Method and system for monitoring moisture content in spray drying process of solid beverage
By combining multimodal fusion features and a confidence-based weighting strategy with drying mechanism correction, the problem of low accuracy in moisture content monitoring during the spray drying process of solid beverages is solved, achieving real-time and reliable online monitoring.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-02
- Publication Date
- 2026-03-10
AI Technical Summary
In existing technologies, the moisture content monitoring during the spray drying process of solid beverages suffers from low accuracy, inability to achieve real-time feedback and control, and the inability of multi-sensor information fusion technology to effectively distinguish multi-scale process characteristics and dynamically adjust weights, leading to abnormal prediction results.
By acquiring process parameters during spray drying, combining near-infrared spectroscopy and acoustic signals, bandpass filtering and time-spectrum splicing are performed to generate multimodal fusion features. Short-time and long-time statistics are calculated, fusion weights are calculated using the coefficient of variation, and weighted fusion is performed by combining energy entropy and Mahalanobis distance. A drying rate limit threshold is introduced to correct the moisture content.
It enables accurate and reliable online monitoring of the water content of solid beverages, avoiding abnormal predictions caused by data noise or model bias, and improving the accuracy and reliability of monitoring.
Smart Images
Figure CN121633006A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of moisture content monitoring, in particular to a solid beverage spray drying process moisture content monitoring method and system. BACKGROUND
[0002] Solid beverages (such as milk powder, protein powder, instant coffee) are widely popular due to their ease of storage, transportation and long shelf life. Spray drying is the core process for producing such products, which atomizes the slurry into fine droplets, allowing it to quickly evaporate water in the hot gas stream, thereby obtaining a powder product. In this process, the moisture content of the product is a key quality control indicator. Too high moisture content will cause the product to caking, shorten the shelf life, and even breed microorganisms; while too low moisture content not only causes energy waste, but also may damage the heat-sensitive nutrients in the product.
[0003] Traditional moisture content detection is mostly offline, and its results are lagging, which cannot provide real-time feedback and control basis for the production process. In order to realize online monitoring, although there are some technologies that use single sensor method, but the harsh working conditions of high dust, large temperature gradient and uneven material distribution in the spray drying tower, which easily leads to the information of single sensor being seriously disturbed. While the multi-sensor information fusion technology can more comprehensively characterize the drying state by integrating the complementary information of different types of sensors, but the multi-sensor information fusion technology in the prior art still has the following problems: first, the spray drying process contains both short-term rapid fluctuations and long-term slow trends, and most of the existing models cannot effectively distinguish and utilize the process characteristics of these two different time scales. Secondly, in actual working conditions, the quality and reliability of different data sources are dynamically changing, but the existing fusion strategy often cannot dynamically adjust the weight of each information source according to the real-time changes of data quality. Finally, purely data-driven prediction models sometimes produce abnormal prediction results that violate the physical mechanism due to data noise or model overfitting. Therefore, how to effectively fuse multi-source heterogeneous information, and combine the multi-scale time-varying characteristics and physical constraints of the process, to realize accurate and reliable online monitoring of the moisture content of solid beverages, is a technical problem to be solved at present. SUMMARY
[0004] The present application provides a solid beverage spray drying process moisture content monitoring method and system to solve the problem of low accuracy of the measurement results of the moisture content of solid beverages in the prior art.
[0005] In a first aspect, the solid beverage spray drying process moisture content monitoring method of the present application comprises the following steps: The process parameters in the spray drying process are acquired, the process parameters including inlet and outlet air temperature, slurry feeding rate and slurry concentration; near-infrared spectrum data and acoustic signals in the drying chamber are collected; a center frequency and a bandwidth are determined according to the slurry feeding rate, and the collected acoustic signals are band-pass filtered; a time-frequency spectrum of the filtered acoustic signals is calculated, and the near-infrared spectrum data are feature-spliced to generate a multi-modal fusion feature; short-time window statistics and long-time window statistics of the process parameters are calculated to obtain a short-time fluctuation vector and a long-time trend vector; a fusion weight is calculated based on a variation coefficient of the inlet air temperature time sequence, and the short-time fluctuation vector and the long-time trend vector are weighted and summed by using the fusion weight to obtain a composite process state vector; the multi-modal fusion feature is input into a first prediction sub-model to obtain a first moisture content prediction value, and the composite process state vector is input into a second prediction sub-model to obtain a second moisture content prediction value; an energy entropy of the multi-modal fusion feature is calculated, and a Mahalanobis distance of a centroid of the current composite process state vector and a historical composite process state vector is calculated; two prediction values are weighted and fused according to the energy entropy and the Mahalanobis distance to obtain a preliminary moisture content; an increment of the preliminary moisture content and a moisture content at a previous moment is calculated; a drying rate limiting threshold is calculated based on the slurry concentration, the inlet air temperature and the slurry feeding rate; when an absolute value of the increment is greater than the drying rate limiting threshold, the moisture content at the current moment is corrected as a sum of the moisture content at the previous moment and the drying rate limiting threshold with a positive or negative sign of the increment.
