A critical point drying control method and device based on online near-infrared monitoring

Abstract

The present application relates to electron microscope sample pre-treatment technical field, disclose a kind of critical point drying control method and device based on online near infrared monitoring, the method comprises: obtaining infrared spectrum data, obtains the characteristic data set of solvent by spectral data unmixing;According to characteristic data set, output the signal to be corrected, and correct with preset data, determine the signal characteristic of solvent;Signal characteristic is dynamically modified with preset characteristic matrix, and the concentration time series data of solvent is output, and steady state determination is carried out in combination with preset convergence threshold, determine the critical state of solvent;According to critical state, obtain control strategy data, analysis is executed response characteristic and carries out multi-channel mapping and correction processing, and output process control signal;Process control signal is controlled gain correction, and the closed-loop feedback control of final drying process is completed.The present application is based on multi-solvent spectrum sensing and concentration dynamic analysis, realizes the adaptive closed-loop control of critical point drying process.

Description

Technical Field

[0001] This invention relates to the field of sample pretreatment technology for electron microscopy, and in particular to a critical point drying control method and apparatus based on online near-infrared monitoring. Background Technology

[0002] Critical point drying is a crucial step in processing fragile samples such as biological soft tissues and hydrogels before scanning electron microscopy (SEM) observation. Its control precision directly determines the preservation quality of the sample's three-dimensional structure. Achieving accurate determination and adaptive control of the critical point during drying is key to improving sample processing success rate, efficiency, and consistency.

[0003] In one existing technology, critical point drying control mainly employs methods based on preset time programs or fixed physical parameter thresholds. Although this method introduces sensors such as temperature and pressure for process monitoring, its control logic essentially still relies on static judgments of single or limited physical quantities, failing to correlate with the dynamic changes in chemical composition information of the multi-solvent system during the drying process. The root cause lies in the limitations of existing methods in their ability to perform real-time, in-situ analysis and fusion of multi-component solvent concentration signals, particularly in their inability to utilize sensor network chips for high-throughput, parallel near-infrared spectral data acquisition and intelligent front-end processing. This makes it impossible to accurately separate and track the real-time concentrations of key solvents in complex mixed systems, resulting in delayed and inaccurate judgment of the critical point.

[0004] In summary, existing technologies lack the ability to perceive key chemical indicators in the drying process in real time and in situ, and the control logic is disconnected from the physical state, lacking closed-loop feedback control capabilities based on real-time chemical indicator perception. Summary of the Invention

[0005] This invention provides a critical point drying control method and apparatus based on online near-infrared monitoring, aiming to solve the problem that existing critical point drying control methods are difficult to achieve closed-loop feedback control of the drying system.

[0006] In a first aspect, to solve the above-mentioned technical problems, the present invention provides a critical point drying control method based on online near-infrared monitoring, comprising:

[0007] The raw infrared spectral data is acquired, and a multidimensional spectral matrix is ​​constructed based on the raw infrared spectral data. Interference correction and spectral demixing are then performed to obtain the characteristic datasets of each solvent.

[0008] The feature dataset is subjected to mode decomposition and feature frequency band localization to output the feature signal to be corrected. The feature signal to be corrected is then fitted and corrected with the preset standard solvent fingerprint data to determine the signal features corresponding to each solvent.

[0009] Based on the signal characteristics, the instantaneous concentration is obtained by inverting the preset solvent absorbance characteristic matrix. The instantaneous concentration is then dynamically corrected and its changing trend is analyzed, and the concentration curves of each solvent are output.

[0010] Acquire the concentration time-series data of the concentration curve, construct a multi-dimensional feature matrix based on the concentration time-series data, and combine it with a preset convergence threshold to determine the steady state and the critical state of each solvent drying process.

[0011] Based on the critical state, an operating state vector is constructed, and the operating state vector is processed based on the feedback compensation adjustment mechanism of the rate deviation to obtain control strategy data for dynamically regulating the drying process.

[0012] The control strategy data is parsed into execution response features, and the execution response features are subjected to multi-channel mapping and correction processing to output a stable process control signal.

[0013] A state timing diagram is generated based on the process control signal. A quality-efficiency correlation feature vector is extracted based on the state timing diagram, and the control gain is corrected to complete the closed-loop feedback control of the final drying process.

[0014] Secondly, the present invention provides a critical point drying control device based on online near-infrared monitoring, characterized in that it comprises:

[0015] The data acquisition module is used to acquire raw infrared spectral data, construct a multidimensional spectral matrix based on the raw infrared spectral data, and perform interference correction and spectral demixing to obtain the characteristic dataset of each solvent.

[0016] The signal feature determination module is used to perform mode decomposition and feature frequency band localization on the feature dataset, output the feature signal to be corrected, fit and correct the feature signal to be corrected with the preset standard solvent fingerprint data, and determine the signal features corresponding to each solvent.

[0017] The concentration curve generation module is used to obtain the instantaneous concentration based on the signal characteristics and a preset solvent absorbance characteristic matrix, perform dynamic correction and trend analysis on the instantaneous concentration, and output the concentration curves of each solvent.

[0018] The critical state determination module is used to acquire the concentration time series data of the concentration curve, construct a multi-dimensional feature matrix based on the concentration time series data, and perform steady state determination in combination with a preset convergence threshold to determine the critical state of each solvent drying process.

[0019] The strategy data generation module is used to construct an operating state vector based on the critical state, and process the operating state vector based on the feedback compensation adjustment mechanism of the rate deviation to obtain control strategy data for dynamically regulating the drying process.

[0020] The process control signal generation module is used to parse the control strategy data into execution response features, perform multi-channel mapping and correction processing on the execution response features, and output a stable process control signal.

[0021] The closed-loop feedback control module is used to generate a state timing diagram based on the process control signal, extract the quality-efficiency correlation feature vector based on the state timing diagram and perform control gain correction to complete the closed-loop feedback control of the final drying process.

[0022] Compared with the prior art, the present invention has the following beneficial effects:

[0023] (1) This invention achieves effective separation and extraction of characteristic signals of each solvent in a multi-solvent mixture system by performing interference correction, spectral demixing, and multi-dimensional feature modeling on near-infrared spectral data acquired online. This method breaks through the limitation of existing methods that rely solely on a single physical parameter or overall spectral intensity change for judgment during the critical point drying process, enabling the key chemical component information during the drying process to be sensed in real time and in situ, thereby providing a stable and reliable data foundation for subsequent critical state determination;

[0024] (2) This invention achieves instantaneous concentration inversion by performing modal decomposition, characteristic frequency band localization, and fingerprint data fitting and correction on solvent characteristic signals, combined with solvent light absorption characteristics, and further outputs continuous concentration change curves. This technology establishes a quantitative correlation between spectral information and solvent concentration, enabling accurate tracking of the dynamic evolution of multiple solvent concentrations during the drying process, effectively avoiding the problem of inaccurate concentration judgment caused by solvent mixing or signal drift in the prior art;

[0025] (3) This invention constructs a multi-dimensional feature matrix based on solvent concentration time-series data and combines it with a convergence threshold for steady-state determination, thereby achieving accurate identification of the critical state of the drying process. This method introduces the solvent concentration change rate and evolution trend into the critical point judgment logic, overcoming the lag and uncertainty caused by relying on fixed time or a single threshold judgment in traditional methods, and improving the accuracy and consistency of critical point judgment;

[0026] (4) Based on the determination of the critical state, this invention generates dynamic control strategy data by constructing an operating state vector and introducing a feedback compensation adjustment mechanism based on rate deviation. Combined with closed-loop feedback, the control gain is adaptively corrected to achieve real-time control of the drying process. This technology tightly couples the chemical state perception results with the control execution process, enabling the drying process to be dynamically adjusted according to actual state changes, thereby significantly improving the control accuracy, process stability, and sample processing quality of critical point drying. Attached Figure Description

[0027] Figure 1This is a schematic flowchart of a critical point drying control method based on online near-infrared monitoring provided in an embodiment of the present invention;

[0028] Figure 2 This is a schematic diagram of a critical point drying control device based on online near-infrared monitoring provided in an embodiment of the present invention. Detailed Implementation

[0029] 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.

