A dynamic quality state monitoring system and method for food processing

Abstract

The application discloses a kind of food processing process dynamic quality state monitoring system and method, it is related to food production quality management technical field, process parameters are collected by multi-modal sensing array distributed in and face, fermentation, forming, wake, steaming whole process, after completing analog-digital conversion, time stamp alignment, abnormal value elimination and feature extraction by data acquisition and transmission module, transmission to dynamic quality state evaluation engine, decision and feedback control module calculates total deviation amount accordingly, locates key adjustment process and parameter and generates adjustment instruction, by being arranged multi-modal sensing array in steamed product processing whole process node, representative deficiency problem caused by local temperature unevenness is eliminated, the accuracy and reliability of original data are improved, and batch-level archival management and whole-process traceability of processing quality data are realized.

Description

Technical Field

[0001] This invention belongs to the field of food production quality management technology, specifically a dynamic quality status monitoring system and method for food processing. Background Technology

[0002] Steamed buns, a traditional staple food in my country, involve multiple continuous and interconnected processes, including dough mixing, fermentation, shaping, proofing, and steaming. The stability of the process parameters at each stage directly determines the sensory quality, food safety, and batch consistency of the final product. Currently, quality control in the steamed bun processing industry relies primarily on manual sensory evaluation and offline sampling of finished products. This approach suffers from inherent drawbacks such as strong subjectivity, low testing efficiency, and delayed results. It fails to achieve real-time quality monitoring and dynamic intervention in the production process, easily leading to fluctuations in batch product quality and the generation of substandard products. Existing automated monitoring systems only collect parameters for single processes, failing to build an integrated quality monitoring system covering the entire production chain. Furthermore, sensor selection is limited, and their placement is often unreasonable, making them susceptible to localized environmental interference and resulting in insufficient data representativeness. Simultaneously, existing systems lack a quantitative mapping relationship for quality transfer between processes, making it impossible to trace the root cause of quality problems and key influencing parameters. They can only provide simple parameter over-limit alarms and lack a closed-loop feedback control mechanism based on quality deviations. In addition, existing systems' quality assessment models often employ fixed parameter designs, failing to adapt to dynamic changes in production conditions such as raw material batch variations and equipment aging. They also lack security protection for production data transmission, posing a risk of data leakage. Moreover, they fail to link quality monitoring data with production batches to create searchable and traceable batch quality archives, rendering quality data unsuitable for batch-level quality traceability and production quality management decisions. Summary of the Invention

[0003] To address the shortcomings of the existing technology, this invention provides a dynamic quality monitoring system and method for food processing.

[0004] To achieve the above objectives, the present invention adopts the following technical solution:

[0005] A dynamic quality monitoring system for a food processing process, wherein the food processing process includes dough mixing, fermentation, shaping, proofing, and steaming processes, comprising:

[0006] A multimodal sensor array is distributed at the nodes of the dough mixing, fermentation, shaping, proofing and steaming processes to collect measured process parameters at each process node.

[0007] A data acquisition and transmission module is communicatively connected to the multimodal sensor array. The data acquisition and transmission module includes an analog-to-digital converter, a timestamp aligner, an outlier remover, and a feature extractor. The analog-to-digital converter converts the measured process parameters into digital sampling sequences. The timestamp aligner aligns the digital sampling sequences of each sensor channel according to a unified timestamp to obtain time-series aligned data. The outlier removes outlier sampling points from the time-series aligned data to obtain cleaned data. The feature extractor extracts statistical features from the cleaned data to obtain standardized time-series data.

[0008] A dynamic quality state evaluation engine is communicatively connected to the data acquisition and transmission module. The dynamic quality state evaluation engine has a built-in quality state mapping model library and a quality state evaluation processor.

[0009] The quality state mapping model library includes a dough kneading sub-model, a fermentation sub-model, a shaping sub-model, a proofing sub-model, and a steaming sub-model that correspond one-to-one with the dough kneading process, fermentation process, shaping process, proofing process, and steaming process. Each sub-model includes an input register and an output register. The input register stores the measured process parameters of the corresponding process, and the output register stores the quality indicators of the corresponding process.

[0010] The sub-models are connected through a parameter transfer register chain, which sequentially connects the output register of the previous process sub-model to the input register of the next process sub-model. The quality index of the previous process is used as the input of the next process sub-model for calculation through the parameter transfer register chain.

[0011] The quality state evaluation processor includes a weighted fusion operator, a temporal smoothing operator, and a normalization operator. The input of the weighted fusion operator is connected to the output of the feature extractor and the output register of each sub-model. The weighted fusion operator assigns weighting coefficients based on the correlation between the feature quantities of each sensor channel and the output quality index of each sub-model and outputs a fused feature vector. The temporal smoothing operator performs temporal smoothing on the fused feature vector to obtain a smoothed value. The normalization operator normalizes the smoothed value to obtain a quality state index.

[0012] The decision and feedback control module is communicatively connected to the dynamic quality state evaluation engine. The decision and feedback control module includes a deviation calculator, a key adjustment process locator, a key adjustment parameter locator, and an instruction issuer. The deviation calculator calculates the total deviation between the target quality state index and the actual quality state index. The key adjustment process locator determines the deviation component value of each process node based on the total deviation and outputs the target process identifier code corresponding to the process node with the largest deviation component value. The key adjustment parameter locator outputs the target parameter identifier code based on the sensitivity coefficient of each input variable of the target process. The instruction issuer generates adjustment instruction data frames and sends them to the execution terminal of the target process.

