Data acquisition method and device

Abstract

The invention provides a data acquisition method and device, and the method comprises the steps: carrying out the dynamic noise suppression of analog input signals of four input channels, and obtaining a primary purification signal; performing multi-stage nonlinear compensation on the primary purification signal based on a preset sensor type feature library to generate a standardized linear signal; performing time domain-space domain joint reconstruction on the standardized linear signal, and outputting a high-fidelity reconstruction signal; obtaining a parameter configuration library of a preset industrial scene mode, performing digital pre-distortion compensation on the high-fidelity reconstruction signal according to the parameter configuration library, and generating a scene adaptation signal; and performing edge feature extraction and data compression on the scene adaptation signal to generate an industrial data packet, and uploading the industrial data packet to the two output channels through a preset communication protocol.

Description

Technical Field

[0001] The present application relates to the field of data acquisition technology, and in particular to a data acquisition method and device. Background Art

[0002] Currently, industrial field environments are complex, with numerous noise interference sources, resulting in unstable signal quality. Traditional fixed-threshold filtering methods struggle to adapt to dynamically changing noise environments, easily causing signal distortion or residual noise. Furthermore, different sensor types have unique nonlinear characteristics and are affected by external factors such as temperature and vibration, resulting in significant deviations between the acquired signal and the actual physical quantity. Existing compensation methods often use a single fixed model, which cannot meet the precise calibration requirements of diverse sensors. During multi-channel synchronous acquisition, timing inconsistencies between channels occur due to hardware differences and signal transmission delays, affecting signal reconstruction accuracy. Traditional independent channel processing ignores inter-channel correlations and fails to fully utilize multi-channel data for complementary verification and optimization. Furthermore, different industrial application scenarios have varying requirements for output signal timeliness, accuracy, and bandwidth, lacking a flexible output strategy adaptation mechanism. Summary of the Invention

[0003] The present application provides a data acquisition method and device for reducing noise interference and efficiently realizing the conversion of various industrial application scenarios.

[0004] In a first aspect, an embodiment of the present application provides a data collection method, the method comprising: Perform dynamic noise suppression processing on the analog input signals of the four input channels to obtain primary purified signals; Performing multi-level nonlinear compensation on the primary purification signal based on a preset sensor type feature library to generate a standardized linear signal; Performing time-domain and space-domain joint reconstruction on the standardized linear signal to output a high-fidelity reconstructed signal; Obtaining a parameter configuration library of a preset industrial scene mode, performing digital predistortion compensation on the high-fidelity reconstructed signal according to the parameter configuration library, and generating a scene adaptation signal; The scene adaptation signal is subjected to edge feature extraction and data compression to generate an industrial data packet, which is then uploaded to two output channels via a preset communication protocol.

[0005] In a second aspect, an embodiment of the present application provides a data acquisition device, wherein the data acquisition device is used to execute the data acquisition method as described in any one of the embodiments of the present application, and the device includes: The noise processing module is used to perform dynamic noise suppression processing on the analog input signals of the four input channels to obtain primary purified signals; A signal compensation module, configured to perform multi-stage nonlinear compensation on the primary purification signal based on a preset sensor type feature library to generate a standardized linear signal; A signal reconstruction module, configured to perform time-domain and space-domain joint reconstruction on the standardized linear signal and output a high-fidelity reconstructed signal; A signal adaptation module is used to obtain a parameter configuration library of a preset industrial scene mode, perform digital predistortion compensation on the high-fidelity reconstructed signal according to the parameter configuration library, and generate a scene adaptation signal; The protocol conversion module is used to extract edge features and compress data of the scene adaptation signal to generate an industrial data packet, and upload the industrial data packet to two output channels through a preset communication protocol.

