Industrial Internet of Things (IoT) device data acquisition methods, systems, equipment and media

Abstract

This invention discloses a method, system, device, and medium for data acquisition of industrial IoT devices, aiming to solve problems such as resource waste or loss of key features caused by fixed sampling frequencies. The method includes: acquiring raw signals and performing standardization processing; calculating the intensity of nonlinear fluctuations and prediction residuals of the signals, and evaluating the sampling completeness coefficient; extracting features to identify the device's operating modes; combining the completeness coefficient and mode identification results, calculating the target sampling frequency through multi-objective optimization and achieving seamless frequency switching; and adaptively compensating the evaluation model based on reconstruction errors. The system includes modules for data perception, sampling evaluation, mode identification, frequency decision-making, acquisition execution, and closed-loop feedback. This application achieves deep adaptation between the sampling frequency and signal characteristics, significantly reducing the communication and storage burden while ensuring reconstruction accuracy, and improving the system's adaptive capability and fault warning accuracy.

Description

Technical Field

[0001] This invention belongs to the field of industrial data processing in the Internet of Things (IoT), specifically, it relates to a method, system, device and medium for data acquisition from industrial IoT devices. Background Technology

[0002] With the rapid development of the Industrial Internet of Things (IIoT) and intelligent manufacturing technologies, real-time monitoring and health management of industrial equipment have become core elements for ensuring the continuous and stable operation of production lines. Industrial equipment generates a large amount of sensor data reflecting its physical state during operation. By collecting, transmitting, and analyzing this massive amount of data in real time, precise control over equipment operating conditions can be achieved, providing crucial data support for predictive maintenance, energy efficiency optimization, and fault early warning.

[0003] Setting and adjusting the data acquisition frequency is a crucial aspect of industrial monitoring systems, directly impacting the accuracy of data representation of the equipment's true operating status. In complex industrial production environments, equipment often operates under variable conditions. An ideal data acquisition mechanism should be able to adjust the sampling strategy in real time based on the dynamic characteristics of the equipment during operation, ensuring data integrity while minimizing system communication overhead and storage burden.

[0004] Existing technologies for data acquisition in industrial equipment suffer from several drawbacks: First, sampling strategies typically employ preset fixed frequencies, making it difficult to balance data accuracy with system resources. This results in a large amount of redundant data during stable equipment operation, while insufficient sampling density is provided during periods of drastic fluctuation in operating conditions. Second, the lack of a dynamic evaluation mechanism for sampling sufficiency prevents real-time assessment of the fit between the current sampling frequency and the rate of change of signal characteristics. Particularly during critical transient moments such as equipment start-up, shutdown, or initial fault occurrence, system response lags, making it difficult to quickly identify the contradiction that current low-frequency sampling is insufficient to capture key transient features. Third, the lack of deep linkage between data acquisition and state recognition leads to the loss of critical data at points of abrupt state changes, severely impacting the timeliness and accuracy of subsequent fault diagnosis and increasing safety risks to equipment operation. Therefore, developing an optimized method and system for data acquisition in industrial IoT devices based on sensor arrays is of significant practical importance, addressing the shortcomings of existing technologies such as lag in sampling frequency adjustment, lack of real-time evaluation of sampling sufficiency, and easy loss of key transient features. Summary of the Invention

[0005] The purpose of this invention is to provide a data acquisition method, system, device, and medium for industrial Internet of Things (IoT) devices. It mainly addresses the core technical contradictions such as resource waste or loss of key features caused by fixed data acquisition frequency under varying operating conditions of industrial equipment, lack of real-time evaluation mechanism for sampling sufficiency, and disconnect between acquisition strategy and equipment status identification.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0007] A method for data acquisition from industrial IoT devices includes the following steps:

[0008] S1, through a sensor array deployed in key parts of industrial equipment, collects raw physical signals reflecting the operating status of the equipment in real time, performs multi-channel synchronous sampling and signal standardization processing, and constructs the raw observation sequence;

[0009] S2, the original observation sequence is segmented using the sliding window technique, and the nonlinear fluctuation intensity and prediction residual of the signal within each window are calculated. Combined with the dynamic constraints of the sampling theorem, the sampling completeness coefficient at the current sampling frequency is evaluated. ;

[0010] S3, extract the multidimensional feature vector of the original observation sequence, identify the current operating mode of the device through a preset state classification model, and determine the accuracy requirement weight coefficients under different modes. and resource consumption weighting coefficient ;

[0011] S4, the sampling completeness coefficient The results of the operational modality recognition are input into the frequency decision model to establish a model based on the sampling completeness coefficient. The accuracy requirement weighting coefficient The resource consumption weighting coefficient The integrated cost function of the operation mode identification results; the target sampling frequency is calculated using a hysteresis comparator under the condition of satisfying the minimum reconstruction accuracy constraint; wherein, the operation mode identification results include stationary operation mode, transient fluctuation mode and fault latent mode;

[0012] S5, reset the timer load value of the hardware acquisition controller according to the target sampling frequency, perform the sampling frequency switching operation using the double buffer structure, and insert a synchronization frame into the frequency conversion sampling data stream. The synchronization frame includes a frequency identifier, a timestamp, and a sampling point number.

[0013] S6. Compare the goodness of fit between the reconstructed signal and the original sampling points. If the goodness of fit is lower than a preset threshold, then decrease the nonlinear fluctuation intensity judgment threshold by a preset step size and perform adaptive compensation of the evaluation model parameters.

[0014] Furthermore, in S1, the specific process of signal normalization includes:

[0015] The analog signals uploaded by the multi-channel sensors are converted from analog to digital, and the conversion accuracy is set to a preset bit width.

[0016] Pulse-type random interference in the original physical state signal is removed using a filtering algorithm;

[0017] The signal is decomposed into approximate components and detail components using a wavelet decomposition algorithm with a preset number of layers. The approximate components correspond to the low-frequency trend term, and the detail components correspond to the high-frequency detail term.

