitude of each data segment; determining multiple clusters based on the degree of difference between each data segment and the time period data of the reference signal, and a preset clustering algorithm; determining the anomaly score of each cluster based on the principal component direction of the data segments in the multiple clusters; and determining the target cluster among the multiple clusters that meets the preset conditions; the preset conditions are: the anomaly score is less than a second preset threshold; and the data segments in the target cluster are the target data segments.

[0008] In conjunction with the first aspect mentioned above, in one possible implementation, the data verification unit includes: a correlation calculation subunit, used to calculate the correlation coefficient between the second vibration data and the target data segment; and a matching degree determination subunit, used to determine the matching degree based on the correlation coefficient.

[0009] In conjunction with the first aspect mentioned above, in one possible implementation, the correlation calculation subunit is specifically used to: calculate the Pearson correlation coefficient between the second vibration data and the target data segment.

[0010] In conjunction with the first aspect mentioned above, in one possible implementation, the data correction unit includes: an error coefficient calculation subunit, used to determine the system error compensation coefficient based on the statistical characteristic differences between the second dataset and the non-abnormal data; and a verification correction subunit, used to multiply the matching degree between the third vibration data and the first dataset by the system error compensation coefficient to determine the corrected matching degree.

[0011] In conjunction with the first aspect mentioned above, in one possible implementation, the error coefficient calculation subunit is specifically used to: calculate multiple first statistical feature values ​​for each data segment in the second dataset, and second statistical feature values ​​for non-abnormal data; the first statistical feature values ​​are used to characterize the dispersion and fluctuation intensity of each data segment; the second statistical feature values ​​are used to characterize the dispersion and fluctuation intensity of non-abnormal data; based on the differences between the multiple first statistical feature values ​​and the second statistical feature values, determine the average feature difference; and based on the normalization processing of the average feature difference using a preset normalization function, determine the system error compensation coefficient.

[0012] In conjunction with the first aspect mentioned above, in one possible implementation, the dataset update unit is specifically used to: add non-abnormal data to the storage queue of the first dataset; remove the oldest historical data from the storage queue, and determine the second dataset.

[0013] In conjunction with the first aspect mentioned above, in one possible implementation, the data acquisition and verification system of the micro data acquisition instrument is communicatively connected to the high-speed micro data acquisition instrument; the high-speed micro data acquisition instrument is used to acquire vibration data of the helicopter rotor blades.

[0014] The present invention has the following beneficial effects:

[0015] By constructing a closed-loop system of "dataset construction - data verification - dataset update - data correction," the fundamental problem of unstable verification benchmarks for micro data acquisition instruments in complex environments is solved. The core benefit lies in achieving dynamic adaptability in the data verification process: the system can not only establish an initial reliable benchmark (first dataset) based on historical data, but also continuously incorporate verified real-time data (non-abnormal data) into the benchmark during operation (updating to the second dataset). This allows the verification benchmark to evolve with the slow changes in system state, avoiding benchmark failure caused by environmental temperature drift, time drift, and other factors. Furthermore, by introducing a system error compensation coefficient and correcting subsequent data judgments, the system can proactively compensate for systematic deviations caused by long-term operation, significantly improving the accuracy and reliability of data verification in long-term monitoring tasks. This solves the technical problem that existing methods relying on fixed benchmarks are unable to adapt to dynamic changes in data characteristics, resulting in insufficient accuracy and long-term reliability in data verification. Attached Figure Description

[0016] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of the data acquisition and verification system architecture of a miniature data acquisition instrument provided in one embodiment of the present invention;

[0018] Figure 2 This is a schematic diagram of the architecture of a dataset construction unit provided in one embodiment of the present invention;

[0019] Figure 3 This is a schematic diagram of a process for a reference signal determination subunit to determine the reference signal with the highest similarity to first vibration data among multiple component signals, according to an embodiment of the present invention.

[0020] Figure 4 This is a schematic diagram of a data filtering subunit based on a reference signal to determine at least one target data segment from first vibration data, as provided in an embodiment of the present invention.

[0021] Figure 5 This is a schematic diagram of the architecture of a data verification unit provided in one embodiment of the present invention;

[0022] Figure 6 This is a schematic diagram of the architecture of a data correction unit provided in one embodiment of the present invention;

[0023] Figure 7 This is a flowchart illustrating a process for determining system error compensation coefficients based on the statistical characteristic differences between a second dataset and non-abnormal data, provided in an embodiment of the present invention.

[0024] Figure 8 This is a schematic diagram illustrating the process of a dataset update unit determining a second dataset based on non-abnormal data, as provided in an embodiment of the present invention. Detailed Implementation

[0025] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a data acquisition and verification system for a miniature data acquisition instrument proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0026] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0027] The specific solution of the data acquisition and verification system for a miniature data acquisition instrument provided by the present invention will be described in detail below with reference to the accompanying drawings.

