On-line detection method for material areal density based on average working current inversion

Abstract

The present application belongs to the technical field of industrial nondestructive online detection, and particularly relates to an online detection method based on average working current inversion of material area density. The present application extracts the average working current in the operation process of the terahertz thickness measuring equipment, and establishes an area density inversion fitting model in combination with the actually measured thickness of the terahertz, without the need of additional external area density detection hardware, thereby avoiding the high cost and safety protection burden caused by the ray detection, and also solving the problem that the optical detection is easily disturbed by surface environmental factors. The present application fully utilizes the electrical data generated by the operation of the terahertz thickness measuring equipment itself, and realizes the online continuous detection of the material area density without increasing the system complexity. The present application only uses the average working current as the core electrical input parameter to establish the model, reduces the influence of the voltage fluctuation and transient interference in the industrial field on the stability of the model, and improves the reliability of the area density inversion result.

Description

Technical Field

[0001] This invention belongs to the field of industrial non-destructive online technology, specifically relating to an online detection method for material surface density based on average operating current inversion. Background Technology

[0002] In the production of lithium battery electrodes, coated films, composite films, and functional materials, material areal density is a key indicator for evaluating coating uniformity, active material loading, film quality, and product consistency. Taking lithium electrode sheets as an example, electrode areal density directly affects battery capacity, internal resistance, consistency, and cycle performance.

[0003] Existing methods for areal density detection mainly include offline sampling and weighing, X-ray inspection, optical inspection, or multi-sensor fusion detection. These methods have the following shortcomings:

[0004] 1. Offline sampling and weighing cannot cover the entire continuous production line, and the test results are delayed, making it difficult to provide timely feedback to process control;

[0005] 2. X-ray inspection equipment is expensive and requires safety protection and maintenance.

[0006] 3. Optical inspection is easily affected by surface color, reflection, coating texture, and dust;

[0007] 4. Multi-sensor fusion detection requires additional hardware, making system integration complex;

[0008] 5. Existing terahertz thickness measurement systems typically only output thickness values, failing to fully utilize the electrical data generated during the equipment's operation.

[0009] During the continuous operation of a terahertz thickness measurement device, the device load, transmission status, reception status, and material response will be reflected in the changes in the device's operating current. If the average operating current of the device can be extracted under steady-state operating conditions and a correspondence can be established between it and the measured terahertz thickness and the material surface density, then the online continuous inversion of the material surface density can be achieved without adding an external surface density sensor.

[0010] Existing electrical parameter modeling schemes typically collect multiple parameters such as voltage and current simultaneously. However, industrial power grids experience voltage fluctuations, transient interference, and noise, making multivariate models prone to complexity and instability. Therefore, an online detection method is needed that uses the average operating current as the core input and combines terahertz measured thickness for surface density inversion. Summary of the Invention

[0011] The purpose of this invention is to provide an online detection method for material surface density based on average operating current inversion, which can achieve high-precision, real-time online detection of material surface density without increasing additional hardware costs.

[0012] The specific technical solution adopted by this invention is as follows:

[0013] Online detection methods for material surface density based on average operating current inversion include:

[0014] The average operating current is obtained by preprocessing the collected equipment operating current and averaging the preprocessed operating current.

[0015] Multiple standard samples with known thickness and known areal density were selected, and the terahertz measured thickness and average operating current of each standard sample were collected respectively.

[0016] A fitting model is established based on the standard areal density of the standard sample, the measured thickness at the terahertz level, and the average operating current.

[0017] The average operating current and terahertz measured thickness of the material under test are acquired in real time and recorded as test data;

[0018] The average operating current and the measured terahertz thickness of the material under test are input into the fitting model for inversion to obtain the real-time surface density of the material under test.

[0019] In a preferred embodiment, the preprocessing step includes:

[0020] The acquired raw equipment operating current signal is filtered and noise-reduced to remove signal spikes caused by equipment vibration and electromagnetic interference.

[0021] Abnormal peaks and valleys in the noise-reduced operating current signal are removed, and a sliding window mechanism is used to smooth the operating current to obtain a smoothed operating current.

[0022] Add timestamps to the smoothed operating current to form a timestamped operating current sequence.