[0018] By using resampling technology, signals with different physical ranges are mapped to a preset normalization interval, ensuring the dimensional consistency of multi-source data.

[0019] Furthermore, the specific process of S2 is as follows:

[0020] S21, within the sliding window, the theoretical values ​​of subsequent sampling points are predicted using an interpolation polynomial;

[0021] S22, the absolute deviation between the actual observed value and the theoretical predicted value is defined as the nonlinear fluctuation intensity of the signal;

[0022] S23. If the ratio of the current sampling frequency to the highest effective frequency component is less than a preset multiple, or the absolute deviation exceeds the preset ratio of the original signal standard deviation, then the sampling completeness coefficient is determined to be lower than the preset safety threshold, triggering a request to increase the sampling frequency.

[0023] Furthermore, in S3, the specific process of identifying the current operating mode of the device is as follows:

[0024] S31, Calculate the statistical characteristics of the signal within the sliding window and make a determination based on the power spectral density distribution:

[0025] When the variance volatility is less than a preset volatility threshold and the power spectrum is concentrated in the low frequency band, the operating mode is determined to be a stationary operating mode.

[0026] When the absolute value of the signal slope exceeds a preset rate of change threshold for a predetermined number of sampling periods, the operating mode is determined to be a transient fluctuation mode.

[0027] When the statistical indicators exceed the preset threshold and abnormal frequency components appear, the operating mode is determined to be a fault latent mode.

[0028] S32, for the stable operation mode, set a first preset accuracy weight; for the transient fluctuation mode and the fault latency mode, set a second preset accuracy weight.

[0029] Further, in S4, in the comprehensive cost function, the accuracy loss term is set to be inversely proportional to the sampling completeness coefficient, and the resource consumption term is set to be directly proportional to the sampling frequency; the weight coefficients of the accuracy loss term and the resource consumption term are dynamically allocated according to the operating mode identification result; the set of sampling frequencies includes multiple preset frequency levels; when the identified operating mode is the transient fluctuation mode or the fault latency mode, the weight coefficient of the accuracy loss term is increased, and the target sampling frequency is increased according to a preset adjustment ratio until the sampling completeness coefficient recovers to above the preset recovery threshold.

[0030] Furthermore, the process of performing the sampling frequency switching operation using the double buffer structure is as follows:

[0031] In the interrupt service subroutine of the hardware acquisition controller, the new frequency division value calculated based on the target sampling frequency is written into the comparison register of the hardware timer, so that the sampling interval of the next batch of data is executed according to the new frequency.

[0032] The packet is encapsulated using a variable-length frame structure, and each packet header contains a frequency identifier field of a preset width and a timestamp of a preset width.

[0033] At the moment of frequency switching, a synchronization frame is inserted into the data stream, and the synchronization frame clearly identifies the time reference of the frequency jump point.

[0034] Furthermore, in S6, the adaptive compensation mechanism specifically means: if the goodness of fit between the reconstructed signal and the original sampling points is lower than a preset fitting threshold within a predetermined number of evaluation periods, the weight coefficient used to calculate the deviation is automatically reduced, and the threshold for triggering frequency increase is lowered; if the communication bandwidth occupancy rate is above a preset high threshold for a long period of time and the data reconstruction accuracy is higher than a preset accuracy threshold, the weight of the resource consumption item in the comprehensive cost function is increased, inducing the system to reduce the sampling frequency.

[0035] This invention also provides an industrial IoT device data acquisition system based on a sensor array, used to implement the above-mentioned data acquisition method, comprising:

[0036] The data sensing and preprocessing module is used to capture multi-source physical state signals of industrial equipment in real time.

[0037] The sampling adequacy dynamic evaluation module includes an interpolation unit, an FFT transformation unit, and a weighted summation circuit. The interpolation unit performs fourth-order Lagrange fitting to obtain the prediction residual, the FFT transformation unit extracts the highest effective frequency of the signal, and the weighted summation circuit calculates and outputs the sampling completeness coefficient according to the preset weights.

[0038] The equipment operation mode recognition module is used to identify the current operation mode of the equipment;

[0039] The sampling frequency dynamic decision module is used to receive the sampling completeness coefficient and the operating mode identification result to find the target sampling frequency;

[0040] The adaptive acquisition and execution module can dynamically reset the timer load value of the hardware acquisition controller to match the target sampling frequency. It achieves seamless switching of sampling frequency by calculating the clock division coefficient of the next cycle when the current buffer is filled through a dual buffer structure, and adds frequency identifiers and timestamps to the frequency-converted sampling data to complete the encapsulation process.

[0041] The closed-loop feedback and performance monitoring module is used to correct the evaluation model parameters in the sampling adequacy dynamic evaluation module online.

[0042] The present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program that can run on the processor, and the processor executes the computer program to implement the above-described data acquisition method.

[0043] On the other hand, the present invention also provides a computer-readable storage medium storing a computer program that can be loaded by a processor and executed by the above-described data method.

[0044] Compared with the prior art, the present invention has the following beneficial effects:

[0045] (1) This invention completely solves the contradiction between “resource waste” and “loss of key features” caused by fixed sampling frequency by constructing a sampling completeness coefficient evaluation model and a multi-objective optimization frequency decision mechanism. The system can dynamically calculate the optimal sampling frequency according to the real-time operating status of industrial equipment: during the stable power generation stage of wind turbine generator, the sampling frequency is automatically reduced from 20kHz to 2kHz, the data compression rate reaches 90%, and the wireless communication link congestion in remote wind farms is greatly reduced; during the fault latency stage, the sampling frequency is immediately increased to 20kHz to fully capture key features such as bearing impact pulses, and a deep adaptation between sampling strategy and signal characteristics is achieved.