[0028] Please see Figure 1 It illustrates a data acquisition and verification system architecture diagram of a miniature data acquisition instrument provided in an embodiment of the present invention, such as... Figure 1 As shown, the system includes:

[0029] Data set construction unit 101 is used to determine the first data set based on the first vibration data.

[0030] The first vibration data is all the vibration data of the target blade within a preset time period; the preset time period is any historical time period.

[0031] In one possible implementation, the dataset construction unit 101 converts long-term, continuous historical vibration signals into a set of clean, reliable benchmark data to generate the first dataset.

[0032] For example, the dataset construction unit 101 performs signal decomposition on the first data, decomposing it into a series of intrinsic mode components from high frequency to low frequency. By calculating the comprehensive similarity between each component and the original signal in terms of frequency and waveform shape, the component that best represents the core periodic characteristics of the signal is selected as the reference signal. Then, based on the reference signal, the first data is divided into multiple data segments using the sliding window technique. The difference between each data segment and the corresponding part of the reference signal is calculated. Combined with the preset clustering algorithm and principal component analysis technique, high-quality data segments with consistent patterns and low interference are selected. These segments together constitute the initial reference dataset for subsequent verification and comparison.

[0033] The data verification unit 102 is used to determine abnormal data in the second vibration data based on the degree of matching between the second vibration data and the first dataset.

[0034] Among them, the abnormal data matches the first dataset to a degree less than the first preset threshold; the second vibration data is the real-time vibration data of the target blade.

[0035] In one possible implementation, the data verification unit 102 performs anomaly detection on the real-time acquired second vibration data based on the first dataset, and performs quantitative evaluation from two dimensions: matching frequency and degree of anomaly, to determine the abnormal data in the second vibration data.

[0036] Understandably, normal data should have a high degree of similarity to the reference dataset in terms of trend. When the overall shape of the second vibration data is significantly different from the average shape of the first dataset, it can be identified as an abnormal data segment.

[0037] The dataset update unit 103 is used to update the first dataset based on the non-abnormal data in the second vibration data to obtain the non-abnormal data of the second dataset.

[0038] Among them, the non-abnormal data refers to the non-abnormal data in the second vibration data; the second dataset is the first dataset updated based on the non-abnormal data.

[0039] In one possible implementation, the dataset update unit 103 dynamically maintains and incrementally learns the first dataset based on the anomaly detection results of the data verification unit 102 to ensure its timeliness.

[0040] The data correction unit 104 is used to correct the matching degree judgment result between the third vibration data and the first dataset based on the system error compensation coefficient.

[0041] The system error compensation coefficient is determined based on the second dataset.

[0042] In one possible implementation, the data correction unit 104 calculates the system error compensation coefficient based on the updated second dataset and the latest third vibration data, and calibrates the subsequent verification judgment according to the error compensation coefficient to improve the robustness of the system in long-term operation.

[0043] Based on the above technical solutions, this application establishes a complete verification process from benchmark construction to dynamic correction, effectively addressing the benchmark inaccuracy problem caused by temperature drift, time drift, and electromagnetic interference in complex environments for micro data acquisition instruments. The innovation of this system lies in its dynamic adaptive verification mechanism: it can not only construct a high-quality initial reference benchmark, i.e., the first dataset, based on the first vibration data, but also continuously incorporate qualified non-abnormal data into the first dataset during operation. This allows the verification benchmark to autonomously evolve with changes in the state of the monitored object and environmental factors, avoiding the failure of traditional fixed benchmarks over time. The data acquisition verification system for micro data acquisition instruments also introduces an error compensation mechanism to correct the matching results of subsequent data in real time, effectively suppressing the accumulated system deviations during long-term operation. This significantly improves the accuracy and reliability of data verification in continuous monitoring tasks, overcoming the technical limitations of traditional fixed threshold methods that are difficult to adapt to dynamic data characteristics.

[0044] Non-abnormal data such as Figure 2 As shown, in one possible implementation, the dataset construction unit 101 includes: a signal decomposition subunit 1011, used to decompose the first vibration data into signals and determine multiple component signals; a reference signal determination subunit 1012, used to determine the reference signal with the highest similarity to the first vibration data among the multiple component signals; and a data filtering subunit 1013, used to determine at least one target data segment from the first vibration data based on the reference signal; the at least one target data segment is used to constitute the first dataset.

[0045] In one possible implementation, the signal decomposition subunit 1011 uses a signal decomposition algorithm to process the first vibration data, decomposing it into a series of eigenmode components with single frequency components, thereby achieving signal separation in different frequency bands.

[0046] In one possible implementation, the reference signal determination subunit 1012 determines the comprehensive similarity between each component signal and the first vibration data in the frequency domain and time domain. By evaluating a combination of the absolute value of the frequency difference and the value of the time domain morphological difference, the component with the highest comprehensive similarity is selected as the reference signal representing the core features of the signal.

