Intelligent Electromagnetic Compatibility Detection System and Method Based on Smart Sensors

Abstract

This invention discloses an intelligent electromagnetic compatibility (EMC) detection system and method based on smart sensors, belonging to the field of EMC detection technology. The system includes an input source acquisition module, a predictive analysis module, and an interference identification module. Its key technical points are: acquiring the ambient temperature of the area to be tested and simultaneously obtaining historical calibration data of the active antenna; wherein the historical calibration data includes at least an electromagnetic near-field distribution map and spectral envelope characteristics; based on the historical calibration data, establishing a logistic regression model to predict the drift law of the active antenna with ambient temperature and obtaining the gain change curve; introducing one-dimensional convolutional analysis of electromagnetic signals to identify electromagnetic interference events, using a built-in rule engine to perform judgment analysis on the gain deviation of each batch of measurements, and triggering adjustment strategies through electromagnetic interference events; this aspect achieves intelligent EMC detection, with real-time calibration and correction during the detection process, improving detection accuracy.

Description

Technical Field

[0001] This invention relates to the field of electromagnetic compatibility testing technology, specifically to an intelligent electromagnetic compatibility testing system and method based on intelligent sensors. Background Technology

[0002] With the development of electronic technology, the application of electrical equipment is constantly increasing. Changes in current and voltage during the power consumption process will generate electromagnetic field radiation, causing electromagnetic interference to be ubiquitous and easily leading to problems such as signal distortion, data loss, and system crashes. Therefore, the requirements for electromagnetic compatibility are constantly increasing.

[0003] Existing electromagnetic compatibility (EMC) testing technologies suffer from the following problems: In complex electromagnetic environments, interference sources are typically multi-source and distributed. Current testing technologies often rely on single-point sensor acquisition, resulting in ambiguous interference source localization. They can only acquire field strength data at a specific location within the testing area, failing to fully reconstruct the spatial distribution characteristics of the electromagnetic field. For outdoor EMC test antennas, the receiving sensitivity of omnidirectional antennas is uniformly distributed in all directions, making it impossible to focus on the electromagnetic signal in the target area or distinguish the spatial direction of the interference source. In traditional testing processes, laboratory testing and field testing equipment are incompatible, making it difficult to meet the rapid on-site screening needs of aerospace, automotive electronics, and other fields. Furthermore, they often rely on manual operation and experience-based analysis, lacking a unified quantitative standard for determining interference types. Additionally, the performance of outdoor EMC test antennas is affected by ambient temperature, leading to significant errors in test results. Moreover, the bias current of existing active antennas is often a factory-fixed value. Over long-term use, gain drift caused by temperature and humidity can only be corrected through offline calibration, not real-time adjustment during on-site testing. This results in decreased testing accuracy over time, and the cumbersome calibration process affects measurement accuracy and fails to promptly reduce electromagnetic interference. Summary of the Invention

[0004] (a) Technical problems to be solved

[0005] To address the shortcomings of existing technologies, this invention provides an intelligent electromagnetic compatibility (EMC) detection system and method based on smart sensors. It establishes a logistic regression model to predict the drift pattern of an active antenna with varying ambient temperature. By analyzing the activity probabilities of normal, linear offset, and nonlinear distortion states, it obtains gain variation curves. Then, it uses one-dimensional convolution to identify transient, intermittent, or continuous EMC events. A built-in rule engine performs judgment analysis on each batch of gain variation curves, triggering corresponding strategies under different runaway states to correct the bias current and receiving angle of the active antenna. This improves the anti-interference capability and detection accuracy of the detection system, solving the problems mentioned in the background technology.

[0006] (II) Technical Solution

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] In a first aspect, this application provides an intelligent electromagnetic compatibility detection system based on smart sensors, the system comprising:

[0009] The input source acquisition module collects the ambient temperature of the area to be detected and simultaneously acquires the historical calibration data of the active antenna; the historical calibration data includes at least the electromagnetic near-field distribution map and spectral envelope characteristics.

[0010] The predictive analysis module, based on historical calibration data, establishes a logistic regression model to predict the drift pattern of the active antenna with ambient temperature and obtain the gain change curve.

[0011] The interference identification module introduces one-dimensional convolutional analysis of electromagnetic signals to identify electromagnetic interference events. It uses a built-in rule engine to perform judgment analysis on the measurement gain deviation of each batch and triggers adjustment strategies through electromagnetic interference events.

[0012] Furthermore, obtain historical calibration data for the active antenna, including:

[0013] The smart sensors form a sensing array through distributed deployment, and each smart sensor constitutes a data collection point;

[0014] Synchronously monitor the static operating point parameters of the transistors inside the active antenna, including induced voltage and induced current;

[0015] The three-dimensional coordinates of the sensing array relative to the power testing equipment are obtained by using the positioning component, and the electromagnetic near-field distribution spectrum is obtained by spatial interpolation of the level signal intensity at each acquisition point.

[0016] Perform broadband frequency sweep across the entire frequency band to extract spectral envelope features.

[0017] Furthermore, the gain variation curve is obtained, including:

[0018] Feature vectors are extracted based on historical calibration data and combined with ambient temperature to construct the input vector of a logistic regression model. The Softmax function is used to calculate the active probability of the active antenna in different operating states under the current temperature environment. The operating states include normal state, linear offset state and nonlinear distortion state, with the priority increasing in that order.

[0019] With ambient temperature as the independent variable, continuous sampling is performed to count the number of times the active probability occurs in the same temperature range. The working state with the most occurrences is selected as the dominant state in that temperature range. Based on the category of the dominant state, the corresponding prediction function is called to calculate the first deviation value in that temperature range. A sliding window weighted average is performed on the first deviation value. By adjusting the step size and weight distribution of the sliding window, the second deviation value is obtained and used as the discrete coordinate point of the gain change curve.

[0020] Furthermore, the prediction function includes:

[0021] If it is a normal state, use R (frequency envelope reference) for prediction;

[0022] If it is a linear offset state, R (induced voltage, ambient temperature) is used for prediction;

[0023] If the distortion state is nonlinear, R(induced current, ambient temperature) is used for prediction; where R(·) is the relational function.