[0046] (2) This invention uses sliding window technology to calculate the intensity of nonlinear fluctuations and prediction residuals of signals in real time, and combines it with an adaptive compensation mechanism triggered by reconstruction error to form a closed-loop system of "sampling evaluation - frequency decision - execution feedback - model correction". When the pump group experiences an initial cavitation failure, the system can sense changes in signal complexity through the sample entropy index and trigger high-frequency sampling requirements in advance without waiting for significant changes in the time domain amplitude. At the same time, by continuously monitoring bandwidth utilization and goodness of fit, the system automatically adjusts the evaluation model parameters, enabling the system to maintain optimal sampling performance in scenarios such as equipment aging and changes in environmental noise, thus improving its adaptive capability compared to traditional systems.

[0047] (3) This invention breaks through the technical bottleneck of the disconnect between traditional acquisition strategies and equipment status identification. By extracting multi-dimensional features in the time domain, frequency domain, and spatial domain to identify equipment operating modes, and inputting the mode results and sampling completeness coefficients into the frequency decision model, it achieves precise linkage between acquisition strategies and equipment status. Under the abnormal chatter mode of CNC machine tools, the system automatically increases the weight of data accuracy requirements to 0.9, prioritizing the acquisition of high-frequency chatter signals; under the steady-state operation mode of pump stations, the weight is reduced to 0.3, prioritizing the saving of system resources. This linkage mechanism enables fault warning time to be several weeks earlier than the fixed sampling mode, and the fault diagnosis accuracy rate is increased to over 98%, providing core technical support for predictive maintenance of industrial equipment.

[0048] (4) This invention achieves uninterrupted switching of sampling frequency through a dual-buffer structure and a synchronization frame mechanism. The timer division value is dynamically updated in the interrupt service subroutine of the hardware acquisition controller. Combined with variable-length frame encapsulation and high-precision timestamps, the data stream is ensured to be uninterrupted during frequency switching. At the same time, the time reference for the frequency jump point is clearly identified by the synchronization frame, which enables the back-end analysis software to accurately reconstruct non-equal interval sampling signals, eliminate the risk of spectrum aliasing, and ensure the integrity and continuity of data acquisition under all operating conditions of industrial equipment. Attached Figure Description

[0049] Figure 1 This is a control flowchart of the method of the present invention.

[0050] Figure 2 This is a hardware architecture diagram of the system of the present invention. Detailed Implementation

[0051] The present invention will be further described below with reference to the accompanying drawings and embodiments. The embodiments of the present invention include, but are not limited to, the following embodiments.

[0052] Example 1

[0053] like Figure 1 As shown, this embodiment discloses a data acquisition method for industrial IoT devices. This method achieves maximum compression of redundant data while ensuring data reconstruction accuracy through real-time sensing of the operating status of industrial equipment and dynamic optimization of the sampling strategy. In practical applications, taking the vibration monitoring of the main shaft of a large wind turbine generator as an example, the frequency components and evolution characteristics of the vibration signal differ significantly during the normal stable power generation phase, the transient adjustment phase due to sudden wind speed changes, and the latent fault phase where bearing fatigue spalling occurs. This invention is based on an IoT system including an object platform, a sensor network platform, and a management platform. The object platform directly acquires the physical signals of the equipment and executes low-level sampling control, serving as the front end of data acquisition. The corresponding modules are as follows:

[0054] Data Sensing and Preprocessing Module: This module captures multi-source physical status signals from industrial equipment in real time. It serves as the entry point for data into the system, directly interfacing with the equipment, and is a core data acquisition module of the object platform.

[0055] Adaptive Acquisition Execution Module: Directly operates the hardware acquisition controller, resets the timer load value, achieves seamless switching of sampling frequency, and encapsulates the sampled data. It belongs to the underlying execution module of the object platform and is responsible for executing acquisition instructions and processing raw sampled data.

[0056] The sensor network platform receives data collected by the target platform, performs preprocessing, computational analysis, modality recognition, and data acquisition frequency decisions. It serves as the core hub connecting the target platform and the management platform. The corresponding modules are as follows:

[0057] The sampling adequacy dynamic evaluation module includes an interpolation unit, an FFT transformation unit, and a weighted summation circuit. It processes the acquired signals and outputs sampling completeness coefficients. It belongs to the data processing and evaluation module of the sensor network platform.

[0058] Equipment operating mode identification module: Identifies the current operating mode of the equipment, providing a basis for frequency acquisition decisions. It belongs to the state identification and analysis module of the sensor network platform.

[0059] The dynamic decision-making module for sampling frequency receives the sampling completeness coefficient and the operating mode recognition results, calculates and determines the target sampling frequency, and belongs to the decision-making module of the sensor network platform, providing instruction support for the data acquisition execution of the target platform.

[0060] The management platform monitors the overall system performance, corrects core model parameters, and coordinates global operations. The corresponding modules are as follows:

[0061] Closed-loop feedback and performance monitoring module: It corrects the evaluation model parameters of the sampling sufficiency dynamic evaluation module online, monitors the sampling and evaluation performance of the system, and belongs to the performance control and model optimization module of the management platform. It is responsible for the stable operation and optimization of the entire system.

[0062] This invention performs efficient data acquisition tasks through the following steps:

[0063] S1, Data Acquisition and Preprocessing. The system uses an array of industrial equipment sensors deployed on the main bearing, gearbox, and key parts of the wind turbine generator to capture raw physical signals reflecting the operating status of the equipment in real time. These sensors include triaxial vibration accelerometers, high-precision displacement sensors, and resistance temperature sensors. The analog current or voltage signals acquired by the sensor array are transmitted to the data sensing and preprocessing module via shielded twisted-pair cables. Inside the data sensing and preprocessing module, a multi-channel synchronous sampling circuit performs high-bit-width analog-to-digital conversion on the analog signals. In this embodiment, the conversion accuracy is set to 24 bits to ensure that weak signal fluctuations are preserved over a wide measurement range. To eliminate high-frequency aliasing interference, a low-pass anti-aliasing filter with an adjustable cutoff frequency is integrated at the front end of the sampling circuit.