[0006] Preferably, the center frequency is calculated by the following formula: ; wherein, is the center frequency, is a preset rate frequency coefficient, is the slurry feeding rate, is a preset base frequency.
[0007] Preferably, the band-pass filtering of the collected acoustic signals comprises: The collected acoustic signals are band-pass filtered by using a fourth-order Butterworth filter.
[0008] Preferably, the calculation of the time-frequency spectrum of the filtered acoustic signals and the feature splicing with the near-infrared spectrum data to generate the multi-modal fusion feature comprises: The time-frequency spectrum of the filtered acoustic signals is calculated by using a short-time Fourier transform; the collected near-infrared spectrum data are preprocessed by using a standard normal variable transformation; a time-frequency spectrum matrix is flattened into a one-dimensional vector, and is column-spliced with a preprocessed near-infrared spectrum data vector to generate the multi-modal fusion feature.
[0009] Preferably, the calculation of the short-time window statistics and the long-time window statistics of the process parameters to obtain the short-time fluctuation vector and the long-time trend vector comprises: Setting a short window length and a long window length; calculating the mean, standard deviation, skewness and kurtosis of the inlet air temperature, slurry feeding rate and slurry concentration in the short window and the long window respectively to form the short-term fluctuation vector and the long-term trend vector.
[0010] Preferably, the fusion weight based on the coefficient of variation of the inlet air temperature time series comprises: Taking the inlet air temperature time series of the past 60 sampling times, calculating the standard deviation of the inlet air temperature time series And the mean ; The coefficient of variation is calculated by the following formula : ; The fusion weight is calculated by the following formula : ; The fusion weight is used to weight the long-term trend vector, and the fusion weight is used to weight the short-term fluctuation vector.
[0011] Preferably, the energy entropy is calculated by the following method: Normalizing the values of each dimension of the multi-modal fusion feature to an energy probability distribution , calculating the energy entropy by the following formula : ; The Mahalanobis distance is calculated by the following method: Storing the composite process state vector of the past 100 times as a historical composite process state vector, calculating the mean vector and the covariance matrix of the historical composite process state vector; calculating the Mahalanobis distance by the following formula : ; wherein, is the current composite process state vector.
[0012] Preferably, the preliminary moisture content is obtained by weighting and fusing the two prediction values according to the energy entropy and the Mahalanobis distance, comprising: The first confidence is calculated by the following formula : ; wherein, is the energy entropy; The second confidence is calculated by the following formula : ; wherein, is the Mahalanobis distance; The formula is The fusion weights of the first moisture content prediction value were calculated. ; Through formula The fusion weights of the second moisture content prediction value were calculated. ; The preliminary moisture content is calculated using the following formula: .
[0013] Preferably, the drying rate limitation threshold is calculated using the following formula: ; in, The drying rate limit threshold, The preset temperature influence coefficient, Intake air temperature, The preset reference temperature, This is the preset feed rate influence coefficient. The slurry feed rate is... The preset concentration influence coefficient, The slurry concentration, This is the preset base rate offset.
[0014] Secondly, the moisture content monitoring system for the spray drying process of solid beverages of the present invention includes a memory and a processor. The memory stores computer instructions, and when the processor executes the computer instructions, it implements the above-mentioned method for monitoring the moisture content of the spray drying process of solid beverages.