[0024] Furthermore, one-dimensional convolution analysis is introduced to identify electromagnetic interference types, including:

[0025] The system acquires real-time electromagnetic signals and uses a one-dimensional convolutional layer to scan the electromagnetic signal features, extracting time-series features including at least the pulse rising edge and signal duty cycle. Spherical clustering is performed on the time-series features, and interference feature points in the electromagnetic environment that do not fall into the radius window are identified by monitoring the radius. With the pulse rising edge as the horizontal axis and the signal duty cycle as the vertical axis, the interference feature points are mapped to transient, intermittent, or continuous electromagnetic interference events.

[0026] Furthermore, the built-in rule engine is used to perform judgment analysis on the gain change curves of each batch, including:

[0027] The identified electromagnetic interference events and gain change curves are integrated into a target dataset. The interference impact is evaluated by chi-square statistics. An interference defect index is generated by weighted summation through internal training. The mean and standard deviation of the interference defect index are extracted to establish a statistical evaluation benchmark for this batch of data, which is defined as a standard threshold under the current ambient temperature.

[0028] The measurement gain deviation of each batch of real-time sampling is subtracted from the gain change curve of the corresponding batch to obtain a residual value sequence. The Westgard rule engine is used to perform runaway judgment. A residual value sequence in which a single measurement value is greater than twice the standard threshold is identified as a random runaway sequence and a random runaway strategy is triggered. A residual value sequence in which four consecutive measurement values ​​have the same sign and are all greater than one standard deviation is identified and a failure runaway strategy is triggered. If none of the above conditions are met, the data of this batch is determined to be normal.

[0029] Furthermore, adjust the strategy, including:

[0030] In response to the random runaway strategy, and when the electromagnetic interference event is identified as transient or intermittent, the receiving angle correction procedure is triggered to correct the receiving angle of the active antenna in real time.

[0031] When responding to a failure and loss of control strategy and identifying a continuous electromagnetic interference event, the antenna coefficient correction procedure is triggered to correct the bias current of the active antenna in real time.

[0032] Furthermore, the evaluation dataset is extracted during the execution cycle of the adjustment strategy, including:

[0033] The evaluation dataset includes residual fluctuation values ​​and ambient temperature fluctuation values;

[0034] The evaluation dataset is normalized and weighted summation is performed to generate a predicted detection performance value. This value is then compared to a standard detection performance threshold. If the predicted value is greater than or equal to the threshold, the adjustment strategy is deemed effective, and the current receiving angle and bias current are locked. If the predicted value is less than the threshold, an optimization adjustment command is triggered.

[0035] If the random runaway strategy is triggered, the bias current step size adjustment command is executed, and the step size is adjusted in combination with the positive or negative sign of the gain offset; if the failure runaway strategy is triggered, the angle adjustment command is executed to adjust the stepper motor's step angle.

[0036] Furthermore, the adjustment strategy also includes:

[0037] Antenna coefficient correction procedure: Extract the mean of the residual sequence corresponding to the failure and runaway sequence, mark it as the gain offset, and identify the positive and negative signs of the gain offset to adjust the bias current until the residual value is less than or equal to twice the standard deviation of the standard threshold.

[0038] Receiving angle correction procedure: retrieve the electromagnetic near-field distribution spectrum, combine it with the identified interference effects, locate the direction of the interference source vector, calculate the angle between the current spatial pointing of the active antenna and the direction of the interference source vector where the interference effects are located, extract the peak value of the residual value sequence corresponding to random loss of control to determine the deflection angle, drive the stepper motor to rotate the receiving angle of the active antenna until the current residual value returns to within ±1 standard deviation of the standard threshold.

[0039] Secondly, this application provides an intelligent electromagnetic compatibility detection method based on smart sensors, the method comprising:

[0040] The ambient temperature of the area to be tested is collected, and the historical calibration data of the active antenna is acquired simultaneously; the historical calibration data includes at least the electromagnetic near-field distribution map and spectral envelope characteristics.

[0041] Based on historical calibration data, a logistic regression model is established to predict the drift pattern of the active antenna with ambient temperature and obtain the gain change curve.

[0042] One-dimensional convolutional analysis is introduced to identify electromagnetic interference events by electromagnetic signals. The built-in rule engine performs judgment analysis on the measurement gain deviation of each batch and triggers adjustment strategies through electromagnetic interference events.

[0043] (III) Beneficial Effects

[0044] This invention provides an intelligent electromagnetic compatibility detection system and method based on smart sensors, which has the following beneficial effects:

[0045] 1. This invention provides the prediction and analysis module with complete input features of ambient temperature and electromagnetic properties by synchronously acquiring ambient temperature and historical calibration data, thus ensuring the accuracy of drift pattern prediction from the data source level.

[0046] 2. This invention uses a logistic regression model to accurately classify the three working states of an active antenna: normal, linear offset, and nonlinear distortion. By determining the dominant state, a mathematical method corresponding to the relational function is used to fit the quantitative law of antenna gain change with ambient temperature. The output gain change curve can intuitively reflect the antenna gain deviation in different temperature ranges, thus improving the objectivity and accuracy of the prediction results.

[0047] 3. This invention utilizes one-dimensional convolution to process temporal features, combined with spherical clustering and two-dimensional interval mapping, to quickly and accurately identify transient, intermittent, and continuous electromagnetic interference events in complex electromagnetic environments, effectively reducing the probability of signal misjudgment. Through a built-in rule engine, the residual value sequence of each batch of measurement gain deviation is automatically judged. After identifying electromagnetic interference events and runaway states, the corresponding adjustment strategy can be directly triggered to actively adjust antenna-related parameters to suppress interference effects, significantly improving the anti-interference capability of the detection system and ensuring the validity of detection data in complex electromagnetic environments. Attached Figure Description

[0048] Figure 1 This is a schematic diagram of the modules of the present invention;

[0049] Figure 2 This is a flowchart illustrating the present invention. Detailed Implementation

[0050] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0051] This invention monitors ambient temperature and historical calibration data, establishes a logistic regression model to predict the drift pattern of an active antenna with ambient temperature, and obtains gain change curves by analyzing the activity probability of normal, linear offset, and nonlinear distortion states. Then, it uses one-dimensional convolution to identify transient, intermittent, or continuous electromagnetic interference events, and uses a built-in rule engine to perform judgment analysis on each batch of gain change curves, triggering corresponding strategies under different runaway states to correct the bias current and receiving angle of the active antenna, thereby improving the anti-interference capability of the detection system.