[0030] To solve the above problems, refer to Figure 1 The first embodiment of the present invention provides a method for intelligent site selection of charging piles based on big data analysis, including the following steps:

[0031] S1. Obtain the raw infrared spectral data, construct a multidimensional spectral matrix based on the raw infrared spectral data, and perform interference correction and spectral demixing to obtain the characteristic dataset of each solvent.

[0032] S2, perform mode decomposition and feature frequency band localization on the feature dataset, output the feature signal to be corrected, fit and correct the feature signal to be corrected with the preset standard solvent fingerprint data, and determine the signal features corresponding to each solvent;

[0033] S3. Based on the signal characteristics, the instantaneous concentration is obtained by inverting the preset solvent absorbance characteristic matrix. The instantaneous concentration is dynamically corrected and its changing trend is analyzed, and the concentration curves of each solvent are output.

[0034] S4, acquire the concentration time series data of the concentration curve, construct a multi-dimensional feature matrix based on the concentration time series data and combine it with a preset convergence threshold to determine the steady state and the critical state of each solvent drying process.

[0035] S5. Construct an operating state vector based on the critical state, and process the operating state vector based on the feedback compensation adjustment mechanism of the rate deviation to obtain control strategy data for dynamically regulating the drying process.

[0036] S6, the control strategy data is parsed into execution response features, the execution response features are subjected to multi-channel mapping and correction processing, and a stable process control signal is output;

[0037] S7. Generate a state timing diagram based on the process control signal, extract the quality-efficiency correlation feature vector based on the state timing diagram and perform control gain correction to complete the closed-loop feedback control of the final drying process.

[0038] In step S1, the raw infrared spectral data is acquired, a multidimensional spectral matrix is ​​constructed based on the raw infrared spectral data, and interference correction and spectral demixing are performed to obtain the feature datasets of each solvent, including:

[0039] The raw infrared spectral data is converted into a multidimensional spectral matrix, and non-stationary baseline fluctuation regions in the multidimensional spectral matrix are identified. An environmental interference correction model is constructed based on the characteristics of the non-stationary baseline fluctuation regions.

[0040] The multidimensional spectral matrix is ​​input into the environmental interference correction model for differential correction to obtain interference-free mixed spectral data;

[0041] The mixed spectral data is decomposed and reconstructed using wavelet transform to obtain enhanced mixed spectral data. The enhanced mixed spectral data is then subjected to blind source separation using an independent component analysis algorithm to demix the independent spectral components.

[0042] Effective spectral components are selected based on the characteristic peak positions of the independent spectral components, and the effective spectral components are mapped to the preliminary characteristic signals of each solvent to obtain the feature dataset after preliminary separation of each solvent.

[0043] In one implementation, this embodiment uses an infrared spectrometer array to perform continuous or periodic spectral scans on a mixed liquid containing multiple solvents to acquire raw infrared spectral data covering a preset wavenumber range, such as a series of wavenumbers in the range of 4000-400 cm⁻¹. -1 The raw absorption spectral data are presented as absorbance versus wavenumber, reflecting the absorption characteristics of different solvents in the mixed system to infrared radiation.

[0044] Furthermore, the raw infrared spectral data are arranged in chronological order of acquisition, and a corresponding multidimensional spectral matrix is ​​constructed using wavenumber, time, and measurement channel dimensions as indexes. This matrix is ​​used to uniformly describe the temporal evolution characteristics of spectral information during the drying process. For example, rows represent different sampling points or times, and columns represent wavenumber points, forming a two-dimensional or higher-dimensional matrix.

[0045] In one implementation, to identify non-stationary baseline fluctuation regions in the multidimensional spectral matrix, the spectral signals of each spectral channel are segmented and statistically analyzed according to time series, calculating the local mean, variance, and first-order difference rate of change. A preset fluctuation threshold is then used to determine the fluctuation amplitude and abrupt change points. By performing clustering or connectivity analysis on the fluctuation characteristics of each channel signal, time periods and bands with significant non-stationary baseline changes are marked, forming an index matrix or mask map of non-stationary baseline fluctuation regions, thereby accurately identifying spectral regions affected by environmental interference.

[0046] It should be noted that the fluctuation threshold is a criterion used to determine whether local fluctuations in the spectral signal are significant. The local mean and standard deviation are calculated for each channel of the multidimensional spectral matrix using a sliding window segmentation method. Then, based on historical environmental noise data or pre-collected stable spectral data, the fluctuation range is statistically analyzed, and the mean variation amplitude and variance range of each channel are calculated. Finally, the upper limit of the mean variation amplitude and variance range is determined using a statistical distribution method and serves as the fluctuation threshold.

[0047] Furthermore, the non-stationary baseline fluctuation regions of each channel in the multidimensional spectral matrix are identified, and features such as amplitude, frequency, and duration are extracted. Based on the extracted non-stationary baseline features, a baseline model is established using either polynomial fitting or asymmetric least squares methods, and the local offset and trend change parameters of each channel are calculated. Finally, the baseline model and channel parameters are combined to form an environmental disturbance correction model.

[0048] In one implementation, for the determined mixed spectral data, this embodiment calculates the deviation between the spectral value of each sampling point in each channel of the multidimensional spectral matrix and the corresponding baseline value, and uses the deviation amplitude, deviation frequency, and deviation duration characteristics to determine the correction weight. Subsequently, the environmental interference correction model performs differential correction on the spectral value of each sampling point according to the correction weight, that is, subtracting the weighted deviation from the spectral value and performing cumulative adjustment, continuously processing all time points and channels. To reduce the influence of high-frequency noise, a moving average or low-pass filter can be applied to the deviation sequence during the correction process so that extreme fluctuations do not excessively affect the correction result. Finally, the spectral values ​​after differential correction for all channels and time points are integrated to form the interference-free mixed spectral data.

[0049] In one implementation, for the determined independent spectral components, this embodiment constructs the enhanced mixed spectral data into an observation signal matrix, centers it, calculates the mean of each channel signal and subtracts it from the corresponding signal to eliminate the influence of the DC component. Subsequently, the centered signal matrix is ​​whitened, and eigenvalue decomposition or singular value decomposition is used to make the channel signals mutually orthogonal and have consistent variances. Based on the whitened signal matrix, an independent component analysis algorithm is used to iteratively solve the unmixing matrix, and the unmixing result is normalized in each iteration. The iteration stops when the change in the unmixing matrix is ​​less than a preset convergence threshold. Specifically, in each iteration, the norm difference between the current unmixing matrix and the previous iteration's unmixing matrix is ​​calculated. When the norm difference is less than the preset convergence threshold, the unmixing matrix is ​​determined to be stable and the iteration terminates. Finally, by applying the unmixing matrix to the enhanced mixed spectral data, several mutually independent spectral components are obtained through unmixing.

[0050] It should be noted that the preset convergence threshold is pre-set to a fixed small value based on numerical stability requirements. This small value is used to limit the minimum change scale of the unmixing matrix update, for example, it is set to no more than 10. -4 Or 10 -5 On the order of magnitude; in another implementation, the preset convergence threshold is normalized according to the amplitude range of the enhanced mixed spectral data, so that the change is negligible relative to the overall signal amplitude, thereby ensuring the convergence of the demixing result in a numerical sense.

[0051] In one implementation, for each selected effective spectral component, this embodiment calculates the spectral amplitude distribution and local extremum locations of each independent spectral component, and extracts the characteristic peak information of the independent spectral component, including the center wavelength, peak amplitude, and peak width parameters of the characteristic peak. Subsequently, the center wavelength of the characteristic peak is compared with a preset solvent characteristic absorption band range. When the center wavelength of the characteristic peak falls within the characteristic absorption band of the corresponding solvent, the independent spectral component is marked as a candidate effective spectral component.