[0023] In a preferred embodiment, the average operating current is output as follows:

[0024] A current sampling point is set in the main power supply circuit of the terahertz thickness measuring equipment to continuously collect the instantaneous working current value at a preset sampling frequency;

[0025] The average operating current within the corresponding time window is obtained by arithmetically averaging the filtered current values ​​over a fixed time window.

[0026] Each average operating current and its corresponding time window information are recorded and synchronized with the thickness data in time.

[0027] In a preferred embodiment, the specific process of establishing the fitting model is as follows:

[0028] The standard surface density of the standard sample was used as the dependent variable, and the measured terahertz thickness and average operating current were used as independent variables for fitting.

[0029] Residual analysis is performed on the fitting results. Abnormal samples with deviations exceeding the deviation threshold are removed and the model is refitted until the fitting error converges to the allowable range, thus obtaining the areal density inversion fitting model.

[0030] In a preferred embodiment, in the step of establishing the fitting model, only the average operating current is used as an electrical input quantity for fitting.

[0031] In a preferred embodiment, the step of acquiring the average operating current and terahertz measured thickness of the material under test in real time and recording them as test data includes:

[0032] Based on the production line speed and material location information, a positional mapping relationship between the current sampling window and the material segment is established, so that each current sampling window corresponds to a specific area of ​​the material being measured.

[0033] The real-time collected device operating current is preprocessed to obtain the current average operating current;

[0034] The average operating current and the corresponding terahertz measured thickness are synchronized in time and position within the same material region to form the data to be measured.

[0035] Each set of test data is appended with the data acquisition time, production line location code, and steady-state identifier, and temporarily stored in the real-time data buffer.

[0036] In a preferred embodiment, the step of inputting the average operating current and the measured terahertz thickness of the material under test into the fitting model for inversion to obtain the real-time areal density of the material under test includes:

[0037] Read the test data from the real-time data buffer and extract the steady-state flag of the test data;

[0038] If the steady-state flag indicates that the device is in the steady-state working window when the current data to be measured is acquired, then the average operating current and the terahertz measured thickness of the data to be measured are substituted into the established surface density inversion fitting model for calculation to obtain the preliminary surface density value.

[0039] If the steady-state flag display device is not in the steady-state working window, the inversion calculation will be paused and the system will wait for the next set of valid steady-state flag data.

[0040] The preliminary areal density values ​​calculated from multiple consecutive sets of steady-state data are weighted and averaged to obtain the final real-time areal density value. The data acquisition time, production line location code, and number of steady-state data sets involved in the calculation are recorded simultaneously for the real-time areal density value.

[0041] In a preferred embodiment, the fitted model further includes a verification and update mechanism, the specific execution process of which is as follows:

[0042] When equipment maintenance, product specification switching, or raw material batch changes occur, new standard samples or offline sampling samples should be selected.

[0043] Substitute the measured data of the new sample into the established fitting model, calculate the predicted areal density and compare it with the actual standard areal density to obtain the prediction error;

[0044] If the prediction error exceeds the allowable range, the parameters of the areal density inversion fitting model will be updated by refitting the model with the new sample data and the original standard sample data.

[0045] If the prediction error is within the allowable range, the parameters of the original fitted model are retained, and the fitted model is verified.

[0046] In a preferred embodiment, the step of refitting using both the new sample data and the original standard sample data includes:

[0047] Extract the average operating current value and the measured terahertz thickness value of the new sample, and record them as positioning anchor points;

[0048] With the positioning anchor point as the center, it extends outward in both the average working current dimension and the terahertz measured thickness to form a two-dimensional local range.

[0049] Determine whether the range of the two-dimensional local interval falls completely within the effective applicable range of the original fitting model;

[0050] If the range of the two-dimensional local interval exceeds the effective applicable range of the original fitting model, the two-dimensional local interval is truncated by the boundary of the effective applicable range to obtain the corrected local interval.

[0051] Traverse all historical standard samples in the original standard sample database and extract the corresponding average operating current and terahertz measured thickness to form a historical data point set;

[0052] Determine whether the average operating current of each historical standard sample falls within the current boundary of the local interval, and whether the measured terahertz thickness falls within the thickness boundary of the local interval.