[0052] Example 1:

[0053] This invention provides an intelligent electromagnetic compatibility detection system based on smart sensors; Figure 1 This is a schematic diagram of the module of the present invention; please refer to it. Figure 1 The system includes an input source acquisition module, a predictive analysis module, and an interference identification module, and the input source acquisition module, predictive analysis module, and interference identification module are connected in communication.

[0054] The following is a detailed explanation of each module:

[0055] Input source acquisition module:

[0056] The ambient temperature of the area to be tested is collected, and the historical calibration data of the active antenna is acquired simultaneously; the historical calibration data includes electromagnetic near-field distribution map and spectral envelope characteristics;

[0057] Acquire historical calibration data of active antennas, including: intelligent sensors forming a sensing array through distributed deployment, with each intelligent sensor constituting a collection point; according to the power consumption test logic, multiple intelligent sensors are integrated around the active antenna and power consumption test equipment, forming a sensing array through distributed deployment; synchronous trigger commands are sent to all sensing arrays through 5G wireless networking, and clock alignment is achieved based on GPS timing to ensure that all sensing arrays start parameter acquisition simultaneously;

[0058] The sensing array incorporates a temperature sensor to acquire the ambient temperature of the area to be detected; it simultaneously monitors the static operating point parameters of the transistors inside the active antenna, including induced voltage and induced current; the sensing array also incorporates voltage and current sensors to simultaneously acquire voltage and current data; it directly connects to the emitter and collector of the transistors inside the active antenna via shielded cables to capture the induced voltage and induced current of the transistors in real time; during the acquisition process, local preprocessing of the static operating point parameters is performed: Kalman filtering is used to remove pulse interference, and moving average filtering is used to smooth the data to ensure the stability of the output parameters; each sensing array uploads the preprocessed induced voltage and induced current data to the main controller in real time via a 5G network, and the main controller integrates the data from sensing arrays at different locations according to timestamps to form a dataset of static operating point parameters;

[0059] The positioning component acquires the three-dimensional coordinates of the sensing array relative to the power testing equipment, and spatially interpolates the level signal strength at each measurement point to obtain an electromagnetic near-field distribution map. The sensing array integrates a UWB positioning unit and an inertial measurement unit to form a positioning component. A three-dimensional rectangular coordinate system is established with the geometric center of the power testing equipment as the origin to obtain the actual position of the power testing equipment. Simultaneously, electric field strength and magnetic field strength at each acquisition point are collected synchronously through electric field sensors and magnetic field sensors, and converted into uniformly quantized level signal strengths. The Kriging interpolation algorithm is used to perform spatial interpolation to obtain an electromagnetic near-field distribution map. For example, electric field strength is represented by a red gradient, and magnetic field strength is represented by a blue gradient.

[0060] A wideband frequency sweep is performed across the entire frequency range to extract spectral envelope features. The sensing array integrates radio frequency sensors to scan the electromagnetic signals in the space surrounding the active antenna and the power testing equipment to obtain spectral data for the entire operating frequency band. By removing invalid data, calibrating amplitude deviations, and subtracting background noise, discrete spectral amplitude points are extracted and converted into continuous and smooth spectral envelope curves. The Hilbert transform method is used to process the spectral envelope curves formed by the electromagnetic signals generated by the active antenna, and the rectification-low-pass filtering method is used to process the spectral envelope curves formed by the electromagnetic signals generated around the power testing equipment. Features strongly correlated with electromagnetic compatibility, such as peak frequency, bandwidth, harmonic order, spectral flatness, frequency band energy ratio, and peak phase jitter, are extracted for use as input features in the subsequent logistic regression model.

[0061] By synchronously acquiring ambient temperature and historical calibration data, the predictive analysis module is provided with complete input features of ambient temperature and electromagnetic properties, ensuring the accuracy of drift pattern prediction from the data source level.

[0062] Predictive analytics module:

[0063] Based on historical calibration data, a logistic regression model is established to predict the drift pattern of the active antenna with ambient temperature and obtain the gain change curve.

[0064] Obtain the gain variation curve, including:

[0065] Feature vectors are extracted from historical calibration data and combined with ambient temperature to construct the input vector of a logistic regression model. The Softmax function is then used to calculate the probability of the active antenna operating in different states under the current temperature environment. It should be noted that the logistic regression model includes an input layer, a linear transformation layer, a Softmax layer, and an output layer. The input layer receives feature vectors extracted from ambient temperature, static operating point parameters, electromagnetic near-field distribution patterns, and spectral envelope features, including but not limited to peak frequency, bandwidth, harmonic order, spectral flatness, frequency band energy proportion, and peak phase jitter. The linear transformation layer uses a preset weight matrix and bias term to calculate scores for different operating states, including setting corresponding weight matrices for the same operating state. The input vector and weight matrix are multiplied together, and the bias term is added to obtain the corresponding score. A Softmax layer is used to calculate the activity probability of different working states under the current temperature environment using the Softmax function. Multi-class cross-entropy loss is employed to ensure that the logistic regression model outputs an activity probability close to the true label. Gradient descent is used to minimize the loss function, such as using the Adam optimizer with a learning rate α set to 0.001. The working states include the normal state (representing the stability of the tested region), the linear offset state (representing the linear displacement of the tested region with temperature), and the nonlinear distortion state (representing the tested region entering the saturation or cutoff region), with priority increasing sequentially, i.e., the nonlinear distortion state has a higher priority than the linear offset state, which in turn has a higher priority than the normal state.

[0066] Continuous sampling is performed using ambient temperature as the independent variable. For example, temperature samples are continuously collected within the antenna's operating temperature range in 0.5℃ increments to obtain a temperature sequence. This continuous temperature sequence is then merged into intervals of 2℃, such as [-40℃, -38℃). The frequency of occurrences within the same temperature interval is counted, and the operating state with the most occurrences is selected as the dominant state for that temperature interval. For example, in the [-40℃, -38℃) temperature interval, if the normal state occurs 0 times, the linear offset state occurs 4 times, and the nonlinear distortion state occurs 0 times, the linear offset state is selected as the dominant state. It should be noted that if the normal state occurs 0 times, the linear offset state occurs 2 times, and the nonlinear distortion state occurs 2 times, and the linear offset and nonlinear distortion states have the same number of occurrences, the determination is based on priority. Since the nonlinear distortion state has a higher priority than the linear offset state, the nonlinear distortion state is selected as the dominant state.