[0047] In one possible implementation, the data filtering subunit 1013 uses the reference signal as a benchmark, divides the first vibration data into continuous data segments using a sliding window method, calculates the statistical feature difference between each data segment and the corresponding time period of the reference signal, performs pattern classification on the data segments using a preset clustering algorithm, and finally selects the target data segments that are consistent with the pattern of the reference signal and have the smallest difference to form the first dataset.

[0048] Based on the above technical solution, this application, through a specific defined dataset construction unit comprising three sub-units—signal decomposition, reference signal determination, and data filtering—signally improves the quality of the first dataset. By extracting core features from the first vibration data through signal decomposition, and then constructing the first dataset by filtering data segments consistent with the core features (reference signal), this method effectively eliminates abnormal segments in the first vibration data caused by accidental interference or transient sensor failures. This ensures that the initial benchmark is pure and highly reliable, laying a solid foundation for subsequent real-time verification and improving the confidence level of the entire verification system from the source.

[0049] like Figure 3 As shown, in one possible implementation, the process by which the reference signal determination subunit 1012 in this application determines the reference signal with the highest similarity to the first vibration data among multiple component signals can be specifically implemented through the following steps:

[0050] S301. Determine the absolute value of the frequency difference between each component signal and the first vibration data in the multiple component signals.

[0051] One possible implementation involves using a frequency consistency comparison analysis strategy, which quantifies the consistency level of the component signals in terms of vibration frequency by calculating the absolute difference between the approximate frequency of each component signal and the approximate frequency of the historical vibration data segment.

[0052] For example, the preset signal decomposition algorithm is the Empirical Mode Decomposition (EMD) algorithm.

[0053] Understandably, the absolute value of the frequency difference can effectively capture the numerical deviation of two signals at their core vibration frequency, while the measure of frequency consistency reflects whether the component signal inherits the main periodic characteristics of the historical data segment. The smaller the difference, the closer the component signal is to the first vibration data in terms of vibration frequency.

[0054] When the absolute value of the frequency difference of a certain component signal is significantly greater than that of other components, it indicates that its vibration frequency is fundamentally different from that of the main data component, and the probability of this component signal being selected as the reference signal decreases accordingly. Through the above process, this embodiment provides an accurate frequency domain comparison basis for subsequent comprehensive evaluation of signal similarity based on frequency and shape differences.

[0055] S302. Based on the absolute value of the frequency difference between each component signal and the first vibration data, determine the similarity value between the multiple component signals and the first vibration data.

[0056] For example, determine the first The absolute value of the frequency difference between the component signal and the first vibration data is , No. Similarity value between the component signal and the first vibration data Satisfy the following formula 1:

[0057] Formula 1

[0058] in, Indicates the first Signal and first vibration data The dynamic time warping (DTW) value is the value of the first vibration data. The smaller the value, the more similar it is to the waveform of the first vibration data. Indicates the first The absolute value of the frequency difference between the signal and the first vibration data. This represents an exponential function with the natural constant as its base. Its function is to map a large, non-negative input value to an output value in the interval (0,1]. It is a linear normalization function, which maps the calculated original difference values ​​to a fixed interval (such as [0, 1]) to facilitate subsequent uniform threshold processing and cluster analysis. The normalized result of the absolute value of the frequency difference is obtained by normalizing the maximum and minimum values ​​to map T′ to the interval [0,1], thereby eliminating the dimension.

[0059] Understandably, the core of the formula is calculation. Differences in morphology The product of these factors takes into account the differences in both the frequency and time domains of the signal. This comprehensive difference is only small when both the frequency and shape are similar. Taking the negative of the comprehensive difference and then applying it through an exponential function... The transformation is performed. The greater the overall difference, the smaller the similarity value obtained after the transformation with a negative sign and an exponential function.

[0060] Understandably, the similarity value calculated by this formula is a dimensionless value between 0 and 1, which comprehensively quantifies the similarity of the first... This formula measures the overall similarity of a component signal to the original historical data in both frequency and morphology dimensions. Using this formula, a reference signal can be determined from multiple decomposed components from an objective and quantitative perspective. This avoids potential misjudgments that might arise from relying solely on a single feature (such as frequency or morphology), laying a solid foundation for building a reliable verification benchmark.

[0061] S303. Determine the component signal with the highest similarity value among multiple component signals as the reference signal.

[0062] In one possible implementation, after calculating the similarity value between each component signal and the first vibration data, a maximum value selection strategy is adopted to determine the reference signal with the largest similarity value from all candidate component signals.

[0063] Understandably, the maximum value selection strategy can effectively lock onto the component that has the highest overall similarity to the original historical data in both frequency and waveform morphology dimensions. This component is considered to inherit the main vibration mode and core variation law contained in the first vibration data to the greatest extent, while filtering out noise or interference components that are inconsistent with the main characteristics.