[0053] Historical standard samples that simultaneously meet the screening conditions of current boundary and thickness boundary are summarized into a local sample subset;

[0054] The number of samples in the local sample subset is counted. If the number of samples is less than the minimum sample number threshold, the range of the two-dimensional local interval is gradually expanded while keeping the positioning anchor point unchanged, and the filtering is re-executed until the number of samples in the local sample subset meets the minimum sample number threshold.

[0055] The new sample data is merged with the local sample subset to form a local fitting dataset;

[0056] A new areal density inversion fitting model is obtained by refitting the local fitting dataset, replacing the corresponding local interval parameters of the original fitting model, and incrementally updating the fitting model to obtain the updated fitting model.

[0057] The updated fitting model is validated. If the validation is successful, the updated fitting model replaces the original fitting model to complete the model update. Otherwise, the original fitting model is retained, the model update failure event is recorded, and sent to the management terminal.

[0058] The present invention also provides a detection system using the above-described online detection method for material surface density based on average operating current inversion, comprising:

[0059] The initialization module is used to preprocess the collected device operating current and calculate the average operating current by averaging the preprocessed operating current.

[0060] The sample selection module is used to select multiple standard samples with known thickness and known areal density, and to collect the terahertz measured thickness and average operating current of each standard sample respectively.

[0061] The model building module is used to establish a fitting model based on the standard surface density, terahertz measured thickness, and average operating current of the standard sample.

[0062] The inversion preparation module is used to acquire the average operating current and terahertz measured thickness of the material under test in real time and record them as test data.

[0063] The inversion module is used to input the average operating current and the measured terahertz thickness of the material under test into the fitting model for inversion, so as to obtain the real-time surface density of the material under test.

[0064] The technical effects achieved by this invention are as follows:

[0065] This invention extracts the average operating current during the operation of a terahertz thickness gauge and combines it with the measured terahertz thickness to establish a surface density inversion fitting model. This eliminates the need for additional external surface density detection hardware, avoiding the high costs and safety burdens associated with X-ray inspection. It also solves the problem of optical inspection being susceptible to interference from surface environmental factors. By fully utilizing the electrical data generated by the terahertz thickness gauge itself, it achieves online continuous detection of material surface density without increasing system complexity. Furthermore, this invention uses only the average operating current as the core electrical input parameter to establish the model, reducing the impact of voltage fluctuations and transient interference in industrial settings on model stability and improving the reliability of the surface density inversion results.

[0066] This invention, by setting a local incremental update mechanism for the fitted model, can quickly complete model adaptation when product specifications change or raw material batches change. The adaptation process only calls local samples to complete the update, reducing the computational cost of full refitting. It can adapt to changes in working conditions during continuous production, ensure the long-term stability of areal density detection accuracy, and meet the online detection needs of continuous industrial production. Attached Figure Description

[0067] Figure 1 This is a schematic diagram of the method flow of the present invention;

[0068] Figure 2 This is a schematic diagram of the system modules of the present invention. Detailed Implementation

[0069] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0070] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0071] Secondly, the term "an embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in a preferred embodiment" appearing in different places throughout this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that mutually excludes other embodiments.

[0072] Please see Figure 1 As shown, this invention provides an online detection method for material surface density based on average operating current inversion, including:

[0073] S1. The average operating current is obtained by preprocessing the collected equipment operating current and calculating the average value of the preprocessed operating current.

[0074] In step S1, the acquisition of the device's operating current is fundamental to the detection process. Its purpose is to obtain basic data reflecting changes in the material's surface density. After the device's operating current is acquired, preprocessing is performed simultaneously to eliminate noise interference. Then, the average value of the preprocessed device operating current is calculated to obtain an average operating current that stably reflects the material's load state. The preprocessing steps include:

[0075] The collected raw equipment operating current signal is filtered and denoised to remove signal spikes caused by equipment vibration and electromagnetic interference. The filtering and denoising methods can be low-pass filtering or wavelet transform, combined with the moving average algorithm to further suppress random fluctuations and ensure the stationarity and continuity of the data sequence. For example, taking wavelet transform as an example, the signal is separated into approximate components and detail components by using the multi-scale decomposition characteristics, eliminating high-frequency noise components and retaining low-frequency trend signals that reflect load changes, thereby achieving preliminary purification of the signal.