[0067] Based on the category of the dominant state, the corresponding prediction function is invoked, including:

[0068] Where R(·) is a relational function;

[0069] If it is a normal state, R (frequency envelope reference) prediction is used: Calculate the deviation between the current interval spectral envelope feature and the frequency envelope reference, mark it as feature deviation, and standardize it, that is: divide the feature difference by the difference between the maximum and minimum values ​​of the feature in the historical data to eliminate the dimensional differences of different features, square the standardized feature differences respectively, add the squared results, and finally take the square root of the sum to obtain the spectral envelope deviation. Multiply the obtained spectral envelope deviation by the pre-fitted slope coefficient and add the pre-fitted intercept (usually the intercept value is 0) to obtain the gain deviation under normal state. It should be noted that this step belongs to univariate linear regression fitting. By using the spectral envelope deviation as the independent variable and the measured gain deviation of all normal state samples in the training set corresponding to the model as the dependent variable, the dependent variable is calculated by multiplying the independent variable by the slope coefficient and adding the intercept. By minimizing the sum of squared errors between the gain deviation predicted by the model and the measured value, the optimal values ​​of the slope coefficient and intercept under normal state are found.

[0070] If it is a linear offset state, R(induced voltage, ambient temperature) prediction is used: take the mean induced voltage of the temperature range, the midpoint temperature of the temperature range, the pre-fitted voltage coefficient, temperature coefficient, and intercept, multiply the mean induced voltage by the voltage coefficient to obtain the first part of the value, multiply the midpoint temperature of the temperature range by the temperature coefficient to obtain the second part of the value, add the two parts of the value, and add the pre-fitted intercept to obtain the gain deviation under the linear offset state; it should be noted that this step belongs to the binary linear regression fitting, constructing two columns of independent variable matrices through the mean induced voltage and ambient temperature, and using the measured gain deviation of all linear offset state samples in the training set of the model as the dependent variable. The dependent variable is calculated as: voltage coefficient multiplied by the mean induced voltage, ambient temperature multiplied by the temperature coefficient, add the calculated results, and add the intercept. By minimizing the sum of squared errors between the gain deviation predicted by the model and the measured value, the optimal values ​​of voltage coefficient, temperature coefficient, and intercept under the linear offset state are found.

[0071] For nonlinear distortion, R(induced current, ambient temperature) prediction is used: The mean induced current over the temperature range, the midpoint temperature of the range, the pre-fitted current square coefficient, the current-temperature crossover coefficient, the temperature square coefficient, and the intercept are taken. The first part of the value is obtained by multiplying the square of the mean induced current by the current square coefficient. The second part of the value is obtained by multiplying the mean induced current by the midpoint temperature of the range, and then by the current-temperature crossover coefficient. The third part of the value is obtained by multiplying the square of the mean temperature by the temperature square coefficient. These three parts are then added together, along with the pre-fitted intercept, to obtain the gain deviation under nonlinear distortion. It should be noted that... The model uses the squared mean of induced current, the current-temperature cross-coefficient, and the squared temperature as independent variables, and the measured gain deviation of all nonlinear distortion state samples in the training set corresponding to the model as the dependent variable. The dependent variable is calculated as follows: the squared current coefficient multiplied by the squared mean of induced current, the squared mean of induced current multiplied by the temperature multiplied by the current-temperature cross-coefficient, the squared temperature coefficient multiplied by the squared temperature, and then the results of the above multiplications are added together, plus the intercept, and the gradient descent method is used until the sum of squared errors converges. The fitted parameters are not static. The system will periodically add new calibration data, re-execute the above fitting process, update the coefficients and intercept, and ensure that the parameters always match the latest operating state of the antenna.

[0072] It should be noted that a mapping table is pre-built to record the data collected at different detection points and the corresponding gain deviations; the ambient temperature and transistor static parameters operating point parameters within the current time period are received, the gain deviation of the next temperature range is predicted, and the mapping relationship between the transistor static parameters operating point parameters and the gain deviation of the next temperature range is saved in the mapping table. According to the time series, the gain change curve is used to perform level inversion in order to initially calibrate the gain prediction deviation.

[0073] For a given temperature range, the midpoint temperature of that range is obtained, and the gain deviation within that range is calculated and marked as the first deviation value. A sliding window weighted average is then applied to the first deviation value, and the second deviation value is obtained by adjusting the step size and weight distribution of the sliding window.

[0074] Step size: The number of intervals that the sliding window slides to the right after completing one calculation; Weight distribution: The weighted proportion of intervals at different positions within the sliding window. The core principle is that the current interval has the largest weight, which decreases towards both sides; Sliding window size: The number of continuous temperature intervals participating in each weighted calculation;

[0075] For the Mth temperature range, the range numbers covered by its sliding window are: M-2, M-1, M, M+1, M+2. Therefore, the sliding window is set to 5. It should be noted that the window coverage may exceed the actual number of ranges, requiring padding: if it exceeds the left range, head padding is performed, replacing it with the first deviation value of the first range; if it exceeds the right range, tail padding is performed, replacing it with the first deviation value of the last range. The second deviation value for the Mth temperature range is the sum of the products of the first deviation values ​​in these 5 windows and their corresponding weights. The window is slid to the right by a preset step size until the calculation of all temperature ranges is completed, ultimately obtaining a series of second deviation values. This eliminates local random fluctuations in the first deviation values, generating a smoother sequence of deviation values ​​that more closely reflects the actual drift pattern of the active antenna, and serves as the discrete coordinate points for the gain change curve. It should be noted that the sliding window value is just an example; the specific value should be set according to the actual situation, and will not be elaborated upon here.