[0064] Once a component signal is identified as the reference signal, it serves as a clean and reliable benchmark template for subsequent, more refined anomaly screening of the historical data segments. Through this process, this embodiment ensures the representativeness and accuracy of the verification benchmark, providing core support for building a highly reliable data acquisition and verification system.

[0065] The above provides a detailed explanation of the process by which the reference signal determination subunit 1012 determines the reference signal among multiple component signals that has the highest similarity to the first vibration data.

[0066] like Figure 4 As shown, in one possible implementation, the process by which the data filtering subunit 1013 in this application determines at least one target data segment from the first vibration data based on a reference signal can be specifically implemented through the following steps:

[0067] S401. Determine the difference between each data segment and the time period data of the reference signal based on the fluctuation amplitude of each data segment.

[0068] In one possible implementation, the statistical characteristics of each data sub-segment in the first vibration data segment are calculated segment by segment, and the differences are fused with the statistical characteristics of the corresponding time period of the reference signal to determine the degree of difference between each data sub-segment and the time period data of the reference signal.

[0069] Each data segment in the first vibration data segment includes at least two statistical characteristics: fluctuation characteristics (variance) and average level (arithmetic mean). The fluctuation characteristics characterize the intensity of the fluctuation of the data segment around its average level, while the average level characterizes the center position of the signal amplitude of the data segment. By comparing the differences between the data segment and the corresponding time period of the reference signal in these two statistical characteristics, the consistency can be comprehensively evaluated from two dimensions: fluctuation amplitude and benchmark level.

[0070] For example, taking data point j in the first vibration data as the center, four data points are taken before and after it in time sequence to form a data segment containing nine data points. The difference between this data segment and the time period data of the reference signal satisfies the following formula 2:

[0071] Formula 2

[0072] in, This represents the degree of difference between the j-th data segment and the reference signal. This value is a normalized, dimensionless scalar; a larger value indicates a more significant deviation between the data segment and the reference template. and These respectively represent the fluctuation amplitude and average level of the j-th data segment; and These represent the fluctuation amplitude and average level of the reference signal in the corresponding time period, respectively; the absolute value is taken to ensure that the difference is non-negative. It is a linear normalized function. A dimensionless, normalized fluctuation difference in the range [0, 1] was determined. A dimensionless, normalized mean difference in the range [0, 1] was determined, thereby transforming the two incomparable features of variance and mean to a unified relative scale.

[0073] Understandably, the calculation of the dissimilarity score comprehensively considers the deviation of a data segment from the reference benchmark in terms of two key statistical characteristics: "volatility" and "average level." This degree of deviation directly reflects whether the data segment contains anomalies. The greater the dissimilarity score, the higher the probability of anomalies in the data segment, and the lower its reference value when constructing a cleanliness verification benchmark. Through the above process, this embodiment transforms data anomalies that are difficult to determine directly into quantifiable dissimilarity scores, providing accurate input for subsequent clustering and anomaly identification.

[0074] S402. Based on the difference between each data segment and the time period data of the reference signal, and a preset clustering algorithm, determine multiple clusters.

[0075] One possible implementation involves using an unsupervised clustering analysis strategy, which divides all data segments into different groups based on their dissimilarity characteristics, in order to initially separate potentially abnormal and normal data.

[0076] For example, the preset clustering algorithm is a hierarchical clustering algorithm. The difference degree of each data segment calculated in step S401 is used as the feature value of the data segment. The preset number of clustering layers is 2, that is, the goal is to divide all data segments into two clusters.

[0077] Understandably, clustering can automatically group data segments with similar degrees of anomaly based on the distribution characteristics of the data itself, without the need to pre-set an absolute anomaly threshold. This provides a more robust method for distinguishing normal fluctuations from true anomalies in complex vibration signals, avoiding misjudgments caused by improper setting of a single threshold.

[0078] S403. Based on the principal component directions of data segments in multiple clusters, determine the anomaly score for each cluster in the multiple clusters.

[0079] Among them, the anomaly score is used to characterize the differences in principal component directions among multiple clusters.

[0080] In one possible implementation, an analysis strategy based on principal component analysis and inter-cluster comparison is used to determine the final anomaly score by analyzing the differences in the main direction of the data distribution of different clusters and their average anomaly levels.

[0081] For example, principal component analysis is performed on a data segment in any one of the multiple clusters to determine the direction of its first principal component. The main characteristic direction of the data distribution in this cluster is used as the basis for calculating the average difference of all data segments within this cluster. Then the first The final outlier values ​​of each cluster satisfy the following formula 3:

[0082] Formula 3

[0083] in, No. Anomaly score for each cluster, which comprehensively reflects the overall difference between this cluster and all other clusters in terms of data distribution direction and average anomaly level; and They represent the first Principal component orientations and average dissimilarity of each cluster; and They represent the first Principal component orientations and average dissimilarity of each cluster; n represents the total number of data segments; This is a linear normalization function, which scales the calculation result to a standard range [0,1]. Used for normalizing the principal component orientation. This is used to normalize the average variability, thereby eliminating the dimensional differences between the principal component orientation and the average variability.