[0015] The beneficial effects of this invention are as follows: This invention combines the spectral characteristics reflecting chemical components with the acoustic characteristics representing the atomization state, and simultaneously constructs a composite process state vector capable of distinguishing between short-term fluctuations and long-term trends, thereby uncovering deeper information in the data. By performing confidence-based weighted fusion of prediction results from different information sources, and using a rate-limiting threshold based on the drying mechanism to correct the results, abnormal prediction values that violate physical laws due to data noise or model bias are avoided. This improves the reliability of online monitoring of the moisture content of solid beverages. Attached Figure Description
[0016] Figure 1 This is a flowchart illustrating the method for monitoring the moisture content of a solid beverage during spray drying, as provided in an embodiment of the present invention. Detailed Implementation
[0017] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.
[0018] likeFigure 1 As shown, an embodiment of the method for monitoring the moisture content of solid beverages during spray drying provided by the present invention includes the following steps: S1. Acquire process parameters during spray drying, including inlet and outlet air temperatures, slurry feed rate, and slurry concentration; collect near-infrared spectral data and acoustic signals from the drying chamber; determine the center frequency and bandwidth based on the slurry feed rate, and perform bandpass filtering on the collected acoustic signals; calculate the time spectrum of the filtered acoustic signals and perform feature splicing with the near-infrared spectral data to generate multimodal fusion features.
[0019] Specifically, the inlet and outlet air temperatures are collected at a frequency of 1Hz using a PT100 resistance thermometer installed in the inlet and outlet air ducts of the spray drying tower; the slurry feed rate is obtained using an electromagnetic flowmeter installed on the slurry conveying pipeline; the slurry concentration is measured using an online density meter; a probe-type Fourier transform near-infrared spectrometer is installed in the middle section of the drying chamber tower wall, with the spectral acquisition range set to 900 to 1700 nanometers, and spectral data is collected every 10 seconds in diffuse reflection mode; a high-sensitivity piezoelectric acoustic sensor is installed on the outer wall of the drying chamber near the atomizer, with the sampling frequency set to 48kHz, and acoustic vibration signals are continuously collected.
[0020] A mapping table between slurry feed rate and atomizing disc characteristic frequency was established based on historical data. For example, when the slurry feed rate is 50 L / h, the characteristic center frequency is determined to be 10 kHz and the bandwidth is 2 kHz by referring to the table. The original acoustic signal was filtered using a sixth-order Butterworth bandpass filter. The filtered signal was then processed using a short-time Fourier transform with a Hanning window, a window length of 1024 sampling points, and an overlap rate of 50%, resulting in a two-dimensional time-frequency spectrum. The acquired raw near-infrared spectral data was subjected to Savitzky-Golay smoothing and first-order derivative preprocessing. Principal component analysis was used to extract the first five principal components as spectral features. The two-dimensional time-frequency spectrum was unfolded into a one-dimensional vector and concatenated with the five principal components of the spectral features to form a long-dimensional one-dimensional vector as a multimodal fusion feature.
[0021] S2, calculate the short-term window statistics and long-term window statistics of the process parameters to obtain the short-term fluctuation vector and the long-term trend vector; calculate the fusion weight based on the coefficient of variation of the inlet air temperature time series, and use the fusion weight to perform a weighted summation of the short-term fluctuation vector and the long-term trend vector to obtain the composite process state vector.
[0022] Specifically, a short-term window is defined as the past 10 minutes, and a long-term window as the past 60 minutes. The standard deviation and range of each process parameter within the short-term window are calculated to form a short-term fluctuation vector. The mean and linear fit slope of each process parameter within the long-term window are calculated to form a long-term trend vector. The mean of the inlet air temperature time series within the long-term window is calculated. and standard deviation coefficient of variation ; Fusion weights , where k and C are preset parameters.
[0023] .
[0024] S3, input the multimodal fusion features into the first prediction sub-model to obtain the first water content prediction value, and input the composite process state vector into the second prediction sub-model to obtain the second water content prediction value; calculate the energy entropy of the multimodal fusion features, and calculate the Mahalanobis distance between the centroids of the current composite process state vector and the historical composite process state vector; perform weighted fusion of the two prediction values based on the energy entropy and the Mahalanobis distance to obtain the preliminary water content.