[0052] Furthermore, the candidate valid spectral components are screened for amplitude stability and signal-to-noise characteristics. Specifically, the amplitude change rate and variance of the candidate valid spectral components are calculated at continuous sampling times. When both the amplitude change rate and variance are within a preset stable range, the independent spectral component is determined to meet the stability requirements. Simultaneously, the signal-to-noise ratio parameter of the independent spectral component in the characteristic peak band is calculated. When the signal-to-noise ratio parameter is higher than a preset signal-to-noise threshold, the independent spectral component is retained as a valid spectral component, and other independent spectral components that do not meet the conditions are eliminated.

[0053] It should be noted that the preset solvent characteristic absorption band range is determined based on pre-acquired standard solvent fingerprint spectra. Specifically, for each solvent to be monitored, its standard near-infrared spectral data is collected under independent conditions. After smoothing the standard spectrum, the center wavelength corresponding to its absorption peak is extracted. Using the center wavelength as a reference, a preset wavelength tolerance range is extended to both sides to form the characteristic absorption band range of the corresponding solvent. The preset stable range is determined based on the reference measurement data of independent spectral components under conditions without external disturbance. Specifically, the amplitude data of the same independent spectral component in multiple consecutive sampling periods are statistically analyzed, and its mean and standard deviation are calculated. Using the mean as the center, a stable range is constructed according to a preset multiple of the standard deviation range. When the amplitude change rate and variance of the independent spectral component during operation fall within the stable range, it is determined that it meets the stability condition. The preset signal-to-noise ratio (SNR) threshold is determined by calibration measurement of the standard solvent spectral data. Specifically, the signal amplitude is calculated in the solvent characteristic absorption band, and the noise amplitude is estimated in the adjacent non-absorption band. The SNR is calculated based on the ratio of the two. By statistically analyzing multiple sets of standard measurement results, the minimum signal-to-noise ratio that can stably distinguish between effective feature signals and noise signals is selected as the preset signal-to-noise judgment threshold.

[0054] In one implementation, for a given feature dataset, this embodiment extracts the corresponding characteristic peak center wavelength and peak shape parameters for each effective spectral component, and matches the characteristic peak center wavelength with a preset solvent characteristic absorption band range. When the characteristic peak center wavelength of a certain effective spectral component falls within the characteristic absorption band range corresponding to a certain solvent, a correspondence is established between the effective spectral component and the solvent, and it is marked as a candidate feature signal of the solvent.

[0055] Furthermore, after band matching is completed, one or more candidate feature signals corresponding to the same solvent are normalized and amplitude aligned. Specifically, based on the relative intensity relationship of corresponding absorption peaks in the standard solvent fingerprint spectrum, the amplitude ratio of the candidate feature signals is adjusted to ensure consistency with the standard solvent fingerprint data in terms of amplitude scale. Subsequently, the normalized candidate feature signals are combined according to time or sampling order to construct the preliminary feature signal of the solvent. By performing the above mapping and combination process on each solvent, a feature dataset containing preliminary feature signals of multiple solvents is formed, which is output as the feature dataset after preliminary separation of each solvent.

[0056] In step S2, the modal decomposition and feature frequency band localization of the feature dataset are performed to output the feature signal to be corrected. The feature signal to be corrected is then fitted and corrected with preset standard solvent fingerprint data to determine the signal features corresponding to each solvent. This includes:

[0057] The feature dataset is decomposed into intrinsic mode components using an adaptive variational mode decomposition algorithm.

[0058] The intrinsic mode components are used to construct the Hilbert marginal spectrum and locate the main characteristic frequency band, thereby generating the characteristic signal to be corrected;

[0059] Calculate the residual sequence between the feature signal to be corrected and the preset standard solvent fingerprint data, and use the least squares method to perform fitting regression processing on the residual sequence to output the corrected feature signal;

[0060] The independent signal waveforms are reconstructed based on the correction feature signals to determine the signal characteristics corresponding to each solvent.

[0061] In one implementation, to determine the intrinsic modal components, this embodiment uses the feature datasets after preliminary separation of each solvent as input signals to construct a signal sequence to be decomposed. First, the parameters of the variational mode decomposition model are initialized according to a preset number of modes, including the initial values ​​of the center frequencies and bandwidth constraint parameters of each mode. The number of modes can be preset according to the spectral structure of the feature dataset, or adjusted during the decomposition process through an adaptive update mechanism. Subsequently, based on the constraint optimization model of variational mode decomposition, the input signal is represented as a superposition of several finite-bandwidth modal signals, and each modal signal and its corresponding center frequency are updated simultaneously through an iterative method.

[0062] Furthermore, in each iteration, the remaining modal signals are fixed, and only the current modal signal is updated to concentrate it near its corresponding center frequency in the frequency domain. Simultaneously, the center frequency is recalculated based on the updated modal signal to reflect the dominant frequency position of the current mode. During the iteration process, the change in each modal signal between two adjacent iterations is calculated. When the change is less than a preset convergence threshold, the iteration operation stops. Finally, the output set of converged modal signals represents the intrinsic modal components, each corresponding to a relatively independent frequency component in the input feature dataset.

[0063] It should be noted that the preset convergence threshold is set by statistically analyzing the difference in energy change of each intrinsic mode component in two adjacent iterations. Specifically, it involves statistically analyzing the energy difference of mode updates during historical or prior spectral decomposition, selecting the maximum energy change value corresponding to the stable phase, and then scaling the value proportionally to determine the convergence threshold.

[0064] In one implementation, after obtaining the intrinsic mode components, this embodiment performs time-frequency characteristic analysis on each intrinsic mode component and constructs a corresponding Hilbert marginal spectrum based on the intrinsic mode components to characterize the distribution of signal energy in different frequency bands. By analyzing the Hilbert marginal spectrum, the system identifies frequency bands with high energy proportions and continuous response in the time dimension, and determines them as the main characteristic frequency bands. Subsequently, signal components located in the main characteristic frequency bands are extracted from each intrinsic mode component, and the extracted signal components are superimposed and reconstructed to form a characteristic signal that can centrally reflect solvent absorption behavior and has effectively suppressed non-characteristic frequency band interference. The reconstructed characteristic signal is used as the characteristic signal to be corrected.

[0065] In one implementation, this embodiment performs point-by-point alignment processing between the feature signal to be corrected and preset standard solvent fingerprint data to ensure their consistency in the wavenumber dimension. Subsequently, the system calculates the difference value between the feature signal to be corrected and the standard solvent fingerprint data at the corresponding wavenumber positions, and arranges the difference values ​​at each wavenumber position into a residual sequence to characterize the offset of the current measurement signal relative to the standard fingerprint.

[0066] Furthermore, after obtaining the residual sequence, this embodiment performs regression fitting processing on the residual sequence to reduce the impact of systematic bias on signal characteristics. The regression fitting processing employs a parameter estimation method based on the principle of error minimization. By performing overall fitting on the residual sequence, correction parameters for correcting the feature signal to be corrected are determined. Subsequently, the system compensates and adjusts the feature signal to be corrected according to the correction parameters, eliminating systematic errors caused by instrument drift, environmental changes, or mixed interference, thereby obtaining the corrected feature signal.

[0067] It should be noted that the standard solvent fingerprint data is pre-established during the system initialization phase. In this embodiment, for each target solvent, infrared spectral data of the single solvent is collected under standard experimental conditions, and preprocessing operations such as baseline correction and noise suppression are performed on the collected spectral data. Subsequently, representative spectral response characteristics of each solvent in the main absorption bands are extracted from the processed spectra, and normalized according to the wavenumber dimension to form the corresponding standard solvent fingerprint data.

[0068] In one implementation, this embodiment uses the correction feature signal as input and performs segmented reconstruction processing on the correction feature signal according to the characteristic frequency band division results corresponding to different solvents. Specifically, the system extracts signal components related to each solvent from the correction feature signal based on the main absorption frequency band position corresponding to each solvent in the standard solvent fingerprint data, and performs superposition and smoothing processing on the extracted signal components to form an independent signal waveform characterizing the absorption behavior of a single solvent. The independent signal waveform is used to reflect the stable spectral response characteristics of the solvent under the current operating conditions and serves as the basis for determining the signal characteristics corresponding to each solvent.