[0084] It is understandable that the summation and averaging in Formula 3 are... Sum of the absolute difference values ​​and divide by , obtained the The average comprehensive deviation of a cluster relative to all clusters; the larger the average comprehensive deviation, the more significant the deviation. This cluster differs from most clusters in both data patterns and the degree of anomaly, therefore its final anomaly score is... The larger; The first one was measured The first cluster and the first The greater the difference in the main change patterns among the clusters, the more unique the pattern and the more likely it is to be abnormal. The first one was measured The first cluster and the first Differences in average anomaly severity among the clusters; It is a normalized evaluation value that measures the "isolation" or "atypicality" of a particular cluster relative to all clusters in the entire dataset.

[0085] Understandably, anomaly scoring considers not only the average anomaly level within a cluster but also the differences in the main direction of data distribution. This allows the scoring to capture hidden anomalies that, while having low average variability, exhibit a different direction of data change compared to the normal group.

[0086] S404. Determine the target cluster that meets the preset conditions among multiple clusters.

[0087] Among them, the data sub-segments in the target cluster are the target data sub-segments.

[0088] One possible implementation involves using a threshold-based decision-making strategy to select target data segments from all clusters that can be used to construct the first dataset based on the calculated anomaly score.

[0089] In one possible implementation, after the first dataset is constructed, the total number of target data segments is... If ≥1, and no target data segment is found, reselect the first vibration data (expand the preset historical time period or adjust the filtering parameters) until... ≥1.

[0090] For example, the preset condition is: the anomaly score of the cluster is less than the second preset threshold of 0.4. All clusters are traversed, and their anomaly scores are compared with 0.4; clusters that meet the condition of anomaly score less than 0.4 are determined as target clusters; target clusters represent data groups whose data distribution pattern is consistent with the mainstream and whose average anomaly level is low, and all data segments contained in these target clusters are determined as target data segments.

[0091] Understandably, threshold determination is a crucial step in ultimately achieving historical data purification. Based on the comprehensive anomaly score calculated in the preceding steps, it explicitly excludes anomalous data clusters from the reference benchmark. The target data segments thus selected constitute a clean and reliable initial reference dataset, laying a solid foundation for the accurate verification of subsequent real-time data and effectively solving the circular dependency problem of "the verification benchmark itself also needing verification."

[0092] The above provides a detailed description of the process by which the data filtering subunit 1013 in this embodiment determines at least one target data segment from the first vibration data based on a reference signal.

[0093] like Figure 5 As shown, in one possible implementation, the data verification unit 102 includes: a correlation calculation subunit 1021, used to calculate the correlation coefficient between the second vibration data and the target data segment; and a matching degree determination subunit 1022, used to determine the matching degree based on the correlation coefficient.

[0094] In one possible implementation, the correlation calculation subunit 1021 determines the degree of similarity between the second vibration data and the target data segment by calculating the similarity coefficient between the second vibration data and each target data segment in the first dataset.

[0095] For example, the second vibration data (i.e., the real-time vibration data segment) is divided into multiple real-time data segments to be verified according to a fixed duration (e.g., 5 seconds). For each real-time data segment, the correlation coefficient between it and each target data sub-segment in the first dataset is calculated.

[0096] In one possible implementation, the correlation coefficient between each real-time data segment in the second vibration data and each target data segment in the first dataset is calculated using the Pearson correlation coefficient calculation formula. It is understood that the Pearson correlation coefficient is an existing correlation calculation method, which will not be elaborated here.

[0097] For example, the first The degree of matching of the second vibration data segment Satisfy the following formula 4

[0098] Formula 4

[0099] in, For the first The degree of matching of each real-time data segment; the greater the degree of matching, the more normal the real-time data segment is. Indicates the first The first real-time data segment and the first The Pearson correlation coefficient between the reference data segments. Its value ranges from [-1, 1]. A larger Pearson correlation coefficient indicates a stronger linear correlation. The total number of target data segments, and ≥1 (as explained in step S404, and will not be repeated here).

[0100] Understandably, taking the average value in Formula 4 comprehensively considers the overall similarity between the real-time data segment and the entire reference dataset, avoiding misjudgments caused by accidental similarity or dissimilarity with a single reference segment, resulting in more accurate results. The larger the correlation coefficient, the more consistent the changing trend of the second vibration data with that of a certain target data segment, and the greater its contribution to the average value. Therefore, the degree of matching is positively correlated with the correlation coefficient.

[0101] Understandably, the matching degree, by averaging the results of comparisons between the second vibration data and multiple target data segments, enhances the robustness of determining normal states. Even if individual target data segments do not match the current data well due to accidental factors, as long as the current data shows a high correlation with most target data segments, the final matching degree will still be high. This effectively avoids the risk of misjudgment due to a single reference data point, making the conclusion more reliable. This matching degree will be used to compare with a first preset threshold to ultimately determine whether the real-time data is normal.