[0076] By analyzing the active probability and dominant state output by the predictive analysis module, a differentiated relational function is triggered to perform gain prediction: If the dominant state is normal, the detection data quality is determined to be high, the gain has no obvious drift, and it has no significant impact on detection accuracy. The relational function R (frequency envelope reference) is called to find the response relationship between the antenna's first deviation value and the feature envelope reference. If the dominant state is linear offset, the detection data quality is determined to be medium, the gain drift is linear, the deviation can be corrected by linear compensation, and the impact is moderate. The relational function R (induced voltage, ambient temperature) is called to find the linear mapping relationship between the antenna's first deviation value and the induced voltage and ambient temperature. If the dominant state is nonlinear distortion, the detection data quality is determined to be low, the antenna gain drifts nonlinearly with temperature, which easily leads to amplitude and phase deviations in electromagnetic signal acquisition, and has the greatest impact on interference identification accuracy. The relational function R (induced current, ambient temperature) is called to find the coupling relationship between the antenna's first deviation value and the induced current and ambient temperature.

[0077] By employing a logistic regression model and introducing the Softmax function, the three operating states of the active antenna—normal, linear offset, and nonlinear distortion—were accurately classified. By determining the dominant state, a mathematical method corresponding to the relational function was used to fit the quantitative law of antenna gain variation with ambient temperature. The output gain variation curve can intuitively reflect the antenna gain deviation in different temperature ranges, improving the objectivity and accuracy of the prediction results.

[0078] Interference identification module:

[0079] One-dimensional convolutional analysis of electromagnetic signals is introduced to identify electromagnetic interference events. The built-in rule engine performs judgment analysis on the measurement gain deviation of each batch and triggers adjustment strategies through electromagnetic interference events.

[0080] One-dimensional convolution analysis is introduced to identify electromagnetic interference types, including:

[0081] The system acquires real-time electromagnetic signals and uses a one-dimensional convolutional layer to scan the features of these signals. The one-dimensional convolutional layer consists of convolutional layer 1 and convolutional layer 2: Convolutional layer 1 has 32 kernels, a kernel size of 16, and a stride of 2. It captures local features of the rising edge of the pulse by scanning local segments of the electromagnetic signal. Convolutional layer 2 has 64 kernels, a kernel size of 8, and a stride of 1. It extracts global features such as the signal duty cycle and frequency energy distribution. Finally, a fully connected layer fuses the local and global features to integrate them into temporal features.

[0082] Spherical clustering is performed on the temporal features. Temporal features of electromagnetic signals under normal conditions are extracted from historical calibration data to form a normal feature dataset. The extreme points of this normal feature dataset are extracted as the initial set of sphere centers for spherical clustering. Interference feature points in the electromagnetic environment that do not fall within the radius window are identified by the monitoring radius; conversely, those falling within the radius window are considered normal feature points. The process involves calculating the currently acquired temporal features and standardizing them to feature points. The Euclidean distance from each feature point to the initial sphere center is then calculated. If the Euclidean distance is greater than the monitoring radius, it is determined that the feature point does not fall within the radius window; if the Euclidean distance is less than or equal to the monitoring radius, it is marked as falling within the radius window. The monitoring radius is the product of the standard deviation of the normal feature dataset and the confidence coefficient. It should be noted that the confidence coefficient is either 2 or 3, typically 3 in the laboratory and 2 in the field. This is just an example; the specific value needs to be set according to the actual situation.

[0083] Using the pulse rising edge as the horizontal axis and the signal duty cycle as the vertical axis, with the horizontal axis divided into high-frequency, mid-frequency, and low-frequency ranges, and the vertical axis divided into small-amplitude, medium-amplitude, and large-amplitude ranges, interference characteristic points are mapped as transient, intermittent, or continuous electromagnetic interference events through cross-combination; Table 1 represents this:

[0084] Table 1 shows the criteria for determining electromagnetic interference events.

[0085]

[0086] It should be noted that low-frequency micro-amplitude represents weak low-frequency interference, classified as intermittent interference events: slow pulse rise, occasional effective signals, low interference amplitude, and minimal impact; mid-frequency micro-amplitude represents weak intermittent interference, classified as intermittent interference events: moderate pulse rise speed, occasional effective signals, and irregular interference occurrence; high-frequency micro-amplitude represents transient interference events: extremely rapid pulse rise, occasional effective signals; low-frequency medium-amplitude represents low-frequency intermittent interference, classified as intermittent interference: slow pulse rise, intermittent effective signals, and moderate interference duration; mid-frequency medium-amplitude represents typical intermittent interference, classified as intermittent interference events. Intermittent interference events are characterized by: moderate pulse rise speed and intermittent effective signal appearance; medium amplitude at high frequency, representing strong transient pulse trains; extremely rapid pulse rise and intermittent effective signal appearance; large amplitude at low frequency, representing typical continuous interference; slow pulse rise and continuous effective signal presence, consistent with steady-state interference characteristics; large amplitude at medium frequency, representing medium-frequency quasi-continuous interference; moderate pulse rise speed and continuous effective signal presence; large amplitude at high frequency, representing high-frequency continuous interference; extremely rapid pulse rise and continuous effective signal presence.

[0087] The division of high-frequency, mid-frequency, and low-frequency ranges, and micro-amplitude, mid-amplitude, and large-amplitude ranges, is based on the following criteria: For the high-frequency, mid-frequency, and low-frequency ranges, based on historical statistics, the pulse rising edge data is sorted in descending order, and the ranges are divided according to the 33rd and 66th percentiles. Values ​​at the 33rd and 66th percentiles are obtained. Values ​​below the 33rd percentile are marked as high-frequency ranges, values ​​above or equal to the 33rd percentile and below the 66th percentile are marked as mid-frequency ranges, and values ​​above or equal to the 66th percentile are marked as low-frequency ranges. To adapt to the computational requirements of spherical clustering, the pulse rising edges are normalized. For the micro-amplitude, mid-amplitude, and large-amplitude ranges, based on historical statistics, the signal duty cycle data is ordered... The data is sorted and divided into intervals based on the 20th and 80th percentiles, and the corresponding values ​​for the 20th and 80th percentiles are obtained. Intervals less than or equal to the 20th percentile are marked as small-amplitude intervals, intervals greater than the 20th percentile and less than or equal to the 80th percentile are marked as medium-amplitude intervals, and intervals greater than the 80th percentile are marked as large-amplitude intervals. It should be noted that when a new type of interference source is detected, its pulse rising edge and duty cycle parameters are automatically included in the statistics, the quantiles are recalculated, and the interval thresholds are updated. The quantile values ​​in this embodiment are just an example; the specific settings should be based on actual conditions. Frequency energy distribution is used for auxiliary verification, which will not be elaborated upon here.