[0064] In the signal standardization process, the system first performs a composite filtering operation, using a five-point median filtering algorithm to remove pulse-type random noise caused by electromagnetic interference, and then smooths the signal baseline using a moving average filter. The core of the standardization process lies in using a wavelet decomposition algorithm with a preset number of levels to strip the signal. Specifically, the system uses the Mallat algorithm structure, selecting the db4 wavelet basis function to perform a three-level decomposition on the original signal, thereby obtaining one approximate component and three detail components. The approximate component represents the fundamental frequency and low-frequency trend term of the spindle rotation, while the detail components capture high-frequency transient terms such as bearing impact and gear meshing. To eliminate the influence of differences in the measurement ranges of different sensors on subsequent calculations, the system performs resampling normalization through a linear mapping function, mapping the original physical quantities to intervals. In addition, to address the potential phase asynchrony problem in multi-channel data, the system calculates the time delay deviation by calculating the cross-correlation function between each channel and performs time delay compensation in the digital domain, ultimately constructing an original observation sequence with spatiotemporal consistency.

[0065] S2, Sampling Completeness Coefficient Evaluation. This process involves using an interpolation unit to perform fourth-order Lagrange fitting to obtain the prediction residual, an FFT transform unit to extract the highest effective frequency of the signal, and a weighted summation circuit to calculate and output the sampling completeness coefficient according to preset weights. Specifically, the system uses a sliding window technique to segment the original observation sequence, constructing sliding window segments. The length of the sliding window is adaptively configured based on the minimum operating speed of the wind turbine main shaft, ensuring that each window covers at least two complete rotation cycles, thereby guaranteeing the stability of the statistical characteristics. The sampling sufficiency dynamic evaluation module performs nonlinear fluctuation intensity / prediction residual calculation within each window. Specifically, within the current window, the system uses a fourth-order Lagrange interpolation polynomial to fit the implicit trajectory between known sampling points and predicts the theoretical value at the next sampling time. The system defines the absolute deviation between the actually observed sampling value and the theoretical predicted value as the nonlinear fluctuation intensity of the signal. This intensity quantifies the severity of the signal deviating from the smooth evolution pattern.

[0066] To further accurately evaluate sampling density, the sampling adequacy dynamic evaluation module also introduces dynamic constraints based on Shannon's sampling theorem. The system obtains the power spectrum distribution of the signal within the current window through Fast Fourier Transform and identifies the highest effective frequency component. The system calculates the difference between the current sampling frequency and twice the highest effective frequency to obtain the sampling redundancy. The sampling completeness coefficient is generated by a weighted fusion algorithm using nonlinear fluctuation intensity, prediction residual, and sampling redundancy. The specific calculation formula is as follows:

[0067]

[0068] in, The sampling completeness coefficient, To predict residuals, The standard deviation of the signal. For nonlinear wave intensity, This is the current sampling frequency; under stable operating conditions, , , These are the weighting coefficients corresponding to the nonlinear fluctuation intensity, prediction residual, and sampling redundancy, respectively, and satisfy the following conditions: + In this embodiment, a weight vector is set. for .

[0069] When the equipment is operating stably, the signal prediction residual is extremely small, and the sampling redundancy is high, resulting in a high sampling completeness coefficient. However, if the equipment experiences a sudden load change or an early, minor impact, the prediction residual increases rapidly. If the sampling frequency is insufficient to reconstruct the second derivative characteristics of the signal, the sampling completeness coefficient is determined to be below a preset safety threshold. The system then generates trigger logic to increase the sampling frequency to prevent the loss of critical transient features due to low-frequency sampling.

[0070] S3, Operational Mode Recognition. The equipment operational mode recognition module simultaneously extracts multi-dimensional feature vectors from the original observation sequence in the time, frequency, and spatial domains. Time-domain features include root mean square (RMS) value, peak factor, impulse index, and kurtosis coefficient, among which the kurtosis coefficient has extremely high sensitivity for identifying early pitting faults in bearings. Frequency-domain features include centroid frequency, MMS frequency, and frequency standard deviation, used to characterize the distribution and evolution of vibration energy in the frequency spectrum. Spatial features are obtained through coherence analysis of signals from different installation positions in the sensor array.

[0071] The system inputs the extracted multidimensional feature vectors into a pre-defined state classification model. This model employs a random forest algorithm, configured with 120 decision trees, a maximum depth of 15 layers, and a minimum of 5 leaf node samples. The model is trained on 3000 sets of vibration feature vectors labeled with modal tags, generating a forest of decision trees that includes node splitting thresholds. The feature vector to be tested is input into this forest for voting, and the mode category with the highest probability is output. The classification model outputs the probability distribution of the current device's mode and determines the required accuracy weight accordingly. Under stable operating modes, the system prioritizes resource conservation, allocating a lower required accuracy weight (first pre-defined accuracy weight). Under transient fluctuation modes (such as wind turbine start-up and shutdown, pitch adjustment) or latent fault modes (such as detecting abnormal harmonic energy at a characteristic frequency), the system allocates a higher required accuracy weight (second pre-defined accuracy weight). This deep linkage between mode and acquisition strategy ensures that the monitoring system can dedicate all computing and transmission resources for refined monitoring at critical moments.

[0072] S4, Frequency Dynamic Decision. The frequency dynamic decision module receives the sampling completeness coefficients and accuracy requirement weights from the previous stage and constructs a comprehensive cost function:

[0073]

[0074] in, For the selected sampling frequency, To support the highest frequency in hardware, Let λ be the switching penalty coefficient, where λ ∈ [0.1, 0.2]. For accuracy requirements, weighting coefficients, This is the resource consumption weighting coefficient. In the steady-state mode: α∈[0.2,0.4], β∈[0.6,0.8], in the transient fluctuation / fault latency mode: α∈[0.8,0.95], β∈[0.05,0.2].