[0006] An embodiment of the present application provides a data acquisition method, which includes: performing dynamic noise suppression processing on analog input signals of four input channels to obtain a primary purified signal; performing multi-level nonlinear compensation on the primary purified signal based on a preset sensor type feature library to generate a standardized linear signal; performing time-domain and spatial-domain joint reconstruction on the standardized linear signal to output a high-fidelity reconstructed signal; obtaining a preset parameter configuration library of an industrial scene mode, performing digital pre-distortion compensation on the high-fidelity reconstructed signal according to the parameter configuration library to generate a scene adaptation signal; performing edge feature extraction and data compression on the scene adaptation signal to generate an industrial data packet, and uploading the industrial data packet to two output channels through a preset communication protocol. Through the above method, a method combining dynamic noise suppression and multi-level nonlinear compensation is adopted to accurately calibrate the analog input signals of the four input channels, effectively improving the acquisition accuracy of the four analog input signals. Through time-domain-spatial domain joint reconstruction and digital pre-distortion compensation technology, high-quality and high-fidelity reconstructed signals of multi-channel signals are ensured. Then, the output characteristics are flexibly adjusted according to the requirements of industrial scenarios to generate scene-adapted signals. Finally, edge feature extraction and dual-channel communication mechanism are introduced to transmit the industrial data packets corresponding to the scene-adapted signals. This significantly improves bandwidth utilization efficiency while ensuring data transmission reliability, effectively balancing the requirements of real-time and reliability. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0008] Figure 1 A schematic flow chart of a data acquisition method provided in an embodiment of the present application; Figure 2A schematic block diagram of a data acquisition device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0009] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0010] The flowcharts shown in the accompanying drawings are for illustrative purposes only and do not necessarily include all contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may be decomposed, combined, or partially merged, so the actual execution order may vary depending on the actual situation.

[0011] It should also be understood that the terms used in this specification are for the purpose of describing specific embodiments only and are not intended to limit the present application. As used in this specification and the appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms unless the context clearly indicates otherwise.

[0012] It should be further understood that the term "and / or" used in this specification and the appended claims refers to and includes any and all possible combinations of one or more of the associated listed items.

[0013] See also Figure 1 , Figure 1 The following is a schematic flow chart of a data acquisition method provided in an embodiment of the present application. Figure 1 As shown, the specific steps of the data acquisition method include: S101-S105.

[0014] S101 , performing dynamic noise suppression processing on analog input signals of four input channels to obtain primary purified signals.

[0015] For example, when dynamic noise suppression is performed on four analog input signals, an adaptive threshold adjustment algorithm is used to calculate the noise baseline and fluctuation characteristics of the signal in real time. The algorithm establishes a dynamic noise model by continuously monitoring the mean, variance and peak distribution of the signal, and combines the wavelet transform multi-scale decomposition technology to perform hierarchical identification of noise components in different frequency bands. After identifying the noise characteristics, an improved Kalman filter is used to adaptively filter the signal. The filter can dynamically adjust the gain coefficient according to the noise intensity to achieve accurate suppression of sudden interference and continuous noise. By introducing a nonlinear prediction compensation mechanism, the phase lag problem of traditional filtering methods when the signal changes rapidly is effectively avoided, ensuring the time domain integrity and frequency domain characteristics of the primary purification signal.

[0016] S102 : Perform multi-level nonlinear compensation on the primary purification signal based on a preset sensor type feature library to generate a standardized linear signal.

[0017] For example, when performing multi-level nonlinear compensation based on a preset sensor type feature library, a feature database containing various industrial sensor characteristic models such as temperature, pressure, and displacement is established. This database records the nonlinear characteristic curves, temperature drift coefficients, and hysteresis characteristic parameters of different types of sensors under various working conditions. The real-time signal slope of the primary purification signal is obtained by calculation, and the most suitable sensor characteristic curve is dynamically matched based on the real-time signal slope. Based on the sensor characteristic curve and the real-time signal slope, a piecewise polynomial fitting method is used to construct a dynamic compensation model. An adaptive weight adjustment mechanism is introduced in the compensation process to automatically adjust the compensation coefficients at each level according to the dynamic range and change rate of the signal to ensure a smooth transition of the compensation effect. For complex nonlinear distortion, an iterative optimization algorithm is used to correct the compensation parameters step by step until the ideal linear output characteristics are obtained.

[0018] S103 , performing time-domain and space-domain joint reconstruction on the standardized linear signal, and outputting a high-fidelity reconstructed signal.

[0019] Exemplarily, when reconstructing the signal through a sliding window mechanism and adjacent channel cross-validation, an overlapping sliding window is used to segment the multi-channel signal. The window size is dynamically adjusted according to the signal bandwidth and sampling rate to ensure that the complete signal characteristics are captured. Within each window, the signal correlation of adjacent channels is used to establish a multi-dimensional feature space, and the common eigenvector is extracted through principal component analysis. The signal is jointly reconstructed by combining time domain interpolation and frequency domain reconstruction technology. In this process, an improved singular spectrum analysis method is applied to effectively separate the periodic component and random component in the signal. By introducing a mutual information evaluation mechanism between channels, the reconstruction results are verified and corrected in real time to ensure the synchronization and consistency of multi-channel signals. The reconstruction algorithm also includes an adaptive delay compensation module to eliminate the sampling delay and transmission time difference between channels.