[0076] Abnormal peaks and valleys in the noise-reduced working current signal are removed, and a sliding window mechanism is used to smooth the working current to obtain a smoothed working current. This process replaces instantaneous extreme values ​​with local averages, which reduces the impact of instantaneous interference on subsequent average calculations while preserving the overall trend of the signal, and avoids the overall average result being shifted by a single outlier.

[0077] A timestamp is added to the smoothed working current. The timestamp is used to identify the timing of data acquisition so that the current data and thickness data can be time-aligned to form a timestamped working current sequence. This ensures that data collected from different dimensions can correspond in the same material region under test, avoiding inversion errors caused by spatiotemporal misalignment.

[0078] Secondly, the output process of the average operating current is as follows:

[0079] A current sampling point is set in the main power supply circuit of the terahertz thickness measuring equipment. A current sensor is deployed under the current sampling point to continuously collect the instantaneous working current value in the circuit. The instantaneous working current value is continuously collected at a preset sampling frequency. The sampling frequency can be adjusted according to the production line speed and the segmented detection accuracy requirements of the material to be tested. For example, it can be set to 100Hz to 500Hz for high-speed production lines, and the sampling frequency can be appropriately reduced for low-speed or high-precision detection scenarios to reduce data redundancy.

[0080] The filtered current values ​​are arithmetically averaged over a fixed time window to obtain the average operating current within the corresponding time window. Specifically, the sum of all preprocessed instantaneous current values ​​within the current time window is calculated and then divided by the number of effective sampling points within the time window to eliminate the influence of instantaneous power fluctuations on the current values, thus obtaining the average operating current that reflects the average load level of the equipment during that period.

[0081] Each average operating current and its corresponding time window information are recorded and synchronized with the thickness data in time to establish the time-series correspondence between current and thickness, providing an accurate data foundation for the subsequent construction of the areal density inversion model.

[0082] S2. Select multiple standard samples with known thickness and known areal density, and collect the terahertz measured thickness and average operating current of each standard sample respectively.

[0083] In step S2, the standard samples cover the normal process range, allowable fluctuation range and typical abnormal range of the target product. For each standard sample, the standard sample number, standard thickness, standard areal density, material type, production batch or sample source, test temperature, terahertz measured thickness and the average operating current of the equipment under the corresponding operating conditions are recorded. The number of standard samples is not less than 15 sets, and includes at least 3 sets of boundary condition samples to ensure the coverage and robustness of the fitting model.

[0084] S3. Establish a fitting model based on the standard surface density, terahertz measured thickness, and average operating current of the standard sample;

[0085] In step S3, after the standard sample is collected, a fitting model is established based on it to form a mapping relationship between the average working current of the standard sample and the standard areal density. Subsequently, the areal density of the material to be tested can be inferred from the measured average working current, realizing non-contact, online areal density detection. The specific process of establishing the fitting model is as follows:

[0086] The fitting model can take one of the following forms:

[0087] Linear model: suitable for scenarios with stable operating conditions, single material type, and small range of areal density variation;

[0088] Piecewise linear model: suitable for scenarios where the slopes differ across different thickness ranges or different coating amounts;

[0089] Polynomial or regression model: suitable for scenarios where there is a nonlinear relationship between average operating current, thickness, and areal density;

[0090] Lookup table model: suitable for scenarios where standard sample data is sufficient and the goal is to reduce online computation.

[0091] Temperature-compensated model: suitable for scenarios where the device current or material response is significantly affected by temperature;

[0092] In the process of fitting a model, the standard areal density of the standard sample is used as the dependent variable, and the measured terahertz thickness and average operating current are used as independent variables. Only the average operating current is used as the electrical input to simplify the model input dimension, avoid introducing additional interference and errors from multiple electrical parameters, and reduce model complexity and online computation. Specifically, the independent variables are first normalized to eliminate the weighting effect of different physical quantities' dimensional differences on the fitting results. Then, the initial parameters are solved using the least squares method. The specific solution formula is as follows: Let the first... The average operating current of the standard samples is The measured thickness at the terahertz level is Standard surface density is Then the fitting model can be expressed as ,in For the selected model form of the function expression, For the random error term, minimize the sum of squared residuals. Solve for the initial fitting parameters of the model;

[0093] Residual analysis is performed on the fitting results. After removing outlier samples whose deviation exceeds the deviation threshold, the model is refitted until the fitting error converges to within the allowable range, thus obtaining the areal density inversion fitting model. The specific process of residual analysis is as follows:

[0094] First, calculate the absolute value of the residual for each sample. Mark samples with an absolute value of residual greater than three times the mean square error of the fit (deviation threshold) as outliers. After removing all marked outliers, refit the parameters using the remaining samples. Repeat the residual analysis and removal process until there are no outliers, and obtain a stable areal density inversion fitting model. At the same time, record the effective applicable range of the fitting model, that is, the value range of the independent variable average operating current and the terahertz measured thickness, to provide boundary basis for subsequent updates of the fitting model.