[0075] The cost function consists of a weighted sum of an accuracy loss term and a resource consumption term. The accuracy loss term is inversely proportional to the sampling completeness coefficient, representing the risk of information distortion due to reduced sampling frequency; the resource consumption term is directly proportional to the sampling frequency, representing communication bandwidth usage and storage costs. The system dynamically adjusts the ratios of these two terms based on the accuracy requirement weights. When a latent fault mode is identified, the system significantly increases the weight of the accuracy loss term, inducing the decision logic to shift towards higher frequency sampling.

[0076] The multi-objective optimization solver integrated within the frequency acquisition dynamic decision module performs a heuristic search within a preset frequency set (e.g., 1kHz, 2kHz, 5kHz, 10kHz, 20kHz) to find the target sampling frequency that minimizes the comprehensive cost function. To prevent frequent oscillations in the sampling frequency under critical operating conditions, a hysteresis comparator based on a switching penalty term is added to the decision logic. This hysteresis comparator sets a switching cost threshold of 10% of the current cost function value. At this time, the hardware controller executes the frequency change command. The frequency change command is only executed when the performance improvement brought by the new frequency exceeds the switching cost threshold. The finally determined target sampling frequency is sent to the execution end.

[0077] S5, Adaptive Acquisition and Execution. The adaptive acquisition and execution module is responsible for translating decision results into hardware-level control actions. The hardware acquisition controller internally employs a hardware timer / dual-buffer structure. When a new target sampling frequency is received, the system calculates the corresponding timer division value in the interrupt service routine and writes it into the hardware timer's compare register. To achieve seamless frequency switching, the system utilizes a dual-buffer mechanism: while the CPU processes the data in the current buffer, the DMA controller stores the newly acquired data into another buffer. At the moment of buffer switching, the new sampling interval takes effect immediately, ensuring that the data stream does not break on the timeline.

[0078] The adaptive acquisition and execution module encapsulates data according to a unified variable-length frame protocol. In the encapsulated frequency-converted data stream, each data packet header includes a frequency identifier field of preset width and a high-precision timestamp. At the instant of a frequency jump, the system inserts a special synchronization frame into the data stream. This frame records the specific sampling point number of the frequency switch and the system's absolute time reference. This encapsulation method allows the backend analysis software to reconstruct non-equally spaced sampled signals based on the frequency identifier when reading the data, thereby obtaining the correct amplitude-frequency correspondence in spectrum analysis.

[0079] S6, Closed-Loop Feedback and Compensation. The closed-loop feedback and performance monitoring module continuously monitors system performance. The system evaluates resource saving effectiveness by statistically analyzing bandwidth utilization and storage increment per unit time. Simultaneously, the system performs signal reconstruction in the backend, using a cubic spline interpolation algorithm to restore the continuous shape of the signal and calculating the goodness of fit between the reconstructed curve and the original high-precision sampling points. This embodiment uses the coefficient of determination as the evaluation index. The system sets a fitting threshold of 0.92. If the goodness of fit is lower than this value for three consecutive periods, the nonlinear fluctuation intensity judgment threshold is decreased by a fixed step size of 5%. The goodness of fit was maintained until it rose above 0.95.

[0080] Conversely, if the bandwidth remains saturated for an extended period and the reconstruction accuracy far exceeds the requirements, the system increases the weight of resource consumption items. Through this closed-loop feedback, the system can automatically calibrate and evaluate parameters based on sensor aging or environmental noise levels, maintaining optimal acquisition performance over the long term.

[0081] The method provided by this invention demonstrates significant technical advantages in practical wind power monitoring systems. During months of operational testing, when the wind turbine is in a constant-speed, stable power generation phase, the system automatically reduces the sampling frequency from the initial 20kHz to 2kHz, achieving a data compression rate of 90%, greatly alleviating the congestion problem of wireless communication links in remote wind farms. Furthermore, when minor fatigue pitting in the bearing causes periodic, weak impacts in the vibration signal, the sampling adequacy dynamic evaluation module detects an abnormal increase in the prediction residual within tens of milliseconds. The sampling frequency dynamic decision module immediately instructs the hardware to switch to a 20kHz high-frequency acquisition mode, successfully capturing the complete impact pulse waveform. This provides high-quality raw data for the subsequent envelope spectrum analysis, advancing the fault warning time by several weeks compared to the fixed sampling mode.

[0082] Furthermore, this invention solves the phase misalignment problem commonly encountered in traditional frequency conversion acquisition processes through a dual-buffer structure in the adaptive acquisition and execution module. At the moment of frequency switching, the introduction of a precise synchronization frame and frequency identifier allows the fast Fourier transform executed by the back-end algorithm to eliminate spurious frequency components caused by variations in the sampling interval, ensuring the reliability of fault diagnosis results. The closed-loop feedback and performance monitoring module ensures that the system maintains consistent evaluation sensitivity across different seasons and wind speeds, reducing the frequency of manual intervention for debugging.

[0083] Example 2

[0084] This embodiment uses vibration monitoring and health management of a large CNC machine tool spindle system as a specific application scenario. During high-speed cutting, the spindle speed changes frequently, and the cutting force changes in real time with the workpiece material hardness and tool wear, resulting in highly non-stationary vibration signals. Traditional fixed-frequency sampling generates massive amounts of redundant data during the stable cutting phase, while in the transient moments when the tool enters the workpiece or chatter occurs, the sampling frequency is insufficient, often failing to capture high-frequency impact characteristics.

[0085] This invention discloses a data acquisition method for industrial IoT devices, such as... Figure 2As shown, the hardware system upon which this method relies includes a data sensing and preprocessing module, whose input is connected to a triaxial piezoelectric accelerometer and an eddy current displacement sensor array mounted on the spindle bearing housing. The data sensing and preprocessing module captures the raw physical state signal at an initial 100kHz reference sampling rate using a 24-bit high-precision analog-to-digital converter chip. In the signal standardization stage, the module first performs multi-channel synchronous alignment to ensure absolute consistency of spatial features across the time axis. Subsequently, a 5-layer wavelet decomposition algorithm is used to denoise the raw signal, employing the db4 wavelet basis function to decompose the signal into approximate and detail components, retaining the low-frequency trend term reflecting spindle rotation accuracy and the high-frequency impact component reflecting early bearing failure. The processed data is mapped to a normalized interval of 0 to 1, forming a standardized data stream to be evaluated.