[0025] Specifically, the first prediction sub-model is a pre-trained convolutional neural network, and the second prediction sub-model is a pre-trained random forest regression model; the energy distribution probability is obtained by normalizing the squared values of each element of the multimodal fusion feature vector. The energy entropy is The first confidence level is calculated using the following formula. : .
[0026] 1000 historical composite process state vectors under normal operating conditions are pre-stored. The mean vector and covariance matrix of the historical composite process state vectors are calculated. The second confidence level is then calculated. Second confidence level The first confidence level is the reciprocal of the Mahalanobis distance between the current composite process state vector and the mean vector. Second confidence level Normalization yields the weights. and .
[0027] .
[0028] S4, calculate the increment of the initial moisture content compared to the moisture content at the previous moment; calculate the drying rate limit threshold based on the slurry concentration, inlet air temperature, and slurry feed rate; when the absolute value of the increment is greater than the drying rate limit threshold, correct the current moisture content to the sum of the moisture content at the previous moment and the drying rate limit threshold with the positive or negative sign of the increment.
[0029] Specifically, ; Drying rate limit threshold = K (Inlet air temperature - Outlet air temperature) / (Slurry feed rate) Slurry concentration); where K is an empirical constant. The drying rate limit threshold represents the maximum physical change in moisture content that may occur within a calculation step; if the absolute value of ΔM is greater than the drying rate limit threshold, then the moisture content at the current moment = the moisture content at the previous moment + the sign of ΔM. Drying rate limit threshold; if the absolute value of ΔM is not greater than the drying rate limit threshold, then the moisture content at the current moment = the initial moisture content at the current moment.
[0030] In an optional embodiment, the center frequency is calculated using the following formula: ;in, For the center frequency, The preset rate frequency coefficient, The slurry feed rate is... This is the preset base frequency.
[0031] The bandpass filtering of the acquired acoustic signal includes: using a fourth-order Butterworth filter to perform bandpass filtering on the acquired acoustic signal.
[0032] For example, setting The slurry feed rate at a certain moment is V = 300 L / h. Substituting this rate value into the center frequency calculation formula, we get the center frequency. =650Hz. The center frequency represents the acoustic signal frequency that needs to be considered under the current operating conditions.
[0033] Based on a fixed 200Hz bandwidth, the required frequency range is determined, centered at the calculated center frequency of 650Hz, extending 100Hz above and below it; that is, the filter's passband range is 550Hz to 750Hz. A fourth-order Butterworth bandpass filter is applied to process the raw acoustic signal. This filter preserves the signal components between 550Hz and 750Hz while attenuating or filtering out all other frequency components outside this range, thereby extracting the acoustic features relevant to the current drying process from the noisy background noise.
[0034] In an optional embodiment, the calculation of the time spectrum of the filtered acoustic signal and the feature concatenation with near-infrared spectral data to generate multimodal fusion features include: The time spectrum of the filtered acoustic signal is calculated using short-time Fourier transform; the acquired near-infrared spectral data is preprocessed using standard normal variable transformation; the time spectrum matrix is flattened into a one-dimensional vector and column-wise concatenated with the preprocessed near-infrared spectral data vector to generate the multimodal fusion feature.
[0035] Specifically, this process is divided into three stages: acoustic signal processing, near-infrared spectral processing, and feature fusion. In the first stage, a short-time Fourier transform is performed on the filtered acoustic signal. This transform uses a Hamming window as the window function, extracting 256 data points for each Fourier transform, with a 50% overlap between adjacent windows (128 data points). This operation transforms the one-dimensional time-domain acoustic signal into a two-dimensional time-frequency matrix, which simultaneously displays the distribution of signal energy in time and frequency.
[0036] In the second stage, the raw spectral data is affected by scattering from the sample surface. This interference can be corrected through a standard normal transformation. For each spectral data point, the mean value is subtracted, and then the result is divided by its standard deviation, thereby enhancing the stability and comparability of the spectral features. In the third stage, the two-dimensional time-spectral matrix, for example, a 129-row, 50-column matrix, is expanded row-wise or column-wise into a single-column vector containing 6450 elements. The acoustic feature vector is then concatenated with the preprocessed near-infrared spectral feature vector, for example, a vector containing 200 elements, to form a multimodal fusion feature vector containing 6650 elements.