[0088] By utilizing one-dimensional convolution to process temporal features, combined with spherical clustering and two-dimensional interval mapping, transient, intermittent, and continuous electromagnetic interference events in complex electromagnetic environments can be quickly and accurately identified, effectively reducing the probability of signal misjudgment.

[0089] The built-in rule engine performs judgment analysis on the gain change curves of each batch, including:

[0090] The identified electromagnetic interference events and gain change curves were integrated into a target dataset, which was then divided into a training set. Temperature range was used as the unique correlation key. For the same temperature range, a 3×3 contingency table was constructed, with electromagnetic interference event type as the row variable and gain change curve as the column variable. The chi-square statistic was calculated by comparing the actual observation frequency and the expected frequency within the training set, with a significance level of 0.05. The impact of interference was assessed by looking up the table. If the calculated chi-square value was greater than a preset threshold, it indicated a significant difference, and the correlation between interference and gain deviation was considered higher. If the calculated chi-square value was less than or equal to the preset threshold, it indicated no significant difference, and the correlation between interference and gain deviation was considered lower. Simultaneously, the target... Records in the dataset are labeled as strongly correlated or weakly correlated. Records with strong correlations are given higher training weights when calculating the interference defect index; records with weak correlations are given lower training weights because gain drift may be due to antenna issues. The interference defect index is generated by internal training and weighted summation. In this embodiment, interference factors include, but are not limited to, interference correlation, interference type, and gain deviation. Weight coefficients for each interference factor are trained using historical data, and these weights are multiplied by the quantized values ​​of the corresponding factors. Finally, the summation is used to obtain the interference defect index. The mean and standard deviation of the interference defect index are extracted to establish a statistical evaluation benchmark for this batch of data, and defined as a standard threshold under the current ambient temperature.

[0091] For each batch of real-time sampling, the measurement gain deviation is calculated by subtracting the corresponding batch's gain change curve. By subtracting the second deviation value from the strategy gain deviation, the residual value is obtained. According to the time series of the gain change curve, the residual value sequence is obtained, and the Westgard rule engine is used to perform runaway judgment: if a residual value sequence in which a single measurement value is greater than twice the standard threshold is identified, it is marked as a random runaway sequence, and the random runaway strategy is triggered; if a residual value sequence in which four consecutive measurement values ​​have the same sign and are all greater than one standard deviation is identified, the failure runaway strategy is triggered; if none of the above conditions are met, the data of this batch is determined to be normal.

[0092] Adjusting strategies, including:

[0093] In response to the random runaway strategy, and upon identifying whether the electromagnetic interference event is transient or intermittent, a receiving angle correction procedure is triggered to correct the receiving angle of the active antenna in real time. This receiving angle correction procedure involves: retrieving the electromagnetic near-field distribution map; combining this with the identified interference effects; locating the spatial direction with the strongest radiated energy; and filtering out the direction corresponding to the maximum energy density by traversing the radiated energy density in all spatial directions of the electromagnetic near-field distribution map. This direction is the direction of the interference source vector. The spatial direction is (azimuth, elevation). For example, if the maximum energy density in the electromagnetic near-field distribution map occurs at an azimuth of 30° and an elevation of 15°, then the direction of the interference source vector is (30°, 15°).

[0094] Calculate the angle between the spatial pointing of the active antenna's current sensing plate and the direction of the interference source vector. Convert the spatial pointing of the antenna's sensing plate into a first unit vector and the direction of the interference source vector into a second unit vector. Calculate the cosine angle between the first and second unit vectors using electrowinning. The larger the angle, the greater the deviation between the antenna's current pointing direction and the direction of the interference source. When the angle is 0, the antenna is directly facing the direction of the interference source, and the received interference is strongest. Extract the residual value with the largest absolute value from the sequence of residual values ​​corresponding to random loss of control and mark it as the peak value. Establish a mapping relationship between the residual peak value and the deflection angle to determine the deflection angle. Drive the stepper motor to rotate the receiving direction of the active antenna until the current residual value returns to within ±1 standard deviation of the standard threshold. In addition, align the antenna's radiation pattern null point with the direction of the interference source vector. In the next batch of sampling after performing the above actions, recalculate the residual value. If the residual value returns to within ±1 standard deviation of the standard threshold, maintain the current parameter result; otherwise, it is still judged as a random loss of control sequence.

[0095] When responding to failure and loss of control strategies, and identifying electromagnetic interference events as persistent,

[0096] The antenna coefficient correction procedure is triggered to correct the bias current of the active antenna in real time. This procedure involves: extracting the mean of the residual sequence corresponding to the failure / runaway sequence; combining this mean with the gain change curve of the logistic regression model to obtain the gain offset and determine its sign. It should be noted that the physical meaning of the residual value is the measured gain minus the predicted gain. Therefore, the mean value at this point is equivalent to the gain offset. If the gain offset is greater than 0, it is a positive sign, indicating that the measured gain is higher than the predicted gain, and the bias current needs to be reduced. If the gain offset is less than 0, it is a negative sign, indicating that the measured gain is lower than the predicted gain, and the bias current needs to be increased. The bias current is adjusted until the residual value regresses to less than or equal to twice the standard threshold standard deviation.

[0097] Extract the evaluation dataset during the execution cycle of the adjustment strategy, including:

[0098] The evaluation dataset includes residual fluctuation values ​​and ambient temperature fluctuation values. The residual fluctuation value reflects the degree of fluctuation of the adjusted residual value; the smaller the fluctuation, the higher the detection stability. The ambient temperature fluctuation value reflects the degree of change in the ambient temperature within the period; the greater the temperature fluctuation, the higher the risk of antenna gain drift. The fluctuation value is calculated by extracting all residual data and ambient temperature data within the execution period of the adjustment strategy, obtaining the maximum and minimum values ​​of the residual data and ambient temperature data through statistical analysis, setting the fluctuation value as the absolute value of the difference between the maximum and minimum values, and taking the average value.