[0086] The sampling completeness dynamic evaluation module is connected to the output of the data sensing and preprocessing module. This module utilizes a sliding window technique, setting the window length to 1024 sampling points with an overlap rate of 50%. Within each window, the module uses a third-order Lagrange interpolation polynomial to predict the theoretical value of the ninth sampling point based on the first eight known sampling points. By calculating the absolute deviation between the actual observed value and the theoretical prediction, the nonlinear fluctuation intensity of the signal is quantified. Simultaneously, the module performs a Fast Fourier Transform on the data within the window to extract its power spectral density distribution. The calculation of the sampling completeness coefficient comprehensively considers reconstruction error and spectral aliasing risk.

[0087] The equipment operation mode recognition module synchronously extracts the time-domain statistical features of the data stream, including variance, skewness, kurtosis, and peak factor. Using a pre-defined random forest classification model, it determines whether the spindle is currently in standby mode, stable cutting mode, or abnormal chatter mode. For stable cutting mode, the system sets the data accuracy requirement weight to 0.3; for abnormal chatter or tool entry transient modes, the weight is increased to 0.9.

[0088] The frequency dynamic decision-making module receives the sampling completeness coefficient and operating mode identification results, and uses a multi-objective optimization algorithm to find a balance point in a preset frequency set. The frequency set includes five magnitudes: 1kHz, 5kHz, 10kHz, 20kHz, and 50kHz. The decision-making process follows the comprehensive cost function:

[0089]

[0090] in, For the overall cost, Let C be the sampling frequency to be selected, and C be the sampling completeness coefficient. The highest frequency supported by the hardware. Indicates the target sampling frequency. Indicates the current sampling frequency. Weights are assigned based on accuracy requirements. Resource consumption weighting coefficient The system has a preset weight mapping table. When the modal identification result is "stable cutting", the weight is set... When the identification result is "abnormal flutter", set At this point, the cost function shifts in the direction of reducing reconstruction error.

[0091] The adaptive acquisition and execution module internally maintains a dual-buffer structure, each with a capacity of 4MB. While filling the current buffer, the controller calculates the clock division factor for the next cycle based on the frequency adjustment instruction. When a buffer full interrupt is triggered, the controller writes the new division value to the hardware timer's compare register in the interrupt service routine, achieving seamless switching of the sampling frequency. Each data packet header contains a 16-bit frequency identifier field and a 64-bit high-precision timestamp, ensuring that the backend can reconstruct an accurate time series using a non-uniform sampling signal reconstruction algorithm.

[0092] The closed-loop feedback and performance monitoring module monitors the system's communication bandwidth utilization in real time. If the machine tool workshop's LAN load exceeds 80% and the current data reconstruction accuracy remains above 98%, the module will automatically increase the weight beta of the resource consumption item, forcing the decision-making module to reduce the sampling frequency while ensuring basic monitoring needs. Conversely, if the goodness of fit between the reconstructed signal and the original sampling points is detected to be below 90%, adaptive compensation of the evaluation model parameters is triggered, lowering the threshold for increasing the trigger frequency.

[0093] The specific steps of the data acquisition method implemented in this method are as follows:

[0094] Step S110 involves acquiring raw physical signals reflecting spindle vibration, displacement, and temperature in real time using a sensor array deployed at key parts of the machine tool spindle. Multi-channel synchronous sampling is performed, with an external clock synchronization signal ensuring the sampling time difference between channels is controlled within 10 nanoseconds. The acquired analog voltage signals are range-scaled, and an anti-aliasing filter removes noise components higher than 50kHz to construct the original observation sequence.

[0095] Step S120: The original observation sequence is segmented using a sliding window of 2048 points. The nonlinear fluctuation intensity of the signal within each window is calculated, specifically by calculating the mean absolute value of the second derivative. Based on the dynamic constraints of the sampling theorem, if the current sampling frequency is less than 5 times the highest effective frequency of the signal, the sampling completeness coefficient is determined to be decreasing.

[0096] Step S130: Extract the kurtosis index in the time domain and the centroid frequency in the frequency domain from the original observation sequence. Identify the spindle operating mode using a support vector machine model. During the stable cutting phase, the signal exhibits obvious periodic characteristics; when the tool breaks or chatter occurs, the kurtosis index increases significantly. Based on the identification results, determine the data accuracy requirement weights for different modes.

[0097] Step S140: Input the sampling completeness coefficient and the operating mode identification result into the frequency decision model. Under the constraint that the reconstruction error is less than 5%, find the target sampling frequency that minimizes the sum of system computational overhead and transmission bandwidth. If the current mode is in a fault-latent state, force the target sampling frequency to be locked at the highest order of magnitude.

[0098] Step S150: Dynamically reset the timer load value of the hardware acquisition controller according to the target sampling frequency. At the moment of frequency switching, the system inserts a synchronization frame to identify the exact time of the frequency jump. The frequency-converted data is encapsulated according to a unified industrial Ethernet protocol, including device identifier, serial number, sampling rate code, and checksum.

[0099] Step S160: Statistical data is stored incrementally in real time at the edge. The goodness of fit between the reconstructed signal and the original sampling points is compared. If the goodness of fit is lower than a preset threshold for five consecutive evaluation periods, the deviation weight coefficient in the sampling completeness coefficient calculation formula is automatically reduced to improve the system's sensitivity to signal fluctuations.