[0020] S104: Obtain a parameter configuration library of a preset industrial scene mode, perform digital predistortion compensation on the high-fidelity reconstructed signal according to the parameter configuration library, and generate a scene-adapted signal.

[0021] For example, when selecting an output strategy based on a preset industrial scenario mode, an output parameter configuration library for different industrial application scenarios is established. The parameter configuration library contains bandwidth requirements, delay requirements and accuracy indicators for various industrial control, monitoring and feedback scenarios. The parameter configuration library analyzes the current working mode and load status corresponding to the high-fidelity reconstructed signal in real time, selects a matching output strategy, and independently optimizes the two target output channels. The digital pre-distortion compensation process adopts a lookup table structure to store the distortion characteristics and compensation coefficients at different working points. By dynamically updating the compensation parameters, accurate linearization of the output signal is achieved. The compensation algorithm also includes an adaptive equalization module to compensate for the distortion caused by changes in channel characteristics and ensure that the output signal maintains stable characteristics under different load conditions.

[0022] S105 , extracting edge features and compressing data on the scene adaptation signal to generate an industrial data packet, and uploading the industrial data packet to two output channels through a preset communication protocol.

[0023] For example, when extracting edge features and compressing data for scene-adapted signals, a multi-level feature extraction strategy is adopted, including time-domain statistical features, frequency-domain energy distribution, and waveform morphology features. By setting dynamic thresholds, key change points and feature segments in the signal are identified and retained, and redundant information is discarded. Data compression adopts a hybrid coding scheme, selecting the optimal compression algorithm for different types of feature data, such as run-length coding, differential coding, or wavelet compression. When generating industrial data packets, a layered encapsulation structure is used to organize feature data, timestamps, and checksum information. The dual-channel communication protocol realizes the parallel transmission of data streams, with the main channel responsible for real-time data transmission and the backup channel for redundant transmission of key data and retransmission of lost packets. The protocol also includes a dynamic bandwidth allocation mechanism that can automatically adjust the data stream allocation ratio according to network conditions to ensure transmission quality.

[0024] An embodiment of the present application provides a data acquisition method, which includes: performing dynamic noise suppression processing on analog input signals of four input channels to obtain a primary purified signal; performing multi-level nonlinear compensation on the primary purified signal based on a preset sensor type feature library to generate a standardized linear signal; performing time-domain and spatial-domain joint reconstruction on the standardized linear signal to output a high-fidelity reconstructed signal; obtaining a preset parameter configuration library of an industrial scene mode, performing digital pre-distortion compensation on the high-fidelity reconstructed signal according to the parameter configuration library to generate a scene adaptation signal; performing edge feature extraction and data compression on the scene adaptation signal to generate an industrial data packet, and uploading the industrial data packet to two output channels through a preset communication protocol. Through the above method, a method combining dynamic noise suppression and multi-level nonlinear compensation is adopted to accurately calibrate the analog input signals of the four input channels, effectively improving the acquisition accuracy of the four analog input signals. Through time-domain-spatial domain joint reconstruction and digital pre-distortion compensation technology, high-quality and high-fidelity reconstructed signals of multi-channel signals are ensured. Then, the output characteristics are flexibly adjusted according to the requirements of industrial scenarios to generate scene-adapted signals. Finally, edge feature extraction and dual-channel communication mechanism are introduced to transmit the industrial data packets corresponding to the scene-adapted signals. This significantly improves bandwidth utilization efficiency while ensuring data transmission reliability, effectively balancing the requirements of real-time and reliability.

[0025] In order to more clearly introduce the technical solution of the present application, the technical solution of the present application will be introduced through specific embodiments below. It should be noted that the specific embodiments are used to expand the technical solution of the present application, but are not intended to limit the present application.

[0026] In some embodiments, dynamic noise suppression processing is performed on analog input signals from four input channels to obtain a primary cleansed signal, including: sampling the analog input signals from the four input channels to obtain an original digital signal sequence; performing a mutation point distribution detection on the original digital signal sequence; and determining a signal sequence to be filtered based on the mutation point distribution detection results. Multi-scale decomposition of the signal sequence to be filtered is performed using a wavelet transform to extract signal characteristic coefficients. Noise energy estimation and dynamic filtering are performed on the signal sequence to be filtered based on the signal characteristic coefficients to generate a primary cleansed signal.