[0095] S4. Real-time acquisition of the average operating current and terahertz measured thickness of the material under test, and recording them as test data;

[0096] In step S4, after obtaining the measured terahertz thickness and average operating current of the material under test, these are recorded as test data. The specific process is as follows:

[0097] Based on the production line speed and material location information, a positional mapping relationship between the current sampling window and the material segment is established, so that each current sampling window corresponds to a specific area of ​​the material being measured. Specifically, based on the production line speed and sampling frequency, the time length of each current sampling window is converted into the corresponding material length, thereby establishing a one-to-one correspondence between the current sampling window and the material segment, and thus establishing a spatial correspondence between the current and the material segment.

[0098] The real-time acquired device operating current is preprocessed to obtain the current average operating current. This process is consistent with the current preprocessing method of the standard sample in step S1 above, including noise reduction, baseline correction and abnormal peak removal, in order to ensure data quality.

[0099] The average operating current and the corresponding terahertz measured thickness are synchronized in time and location according to the same material region to form the test data. Specifically, the timestamp of the current sampling window is aligned with the acquisition timestamp of the terahertz measured thickness to ensure that the two correspond to the same material region, thereby forming multiple sets of one-to-one test data.

[0100] Each set of test data is assigned a data acquisition time, production line location code, and steady-state identifier, and temporarily stored in a real-time data buffer. The purpose is to ensure data traceability and facilitate data filtering during subsequent anomaly backtracking and model updates.

[0101] S5. Input the average operating current and the measured terahertz thickness of the material under test into the fitting model for inversion to obtain the real-time surface density of the material under test.

[0102] In step S5, after the data to be measured is collected, it is used as an input parameter to call the areal density inversion fitting model for calculation, thereby obtaining the real-time areal density of the material to be measured. The specific process is as follows:

[0103] Read the test data from the real-time data buffer and extract the steady-state identifier of the test data. Only select the data group with valid steady-state identifier and remove invalid data in non-steady-state stages such as start-stop and speed change to ensure that the data input to the model meets the applicable boundary conditions of the fitted model.

[0104] If the steady-state flag indicates that the device is within the steady-state operating window when the current data to be measured is acquired, then the average operating current and the measured terahertz thickness of the data to be measured are substituted into the established areal density inversion fitting model for calculation to obtain a preliminary areal density value. Specifically, the average operating current and the measured terahertz thickness are substituted into the fitting formula to calculate a preliminary areal density value that matches the current operating condition. The fitting formula is as follows: ,in, Represents surface density, Indicates the average operating current. This indicates the measured thickness at the terahertz level.

[0105] If the steady-state indicator display device is not in the steady-state working window, the inversion calculation will be paused and the system will wait for the next set of valid steady-state indicator data. This is to avoid detection errors caused by current fluctuations under non-steady-state conditions such as production line start-up and shutdown and speed adjustment, and to improve the reliability of the final detection results.

[0106] The preliminary areal density values ​​calculated from multiple consecutive sets of steady-state data are weighted and averaged to obtain the final real-time areal density value. During the weighting process, weights are assigned according to the signal-to-noise ratio (SNR) of the data to reduce the interference of low SNR data on the final result and improve the accuracy of the areal density detection value. In addition, the data acquisition time, production line location code, and number of steady-state data sets involved in the calculation are recorded simultaneously for the real-time areal density value, providing complete information support for subsequent data traceability, anomaly diagnosis, and model optimization.

[0107] In addition, the fitted model also includes a verification and update mechanism, the specific execution process of which is as follows:

[0108] When equipment maintenance, product specification switching, or raw material batch changes occur, new standard samples or offline sampling samples are selected to ensure the applicability of the fitting model under the current working conditions.