[0069] In step S3, the instantaneous concentration is obtained by inverting the signal characteristics and the preset solvent absorbance characteristic matrix, and the instantaneous concentration is dynamically corrected and its trend is analyzed to output the concentration curves of each solvent, including:

[0070] Based on the signal characteristics, the instantaneous concentration of each solvent is obtained by inversion using a preset solvent absorption characteristic matrix;

[0071] Based on the preset window length and recursion strategy, a real-time dynamic response window is created, and the instantaneous concentration is input into the real-time dynamic response window for smoothing and filtering correction to obtain the corrected component content data.

[0072] Based on the differential calculation results of the component content data, the direction and rate of increase or decrease of each solvent component are determined, and based on the direction and rate of increase or decrease, a component change trend description vector of each solvent is generated;

[0073] Based on the component change trend description vector, the component content data is time-domain mapped and interpolated to output the concentration curves of each solvent's dynamic changes.

[0074] In one implementation, to determine the instantaneous concentration of each solvent, this embodiment combines a preset solvent absorbance characteristic matrix to perform concentration inversion processing on the signal features. Specifically, the solvent absorbance characteristic matrix describes the unit concentration response relationship of different solvents in each characteristic frequency band, reflecting the correspondence between signal amplitude changes and solvent concentrations. The system matches the response intensity of the signal features in the corresponding characteristic frequency band with the solvent absorbance characteristic matrix, and, assuming that the absorption contributions of each solvent have a linear superposition relationship, solves for the estimated concentration of each solvent at the current sampling time, thereby obtaining the instantaneous concentration of each solvent.

[0075] It should be noted that the solvent absorbance characteristic matrix is ​​pre-established during the system initialization phase. In this embodiment, for each target solvent, infrared spectral data at different known concentration levels are collected under standard experimental conditions, and the corresponding signal response intensity is extracted within the characteristic frequency band consistent with step S2. Subsequently, based on the collected multiple sets of concentration-response data, the unit concentration response relationship of each solvent in each characteristic frequency band is statistically determined, and the response relationship is organized according to solvent category and characteristic frequency band dimension to form the solvent absorbance characteristic matrix.

[0076] In one implementation, the instantaneous concentration is dynamically corrected. This embodiment introduces a real-time dynamic response window to process the retrieved instantaneous concentration sequence. A preset window length (e.g., covering data from the most recent 20 sampling times) is used, and this window is continuously updated in a sliding manner. The instantaneous concentration data within the window first undergoes a smoothing filter, such as using a moving average algorithm to suppress random noise. Subsequently, outliers within the window are detected and corrected. For example, if the concentration value at a certain moment changes drastically due to instantaneous disturbances, it is corrected to a value consistent with the trend of the preceding and following moments. After this processing, a more stable and reliable set of corrected component content data is output.

[0077] For example, assuming that the instantaneous concentration of methanol fluctuates greatly in a short period of time during continuous monitoring, such as 0.019, 0.021, 0.018 mol / L, etc., by taking the average of 3 time points through a moving average window, the smoothed component concentration data can be obtained as 0.0193 mol / L.

[0078] It should be noted that the window length can be dynamically adjusted according to the drying process. In one implementation, the system uses a shorter window (e.g., 10 points) in the early stages of drying (when concentration changes rapidly) to improve response speed. As the system approaches the critical point (when concentration changes are minimal), it automatically switches to a longer window (e.g., 30 points) to enhance smoothing and accurately capture subtle trends. This adaptive mechanism ensures that the corrected data remains undistorted and effectively filters out interference throughout the drying process.

[0079] In one implementation, this embodiment performs differential calculations on the smoothed component content data to quantitatively analyze the real-time changes in the concentration of each solvent, considering the direction and rate of increase or decrease of each solvent component. Specifically, the system calculates the concentration difference of each solvent at two consecutive sampling times. The direction of increase or decrease in solvent concentration is determined by the sign of the difference: a negative difference indicates a decrease in concentration, and a positive difference indicates an increase in concentration. Simultaneously, the concentration difference is divided by the sampling time interval to obtain the instantaneous rate of change of the solvent at the current moment, which quantifies the speed of concentration change.

[0080] For example, the difference operation is positioned as follows:

[0081]

[0082] in, At the current sampling time, This refers to the previous sampling time. Indicates the first The instantaneous change in the concentration of a solvent. Indicates at time The measured number The concentration of the solvent.

[0083] The instantaneous rate of change is defined as follows:

[0084]

[0085] in, The time interval between adjacent sampling points. Indicates the first The instantaneous rate of change of a solvent is obtained by absolute value calculation, which is the magnitude of the concentration change rate and is used to quantify the severity of the change; the specific direction of the concentration change, such as increase or decrease, is directly characterized by the sign of the result of the difference operation.

[0086] In one implementation, for generating a component change trend description vector, this embodiment constructs a component change trend description vector based on the change direction and rate information of the current period and a recent historical period. This vector integrates multiple trend features, including but not limited to: the current instantaneous change rate, the average change rate calculated based on short-term history, and the change trend of the rate (e.g., accelerating or decelerating).

[0087] In one implementation, for generating and outputting the concentration curve, this embodiment uses time as the horizontal axis and the corrected component content data as the vertical axis. Information from the component change trend description vector guides the interpolation process between data points, performing curve fitting to generate a continuous and smooth concentration change curve, which is then output. This curve clearly shows the complete evolution trajectory of each solvent concentration from the start of drying to the current moment.

[0088] It should be noted that the final output concentration curve is the result of dynamic correction and trend optimization. It not only provides the instantaneous concentration reading but also reveals the dynamic change process of the concentration. This high-fidelity process curve is the core input for subsequent intelligent determination of the drying critical state.

[0089] In step S4, acquiring the concentration time-series data of the concentration curve, constructing a multi-dimensional feature matrix based on the concentration time-series data, and performing steady-state determination in conjunction with a preset convergence threshold to determine the critical state of each solvent drying process includes:

[0090] Acquire the concentration time-series data of the concentration curve, and calculate the instantaneous decay rate and acceleration feature vector of the concentration time-series data;

[0091] The instantaneous decay rate and the acceleration feature vector are combined to construct a multidimensional feature matrix. The multidimensional feature matrix is ​​input into a preset critical point judgment model, and the Euclidean distance between the current state point of the solvent and the preset steady-state hyperplane is calculated.

[0092] If the Euclidean distance is less than a preset convergence threshold, the root mean square deviation of the residual sequence is calculated. If the root mean square deviation remains within a preset steady-state confidence interval, the current solvent drying process is determined to have reached a critical state.

[0093] In one implementation, to determine the instantaneous decay rate and acceleration feature vector, this embodiment acquires the solvent concentration curves output from step S3 in real time and samples them at fixed time intervals to form a discrete concentration time-series data sequence. Based on this sequence, the instantaneous decay rate at each sampling point is calculated, i.e., the first derivative of concentration with respect to time, to characterize the rate of concentration change. Further, the system calculates the rate of change of the decay rate, i.e., the second derivative of concentration with respect to time, as the acceleration feature vector to characterize whether the concentration change trend is accelerating, decelerating, or trending towards stability.

[0094] For example, suppose that over a period of time, the concentration time-series data of methanol is recorded as one point per minute, with data points of 0.020, 0.019, and 0.018 mol / L, while the data for ethanol are 0.010, 0.011, and 0.012 mol / L. When differentiating these discretized concentration time-series data to obtain the instantaneous decay rate and acceleration feature vector, the rate can be estimated by the difference in concentration values ​​between adjacent time points. Taking methanol as an example, the concentration difference between two adjacent points shows that its rate exhibits a negative trend, indicating that the concentration is decreasing. Further analysis of the rate change trend can yield acceleration characteristics, reflecting the speed of concentration decrease.