[0027] For example, a four-channel synchronous sampling analog-to-digital converter (ADC) was used, with a sampling frequency of 200kHz and a resolution of 24 bits. The analog signals from the four input channels were sampled at equal intervals. Each channel generated a continuous digital signal sequence with a sampling interval of 5μs and a data storage depth of 1MB. This resulted in four sets of time-aligned raw digital signal sequences, generating 2.4MB of raw data per second per channel. A CUSUM (cumulative sum) algorithm was used to detect signal mutation characteristics, with a sliding window length of 128 sampling points and a threshold of three standard deviations. Abnormal mutation points exceeding the threshold were identified by calculating the rate of change of the signal slope and the energy jump within the window. A binary sequence containing mutation location markers was generated for each channel, with marker accuracy reaching the level of a single sampling point. An average of 50 mutation points were detected per channel. A 5-layer multiresolution decomposition was performed using the db4 wavelet basis function, with the number of decomposition layers adaptively adjusted between 3 and 7 layers. For each sampling point, a feature vector containing approximate and detail coefficients is generated. By calculating the energy contribution at each scale, the principal signal component is separated from the noise component. The feature coefficients are compressed to 1 / 32 of the original signal dimension, generating a feature matrix containing the frequency-energy distribution for each channel. The noise energy baseline is calculated using the Stein unbiased risk estimation method, with a dynamically adjusted threshold factor ranging from 0.2 to 1.5. Nonlinear soft thresholding is applied to the wavelet coefficients at each scale, with a strict threshold of 1.2 for high frequencies and a threshold factor of 0.3 for low frequencies to preserve coefficient integrity. After filtering, the signal-to-noise ratio is improved by over 20 dB, and noise energy is reduced by 85%. Four channels of primary cleaned signals are output, maintaining a data rate of 1.8 MB / s and waveform distortion less than 0.5%.

[0028] In some embodiments, multi-level nonlinear compensation is performed on the primary cleanup signal based on a preset sensor type feature library to generate a standardized linear signal. This includes calculating the real-time signal slope based on the primary cleanup signal and extracting the corresponding sensor characteristic curve from the preset sensor type feature library. A dynamic compensation model is established based on the sensor characteristic curve and the real-time signal slope. In this dynamic compensation model, a first compensation signal is obtained by performing a first-level compensation on the primary cleanup signal using an adaptive weight adjustment mechanism. A second compensation signal is generated by performing a second-level compensation on the base compensation signal through residual analysis. The second compensation signal is then fitted and calibrated with the primary cleanup signal to output a standardized linear signal.

[0029] For example, a sliding difference calculation with a window width of 100 sampling points and a step size of 10 sampling points is used to calculate the real-time signal slope and normalize it to the range [-1, 1]. A preset sensor type feature library stores sensor characteristic curves for 50 types of industrial sensors, each containing 1000 discrete data points. A dynamic time warping algorithm is used to match the target characteristic curve with a similarity of 98%, with a matching response time of less than 2ms. Based on the target characteristic curve, a compensation function is established using a third-order piecewise polynomial fit, parameterizing the curve using seven feature nodes. A normalized weight matrix is ​​constructed, dynamically adjusting the compensation coefficient weights based on the signal change rate, with a weight adjustment step size of 0.01 and a compensation residual threshold set at 0.05% of the range. After primary compensation, the nonlinear error is reduced to 0.15%FS, and the efficiency of temperature drift coefficient compensation is improved by 35%. Residual analysis (secondary compensation) is performed on the signal after primary compensation, and a residual prediction model is constructed using recursive least squares, with a forgetting factor of 0.99 and a window length of 200 points. By establishing a residual-compensation mapping table and dynamically modifying compensation parameters, the residual fluctuation range is reduced from ±1.2% to ±0.3%, with no more than five compensation iterations. Final calibration is performed using the Levenberg-Marquardt optimization algorithm, with a maximum number of iterations set to 20 and a convergence threshold of 0.001%FS. The calibration process simultaneously optimizes both time and frequency domain characteristics, resulting in a standardized linear signal with a linearity of 0.02, a hysteresis error of less than 0.01%, and a full-scale temperature drift coefficient stable within 5ppm / °C. After four-channel compensation, signal delay variation is controlled to ±0.5μs, meeting industrial-grade signal standardization requirements.