[0100] Example 3

[0101] This embodiment addresses the scenario of monitoring rotating machinery in a remote, unmanned pumping station. This scenario is characterized by limited communication bandwidth and the equipment operating in a steady state for extended periods, but signal characteristics evolve extremely rapidly once a fault occurs. Compared to Embodiment 1, this embodiment introduces signal entropy features into the sampling adequacy evaluation algorithm to more sensitively perceive nonlinear dynamic changes within the signal.

[0102] This embodiment provides an industrial IoT device data acquisition and optimization system based on a sensor array, wherein the data sensing and preprocessing module is connected to the pressure, flow and vibration sensors of the pump group.

[0103] In step S210, the system acquires multi-source physical quantity signals and uses a median filtering algorithm to eliminate electromagnetic pulse interference in the environment.

[0104] In step S220, the system constructs a dynamic evaluation model for sampling sufficiency based on sample entropy. Sample entropy, as a measure of signal complexity, can effectively reflect the unpredictability of a signal. The formula for calculating sample entropy is as follows:

[0105]

[0106] Where m is the embedding dimension, r is the similarity tolerance, N is the data length, and A and B are the number of matching template vectors, respectively. When cavitation or bearing spalling occurs in the pump unit in its early stages, the signal's sample entropy increases significantly, indicating that the signal contains more nonlinear dynamic information. At this time, the system automatically lowers the sampling completeness coefficient, thereby triggering a higher frequency sampling requirement, even if the signal's time-domain amplitude has not yet changed significantly.

[0107] Step S230: The signal features are classified in real time using a classification algorithm to determine whether the equipment is in a startup, shutdown, steady-state operation, or abnormal fluctuation mode. For the unique startup and shutdown process of the pump station, the system has a pre-set transition state identification logic that predicts state switching in advance by monitoring the frequency change rate of the motor current.

[0108] In step S240, when determining the optimal sampling frequency, the remaining energy of the sensor nodes and the current congestion level of the 4G / 5G network are additionally considered. When the network packet loss rate exceeds 5%, the system initiates a secondary frequency reduction strategy. While ensuring the minimum requirements for equipment safety monitoring, the sampling frequency is further reduced for devices operating in a steady-state mode, prioritizing the data transmission bandwidth of devices operating in abnormal fluctuation modes.

[0109] Step S250: Drive the hardware acquisition circuit to perform frequency switching. The data encapsulation protocol adopts a variable-length frame structure, and the metadata defined in the packet header includes a 4-bit sampling frequency code. At the moment of frequency switching, the system inserts a synchronization frame to clearly identify the time reference of the frequency jump point, ensuring that the time axis of multi-channel data is absolutely consistent during back-end fusion.

[0110] Step S260: Perform a data quality assessment at the monitoring center. The effectiveness of the current sampling strategy is evaluated by calculating the cross-correlation coefficient between the reconstructed signal and the actual trajectory. If the cross-correlation coefficient remains above 0.99 for an extended period, it indicates redundancy in the current sampling frequency. The system will then issue a control command to attempt to further reduce the reference sampling frequency in the next steady-state cycle.

[0111] The edge computing gateway in this embodiment adopts a multi-core parallel processing architecture. Two cores are dedicated to real-time acquisition of high-frequency signals, DMA transfer, and interrupt handling, ensuring no data loss during the acquisition process. The other two cores are responsible for sample entropy calculation, modality recognition algorithm, and frequency decision calculation. This architecture guarantees the determinism of the acquisition process, controlling the delay from signal characteristic change to frequency adjustment command taking effect to within 10 milliseconds.

[0112] The high-reliability storage array categorizes data according to frequency characteristics. High-frequency transient data (such as data within 5 seconds before and after a failure) is stored in the high-speed solid-state storage area, serving as high-precision evidence for fault tracing. Low-frequency steady-state data, after Huffman compression, is stored in the mechanical hard disk area for long-term trend analysis. The visual configuration terminal allows maintenance personnel to set the baseline sampling frequency range from 100Hz to 20kHz for different pump models.

[0113] In summary, this invention fundamentally resolves the technical contradiction between accuracy and efficiency in industrial equipment data acquisition by constructing a closed-loop sampling optimization system with self-sensing and adjustment capabilities. In scenarios such as high-speed cutting machine tools and remotely operating pump stations, this invention provides excellent data acquisition quality and system operating efficiency.

[0114] The above embodiments are merely one of the preferred embodiments of the present invention and should not be used to limit the scope of protection of the present invention. Any modifications or refinements made to the main design concept and spirit of the present invention that are not of substantial significance, but solve the same technical problem as the present invention, should be included within the scope of protection of the present invention.

Claims

1. A method for data acquisition from industrial Internet of Things (IoT) devices, characterized in that, Includes the following steps: S1, through a sensor array deployed in key parts of industrial equipment, collects raw physical signals reflecting the operating status of the equipment in real time, performs multi-channel synchronous sampling and signal standardization processing, and constructs the raw observation sequence; S2, the original observation sequence is segmented using the sliding window technique, and the nonlinear fluctuation intensity and prediction residual of the signal within each window are calculated. Combined with the dynamic constraints of the sampling theorem, the sampling completeness coefficient at the current sampling frequency is evaluated. ; S3, extract the multidimensional feature vector of the original observation sequence, identify the current operating mode of the device through a preset state classification model, and determine the accuracy requirement weight coefficients under different modes. and resource consumption weighting coefficient ; S4, the sampling completeness coefficient The results of the operational modality recognition are input into the frequency decision model to establish a model based on the sampling completeness coefficient. The accuracy requirement weighting coefficient The resource consumption weighting coefficient The integrated cost function of the operation mode identification results; the target sampling frequency is calculated using a hysteresis comparator under the condition of satisfying the minimum reconstruction accuracy constraint; wherein, the operation mode identification results include stationary operation mode, transient fluctuation mode and fault latent mode; S5, reset the timer load value of the hardware acquisition controller according to the target sampling frequency, perform the sampling frequency switching operation using the double buffer structure, and insert a synchronization frame into the frequency conversion sampling data stream. The synchronization frame includes a frequency identifier, a timestamp, and a sampling point number. S6. Compare the goodness of fit between the reconstructed signal and the original sampling points. If the goodness of fit is lower than a preset threshold, then decrease the nonlinear fluctuation intensity judgment threshold by a preset step size and perform adaptive compensation of the evaluation model parameters.