[0037] In an optional embodiment, the calculation process parameters, including short-term window statistics and long-term window statistics, yield a short-term fluctuation vector and a long-term trend vector, comprising: Set the short-term window length and the long-term window length; within the short-term window and the long-term window respectively, calculate the mean, standard deviation, skewness and kurtosis of the inlet and outlet air temperatures, slurry feed rate and slurry concentration to form the short-term fluctuation vector and the long-term trend vector.
[0038] Specifically, four key process parameters are inlet air temperature, outlet air temperature, slurry feed rate, and slurry concentration. A short-term window, composed of data from the most recent 10 sampling points, is used to monitor immediate fluctuations in the process; a long-term window, composed of data from the most recent 60 sampling points, is used to reflect the long-term trend of the process. The calculation of the short-term fluctuation vector involves extracting the numerical sequence of each process parameter over the past 10 sampling times. For example, for inlet air temperature, the most recent 10 temperature readings are taken, and four statistical indicators are calculated based on these readings: mean, standard deviation, skewness, and kurtosis. The same calculation is performed on the other three parameters. Each of the four process parameters generates four statistical values, totaling sixteen values, which together form a sixteen-dimensional short-term fluctuation vector. Similarly, to calculate the long-term trend vector, the same statistical calculation is performed on the numerical sequence of each process parameter over the past 60 sampling times, also yielding a long-term trend vector containing sixteen statistical values.
[0039] To adjust long-term trend and short-term fluctuation information based on process stability, in an optional embodiment, the calculation of fusion weights based on the coefficient of variation of the inlet air temperature time series includes: extracting the inlet air temperature time series from the past 60 sampling times and calculating the standard deviation of the inlet air temperature time series. and mean The coefficient of variation is calculated using the following formula. : ; The fusion weight is calculated using the following formula. : ; Fusion weights Used to weight the long-term trend vector, fusion weight Used to weight the short-term fluctuation vector.
[0040] Specifically, the intake air temperature data for the past 60 sampling times is obtained. Assume the mean μ of the intake air temperature for these 60 sampling times is calculated to be 180℃, and the standard deviation σ is calculated to be 3.6℃. The coefficient of variation (CV) of the intake air temperature is calculated using these two statistics. The CV is the ratio of the standard deviation to the mean, representing the relative volatility of the data. In this example, the CV is 0.02. Substituting the CV into the formula for calculating the fusion weights yields... Approximately 0.047, This will be used to weight the long-term trend vector. Meanwhile, the weight of the short-term fluctuation vector is 0.953. When the inlet air temperature is very stable, i.e., the coefficient of variation is very small, the focus is on reflecting short-term fluctuation information that reflects immediate changes.
[0041] In an optional embodiment, the energy entropy is calculated in the following manner: The values of each dimension of the multimodal fusion feature are normalized to an energy probability distribution. The energy entropy is calculated using the following formula. : ; The Mahalanobis distance is calculated in the following way: Store the composite process state vectors from the past 100 time steps as historical composite process state vectors, and calculate the mean vector of the historical composite process state vectors. Covariance Matrix The Mahalanobis distance is calculated using the following formula. : ;in, This is the current state vector of the composite process.
[0042] The first confidence level is calculated using the following formula. : ;in, Let be the energy entropy.
[0043] The second confidence level is calculated using the following formula. : ;in, Let be the Mahalanobis distance.
[0044] Specifically, the calculation of the first confidence level focuses on the information content of the current multimodal fused features. All element values in the fused feature vector are processed to make them all positive and sum to 1, for example, by normalizing after taking the absolute value, forming an energy probability distribution. The energy entropy H of this distribution is calculated using the information entropy formula. If the feature energy is concentrated in a few dimensions, the entropy value will be low; conversely, if the energy distribution is uniform, the entropy value will be high.
[0045] The second confidence level is used to represent the degree of deviation between the current process state and the normal historical state. A database containing 100 composite process state vectors from the past is maintained, and the mean vector and covariance matrix of the historical normal states are calculated based on this database. When a new current composite process state vector x is obtained, its distance to the historical data point is calculated using the Mahalanobis distance formula. Mahalanobis distance considers the correlation between variables and can measure the degree of anomaly. For example, the calculated Mahalanobis distance... The confidence level is 0.5. It is approximately 0.606. The smaller the distance, the closer the current state is to the historical normal state, and the higher the second confidence level.