[0030] In some embodiments, a time-domain-spatial-domain joint reconstruction is performed on a standardized linear signal to output a high-fidelity reconstructed signal, including: segmenting the standardized linear signal based on a dynamic sliding window of a preset width to obtain segmented signal intervals. According to the signal change rate of the segmented signal interval, a first change interval and a second change interval are determined, and a time-domain segmented sequence is generated based on the first change interval and the second change interval, respectively. Correlation analysis is performed on adjacent channel signals in the time-domain segmented sequence to obtain an outlier mapping table. Using the outlier mapping table, a correction signal is extracted from the adjacent channel signal, and the time-domain segmented sequence is interpolated and replaced based on the correction signal to generate a preliminary correction signal. Key feature points in the preliminary correction signal are extracted to form a feature point sequence. Cubic spline interpolation reconstruction is performed based on the feature point sequence to generate a high-fidelity reconstructed signal.

[0031] For example, segmented processing of the standardized linear signal was performed using a dynamic sliding window, with a window length of 256 sampling points, a step size of 64 points, and a window overlap of 25%. The segmented signal interval was divided into a first variation interval (rate > threshold) and a second variation interval (rate ≤ threshold) based on a standard deviation threshold of 0.5 V / ms. Time-domain segmented sequences were generated using a K-means clustering algorithm, with two cluster centers set. The first variation interval was labeled as a high-dynamic segment, and the second variation interval was labeled as a steady-state segment. The segment boundary positioning error was less than ±0.1 ms. Correlation analysis of adjacent channel signals was performed using a sliding window Pearson correlation coefficient, with a window length of 128 points and a correlation coefficient threshold of 0.92. Regions below the threshold were marked as candidate outliers. An adaptive threshold model was used to construct the outlier mapping table, with a dynamic adjustment range of 2.8-3.2σ. Approximately 200 outliers were detected per hour, with each anomaly associated with a timestamp (with an accuracy of 10 μs) and an adjacent channel number (1-4). Corrected signal extraction selects signals from two adjacent channels and applies linear weighting. The weighting coefficient is dynamically adjusted (0.3-0.7) based on signal quality assessment. Interpolation replacement uses a 5-point cubic Hermite interpolation method with a 32-point interpolation window, achieving an 85% outlier repair rate. Key feature point extraction uses sliding curvature analysis with a curvature threshold of 0.05 mm⁻¹ and a duration threshold of 10 ms. Approximately 1,500 feature points (including mutation points and inflection points) are extracted per hour. Cubic spline interpolation reconstruction uses natural boundary conditions and adaptively adjusts node spacing (10-100 points). This method produces a high-fidelity reconstructed signal with a time domain error of less than 0.05% FS, a frequency domain harmonic distortion of less than -60 dB, and a delay variation between the four channels within ±0.8 μs, meeting industrial-grade high-fidelity reconstruction requirements.

[0032] In some embodiments, correlation analysis is performed on adjacent channel signals in a time-domain segmented sequence to obtain an outlier mapping table, including: calculating a time-series correlation coefficient between adjacent channel signals and generating a channel correlation matrix based on the time-series correlation coefficient. Data points in the channel correlation matrix are preliminarily labeled based on a preset adaptive threshold to obtain a set of candidate outlier points. Time-window clustering analysis is performed on the set of candidate outlier points to obtain an anomaly feature distribution map, which includes continuous anomaly intervals and a first outlier point. Signal trends of adjacent channel signals are calculated, and second outliers are determined from discrete outliers based on the signal trends. The second outlier points are then associated and mapped with corresponding timestamps and channel numbers to generate an outlier mapping table.