[0095] In one implementation, to comprehensively describe the system's state at any given time, this embodiment combines and standardizes the concentration values, instantaneous decay rates, and acceleration feature vectors of each solvent at the same sampling time to construct a multi-dimensional feature matrix. Each row of this matrix represents a sampling time, and each column represents a state feature variable (such as solvent A concentration, solvent A decay rate, solvent A acceleration, solvent B concentration, etc.). This matrix maps the dynamic changes of the drying process into a high-dimensional feature space.

[0096] In one implementation, a pre-defined critical point judgment model is used, which is a distance-based rule-based model. Its core lies in calculating the Euclidean distance from the current state point to a pre-defined steady-state hyperplane, using this distance as a quantitative indicator of state proximity. This model is a binary classification machine learning model (e.g., support vector machine or neural network) trained through supervised learning. The model uses historical drying process data as its training set, with data samples labeled as "non-critical states" and "critical states." The trained model can directly perform pattern recognition on the input multi-dimensional feature vector and output a probability score representing the likelihood of belonging to a critical state. In this implementation, the calculation of the Euclidean distance can be part of the model's internal feature extraction or decision-making process.

[0097] It should be noted that the steady-state hyperplane is a mathematical reference plane defined in the multidimensional feature space, representing the theoretical state where all solvents have evaporated to reach dynamic equilibrium. This hyperplane is defined directly analytically. Assume the multidimensional feature vector is X = [x1, x2, ..., xn], where the first m dimensions (e.g., the concentration values ​​of each solvent) are state variables, and the last nm dimensions (e.g., the instantaneous decay rate and acceleration of each solvent) are rate-of-change variables. Then the steady-state hyperplane S can be defined by a set of simple linear equations: for all dimensions i (i > m) representing the rate of change, x_i = 0.

[0098] It should be noted that the convergence threshold is calculated by taking the Euclidean distance between the state point and the steady-state hyperplane corresponding to the drying endpoint of multiple sets of experiments, and using a specific high-order quantile (such as the 95th quantile) of this distance dataset as the threshold. This method ensures the objectivity of the threshold and keeps the judgment criterion consistent with the high-probability statistical interval of historical process endpoints.

[0099] In one implementation, the steady-state confidence interval is preset by analyzing the signal residual characteristics of historical drying processes during the steady-state phase. The specific preset method is as follows: The system collects multiple sets of historical experimental data and precisely extracts the spectral residual sequences that are determined to have reached the steady-state phase (i.e., after the critical state). These residual sequences originate from the data generated during the signal correction or fitting process in step S2 or S3, representing the deviation between measured values ​​and model predicted values. For each residual sequence, its root mean square deviation (RMSD) value is calculated as a quantitative indicator of the inherent fluctuation intensity under the steady-state condition of that experiment. Subsequently, statistical analysis is performed on the RMS deviation values ​​calculated from all historical experiments to determine their central tendency and dispersion. Finally, a numerical interval is set based on the statistical results. For example, the upper limit of the interval is the average value of the batch of data plus twice the standard deviation, and the lower limit is the average value minus twice the standard deviation, thus constituting the steady-state confidence interval.

[0100] In step S5, the process of constructing an operating state vector based on the critical state and processing the operating state vector based on a feedback compensation adjustment mechanism for rate deviation to obtain control strategy data for dynamically regulating the drying process includes:

[0101] An operating state vector is constructed based on the drying process at the critical state, and a rate deviation feedback signal of the operating state vector is calculated using a proportional-integral-derivative control algorithm. The rate deviation feedback signal characterizes the difference in concentration decay rate.

[0102] The rate deviation feedback signal is convolved using a preset system response hysteresis function to obtain a time-compensated deviation sequence.

[0103] The time-compensated deviation sequence is mapped to a preset fuzzy control rule table to generate a dynamic adjustment gain matrix. The dynamic adjustment gain matrix is ​​then superimposed on the operating state vector to generate control strategy data for dynamically regulating the drying process.

[0104] In one implementation, for the generation rate deviation feedback signal, this embodiment constructs an operating state vector based on critical state and real-time monitoring data, integrating key process variables (including solvent concentration, its instantaneous decay rate, system temperature, and pressure). Subsequently, the system extracts the actual solvent decay rate from this vector, compares it with the preset expected target rate of the current drying stage to obtain the instantaneous deviation, and applies a proportional-integral-derivative (PID) control algorithm to process the deviation. The current deviation is multiplied by the proportional gain to obtain the proportional term, the historical deviations are accumulated and multiplied by the integral gain to obtain the integral term, the deviation change rate is calculated and multiplied by the derivative gain to obtain the derivative term, and finally the three outputs are summed to generate a rate deviation feedback signal that comprehensively reflects the magnitude of the deviation, historical accumulation, and changing trend.

[0105] For example, the rate deviation feedback signal is defined as follows:

[0106]

[0107] in, This is represented as a rate deviation feedback signal. Represented as a proportional term, Represented as an integral term, Represented as differential terms, all three are obtained by performing proportional, integral, and differential operations on the current speed deviation value.

[0108] In one implementation, this embodiment performs a convolution operation on the discrete rate deviation feedback signal sequence and a preset system response hysteresis function. This operation essentially simulates the shape of the deviation signal after experiencing the inherent hysteresis of the system, and its output is a new sequence, namely the time-compensated deviation sequence.

[0109] It should be noted that the preset system response hysteresis function is obtained by performing a standard step response test on the drying system (e.g., stepping the heating power and recording the temperature sensor changes at high frequency) and collecting the system output time series data. Based on this data, a first-order or second-order inertial model with a pure hysteresis element is fitted using the least squares method. The mathematical expression of this model is the preset system response hysteresis function.

[0110] In one implementation, this embodiment uses a pre-defined fuzzy control rule table to generate control strategy data. This rule table defines the nonlinear mapping relationship between input variables (such as the magnitude and trend of time compensation deviation) and output variables (such as heating power adjustment coefficient, intake valve opening adjustment coefficient, etc.). The system maps the current feature values ​​of the time compensation deviation sequence to this rule table and generates a dynamic adjustment gain matrix through fuzzy inference and defuzzification calculation. Each gain coefficient in this matrix corresponds to the adjustment intensity and direction of a control execution channel.

[0111] Furthermore, the adjustment values ​​corresponding to each control variable in the gain matrix are superimposed onto the corresponding state components in the operating state vector (for example, the heating power adjustment gain is superimposed onto the current temperature state value), thereby generating a new vector containing the target control command, which is the control strategy data used to dynamically regulate the drying process.

[0112] It should be noted that the control strategy data is not a drive signal directly sent to the actuator, but rather a set of intermediate instructions containing the target setpoint and adjustment logic. The core of this step lies in transforming the critical state information and real-time dynamics of the drying process into preliminary control decisions with predictive and adaptive capabilities through rate deviation feedback and system hysteresis compensation, laying the foundation for the final generation of stable and accurate process control signals.

[0113] In step S6, parsing the control strategy data into execution response features, performing multi-channel mapping and correction processing on the execution response features, and outputting a stable process control signal includes:

[0114] Acquire the background thermal noise signal monitored in real time by the temperature sensor;

[0115] The control strategy data is parsed into execution response features, and a multi-channel instruction mapping table is constructed based on the execution response features.

[0116] The theoretical thermal distribution value is calculated based on the multi-channel command mapping table. If the theoretical thermal distribution value deviates from the background thermal noise signal, the multi-channel command mapping table is corrected to generate the estimated control quantity.

[0117] The estimated control quantity is subjected to residual evaluation to obtain an accuracy correction coefficient, and the estimated control quantity is subjected to waveform shaping based on the accuracy correction coefficient to obtain a voltage drive sequence;

[0118] The voltage drive sequence is filtered and smoothed to output a stable process control signal.

[0119] In one implementation, regarding the parsing of control strategy data and the acquisition of environmental background signals, this embodiment receives control strategy data from step S5. This data is a set containing target setpoints and adjustment commands. The system first parses this dataset and extracts its core execution response features, such as total thermal power demand, intake flow setpoint, and pressure adjustment direction. Simultaneously, the system reads background thermal noise signals in real time from sensor network chips installed in key parts of the equipment. These signals characterize the uncontrolled inherent thermal fluctuations or environmental thermal interference currently present in the system.