[0099] The evaluation dataset is normalized and weighted summation is performed. The estimated detection effect is obtained by multiplying the evaluation dataset by its corresponding weight coefficient and then summing the results. Specifically, the residual fluctuation value is multiplied by its corresponding weight coefficient, and the ambient temperature fluctuation value is multiplied by its corresponding weight coefficient. It should be noted that the weight coefficients are dynamic values, automatically assigned by the system. The acquisition process involves collecting relevant raw data from historical evaluation records that triggered optimization adjustment commands, including residual fluctuation values, ambient temperature fluctuation values, and timestamps indicating the triggering of optimization adjustments. Each execution cycle and process is marked as an evaluation group. Based on expert experience rules, historical evaluation groups are assigned a true correlation validity label, set in a 0-1 range, where 1 represents a better adjustment effect and the parameters are directly locked after adjustment, and 0 represents... The table adjustment effect was poor; all historical candidate groups were sorted from low to high according to the true association validity label to form a historical association dataset; every 4 positions, the historical association dataset was sampled, and the resulting dataset was used as the test set. A lightweight fully connected neural network was preset as the weight learning model. The model input layer has 2 feature nodes, corresponding to the residual fluctuation value and the ambient temperature fluctuation value, respectively. There are 2 hidden layers, each with 16 neurons, both using the ReLU activation function. The output layer has 2 weight nodes, corresponding to the two dynamic weights of the evaluation dataset, and the sum of the weights is 1. At the same time, by adjusting the dynamic weights according to the mapping relationship between the residual fluctuation value and the ambient temperature fluctuation value and the true association validity label, the error between the calculated result of the detection effect prediction and the true association validity label is minimized.

[0100] The system compares the predicted detection effect with the standard detection effect threshold. If the predicted detection effect is greater than or equal to the standard detection effect threshold, the adjustment strategy is deemed effective, and the current receiving angle and bias current are locked. If the predicted detection effect is less than the standard detection effect threshold, an optimization adjustment command is triggered. If a random runaway strategy is triggered, a bias current step size adjustment command is executed, and the step size is adjusted in conjunction with the positive or negative sign of the gain offset. If a failure runaway strategy is triggered, an angle adjustment command is executed, reducing the step angle of the stepper motor, and secondary positioning is performed in conjunction with the electromagnetic near-field distribution map.

[0101] The standard detection effect threshold is derived from historical data. Analysis of past operational detection effect predictions is performed, and statistical analysis is conducted on the collected data to determine the mean and standard deviation of the detection effect predictions under effective adjustment strategies. Based on the statistical results, a standard detection effect threshold is set as the mean of the detection effect predictions under effective adjustment strategies plus two or three times the standard deviation. It should be noted that the multiple is merely an example; the specific value should be set according to the actual situation, and will not be elaborated upon here.

[0102] The built-in rule engine automatically determines the residual value sequence of each batch of measurement gain deviation without manual intervention, significantly reducing the time required for the determination process. At the same time, the rule engine's judgment criteria are uniform, avoiding subjective differences among different testing personnel and improving the repeatability and consistency of the test results. After identifying electromagnetic interference events and runaway states, corresponding adjustment strategies can be directly triggered, including triggering a receiving angle correction program for transient intermittent interference and a bias current correction program for continuous interference. By actively adjusting antenna parameters, the interference effect is suppressed, and the residual value quickly returns to the normal range, greatly improving the anti-interference capability of the testing system and ensuring the validity of the test data in complex electromagnetic environments.

[0103] Both one-dimensional convolution and rule engine are lightweight algorithms that can be deployed in portable field detection devices without relying on large analytical instruments in laboratories. At the same time, the algorithm's judgment criteria are consistent with those in laboratories, and the output evaluation results and adjusted parameters can be seamlessly integrated with laboratory data, meeting the needs of rapid on-site screening in multiple application scenarios.

[0104] Example 2:

[0105] This invention provides an intelligent electromagnetic compatibility detection method based on smart sensors; Figure 2 This is a flowchart illustrating the invention; please refer to [link / reference]. Figure 2 The method includes the following steps:

[0106] The ambient temperature of the area to be tested is collected, and the historical calibration data of the active antenna is acquired simultaneously; the historical calibration data includes at least the electromagnetic near-field distribution map and spectral envelope characteristics.

[0107] Based on historical calibration data, a logistic regression model is established to predict the drift pattern of the active antenna with ambient temperature and obtain the gain change curve.

[0108] One-dimensional convolutional analysis is introduced to identify electromagnetic interference events by electromagnetic signals. The built-in rule engine performs judgment analysis on the measurement gain deviation of each batch and triggers adjustment strategies through electromagnetic interference events.

[0109] In the application, the various formulas mentioned are all calculated by removing dimensions and taking their numerical values. The formulas are derived from the most recent real-world situation by collecting a large amount of data and conducting software simulations. The formulas are set by those skilled in the art according to the actual situation.

[0110] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0111] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0112] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. An intelligent electromagnetic compatibility detection system based on smart sensors, characterized in that, The system includes: The input source acquisition module collects the ambient temperature of the area to be detected and simultaneously acquires the historical calibration data of the active antenna; the historical calibration data includes at least the electromagnetic near-field distribution map and spectral envelope characteristics. The predictive analysis module, based on historical calibration data, establishes a logistic regression model to predict the drift pattern of the active antenna with ambient temperature and obtain the gain change curve. The interference identification module introduces one-dimensional convolutional analysis of electromagnetic signals to identify electromagnetic interference events. It uses the built-in rule engine to perform judgment analysis on the measurement gain deviation of each batch and triggers adjustment strategies through electromagnetic interference events. The built-in rule engine is used to perform judgment analysis on the gain change curves of each batch, including: The identified electromagnetic interference events and gain change curves are integrated into a target dataset. The interference impact is evaluated by chi-square statistics. An interference defect index is generated by weighted summation through internal training. The mean and standard deviation of the interference defect index are extracted to establish a statistical evaluation benchmark for this batch of data, which is defined as a standard threshold under the current ambient temperature. The measurement gain deviation of each batch of real-time sampling is subtracted from the gain change curve of the corresponding batch to obtain a residual value sequence. The Westgard rule engine is used to perform runaway judgment. A residual value sequence in which a single measurement value is greater than twice the standard threshold is identified as a random runaway sequence and a random runaway strategy is triggered. A residual value sequence in which four consecutive measurement values ​​have the same sign and are all greater than one standard deviation is identified and a failure runaway strategy is triggered. If none of the above conditions are met, the data of this batch is determined to be normal.