[0033] For example, a sliding window Pearson correlation coefficient algorithm is used, with a window length of 256 sampling points and a step size of 64 points, to calculate the pairwise correlation coefficients between the four channels. The dimension of the correlation matrix is ​​4×4×N, and each time slice contains 6 groups of correlation coefficient values. The correlation coefficient threshold is 0.85, and the area below the threshold is marked as a weak correlation area. The data update frequency is 1kHz, forming a continuous time series correlation map. An adaptive threshold model based on the sliding standard deviation is established, with a window length of 512 points and a dynamic adjustment range of the threshold coefficient of 2.5-3.5σ. The correlation coefficient matrix is ​​detected point by point, and the area with 5 consecutive sampling points below the threshold is marked as a candidate anomaly segment. On average, 120 candidate anomaly intervals are detected per channel per hour, with a marking accuracy of ±2 sampling points. The DBSCAN clustering algorithm is used, with a neighborhood radius of 50ms and a minimum number of samples of 3. Candidate outliers were spatiotemporally clustered. Continuous outlier intervals lasting longer than 100ms were identified as first-level anomalies, while discrete outliers were classified as second-level anomalies. After clustering, the number of anomalies was reduced to 35% of the original candidate set, resulting in a heat map of the anomaly distribution, including spatiotemporal coordinates. A fifth-order polynomial trend curve was fitted using the least squares method with a window length of 1024 points. The residual standard deviation between the actual signal and the trend curve was calculated, with 2.8σ set as the anomaly cutoff. Trend consistency checks were performed on discrete outliers, and outliers with deviations from the trend line exceeding 3σ were identified as second-level anomalies, reducing the false alarm rate to below 0.5%. The timestamp (1μs accuracy), channel number (1-4), and anomaly intensity value (normalized to 0-1) of the second-level anomaly point were three-dimensionally encoded. A mapping table was created using a hierarchical storage structure with a temporal resolution of 10ms and a spatial resolution of 0.1%. A generated outlier mapping table containing 450 anomaly records was generated every hour, each containing a 24-dimensional feature vector, enabling millisecond-level fast retrieval and visual location.

[0034] In some embodiments, a parameter configuration library for a preset industrial scene mode is obtained, and digital pre-distortion compensation is performed on the high-fidelity reconstructed signal according to the parameter configuration library to generate a scene-adapted signal, including: selecting a parameter configuration library corresponding to the current application scenario, including: signal bandwidth requirements, dynamic range, and distortion compensation coefficient. Based on the parameter configuration library, spectrum analysis and bandwidth adjustment are performed on the high-fidelity reconstructed signal to generate a first matching signal. Based on the distortion compensation coefficient, nonlinear pre-distortion processing is performed on the initial bandwidth matching signal to eliminate the inherent distortion in the output channel to obtain a second matching signal. The amplitude of the second matching signal is adjusted to within the target dynamic range, and the scene-adapted signal is output.

[0035] For example, the parameter configuration library integrates 12 industrial scenario models, each pre-stored with 48 dynamic range configurations (adjustable from 10dB to 90dB) and a 256-dimensional distortion compensation vector, including signal bandwidth requirements, dynamic range, and distortion compensation coefficients. Spectral analysis uses a 4096-point FFT to calculate power spectral density, while bandwidth adjustment employs digital filter bank technology, designing a 128th-order FIR filter to achieve ±2% bandwidth accuracy. Nonlinear pre-distortion processing is based on a memory polynomial model with a fifth-order nonlinear term and a third-order memory depth. The compensation coefficients are dynamically updated using a least-mean-square algorithm with a 1ms update cycle, resulting in a harmonic distortion suppression ratio of up to 55dB. Dynamic range adjustment utilizes segmented automatic gain control, divided into eight amplitude ranges, each with an independent gain coefficient (adjustable from 0.8 to 1.2). The output signal peak-to-average ratio is optimized to 4.8dB, with an in-band flatness error of less than ±0.03dB.

[0036] In some embodiments, edge feature extraction and data compression are performed on the scene adaptation signal to generate an industrial data packet, which is then uploaded to two output channels via a preset communication protocol. This includes: edge feature extraction of the scene adaptation signal to obtain a feature point sequence, including: the signal's mutation point, inflection point, and steady-state interval. Based on the feature point sequence and the scene adaptation signal, a feature description table is determined, the feature description table including: mean, variance, and trend coefficient. Data redundancy analysis is performed using the feature description table to obtain a first data packet. The first data set is grouped and packaged according to timestamps, and a channel identifier and data type tag are added to generate a second data packet. The second data packet is differentially encoded and run-length encoded to generate an industrial data packet. Through a preset communication protocol, the industrial data packet is divided into main data and verification data, and transmitted in parallel through dual channels.