2. The industrial IoT device data acquisition method according to claim 1, characterized in that, In S1, the specific process of signal normalization includes: The analog signals uploaded by the multi-channel sensors are converted from analog to digital, and the conversion accuracy is set to a preset bit width. Pulse-type random interference in the original physical state signal is removed using a filtering algorithm; The signal is decomposed into approximate components and detail components using a wavelet decomposition algorithm with a preset number of layers. The approximate components correspond to the low-frequency trend term, and the detail components correspond to the high-frequency detail term. By using resampling technology, signals with different physical ranges are mapped to a preset normalization interval, ensuring the dimensional consistency of multi-source data.

3. The industrial IoT device data acquisition method according to claim 1, characterized in that, The specific process of S2 is as follows: S21, within the sliding window, the theoretical values ​​of subsequent sampling points are predicted using an interpolation polynomial; S22, the absolute deviation between the actual observed value and the theoretical predicted value is defined as the nonlinear fluctuation intensity of the signal; S23. If the ratio of the current sampling frequency to the highest effective frequency component is less than a preset multiple, or the absolute deviation exceeds the preset ratio of the original signal standard deviation, then the sampling completeness coefficient is determined to be lower than the preset safety threshold, triggering a request to increase the sampling frequency.

4. The industrial IoT device data acquisition method according to claim 1, characterized in that, In step S3, the specific process of identifying the current operating mode of the device is as follows: S31, Calculate the statistical characteristics of the signal within the sliding window and make a determination based on the power spectral density distribution: When the variance volatility is less than a preset volatility threshold and the power spectrum is concentrated in the low frequency band, the operating mode is determined to be a stationary operating mode. When the absolute value of the signal slope exceeds a preset rate of change threshold for a predetermined number of sampling periods, the operating mode is determined to be a transient fluctuation mode. When the statistical indicators exceed the preset threshold and abnormal frequency components appear, the operating mode is determined to be a fault latent mode. S32, for the stable operation mode, set a first preset accuracy weight; for the transient fluctuation mode and the fault latency mode, set a second preset accuracy weight.

5. The industrial IoT device data acquisition method according to claim 1, characterized in that, In step S4, the comprehensive cost function sets the accuracy loss term to be inversely proportional to the sampling completeness coefficient and the resource consumption term to be directly proportional to the sampling frequency. The weight coefficients of the accuracy loss term and the resource consumption term are dynamically allocated according to the operating mode identification results. The set of sampling frequencies includes multiple preset frequency levels. When the identified operating mode is the transient fluctuation mode or the fault latency mode, the weight coefficient of the accuracy loss term is increased, and the target sampling frequency is increased according to a preset adjustment ratio until the sampling completeness coefficient recovers to above the preset recovery threshold.

6. The industrial IoT device data acquisition method according to claim 1, characterized in that, The process of performing sampling frequency switching operation using a double buffer structure is as follows: In the interrupt service subroutine of the hardware acquisition controller, the new frequency division value calculated based on the target sampling frequency is written into the comparison register of the hardware timer, so that the sampling interval of the next batch of data is executed according to the new frequency. The packet is encapsulated using a variable-length frame structure, and each packet header contains a frequency identifier field of a preset width and a timestamp of a preset width. At the moment of frequency switching, a synchronization frame is inserted into the data stream, and the synchronization frame clearly identifies the time reference of the frequency jump point.

7. The industrial IoT device data acquisition method according to claim 1, characterized in that, In S6, the adaptive compensation mechanism is specifically as follows: if the goodness of fit between the reconstructed signal and the original sampling points is lower than a preset fitting threshold within a predetermined number of evaluation periods, the weight coefficient used to calculate the deviation is automatically reduced, and the threshold for triggering frequency increase is lowered; if the communication bandwidth occupancy rate is above a preset high threshold for a long period of time and the data reconstruction accuracy is higher than a preset accuracy threshold, the weight of the resource consumption item in the comprehensive cost function is increased, inducing the system to reduce the sampling frequency.

8. An industrial Internet of Things (IoT) device data acquisition system, characterized in that, The method for implementing the industrial Internet of Things (IoT) device data acquisition method as described in claim 7 includes: The data sensing and preprocessing module is used to capture multi-source physical state signals of industrial equipment in real time. The sampling adequacy dynamic evaluation module includes an interpolation unit, an FFT transformation unit, and a weighted summation circuit. The interpolation unit performs fourth-order Lagrange fitting to obtain the prediction residual, the FFT transformation unit extracts the highest effective frequency of the signal, and the weighted summation circuit calculates and outputs the sampling completeness coefficient according to the preset weights. The equipment operation mode recognition module is used to identify the current operation mode of the equipment; The sampling frequency dynamic decision module is used to receive the sampling completeness coefficient and the operating mode identification result to find the target sampling frequency; The adaptive acquisition and execution module can dynamically reset the timer load value of the hardware acquisition controller to match the target sampling frequency. It achieves seamless switching of sampling frequency by calculating the clock division coefficient of the next cycle when the current buffer is filled through a dual buffer structure, and adds frequency identifiers and timestamps to the frequency-converted sampling data to complete the encapsulation process. The closed-loop feedback and performance monitoring module is used to correct the evaluation model parameters in the sampling adequacy dynamic evaluation module online.

9. A computer device, characterized in that, The method includes a memory and a processor, wherein the memory stores a computer program that can run on the processor, and the processor executes the computer program to implement the method of any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer program is stored that can be loaded by a processor and execute the method of any one of claims 1 to 7.

Images