[0102] For example, the first preset threshold is 0.55. When the matching degree of the second vibration data is greater than 0.55, it indicates that the second vibration data is not abnormal. It should be noted that the first preset threshold of 0.55 here is only an exemplary value. In practical applications, this threshold should be calibrated based on experience or experimental data.

[0103] like Figure 6 As shown, in one possible implementation, the data correction unit 104 includes: an error coefficient calculation subunit 1041, used to determine the system error compensation coefficient based on the statistical characteristic differences between the second dataset and the non-abnormal data; and a verification correction subunit 1042, used to multiply the matching degree between the third vibration data and the first dataset by the system error compensation coefficient to determine the corrected matching degree.

[0104] In one possible implementation, a scalar value, namely the system error compensation coefficient, is calculated by comparing and analyzing the statistical characteristics of all data segments contained in the second dataset (i.e., the dynamically updated first dataset) with the statistical characteristics of the newly added non-abnormal data (i.e., the second vibration data segment judged to be normal).

[0105] Understandably, the second dataset represents a dynamically evolving "normal state" benchmark library as the system operates. The newly added non-abnormal data consists of samples deemed "normal" by the system in its latest state. By calculating the group differences between these two datasets in key statistical characteristics (such as the variance representing data volatility and the mean representing the average data level), the systemic change trends caused by factors such as equipment heating, mechanical wear, and cumulative environmental effects can be effectively captured. The system error compensation coefficient is a quantified output of this systemic change trend. This coefficient will be used by the subsequent verification and correction subunit to calibrate the matching degree in real time, thereby making the threshold benchmark for anomaly judgment adaptive and avoiding the problem of increased false positive rates due to slow system drift. Through this process, this embodiment ensures the robustness and reliability of the data acquisition verification system of the micro data acquisition instrument during long-term continuous operation.

[0106] In one possible implementation, the verification correction subunit 1042 is used to receive two input parameters: and ;in, Indicates the first The degree of matching between the third vibration data and the first dataset; Indicates the first The system error compensation coefficient corresponding to the third vibration data segment.

[0107] For example, the corrected matching degree The following formula 5 is satisfied:

[0108] Formula 5

[0109] in, For the first The degree of matching after correction of the third vibration data segment, which is a similarity measure used for the final anomaly determination after system error calibration; Indicates the first The degree of matching of the third vibration data segment before correction, which reflects the correlation between the third vibration data and the first dataset in terms of waveform change trend; Indicates the first The system error compensation coefficient corresponding to the third vibration data segment is a coefficient calculated based on the updated reference dataset, used to characterize the degree of deviation of the current system state from the historical benchmark. This constitutes a suppression factor. System error compensation coefficient. The larger the value, the smaller the inhibitory factor, and the lower the degree of original matching. The stronger the inhibitory effect, the better.

[0110] It should be noted that the data segment difference calculation in this step is performed based on the reference signal already selected through signal decomposition and similarity analysis in the previous stage. The reference signal already represents the core characteristics of historical data in terms of frequency and overall shape. Therefore, the focus of this step is on verifying the consistency of local statistical characteristics, aiming to eliminate data segments that, although the macroscopic frequency pattern is consistent, have abnormal fluctuations or baseline drift in local areas. This is a refined denoising process.

[0111] Understandably, the multiplication operation in Formula 5 constitutes a concise and effective dynamic calibrator. When the system state changes slowly (such as the vibration signal baseline drift caused by the blade heating up during long-term operation), increasing the absolute difference between the third vibration data and the first dataset, the system error compensation coefficient will decrease accordingly. At this time, multiplying the matching degree by the system error compensation coefficient can moderately suppress the matching degree, thereby compensating for the correlation attenuation caused by the system baseline drift and effectively reducing the risk of misjudging normal data as abnormal.

[0112] The corrected matching degree will be used for the final anomaly determination. For example, it will be compared with a first preset threshold: if the corrected matching degree is greater than the first preset threshold, the current third vibration data is determined to be normal; otherwise, it is determined to be abnormal.

[0113] Through the above process, this embodiment enables the data acquisition and verification system of the micro data acquisition instrument to have adaptive capabilities, dynamically adjusting its verification sensitivity according to the state evolution of the monitored object, thereby maintaining a high-precision anomaly identification capability during long-term operation.

[0114] like Figure 7As shown, in one possible implementation, the error coefficient calculation subunit 1041 in this application determines the system error compensation coefficient based on the statistical characteristic differences between the second dataset and the non-abnormal data. This process can be implemented through the following steps:

[0115] S701. Calculate multiple first statistical feature values ​​for each data segment in the second dataset, as well as second statistical feature values ​​for non-abnormal data.

[0116] For example, the variance is calculated using the standard statistical variance formula; the mean is calculated using the arithmetic mean formula.