[0120] In one implementation, to generate the estimated control quantity, this embodiment queries a preset multi-channel command mapping table based on the parsed execution response characteristics, initially allocating the abstract adjustment requirements to specific control channels (such as heating rod channels, circulating fan channels, and intake valve channels). This mapping table defines the correspondence between response characteristic values ​​and the basic drive commands (such as duty cycle and opening degree) for each channel. Subsequently, based on the initially allocated drive commands for each channel, the system calculates the expected theoretical thermal distribution value (such as the estimated temperature of each region of the cavity) according to the equipment's thermodynamic model. The system compares this theoretical value with the actual monitored background thermal noise signal. If a significant deviation exists, indicating that the current environmental interference or equipment state deviates from the model assumptions, the system dynamically corrects the output of the multi-channel command mapping table, generating a set of estimated control quantities to preemptively offset the impact of this deviation.

[0121] It should be noted that the preset multi-channel command mapping table is preset through a combination of equipment characteristic calibration and process data optimization. First, by independently testing each actuator channel, the basic static characteristic relationship of its "drive command-output response" is established to determine the initial mapping. Subsequently, based on the optimized control combination of each channel from historical successful process data, the initial mapping relationship is collaboratively corrected and fine-tuned to form a preset lookup table or function that maps the control strategy vector (such as total heat demand) to the specific, collaborative drive command vector of each channel.

[0122] In one implementation, for residual evaluation and waveform shaping to generate the drive sequence, this embodiment sends the estimated control quantity to a high-fidelity equipment simulation model or tests it through a small-amplitude actual output to quickly obtain the residual between the estimated output and the actual requirement. Based on this residual, an evaluation is performed, and a precision correction coefficient is calculated. Subsequently, the system uses this coefficient to finely adjust the estimated control quantity and uses waveform shaping techniques (such as gradient constraints or smoothing filters) to convert it into a voltage drive sequence or pulse width modulation sequence with a smooth timing suitable for the actuator response. This step ensures that the control command is not only statically accurate but also dynamically stable, avoiding impact on the actuator.

[0123] For example, when evaluating the residual of the estimated control quantity, the actual temperature data is collected in real time by the sensor network chip integrated into the system and compared with the model's estimated data. For instance, within a 10-second window, the average residual is calculated to be 0.5℃ and the standard deviation is 0.12℃. Based on this, the accuracy correction coefficient of 1.005 is calculated using the formula k=1+0.5×0.5 / 50. Here, the first 0.5 is a preset dimensionless empirical coefficient, whose value is usually distributed between 0.1 and 0.5. If the temperature sensor accuracy is high (e.g., ±0.1℃), the lower limit of 0.1–0.2 can be taken; if the environmental interference is large (e.g., ±1℃ fluctuation), the upper limit of 0.3–0.5 can be taken. 0.5 / 50 is the ratio of the average residual to the reference temperature, where the reference temperature is in ℃. Subsequently, the control quantity sequence is scaled proportionally using this coefficient, and high-frequency spikes are eliminated by three-point moving average filtering. Finally, it is converted into a smooth voltage drive sequence of 0-10V through PWM modulation, so that the control output is stable and the solvent decay rate is kept constant.

[0124] In one implementation, for a stable output process control signal, this embodiment performs final filtering and smoothing on the generated voltage drive sequence. A low-pass filter is used to remove noise that may be introduced by digital calculation or high-frequency correction, outputting a set of highly stable and accurate process control signals. These signals are directly sent to the drivers of each actuator, thereby achieving precise and stable control of the drying process.

[0125] In step S7, generating a state timing diagram based on the process control signal, extracting quality-efficiency correlation feature vectors based on the state timing diagram, and performing control gain correction to complete the closed-loop feedback control of the final drying process includes:

[0126] Based on the instantaneous energy consumption rate obtained by the real-time monitoring system, the state timing diagram is obtained by analyzing the process control signal using a preset multi-dimensional state observer.

[0127] Based on the state time series diagram and the product quality feedback data of the predicted quantity, extract the moisture content deviation feature, and generate a quality-efficiency correlation feature vector based on the moisture content deviation feature;

[0128] The quality-efficiency correlation feature vector is input into a preset closed-loop verification model to calculate the mutual information entropy between the moisture content deviation feature and the instantaneous energy consumption rate, and the response sensitivity of the current control strategy is quantified based on the mutual information entropy.

[0129] Based on the control gain matrix currently used by the response sensitivity correction system, a corrected control gain sequence is output to complete the closed-loop feedback control of the final drying process.

[0130] In one implementation, for generating the state-time sequence diagram, this embodiment acquires the process control signal output in step S6 in real time, and simultaneously reads the instantaneous energy consumption rate from the monitoring unit. These two sets of signals are input into a preset multi-dimensional state observer. Based on the system's state-space model, this observer analyzes the internal key state variables that cannot be directly measured (such as the internal temperature gradient of the material, the effective mass transfer coefficient, etc.), and aligns all state variables with the control signal and energy consumption rate in time, integrating them to generate a comprehensive state-time sequence diagram. This diagram, with time as the horizontal axis, fully depicts the trajectory of each process variable's evolution over time and their interrelationships.

[0131] It should be noted that the preset multidimensional state observer is constructed based on the physicochemical mechanism model of the drying process. First, a state-space equation describing the internal state of the system (such as the internal temperature and concentration gradient of the material) is established based on the principles of heat and mass transfer. Then, observer design theory (such as the Romberg observer design method) is applied to configure a state estimator and its gain matrix for the model, forming a multidimensional state observer.

[0132] In one implementation, to extract the quality-efficiency correlation feature vector, this embodiment synchronously receives product quality feedback data from online quality detection units (such as near-infrared spectroscopy or humidity sensors), for example, real-time moisture content. This product quality data is compared and analyzed with the process variables at the corresponding time points in the state-time sequence diagram to calculate the deviation between the measured value and the target value of key quality indicators (such as moisture content), i.e., the moisture content deviation feature. The system analyzes the correlation pattern between this deviation and multiple process variables (such as temperature, pressure, and concentration change rate) in the state-time sequence diagram, extracting and constructing a quality-efficiency correlation feature vector that can characterize the coupling relationship between "process control effect" and "product quality result."

[0133] In one implementation, to quantify response sensitivity by calculating mutual information entropy, this embodiment inputs the aforementioned quality-efficiency correlation feature vector into a preset closed-loop verification model. The core operation of this model is to calculate the mutual information entropy between the water content deviation feature and the instantaneous energy consumption rate sequences. Mutual information entropy is an information-theoretic metric used to quantify the strength of the statistical dependency between two variables. Specifically, the mutual information entropy... Calculated using the following formula,

[0134]

[0135] in, The discretized sequence representing the moisture content deviation characteristics. Represents the discretized sequence of instantaneous energy consumption rate. For their joint probability distribution, and Its marginal probability distribution; through this calculation, the efficiency of adjusting system energy consumption on the final product quality deviation under the current control strategy can be evaluated. This efficiency value is quantified as response sensitivity. High sensitivity means that the control action can be efficiently reflected in quality improvement. Low sensitivity indicates that the control effect is buffered by system inertia or disturbances.

[0136] It should be noted that the pre-set closed-loop verification model is pre-set through statistical learning based on historical data. The system collects time-series data on moisture content deviation and instantaneous energy consumption rate recorded synchronously during historical drying processes as training samples. The core of the model is a pre-set computational framework. This framework first calculates the mutual information entropy between the two sets of time-series data to quantify their statistical dependence. Subsequently, the calculated mutual information entropy value is correlated with the actual "response sensitivity" level of the drying process for that batch, determined manually or by the process, for training (e.g., through regression analysis). The trained model internally solidifies the mapping relationship between mutual information entropy and specific response sensitivity quantification values, thus becoming a verification model capable of automatically and accurately evaluating the effectiveness of the current control strategy based on two types of real-time input feature data.