[0109] Substitute the measured data of the new sample into the established fitting model, calculate the predicted areal density and compare it with the actual standard areal density to obtain the prediction error, which is the difference between the predicted areal density and the actual standard areal density.

[0110] If the prediction error exceeds the allowable range, the new sample data and the original standard sample data are used together to refit and update the parameters of the areal density inversion fitting model. The allowable range is set based on the standard deviation of historical detection data. Specifically, the standard deviation of historical detection data is multiplied by the corresponding confidence coefficient as the allowable error threshold. The confidence coefficient is generally in the range of 1.5 to 3.0 to balance the sensitivity and stability of the model update, thereby ensuring that the model can adapt to new working conditions while avoiding the loss of generalization ability to historical stable working conditions due to overfitting a small amount of new data.

[0111] If the prediction error is within the allowable range, the parameters of the original fitted model are retained, the fitted model is verified, and a verification pass record is generated as the basis for the continuous and stable operation of the model. The subsequently collected data can be directly substituted into the original model for surface density inversion calculation.

[0112] It should be noted that the step of refitting using both new sample data and existing standard sample data includes:

[0113] The average operating current value and the measured terahertz thickness value of the new sample are extracted and recorded as positioning anchor points. Specifically, the detection data of the new standard sample or offline sample collected during the verification stage are read, the measured average operating current value and the measured terahertz thickness value of the sample are extracted, and they are used together as positioning anchor points for the fitting model update. The positioning anchor points correspond to the combination of current and thickness characteristics under the new operating conditions and are used to define the local range of action of the fitting model update.

[0114] Centered on the positioning anchor point, extending outwards along both the average operating current dimension and the measured terahertz thickness, a two-dimensional local range is formed. The boundary in the current dimension is represented as follows: , Take 10% to 20% of the current fluctuation range in the original fitted model, and the same applies to the thickness dimension, to define the local range of model parameter updates.

[0115] Determine whether the range of the two-dimensional local interval falls completely within the effective applicable range of the original fitting model;

[0116] If the range of the two-dimensional local interval exceeds the effective applicable range of the original fitting model, the two-dimensional local interval is truncated at the boundary of the effective applicable range to obtain the corrected local interval; otherwise, the original two-dimensional local interval is directly retained as the basis for subsequent screening of historical data.

[0117] Traverse all historical standard samples in the original standard sample database and extract the corresponding average operating current and terahertz measured thickness to form a historical data point set;

[0118] Each historical standard sample is judged to determine whether its average operating current falls within the current boundary of the local interval and whether its terahertz measured thickness falls within the thickness boundary of the local interval. Historical standard samples that simultaneously meet the current boundary and thickness boundary screening conditions are summarized into a local sample subset. The samples in the local sample subset have the highest matching degree with the current and thickness characteristics of the new operating conditions and can be used as the basic data for local updates.

[0119] The number of samples in the local sample subset is counted. If the number of samples is less than the minimum sample size threshold, the range of the two-dimensional local interval is gradually expanded while keeping the positioning anchor point unchanged. The screening is then repeated until the number of samples in the local sample subset meets the minimum sample size threshold. The minimum sample size threshold is set to ensure the statistical significance and generalization ability of the local fit. It is usually set to 5% to 10% of the total number of original standard samples.

[0120] The new sample data is merged with the local sample subset to form a local fitting dataset. The local fitting dataset contains both new working condition samples and matched historical working condition samples, taking into account both the working condition adaptability and historical generalization ability of the model.

[0121] A new areal density inversion fitting model is obtained by refitting the local fitting dataset, replacing the corresponding local interval parameters of the original fitting model, and incrementally updating the fitting model to obtain the updated fitting model. Specifically, only the model parameters in the truncated local interval are recalculated and optimized, without changing the parameters of the other intervals of the original fitting model. This incremental update method can adapt to the changes in new working conditions and retain the parameters of the original model that have been verified to be effective in other intervals, avoiding the overall model drift caused by full fitting and reducing the impact of model updates on the normal production and testing process.