[0117] For example, the first in the second dataset The first statistical characteristic value of each data segment is: and ;in, For the second dataset, the first The variance of the amplitude of each data segment is used to characterize the dispersion and fluctuation intensity of the data within that data segment. For the second dataset, the first The mean of the magnitudes of the first data segment is used to characterize the central tendency and average level of the data within that segment; in non-abnormal data, the first... The second statistical characteristic value of each data sub-period is: and ;in, The first non-abnormal data The variance of each non-abnormal data period is used to characterize the dispersion and volatility of the data within that non-abnormal data period. The first non-abnormal data period The mean of the amplitudes of each non-abnormal data period is used to characterize the central tendency and average level of the data within that non-abnormal data period.

[0118] Understandably, by calculating the variance and mean of each data segment and non-abnormal data in the second dataset separately, the statistical distribution characteristics of the data can be determined from two dimensions: fluctuation characteristics and average level. The differences in these statistical characteristic values ​​will directly reflect the overall deviation between the newly added normal data and the existing reference dataset, providing a reliable quantitative basis for accurately assessing changes in system state.

[0119] S702. Determine the average characteristic difference based on the differences between multiple first statistical characteristic values ​​and second statistical characteristic values.

[0120] In one possible implementation, the variance difference between the first statistical characteristic value and the second statistical characteristic value is determined. and mean The variance difference is multiplied by the mean difference. This ensures that the average characteristic difference is highly sensitive to differences in both types of statistical characteristics. The average characteristic difference will be small only when both types of differences are small; a large difference in either type will lead to a significant increase in the average characteristic difference value, thus more comprehensively capturing abnormal drifts in the system state.

[0121] S703. Based on the preset normalization function, the average feature difference is normalized to determine the system error compensation coefficient.

[0122] For example, the system error compensation coefficient Satisfy the following formula 6:

[0123] Formula 6

[0124] in, In the first The system error compensation coefficient calculated for each non-abnormal data point is a dimensionless scalar. The first non-abnormal data The variance of each non-abnormal data period; For the second dataset, the first The variance of the magnitude of each data segment; For the second dataset, the first The mean of the amplitudes of each data segment; The first non-abnormal data period The mean of the amplitudes for each non-abnormal data period; 0.001 is the total number of data segments in the second dataset; 0.001 is a smoothing term to prevent the denominator from being zero. 0.001 is much smaller than the total number of data segments in the second dataset by a meaningful physical order of magnitude, so its impact on the calculation result is negligible. For the normalization function, respectively for and Individual normalization ensures that the variance and mean differences contribute equally to the final coefficients, avoiding calculation errors caused by different sensor ranges or signal amplitudes; the overall normalization process stabilizes the coefficients within a fixed range (e.g., [0,1]), making them easy to use as multipliers to correct the degree of matching. This is used to sum all the overall dissimilarity and calculate the arithmetic mean, which is used to characterize the average statistical characteristics of the non-outlier data and the second reference dataset, in terms of volatility and average level.

[0125] Understandable This is a normalized coefficient that quantifies the systematic statistical differences between the non-abnormal data and the second dataset, reflecting the current degree of system deviation. This coefficient is crucial for the system's adaptive correction, enabling it to detect slow changes caused by long-term operation and cumulative environmental effects. Using this coefficient to correct subsequent verification judgments is equivalent to dynamically adjusting the sensitivity of anomaly detection. When system deviation is large, the judgment criteria are appropriately relaxed to prevent misjudgments of data caused by system deviation, thereby significantly improving the system's robustness and accuracy in long-term monitoring tasks.

[0126] like Figure 8 As shown, in one possible implementation, the process by which the dataset update unit 103 updates the first dataset based on the non-abnormal data in the second vibration data to obtain the non-abnormal data in the second dataset can be specifically implemented through the following steps:

[0127] S801. Add non-abnormal data to the storage queue of the first dataset.

[0128] The vibration data in the storage queue is arranged in chronological order.

[0129] Understandably, the queue management mechanism enables dynamic updates to the reference dataset, ensuring that the verification benchmark can continuously track the latest normal state of the system while maintaining the stability of the dataset size and avoiding unlimited growth of storage space.

[0130] S802. Remove the oldest historical data from the storage queue and determine the second dataset.

[0131] One possible implementation involves maintaining a fixed-size dataset, incorporating new normal data while discarding the oldest historical data, thereby ensuring that the verification benchmark can both track the latest changes in the system state and maintain the stability of the dataset size.

[0132] In one possible implementation, the data acquisition and verification system of the micro data acquisition instrument is communicatively connected to a high-speed micro data acquisition instrument, which is used to acquire vibration data of helicopter rotor blades.

[0133] Among them, the vibration data of helicopter rotor blades refers to the time-series signal that reflects the vibration state of the blades, which is obtained by sensing through vibration sensors and digitizing through a high-speed micro data acquisition instrument.