2. The intelligent electromagnetic compatibility detection system based on intelligent sensors according to claim 1, characterized in that, Obtain historical calibration data for active antennas, including: The smart sensors form a sensing array through distributed deployment, and each smart sensor constitutes a data collection point; Synchronously monitor the static operating point parameters of the transistors inside the active antenna, including induced voltage and induced current; The three-dimensional coordinates of the sensing array relative to the power testing equipment are obtained by using the positioning component, and the electromagnetic near-field distribution spectrum is obtained by spatial interpolation of the level signal intensity at each acquisition point. Perform broadband frequency sweep across the entire frequency band to extract spectral envelope features.

3. The intelligent electromagnetic compatibility detection system based on intelligent sensors according to claim 1, characterized in that, Obtain the gain variation curve, including: Feature vectors are extracted based on historical calibration data and combined with ambient temperature to construct the input vector of a logistic regression model. The Softmax function is used to calculate the active probability of the active antenna in different operating states under the current temperature environment. The operating states include normal state, linear offset state and nonlinear distortion state, with the priority increasing in that order. With ambient temperature as the independent variable, continuous sampling is performed to count the number of times the active probability occurs in the same temperature range. The working state with the most occurrences is selected as the dominant state in that temperature range. Based on the category of the dominant state, the corresponding prediction function is called to calculate the first deviation value in that temperature range. A sliding window weighted average is performed on the first deviation value. By adjusting the step size and weight distribution of the sliding window, the second deviation value is obtained and used as the discrete coordinate point of the gain change curve.

4. The intelligent electromagnetic compatibility detection system based on intelligent sensors according to claim 3, characterized in that, Prediction functions include: If it is a normal state, use R (frequency envelope reference) for prediction; If it is a linear offset state, R (induced voltage, ambient temperature) is used for prediction; If the distortion state is nonlinear, R(induced current, ambient temperature) is used for prediction; where R(·) is the relational function.

5. The intelligent electromagnetic compatibility detection system based on intelligent sensors according to claim 1, characterized in that, One-dimensional convolution analysis is introduced to identify electromagnetic interference types, including: The system acquires real-time electromagnetic signals and uses a one-dimensional convolutional layer to scan the electromagnetic signal features, extracting time-series features including at least the pulse rising edge and signal duty cycle. Spherical clustering is performed on the time-series features, and interference feature points in the electromagnetic environment that do not fall into the radius window are identified by monitoring the radius. With the pulse rising edge as the horizontal axis and the signal duty cycle as the vertical axis, the interference feature points are mapped to transient, intermittent, or continuous electromagnetic interference events.

6. The intelligent electromagnetic compatibility detection system based on intelligent sensors according to claim 1, characterized in that, Adjusting strategies, including: In response to the random runaway strategy, and when the electromagnetic interference event is identified as transient or intermittent, the receiving angle correction procedure is triggered to correct the receiving angle of the active antenna in real time. When responding to a failure and loss of control strategy and identifying a continuous electromagnetic interference event, the antenna coefficient correction procedure is triggered to correct the bias current of the active antenna in real time.

7. The intelligent electromagnetic compatibility detection system based on intelligent sensors according to claim 6, characterized in that, Extract the evaluation dataset during the execution cycle of the adjustment strategy, including: The evaluation dataset includes residual fluctuation values ​​and ambient temperature fluctuation values; The evaluation dataset is normalized and weighted summation is performed to generate a predicted detection performance value. This value is then compared to a standard detection performance threshold. If the predicted value is greater than or equal to the threshold, the adjustment strategy is deemed effective, and the current receiving angle and bias current are locked. If the predicted value is less than the threshold, an optimization adjustment command is triggered. If the random runaway strategy is triggered, the bias current step size adjustment command is executed, and the step size is adjusted in combination with the positive or negative sign of the gain offset; if the failure runaway strategy is triggered, the angle adjustment command is executed to adjust the stepper motor's step angle.

8. The intelligent electromagnetic compatibility detection system based on intelligent sensors according to claim 7, characterized in that, The adjustment strategy also includes: Antenna coefficient correction procedure: Extract the mean of the residual sequence corresponding to the failure and runaway sequence, mark it as the gain offset, and identify the positive and negative signs of the gain offset to adjust the bias current until the residual value is less than or equal to twice the standard deviation of the standard threshold. Receiving angle correction procedure: retrieve the electromagnetic near-field distribution spectrum, combine it with the identified interference effects, locate the direction of the interference source vector, calculate the angle between the current spatial pointing of the active antenna and the direction of the interference source vector where the interference effects are located, extract the peak value of the residual value sequence corresponding to random loss of control to determine the deflection angle, drive the stepper motor to rotate the receiving angle of the active antenna until the current residual value returns to within ±1 standard deviation of the standard threshold.

9. An intelligent electromagnetic compatibility detection method based on smart sensors, characterized in that, The method includes: The ambient temperature of the area to be tested is collected, and the historical calibration data of the active antenna is acquired simultaneously; the historical calibration data includes at least the electromagnetic near-field distribution map and spectral envelope characteristics. Based on historical calibration data, a logistic regression model is established to predict the drift pattern of the active antenna with ambient temperature and obtain the gain change curve. One-dimensional convolutional analysis of electromagnetic signals is introduced to identify electromagnetic interference events. The built-in rule engine performs judgment analysis on the measurement gain deviation of each batch and triggers adjustment strategies through electromagnetic interference events. The built-in rule engine is used to perform judgment analysis on the gain change curves of each batch, including: The identified electromagnetic interference events and gain change curves are integrated into a target dataset. The interference impact is evaluated by chi-square statistics. An interference defect index is generated by weighted summation through internal training. The mean and standard deviation of the interference defect index are extracted to establish a statistical evaluation benchmark for this batch of data, which is defined as a standard threshold under the current ambient temperature. The measurement gain deviation of each batch of real-time sampling is subtracted from the gain change curve of the corresponding batch to obtain a residual value sequence. The Westgard rule engine is used to perform runaway judgment. A residual value sequence in which a single measurement value is greater than twice the standard threshold is identified as a random runaway sequence and a random runaway strategy is triggered. A residual value sequence in which four consecutive measurement values ​​have the same sign and are all greater than one standard deviation is identified and a failure runaway strategy is triggered. If none of the above conditions are met, the data of this batch is determined to be normal.

Images