[0122] The updated fitting model is validated by selecting new sample data not used in the current fitting and historical standard sample data outside the local sample subset as validation sets. The average operating current and terahertz measured thickness of the validation sets are substituted into the updated model to calculate the error between the predicted areal density and the standard areal density, which is recorded as the validation error. The validation error is then compared with the aforementioned allowable error threshold. If the validation error is still within the allowable range, the validation is considered successful, and the updated fitting model replaces the original fitting model to complete the model update. If the validation error exceeds the allowable range, the original fitting model is retained, the model update failure event is recorded, an alarm notification is triggered, and a notification to the management terminal is sent to request manual intervention.

[0123] Please see Figure 2 A detection system, using the above-described online detection method based on average operating current inversion of material surface density, includes:

[0124] The initialization module is used to preprocess the collected device operating current and calculate the average operating current by averaging the preprocessed operating current.

[0125] The sample selection module is used to select multiple standard samples with known thickness and known areal density, and to collect the terahertz measured thickness and average operating current of each standard sample respectively.

[0126] The model building module is used to establish a fitting model based on the standard surface density, terahertz measured thickness, and average operating current of the standard sample.

[0127] The inversion preparation module is used to acquire the average operating current and terahertz measured thickness of the material under test in real time and record them as test data.

[0128] The inversion module is used to input the average operating current and the measured terahertz thickness of the material under test into the fitting model for inversion, so as to obtain the real-time surface density of the material under test.

[0129] The execution process of the above detection system corresponds to the process of the aforementioned method, and will not be repeated here.

[0130] The above description is merely a preferred embodiment of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention. Structures, devices, and operating methods not specifically described or explained in this invention are implemented according to conventional methods in the art unless otherwise specified or limited.

Claims

1. An online detection method for material surface density based on average operating current inversion, characterized in that: include: The average operating current is obtained by preprocessing the collected equipment operating current and averaging the preprocessed operating current. Multiple standard samples with known thickness and known areal density were selected, and the terahertz measured thickness and average operating current of each standard sample were collected respectively. A fitting model is established based on the standard areal density of the standard sample, the measured thickness at the terahertz level, and the average operating current. The average operating current and terahertz measured thickness of the material under test are acquired in real time and recorded as test data; The average operating current and the measured terahertz thickness of the material under test are input into the fitting model for inversion to obtain the real-time surface density of the material under test.

2. The online detection method for material surface density based on average operating current inversion according to claim 1, characterized in that: The preprocessing steps include: The acquired raw equipment operating current signal is filtered and noise-reduced to remove signal spikes caused by equipment vibration and electromagnetic interference. Abnormal peaks and valleys in the noise-reduced operating current signal are removed, and a sliding window mechanism is used to smooth the operating current to obtain a smoothed operating current. Add timestamps to the smoothed operating current to form a timestamped operating current sequence.

3. The online detection method for material surface density based on average operating current inversion according to claim 1, characterized in that: The output process of the average operating current is as follows: A current sampling point is set in the main power supply circuit of the terahertz thickness measuring equipment to continuously collect the instantaneous working current value at a preset sampling frequency; The average operating current within the corresponding time window is obtained by arithmetically averaging the filtered current values ​​over a fixed time window. Each average operating current and its corresponding time window information are recorded and synchronized with the thickness data in time.

4. The online detection method for material surface density based on average operating current inversion according to claim 1, characterized in that: The specific process for establishing the fitting model is as follows: The standard surface density of the standard sample was used as the dependent variable, and the measured terahertz thickness and average operating current were used as independent variables for fitting. Residual analysis is performed on the fitting results. Abnormal samples with deviations exceeding the deviation threshold are removed and the model is refitted until the fitting error converges to the allowable range, thus obtaining the areal density inversion fitting model.

5. The online detection method for material surface density based on average operating current inversion according to claim 4, characterized in that: In the step of establishing the fitting model, only the average operating current is used as the electrical input quantity for fitting.

6. The online detection method for material surface density based on average operating current inversion according to claim 1, characterized in that: The step of acquiring the average operating current and terahertz measured thickness of the material under test in real time and recording them as test data includes: Based on the production line speed and material location information, a positional mapping relationship between the current sampling window and the material segment is established, so that each current sampling window corresponds to a specific area of ​​the material being measured. The real-time collected device operating current is preprocessed to obtain the current average operating current; The average operating current and the corresponding terahertz measured thickness are synchronized in time and position within the same material region to form the data to be measured. Each set of test data is appended with the data acquisition time, production line location code, and steady-state identifier, and temporarily stored in the real-time data buffer.