[0134] In one possible implementation, the data acquisition and verification system of the micro data acquisition instrument sends parameter configuration instructions and acquisition control signals to the high-speed micro data acquisition instrument; the high-speed micro data acquisition instrument transmits the acquired vibration data to the verification system in real time.

[0135] By establishing a communication connection between the data acquisition and verification system of the micro data acquisition instrument and the high-speed micro data acquisition instrument, the real-time acquisition and online verification of vibration data were realized. The high-speed micro data acquisition instrument can accurately capture the vibration signals of helicopter rotor blades under complex working conditions, providing a high-quality data foundation for subsequent data verification.

[0136] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0137] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A data acquisition and verification system for a miniature data acquisition instrument, characterized in that, The system includes: A dataset construction unit is used to perform signal decomposition on the first vibration data to determine multiple component signals; the first vibration data is all vibration data of the target blade within a preset time period; the preset time period is any historical time period. A dataset construction unit is used to determine the reference signal among the plurality of component signals that has the highest similarity to the first vibration data; A dataset construction unit is configured to determine at least one target data segment from the first vibration data based on the reference signal; the at least one target data segment is used to constitute a first dataset; the first dataset includes at least one first vibration data. A data verification unit is used to determine abnormal data in the second vibration data based on the degree of matching between the second vibration data and the first dataset; the degree of matching between the abnormal data and the first dataset is less than a first preset threshold; the second vibration data is the real-time vibration data of the target blade; A dataset update unit is used to update the first dataset based on non-abnormal data in the second vibration data to obtain a second dataset. The data correction unit is used to determine the system error compensation coefficient based on the statistical characteristic differences between the second dataset and the non-abnormal data; The data correction unit is used to multiply the degree of matching between the third vibration data and the first dataset by the system error compensation coefficient to determine the corrected degree of matching; the third vibration data is all vibration data of the target blade within the target time period.

2. The data acquisition and verification system for the miniature data acquisition instrument according to claim 1, characterized in that, The dataset construction unit is specifically used for: Determine the absolute value of the frequency difference between each of the plurality of component signals and the first vibration data; Based on the absolute value of the frequency difference between each component signal and the first vibration data, a similarity value between multiple component signals and the first vibration data is determined. The component signal with the highest similarity value among the multiple component signals is determined as the reference signal.

3. The data acquisition and verification system for the miniature data acquisition instrument according to claim 1, characterized in that, The dataset construction unit is specifically used for: The degree of difference between each data segment and the time period data of the reference signal is determined based on the fluctuation amplitude of each data segment; Based on the difference between each data segment and the time period data of the reference signal, and a preset clustering algorithm, multiple clusters are determined; Based on the principal component directions of the data segments in the multiple clusters, an anomaly score is determined for each of the multiple clusters; the anomaly score is used to characterize the differences in principal component directions and average differences among the multiple clusters, and comprehensively reflects the degree of anomaly of the clusters. Identify a target cluster among the plurality of clusters that meets a preset condition; the preset condition is: the anomaly score is less than a second preset threshold; the data segment in the target cluster is the target data segment.

4. The data acquisition and verification system for the miniature data acquisition instrument according to claim 3, characterized in that, The data verification unit includes: The correlation calculation subunit is used to calculate the correlation coefficient between the second vibration data and the target data segment; The matching degree determination subunit is used to determine the matching degree based on the correlation coefficient.

5. The data acquisition and verification system for the miniature data acquisition instrument according to claim 4, characterized in that, The correlation calculation subunit is specifically used for: Calculate the Pearson correlation coefficient between the second vibration data and the target data segment.

6. The data acquisition and verification system for a miniature data acquisition instrument according to claim 5, characterized in that, The data correction unit is specifically used for: Calculate multiple first statistical feature values ​​for each data segment in the second dataset, and second statistical feature values ​​for the non-abnormal data; the first statistical feature values ​​are used to characterize the dispersion and fluctuation intensity of each data segment. The second statistical characteristic value is used to characterize the dispersion and fluctuation intensity of the non-abnormal data; Based on the differences between the plurality of first statistical feature values ​​and the second statistical feature values, the average feature difference is determined; The average feature difference is normalized based on a preset normalization function to determine the system error compensation coefficient.

7. The data acquisition and verification system for a miniature data acquisition instrument according to claim 6, characterized in that, The dataset update unit is specifically used for: The non-abnormal data is added to the storage queue of the first dataset; the vibration data in the storage queue are arranged in chronological order. Remove the oldest historical data from the storage queue to determine the second dataset.

8. The data acquisition and verification system for a miniature data acquisition instrument according to claim 1, characterized in that, The data acquisition and verification system of the micro data acquisition instrument is communicatively connected to the high-speed micro data acquisition instrument; the high-speed micro data acquisition instrument is used to acquire vibration data of helicopter rotor blades.

Images