[0137] Specifically, the closed-loop verification model needs to collect time-series data on moisture content deviation and instantaneous energy consumption rate recorded synchronously in multiple complete drying batches as training sample sets; the true value of the response sensitivity corresponding to each batch needs to be pre-calibrated after comprehensive evaluation based on the final product quality of the batch (such as structural integrity and moisture content compliance) and the stability of the control process (e.g., divided into high, medium, and low levels or given specific values); the goal of model training is to learn the statistical mapping relationship between mutual information entropy and the calibrated response sensitivity; when the historical data is sufficiently diverse and comprehensive, the trained model can effectively evaluate the sensitivity of the real-time process.

[0138] In one implementation, to correct the control gain and complete the closed loop, this embodiment uses the calculated response sensitivity to perform online correction on the core control parameters used in steps S5 and S6—the control gain matrix (such as PID gain and fuzzy rule table output gain). If the response sensitivity is too low, the value of the corresponding element in the gain matrix is ​​increased according to a predetermined rule to enhance the strength of the control action; if the sensitivity is too high or an oscillating trend occurs, the gain is appropriately reduced. The system outputs this corrected control gain sequence and feeds it back to the preceding control strategy generation and control signal analysis module to update its internal parameters in real time, thereby forming an adaptive closed-loop control system based on dual feedback of final product quality and process energy efficiency.

[0139] It should be noted that the predetermined rules are defined based on the comparison between the response sensitivity and a preset benchmark sensitivity range. Specifically, the system presets an ideal response sensitivity benchmark range determined by process requirements. When the calculated actual response sensitivity is lower than the lower limit of this range, it indicates that the control action has too sluggish an impact on quality. In this case, the predetermined rule is to multiply the relevant elements in the control gain matrix by a gain enhancement coefficient greater than 1 (e.g., 1.1 to 1.3). When the actual response sensitivity is higher than the upper limit of this range, it indicates that the control action may be too aggressive and prone to oscillation. In this case, the predetermined rule is to multiply the relevant elements in the control gain matrix by a gain attenuation coefficient less than 1 (e.g., 0.7 to 0.9). If the actual response sensitivity is within the benchmark range, the gain matrix remains unchanged. The specific values ​​of the gain enhancement coefficient and attenuation coefficient are a set of empirical parameters pre-calibrated through system simulation or historical debugging data.

[0140] Reference Figure 2 The second embodiment of the present invention provides a critical point drying control device based on online near-infrared monitoring, comprising:

[0141] The data acquisition module is used to acquire raw infrared spectral data, construct a multidimensional spectral matrix based on the raw infrared spectral data, and perform interference correction and spectral demixing to obtain the characteristic dataset of each solvent.

[0142] The signal feature determination module is used to perform mode decomposition and feature frequency band localization on the feature dataset, output the feature signal to be corrected, fit and correct the feature signal to be corrected with the preset standard solvent fingerprint data, and determine the signal features corresponding to each solvent.

[0143] The concentration curve generation module is used to obtain the instantaneous concentration based on the signal characteristics and a preset solvent absorbance characteristic matrix, perform dynamic correction and trend analysis on the instantaneous concentration, and output the concentration curves of each solvent.

[0144] The critical state determination module is used to acquire the concentration time series data of the concentration curve, construct a multi-dimensional feature matrix based on the concentration time series data, and perform steady state determination in combination with a preset convergence threshold to determine the critical state of each solvent drying process.

[0145] The strategy data generation module is used to construct an operating state vector based on the critical state, and process the operating state vector based on the feedback compensation adjustment mechanism of the rate deviation to obtain control strategy data for dynamically regulating the drying process.

[0146] The process control signal generation module is used to parse the control strategy data into execution response features, perform multi-channel mapping and correction processing on the execution response features, and output a stable process control signal.

[0147] The closed-loop feedback control module is used to generate a state timing diagram based on the process control signal, extract the quality-efficiency correlation feature vector based on the state timing diagram and perform control gain correction to complete the closed-loop feedback control of the final drying process.

[0148] It should be noted that the critical point drying control device based on online near-infrared monitoring provided in this embodiment of the invention is used to execute all the process steps of the critical point drying control method based on online near-infrared monitoring in the above embodiment. The working principles and beneficial effects of the two are one-to-one, so they will not be described again.

[0149] In summary, this invention utilizes online near-infrared spectroscopy for real-time monitoring of the drying process, constructs a multidimensional spectral matrix, and performs interference correction and spectral demixing. Based on this, it combines adaptive mode decomposition and solvent fingerprint fitting to achieve precise extraction of solvent signal characteristics. Furthermore, it generates high-fidelity concentration curves through absorption characteristic matrix inversion and dynamic trend analysis, and intelligently identifies the critical drying state based on a multidimensional feature matrix and convergence threshold determination mechanism. Simultaneously, it generates a forward-looking control strategy by combining PID feedback compensation for rate deviation and system hysteresis compensation, and outputs a stable process control signal through multi-channel command mapping and environmental thermal noise correction technology. Furthermore, it uses a multidimensional state observer to generate a panoramic view of the process state and quantifies the control response sensitivity through mutual information entropy, thereby achieving adaptive correction of the control gain. Through the above multi-step, multi-model collaborative perception, decision-making, and execution closed loop, this invention effectively overcomes the insufficient accuracy and slow response problems caused by existing technologies relying on static models and hysteresis control. It can achieve an overall improvement in the control accuracy, stability, and energy efficiency of the critical point drying process under complex solvent systems and dynamic process conditions.

[0150] This invention also provides an electronic device. The electronic device includes a processor, a memory, and a computer program stored in the memory and executable on the processor, such as a critical point drying control program based on online near-infrared monitoring. When the processor executes the computer program, it implements the steps described in the embodiments of the critical point drying control method based on online near-infrared monitoring, for example... Figure 1 Step S1 is shown. Alternatively, when the processor executes the computer program, it implements the functions of each module / unit in the above-described device embodiments, such as the data acquisition module.

[0151] For example, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the electronic device.

[0152] The electronic device may be a desktop computer, laptop, handheld computer, or smart tablet, etc. The electronic device may include, but is not limited to, a processor and memory. Those skilled in the art will understand that the above components are merely examples of electronic devices and do not constitute a limitation on the electronic device. It may include more or fewer components than described above, or combine certain components, or different components. For example, the electronic device may also include input / output devices, network access devices, buses, etc.

[0153] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the electronic device, connecting all parts of the electronic device via various interfaces and lines.

[0154] The memory can be used to store the computer programs and / or modules. The processor implements various functions of the electronic device by running or executing the computer programs and / or modules stored in the memory and by calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0155] Wherein, if the modules / units integrated in the electronic device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0156] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0157] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A critical point drying control method based on online near-infrared monitoring, characterized in that, include: The raw infrared spectral data is acquired, and a multidimensional spectral matrix is ​​constructed based on the raw infrared spectral data. Interference correction and spectral demixing are then performed to obtain the characteristic datasets of each solvent. The feature dataset is subjected to mode decomposition and feature frequency band localization to output the feature signal to be corrected. The feature signal to be corrected is then fitted and corrected with the preset standard solvent fingerprint data to determine the signal features corresponding to each solvent. Based on the signal characteristics, the instantaneous concentration is obtained by inverting the preset solvent absorbance characteristic matrix. The instantaneous concentration is then dynamically corrected and its changing trend is analyzed, and the concentration curves of each solvent are output. Acquire the concentration time-series data of the concentration curve, construct a multi-dimensional feature matrix based on the concentration time-series data, and combine it with a preset convergence threshold to determine the steady state and the critical state of each solvent drying process. Based on the critical state, an operating state vector is constructed, and the operating state vector is processed based on the feedback compensation adjustment mechanism of the rate deviation to obtain control strategy data for dynamically regulating the drying process. The control strategy data is parsed into execution response features, and the execution response features are subjected to multi-channel mapping and correction processing to output a stable process control signal. A state timing diagram is generated based on the process control signal. A quality-efficiency correlation feature vector is extracted based on the state timing diagram, and the control gain is corrected to complete the closed-loop feedback control of the final drying process.