[0037] For example, edge feature extraction utilizes a three-stage detection strategy: A slope threshold of 0.8V / ms is set for mutation point detection; inflection point identification utilizes the cubic spline interpolation derivative method (curvature threshold of 0.05mm⁻¹); and steady-state intervals are defined as fluctuations lasting 100ms with a variance less than 0.01%FS. Feature description tables are constructed, including time-domain statistics (mean, peak-to-peak value, and crest factor) and frequency-domain features (dominant frequency energy fraction and spectral entropy). The feature dimensions are compressed to 18. Data redundancy analysis utilizes a similarity assessment based on KL divergence, and dictionary compression is applied to data segments with a similarity exceeding 95%. Packing is performed using a 50ms time window, with each frame containing 1024 samples. The frame header includes an 8-byte timestamp (with μs-level accuracy) and a 4-byte channel status word. The compression process begins with 16-bit signed differential encoding, followed by adaptive Huffman coding. The reconstruction error rate is 0.08% when the compression ratio reaches 5:1. Dual-channel transmission adopts a master-slave redundant architecture. The main channel transmits compressed data streams (rate 2.4Mbps), and the backup channel transmits feature description tables (rate 200kbps). The transmission protocol integrates forward error correction coding (RS(255,223)), and the packet loss rate is suppressed to 10⁻ 6 Below, the end-to-end delay jitter is less than ±3ms.

[0038] See also Figure 2 , Figure 2 The embodiment of the present application further provides a schematic block diagram of a data acquisition device, wherein the data acquisition device 200 is used to execute the aforementioned data acquisition method. The data acquisition device can be configured in a server or a terminal.

[0039] The server can be a standalone server or a server cluster, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms. The terminal can be an electronic device such as a mobile phone, tablet computer, laptop computer, desktop computer, user digital assistant, and wearable device.

[0040] like Figure 2 As shown, the data acquisition device 200 includes: a noise processing module 201 , a signal compensation module 202 , a signal reconstruction module 203 , a signal adaptation module 204 and a protocol conversion module 205 .

[0041] The noise processing module 201 is used to perform dynamic noise suppression processing on the analog input signals of the four input channels to obtain primary purified signals.

[0042] The signal compensation module 202 is configured to perform multi-level nonlinear compensation on the primary purification signal based on a preset sensor type feature library to generate a standardized linear signal.

[0043] The signal reconstruction module 203 is used to perform time-domain and space-domain joint reconstruction on the standardized linear signal and output a high-fidelity reconstructed signal.

[0044] The signal adaptation module 204 is configured to obtain a parameter configuration library of a preset industrial scene mode, perform digital predistortion compensation on the high-fidelity reconstructed signal according to the parameter configuration library, and generate a scene-adapted signal.

[0045] The protocol conversion module 205 is used to extract edge features and compress data of the scene adaptation signal to generate an industrial data packet, and upload the industrial data packet to two output channels through a preset communication protocol.

[0046] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present application, and such modifications or substitutions should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A data collection method, characterized in that: The method comprises: Perform dynamic noise suppression processing on the analog input signals of the four input channels to obtain primary purified signals; Performing multi-level nonlinear compensation on the primary purification signal based on a preset sensor type feature library to generate a standardized linear signal; Performing time-domain and space-domain joint reconstruction on the standardized linear signal to output a high-fidelity reconstructed signal; Obtaining a parameter configuration library of a preset industrial scene mode, performing digital predistortion compensation on the high-fidelity reconstructed signal according to the parameter configuration library, and generating a scene adaptation signal; The scene adaptation signal is subjected to edge feature extraction and data compression to generate an industrial data packet, which is then uploaded to two output channels via a preset communication protocol.

2. The data collection method according to claim 1, wherein: The step of performing dynamic noise suppression processing on the analog input signals of the four input channels to obtain primary purified signals includes: Sampling the analog input signals of the four input channels to obtain original digital signal sequences; Performing a mutation point distribution detection on the original digital signal sequence, and determining a signal sequence to be filtered according to a detection result of the mutation point distribution detection; Using wavelet transform to perform multi-scale decomposition on the signal sequence to be filtered and extract signal characteristic coefficients; Noise energy estimation and dynamic filtering are performed on the signal sequence to be filtered according to the signal characteristic coefficient to generate a primary purified signal.

3. The data collection method according to claim 1, wherein: The method of performing multi-stage nonlinear compensation on the primary purification signal based on a preset sensor type feature library to generate a standardized linear signal includes: Calculating a real-time signal slope based on the primary purification signal and extracting a corresponding sensor characteristic curve from a preset sensor type feature library; Establishing a dynamic compensation model based on the sensor characteristic curve and the real-time signal slope; In the dynamic compensation model, the primary purification signal is compensated for at least one level using an adaptive weight adjustment mechanism to obtain a first compensation signal; Performing secondary compensation on the basic compensation signal through residual analysis to generate a second compensation signal; The second compensation signal is fitted and calibrated with the primary purification signal to output a standardized linear signal.