7. The online detection method for material surface density based on average operating current inversion according to claim 1, characterized in that: The step of inputting the average operating current and the measured terahertz thickness of the material under test into the fitting model for inversion to obtain the real-time areal density of the material under test includes: Read the test data from the real-time data buffer and extract the steady-state flag of the test data; If the steady-state flag indicates that the device is in the steady-state working window when the current data to be measured is acquired, then the average operating current and the terahertz measured thickness of the data to be measured are substituted into the established surface density inversion fitting model for calculation to obtain the preliminary surface density value. If the steady-state flag display device is not in the steady-state working window, the inversion calculation will be paused and the system will wait for the next set of valid steady-state flag data. The preliminary areal density values ​​calculated from multiple consecutive sets of steady-state data are weighted and averaged to obtain the final real-time areal density value. The data acquisition time, production line location code, and number of steady-state data sets involved in the calculation are recorded simultaneously for the real-time areal density value.

8. The online detection method for material surface density based on average operating current inversion according to claim 1, characterized in that: The fitted model also includes a verification and update mechanism, the specific execution process of which is as follows: When equipment maintenance, product specification switching, or raw material batch changes occur, new standard samples or offline sampling samples should be selected. Substitute the measured data of the new sample into the established fitting model, calculate the predicted areal density and compare it with the actual standard areal density to obtain the prediction error; If the prediction error exceeds the allowable range, the parameters of the areal density inversion fitting model will be updated by refitting the model with the new sample data and the original standard sample data. If the prediction error is within the allowable range, the parameters of the original fitted model are retained, and the fitted model is verified.

9. The online detection method for material surface density based on average operating current inversion according to claim 1, characterized in that: The step of refitting using new sample data and existing standard sample data includes: Extract the average operating current value and the measured terahertz thickness value of the new sample, and record them as positioning anchor points; With the positioning anchor point as the center, it extends outward in both the average working current dimension and the terahertz measured thickness to form a two-dimensional local range. Determine whether the range of the two-dimensional local interval falls completely within the effective applicable range of the original fitting model; If the range of the two-dimensional local interval exceeds the effective applicable range of the original fitting model, the two-dimensional local interval is truncated by the boundary of the effective applicable range to obtain the corrected local interval. Traverse all historical standard samples in the original standard sample database and extract the corresponding average operating current and terahertz measured thickness to form a historical data point set; Determine whether the average operating current of each historical standard sample falls within the current boundary of the local interval, and whether the measured terahertz thickness falls within the thickness boundary of the local interval. Historical standard samples that simultaneously meet the screening conditions of current boundary and thickness boundary are summarized into a local sample subset; The number of samples in the local sample subset is counted. If the number of samples is less than the minimum sample number threshold, the range of the two-dimensional local interval is gradually expanded while keeping the positioning anchor point unchanged, and the filtering is re-executed until the number of samples in the local sample subset meets the minimum sample number threshold. The new sample data is merged with the local sample subset to form a local fitting dataset; A new areal density inversion fitting model is obtained by refitting the local fitting dataset, replacing the corresponding local interval parameters of the original fitting model, and incrementally updating the fitting model to obtain the updated fitting model. The updated fitting model is validated. If the validation is successful, the updated fitting model replaces the original fitting model to complete the model update. Otherwise, the original fitting model is retained, the model update failure event is recorded, and sent to the management terminal.

10. A detection system, characterized in that: The online detection method for material surface density based on average operating current inversion as described in any one of claims 1 to 9 includes: The initialization module is used to preprocess the collected device operating current and calculate the average operating current by averaging the preprocessed operating current. The sample selection module is used to select multiple standard samples with known thickness and known areal density, and to collect the terahertz measured thickness and average operating current of each standard sample respectively. The model building module is used to establish a fitting model based on the standard surface density, terahertz measured thickness, and average operating current of the standard sample. The inversion preparation module is used to acquire the average operating current and terahertz measured thickness of the material under test in real time and record them as test data. The inversion module is used to input the average operating current and the measured terahertz thickness of the material under test into the fitting model for inversion, so as to obtain the real-time surface density of the material under test.

Images