2. The critical point drying control method based on online near-infrared monitoring according to claim 1, characterized in that, The process involves acquiring raw infrared spectral data, constructing a multidimensional spectral matrix based on the raw infrared spectral data, and performing interference correction and spectral demixing to obtain a feature dataset for each solvent, including: The raw infrared spectral data is converted into a multidimensional spectral matrix, and non-stationary baseline fluctuation regions in the multidimensional spectral matrix are identified. An environmental interference correction model is constructed based on the characteristics of the non-stationary baseline fluctuation regions. The multidimensional spectral matrix is ​​input into the environmental interference correction model for differential correction to obtain interference-free mixed spectral data; The mixed spectral data is decomposed and reconstructed using wavelet transform to obtain enhanced mixed spectral data. The enhanced mixed spectral data is then subjected to blind source separation using an independent component analysis algorithm to demix the independent spectral components. Effective spectral components are selected based on the characteristic peak positions of the independent spectral components, and the effective spectral components are mapped to the preliminary characteristic signals of each solvent to obtain the feature dataset after preliminary separation of each solvent.

3. The critical point drying control method based on online near-infrared monitoring according to claim 2, characterized in that, The process of performing mode decomposition and feature frequency band localization on the feature dataset, outputting the feature signal to be corrected, and fitting and correcting the feature signal to be corrected with preset standard solvent fingerprint data to determine the signal features corresponding to each solvent includes: The feature dataset is decomposed into intrinsic mode components using an adaptive variational mode decomposition algorithm. The intrinsic mode components are used to construct the Hilbert marginal spectrum and locate the main characteristic frequency band, thereby generating the characteristic signal to be corrected; Calculate the residual sequence between the feature signal to be corrected and the preset standard solvent fingerprint data, and use the least squares method to perform fitting regression processing on the residual sequence to output the corrected feature signal; The independent signal waveforms are reconstructed based on the correction feature signals to determine the signal characteristics corresponding to each solvent.

4. The critical point drying control method based on online near-infrared monitoring according to claim 1, characterized in that, The instantaneous concentration is obtained by inverting the signal characteristics and a preset solvent absorbance characteristic matrix. The instantaneous concentration is then dynamically corrected and its trend analyzed to output concentration curves for each solvent, including: Based on the signal characteristics, the instantaneous concentration of each solvent is obtained by inversion using a preset solvent absorption characteristic matrix; Based on the preset window length and recursion strategy, a real-time dynamic response window is created, and the instantaneous concentration is input into the real-time dynamic response window for smoothing and filtering correction to obtain the corrected component content data. Based on the differential calculation results of the component content data, the direction and rate of increase or decrease of each solvent component are determined, and based on the direction and rate of increase or decrease, a component change trend description vector of each solvent is generated; Based on the component change trend description vector, the component content data is time-domain mapped and interpolated to output the concentration curves of each solvent's dynamic changes.

5. The critical point drying control method based on online near-infrared monitoring according to claim 3, characterized in that, The process of acquiring the concentration time-series data of the concentration curve, constructing a multi-dimensional feature matrix based on the concentration time-series data, and performing steady-state determination in conjunction with a preset convergence threshold to determine the critical state of each solvent drying process includes: Acquire the concentration time-series data of the concentration curve, and calculate the instantaneous decay rate and acceleration feature vector of the concentration time-series data; The instantaneous decay rate and the acceleration feature vector are combined to construct a multidimensional feature matrix. The multidimensional feature matrix is ​​input into a preset critical point judgment model, and the Euclidean distance between the current state point of the solvent and the preset steady-state hyperplane is calculated. If the Euclidean distance is less than a preset convergence threshold, the root mean square deviation of the residual sequence is calculated. If the root mean square deviation remains within a preset steady-state confidence interval, the current solvent drying process is determined to have reached a critical state.

6. The critical point drying control method based on online near-infrared monitoring according to claim 5, characterized in that, The step of constructing an operating state vector based on the critical state and processing the operating state vector based on a feedback compensation adjustment mechanism for rate deviation to obtain control strategy data for dynamically regulating the drying process includes: An operating state vector is constructed based on the drying process at the critical state, and a rate deviation feedback signal of the operating state vector is calculated using a proportional-integral-derivative control algorithm. The rate deviation feedback signal characterizes the difference in concentration decay rate. The rate deviation feedback signal is convolved using a preset system response hysteresis function to obtain a time-compensated deviation sequence. The time-compensated deviation sequence is mapped to a preset fuzzy control rule table to generate a dynamic adjustment gain matrix. The dynamic adjustment gain matrix is ​​then superimposed on the operating state vector to generate control strategy data for dynamically regulating the drying process.

7. The critical point drying control method based on online near-infrared monitoring according to claim 1, characterized in that, The step of parsing the control strategy data into execution response features, performing multi-channel mapping and correction processing on the execution response features, and outputting a stable process control signal includes: Acquire the background thermal noise signal monitored in real time by the temperature sensor; The control strategy data is parsed into execution response features, and a multi-channel instruction mapping table is constructed based on the execution response features. The theoretical thermal distribution value is calculated based on the multi-channel command mapping table. If the theoretical thermal distribution value deviates from the background thermal noise signal, the multi-channel command mapping table is corrected to generate the estimated control quantity. The estimated control quantity is subjected to residual evaluation to obtain an accuracy correction coefficient, and the estimated control quantity is subjected to waveform shaping based on the accuracy correction coefficient to obtain a voltage drive sequence; The voltage drive sequence is filtered and smoothed to output a stable process control signal.

8. The critical point drying control method based on online near-infrared monitoring according to claim 7, characterized in that, The step of generating a state timing diagram based on the process control signal, extracting quality-efficiency correlation feature vectors based on the state timing diagram and performing control gain correction to complete the closed-loop feedback control of the final drying process includes: Based on the instantaneous energy consumption rate obtained by the real-time monitoring system, the state timing diagram is obtained by analyzing the process control signal using a preset multi-dimensional state observer. Based on the state time series diagram and the product quality feedback data of the predicted quantity, extract the moisture content deviation feature, and generate a quality-efficiency correlation feature vector based on the moisture content deviation feature; The quality-efficiency correlation feature vector is input into a preset closed-loop verification model to calculate the mutual information entropy between the moisture content deviation feature and the instantaneous energy consumption rate, and the response sensitivity of the current control strategy is quantified based on the mutual information entropy. Based on the control gain matrix currently used by the response sensitivity correction system, a corrected control gain sequence is output to complete the closed-loop feedback control of the final drying process.

9. A critical point drying control device based on online near-infrared monitoring, characterized in that, include: The data acquisition module is used to acquire raw infrared spectral data, construct a multidimensional spectral matrix based on the raw infrared spectral data, and perform interference correction and spectral demixing to obtain the characteristic dataset of each solvent. The signal feature determination module is used to perform mode decomposition and feature frequency band localization on the feature dataset, output the feature signal to be corrected, fit and correct the feature signal to be corrected with the preset standard solvent fingerprint data, and determine the signal features corresponding to each solvent. The concentration curve generation module is used to obtain the instantaneous concentration based on the signal characteristics and a preset solvent absorbance characteristic matrix, perform dynamic correction and trend analysis on the instantaneous concentration, and output the concentration curves of each solvent. The critical state determination module is used to acquire the concentration time series data of the concentration curve, construct a multi-dimensional feature matrix based on the concentration time series data, and perform steady state determination in combination with a preset convergence threshold to determine the critical state of each solvent drying process. The strategy data generation module is used to construct an operating state vector based on the critical state, and process the operating state vector based on the feedback compensation adjustment mechanism of the rate deviation to obtain control strategy data for dynamically regulating the drying process. The process control signal generation module is used to parse the control strategy data into execution response features, perform multi-channel mapping and correction processing on the execution response features, and output a stable process control signal. The closed-loop feedback control module is used to generate a state timing diagram based on the process control signal, extract the quality-efficiency correlation feature vector based on the state timing diagram and perform control gain correction to complete the closed-loop feedback control of the final drying process.

Images