[0046] To combine the outputs of two independent moisture content prediction models and intelligently weight them according to their respective confidence levels to obtain a preliminary moisture content result, in an optional embodiment, the step of weighting and fusing the two predicted values based on the energy entropy and the Mahalanobis distance to obtain the preliminary moisture content includes: Through formula The fusion weights of the first moisture content prediction value were calculated. ; through formula The fusion weights of the second moisture content prediction value were calculated. The preliminary moisture content is calculated using the following formula: .
[0047] In an optional embodiment, the drying rate limitation threshold is calculated using the following formula: ; in, The drying rate limit threshold, The preset temperature influence coefficient, Intake air temperature, The preset reference temperature, This is the preset feed rate influence coefficient. The slurry feed rate is... The preset concentration influence coefficient, The slurry concentration, This is the preset base rate offset.
[0048] Specifically, an empirical formula is used to estimate the maximum possible change in product moisture content per unit time under current operating conditions; this is known as the drying rate limit threshold. The drying rate limit threshold can be used to constrain or correct the model's predictions, preventing outputs that violate physical laws. Three real-time process parameters need to be obtained as input: inlet air temperature, slurry feed rate, and slurry concentration.
[0049] set up , , , , It is 0.02%. Assume that at a certain moment, the measured inlet air temperature... The temperature is 185℃, and the slurry feed rate is... The slurry concentration is 280 L / h. The figure is 48%. The drying rate limiting threshold is calculated using the formula. It is -0.865%.
[0050] The implementation principle of the solid beverage spray drying process moisture content monitoring method of this invention is as follows: This invention constructs a multimodal sensing capability by fusing spectral characteristics reflecting chemical components with acoustic characteristics characterizing the atomization state; simultaneously, it deeply mines data information at different time scales by constructing a composite process state vector that can distinguish between short-term fluctuations and long-term trends. Furthermore, this invention employs a confidence-based weighted fusion strategy to dynamically evaluate the reliability of different information sources; and introduces a rate limiting threshold based on the drying mechanism to correct the prediction results. This design effectively avoids abnormal prediction values that violate physical laws due to data noise or model bias, significantly improving the accuracy and reliability of online monitoring of solid beverage moisture content.
[0051] An embodiment of the moisture content monitoring system for the spray drying process of solid beverages provided by the present invention includes a memory and a processor. The memory stores computer instructions, and when the processor executes the computer instructions, it implements the moisture content monitoring method for the spray drying process of solid beverages in the above embodiment.
[0052] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.
Claims
1. A method for monitoring the moisture content of a solid beverage spray drying process, characterized in that, The method comprises the following steps: acquiring process parameters in a spray drying process, the process parameters comprising inlet and outlet air temperatures, slurry feeding rate and slurry concentration; collecting near-infrared spectrum data and acoustic signals in a drying chamber; determining a center frequency and a bandwidth according to the slurry feeding rate, and performing band-pass filtering on the collected acoustic signals; calculating a time-frequency spectrum of the filtered acoustic signals, and performing feature splicing on the near-infrared spectrum data to generate a multi-modal fusion feature; calculating short-time window statistics and long-time window statistics of the process parameters to obtain a short-time fluctuation vector and a long-time trend vector; calculating a fusion weight based on a variation coefficient of an inlet air temperature time sequence, and performing weighted summation on the short-time fluctuation vector and the long-time trend vector by using the fusion weight to obtain a composite process state vector; inputting the multi-modal fusion feature into a first prediction sub-model to obtain a first moisture content prediction value, and inputting the composite process state vector into a second prediction sub-model to obtain a second moisture content prediction value; calculating an energy entropy of the multi-modal fusion feature, and calculating a Mahalanobis distance of a centroid of a current composite process state vector and a historical composite process state vector; performing weighted fusion on the two prediction values according to the energy entropy and the Mahalanobis distance to obtain a preliminary moisture content. calculating an increment of the preliminary moisture content and a moisture content at a previous time; calculating a drying rate limiting threshold based on the slurry concentration, the inlet air temperature and the slurry feeding rate; when an absolute value of the increment is greater than the drying rate limiting threshold, correcting the moisture content at the current time as a sum of the moisture content at the previous time and the drying rate limiting threshold with a positive or negative sign of the increment.