4. The data collection method according to claim 1, wherein: The step of performing time-domain and space-domain joint reconstruction on the standardized linear signal and outputting a high-fidelity reconstructed signal includes: Segmenting the standardized linear signal based on a dynamic sliding window of a preset width to obtain segmented signal intervals; determining a first change interval and a second change interval according to a signal change rate of the segmented signal interval, and generating a time domain segment sequence according to the first change interval and the second change interval respectively; Performing correlation analysis on adjacent channel signals in the time domain segmented sequence to obtain an abnormal point mapping table; Extracting a correction signal from the adjacent channel signal using the outlier mapping table, and performing interpolation replacement on the time domain segment sequence according to the correction signal to generate a preliminary correction signal; Extracting key feature points from the preliminary corrected signal to form a feature point sequence; Cubic spline interpolation reconstruction is performed based on the feature point sequence to generate a high-fidelity reconstructed signal.

5. The data collection method according to claim 4, wherein: The performing correlation analysis on adjacent channel signals in the time domain segmented sequence to obtain an abnormal point mapping table includes: Calculating a time series correlation coefficient between the adjacent channel signals, and generating a channel correlation matrix according to the time series correlation coefficient; Preliminarily marking the data points of the channel correlation matrix based on a preset adaptive threshold to obtain a set of candidate outlier points; Performing a time window cluster analysis on the candidate abnormal point set to obtain an abnormal feature distribution map, wherein the abnormal distribution map includes: continuous abnormal intervals and the first abnormal point; Calculating a signal trend of the adjacent channel signal, and determining a second abnormal point from the discrete abnormal points according to the signal trend; The second abnormal point is associated with the corresponding timestamp and channel number and mapped to generate an abnormal point mapping table.

6. The data collection method according to claim 1, wherein: The obtaining of a parameter configuration library of a preset industrial scene mode, performing digital predistortion compensation on the high-fidelity reconstructed signal according to the parameter configuration library, and generating a scene adaptation signal includes: Select the parameter configuration library corresponding to the current application scenario, including: signal bandwidth requirements, dynamic range and distortion compensation coefficient; Based on the parameter configuration library, performing spectrum analysis and bandwidth adjustment on the high-fidelity reconstructed signal to generate a first matching signal; performing nonlinear predistortion processing on the initial bandwidth matching signal according to the distortion compensation coefficient to eliminate inherent distortion in the output channel and obtain a second matching signal; The amplitude of the second matching signal is adjusted to be within a target dynamic range, and a scene adaptation signal is output.

7. The data collection method according to claim 1, wherein: The step of extracting edge features and compressing data on the scene adaptation signal to generate an industrial data packet, and uploading the industrial data packet to two output channels through a preset communication protocol includes: Extracting edge features from the scene adaptation signal to obtain a feature point sequence, including: a mutation point, an inflection point, and a steady-state interval of the signal; Determine a feature description table according to the feature point sequence and the scene adaptation signal, wherein the feature description table includes: a mean, a variance, and a trend coefficient; Performing data redundancy analysis using the feature description table to obtain a first data packet; Grouping and packaging the first data set according to timestamps, and adding channel identifiers and data type tags to generate second data packets; performing differential encoding and run-length encoding compression on the second data packet to generate an industrial data packet; Through a preset communication protocol, the industrial data packet is divided into main data and verification data, and transmitted in parallel through dual channels.

8. A data acquisition device, characterized in that: The data acquisition device is used to execute the data acquisition method according to any one of claims 1 to 7, and the device includes: The noise processing module is used to perform dynamic noise suppression processing on the analog input signals of the four input channels to obtain primary purified signals; A signal compensation module, configured to perform multi-stage nonlinear compensation on the primary purification signal based on a preset sensor type feature library to generate a standardized linear signal; A signal reconstruction module, configured to perform time-domain and space-domain joint reconstruction on the standardized linear signal and output a high-fidelity reconstructed signal; A signal adaptation module is used to obtain a parameter configuration library of a preset industrial scene mode, perform digital predistortion compensation on the high-fidelity reconstructed signal according to the parameter configuration library, and generate a scene adaptation signal; The protocol conversion module is used to extract edge features and compress data of the scene adaptation signal to generate an industrial data packet, and upload the industrial data packet to two output channels through a preset communication protocol.

Images