2. The method of claim 1, wherein the solid beverage spray drying process moisture content is monitored by, The center frequency is calculated by the following formula: ; wherein, is a center frequency, is a preset rate frequency coefficient, is a slurry feed rate, is a preset base frequency.
3. The method of claim 1, wherein the solid beverage spray drying process moisture content is monitored by, The band-pass filtering on the collected acoustic signals comprises: performing band-pass filtering processing on the collected acoustic signals by using a fourth-order Butterworth filter.
4. The method of claim 1, wherein the solid beverage spray drying process moisture content is monitored by, The calculation of the time-frequency spectrum of the filtered acoustic signals and the feature splicing on the near-infrared spectrum data to generate the multi-modal fusion feature comprises: calculating a time-frequency spectrum of the filtered acoustic signals by using a short-time Fourier transform; performing standard normal variable transformation preprocessing on the collected near-infrared spectrum data; flattening a time-frequency spectrum matrix into a one-dimensional vector, and performing column splicing on the one-dimensional vector and a near-infrared spectrum data vector after preprocessing to generate the multi-modal fusion feature.
5. The method of claim 1, wherein the solid beverage spray drying process moisture content is monitored by, The calculation of the short-time window statistics and the long-time window statistics of the process parameters to obtain the short-time fluctuation vector and the long-time trend vector comprises: setting a short-time window length and a long-time window length; calculating mean values, standard deviations, skewness and kurtosis of the inlet and outlet air temperatures, the slurry feeding rate and the slurry concentration in the short-time window and the long-time window respectively to constitute the short-time fluctuation vector and the long-time trend vector.
6. The method of claim 1, wherein the solid beverage spray drying process moisture content is monitored by, The calculation of the fusion weight based on the variation coefficient of the inlet air temperature time sequence comprises: intercepting a time series of the air inlet temperature at the last 60 sampling instants, calculating a standard deviation of the time series of the air inlet temperature and a mean value ; The coefficient of variation is calculated by the following formula : ; The fusion weight is calculated by the following equation : ； fusion weights for weighting the long-term trend vector, fusion weights for weighting the short-term fluctuation vector.
7. The method of claim 1, wherein the solid beverage spray drying process moisture content is monitored by, The energy entropy is calculated by the following method: normalizing values of each dimension of the multi-modal fusion feature into an energy probability distribution , the energy entropy is calculated by the following formula : ; The Mahalanobis distance is calculated by the following method: store the composite process state vectors of the past 100 time instants as historical composite process state vectors, calculate a mean vector of the historical composite process state vectors and a covariance matrix ; The Mahalanobis distance is calculated by the following equation : ; wherein, is the current composite process state vector.
8. The method of claim 7, wherein the solid beverage spray drying process moisture content is monitored by, The calculation of the preliminary moisture content by using the energy entropy and the Mahalanobis distance to perform weighted fusion on the two prediction values comprises: The first confidence is calculated by the following equation : ; wherein, is the energy entropy; The second confidence is calculated by the following equation : ; wherein, is the Mahalanobis distance; The fusion weight of the first water cut prediction value is calculated by the formula ; The fusion weight of the second water cut prediction value is calculated by the formula ; The preliminary moisture content is calculated by the following formula: 。 9. The method of claim 1, wherein the solid beverage spray drying process moisture content is monitored by, The drying rate limiting threshold is calculated by the following formula: ； wherein, is a drying rate limiting threshold, is a preset temperature influence coefficient, is an inlet air temperature, is a preset reference temperature, is a preset feed rate influence coefficient, is a slurry feed rate, is a preset concentration influence coefficient, is a slurry concentration, is a preset base rate offset.
10. A solid beverage spray drying process moisture content monitoring system characterized by, The application discloses a solid beverage spray drying process moisture content monitoring method, and relates to the technical field of solid beverage spray drying process moisture content monitoring. The application discloses a solid beverage spray drying process moisture content monitoring method, and relates to the technical field of solid beverage spray drying process moisture content monitoring.