[0013] The quality data management module is communicatively connected to the decision and feedback control module. The quality data management module includes a batch identifier generator, a quality file storage, and a traceability query processor. The batch identifier generator generates a unique batch identifier code for each processing batch. The quality file storage associates and stores the quality status index, total deviation, target process identifier code, target parameter identifier code, and adjustment instruction data frame of each batch with the batch identifier code to form a batch quality file. The traceability query processor retrieves the corresponding batch quality file based on the input batch identifier code and outputs the traceability query result.

[0014] The quality management terminal is communicatively connected to the dynamic quality status evaluation engine, the decision and feedback control module, and the quality data management module. The quality management terminal includes a display processor, which receives the quality status index, the total deviation, the target process identifier code, the target parameter identifier code, the adjustment instruction data frame, and the traceability query result, and outputs a display signal.

[0015] The technical solution provided by this invention brings at least the following beneficial effects:

[0016] By arranging a multimodal sensor array at all process nodes in the steamed bun product processing, real-time synchronous acquisition of dough temperature, ambient humidity, dough pH value, dough viscoelasticity, surface optical properties and vapor dew point temperature was achieved. The triangular distribution of temperature sensors eliminated the problem of insufficient representativeness caused by local temperature unevenness in single-point measurement, and improved the accuracy and reliability of the raw data.

[0017] By constructing a process-level quality state mapping model library that includes a dough-making model, a fermentation model, a shaping model, a proofing model, and a steaming model, and by using a parameter transfer register chain to realize the orderly correlation and transfer of quality indicators of each process, the contribution of different processes to the quality of the final product is quantified, and the root cause process and key adjustment parameters of quality deviation can be accurately located.

[0018] The dynamic quality assessment engine performs correlation-weighted fusion, time-series smoothing, and normalization on multi-source data, outputting a standardized quality state index. Combined with the built-in model self-learning module, it can automatically adjust the weighting coefficients and normalization extreme value parameters based on historical production data, improving the accuracy and environmental adaptability of quality assessment. The decision and feedback control module realizes the automatic calculation of total deviation, the location of key adjustment processes and parameters, and the generation and issuance of adjustment instructions, forming a closed-loop feedback control mechanism based on quality deviation. It can adjust the corresponding process parameters in real time according to the degree of quality deviation, reducing the rate of defective products. By setting four-level quality warning thresholds and matching corresponding intensity adjustment mechanisms, it realizes hierarchical management of quality risks, avoiding over-adjustment and untimely response. By performing local data caching, outlier removal, and feature compression on edge computing nodes, it reduces the amount of data transmission, and encryption technology ensures the security of production data transmission.

[0019] The quality data management module associates and stores the quality status index, deviation data, and adjustment records of each batch with the batch identification code to form a batch quality archive. It also supports traceability queries by batch identification code, realizing batch-level archiving management and full-process traceability of processing quality data, and providing data support for production quality management decisions. The quality management terminal displays the time-series change curve of the quality status index, the distribution of deviation components of each process, adjustment instruction details, and batch quality archives, making it easy for operators to keep track of the production status in real time. Attached Figure Description

[0020] Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0021] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. 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 of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0022] Please see Figure 1 This invention provides a dynamic quality monitoring system for a food processing process, the food processing process including dough mixing, fermentation, shaping, proofing, and steaming processes, comprising:

[0023] A multimodal sensor array is distributed at the nodes of the dough mixing, fermentation, shaping, proofing and steaming processes to collect measured process parameters at each process node.

[0024] A data acquisition and transmission module is communicatively connected to the multimodal sensor array. The data acquisition and transmission module includes an analog-to-digital converter, a timestamp aligner, an outlier remover, and a feature extractor. The analog-to-digital converter converts the measured process parameters into digital sampling sequences. The timestamp aligner aligns the digital sampling sequences of each sensor channel according to a unified timestamp to obtain time-series aligned data. The outlier removes outlier sampling points from the time-series aligned data to obtain cleaned data. The feature extractor extracts statistical features from the cleaned data to obtain standardized time-series data.

[0025] A dynamic quality state evaluation engine is communicatively connected to the data acquisition and transmission module. The dynamic quality state evaluation engine has a built-in quality state mapping model library and a quality state evaluation processor.

[0026] The quality state mapping model library includes a dough kneading sub-model, a fermentation sub-model, a shaping sub-model, a proofing sub-model, and a steaming sub-model that correspond one-to-one with the dough kneading process, fermentation process, shaping process, proofing process, and steaming process. Each sub-model includes an input register and an output register. The input register stores the measured process parameters of the corresponding process, and the output register stores the quality indicators of the corresponding process.

[0027] The sub-models are interconnected through a parameter transfer register chain, which includes a first transfer register, a second transfer register, a third transfer register, and a fourth transfer register. The input of the first transfer register is connected to the output register of the dough kneading sub-model, and the output is connected to the input register of the fermentation sub-model. The input of the second transfer register is connected to the output register of the fermentation sub-model, and the output is connected to the input register of the shaping sub-model. The input of the third transfer register is connected to the output register of the shaping sub-model, and the output is connected to the input register of the proofing sub-model. The input of the fourth transfer register is connected to the output register of the proofing sub-model, and the output is connected to the input register of the steaming sub-model. The quality indicators of the previous process are used as inputs to the sub-model of the next process for calculation via the parameter transfer register chain.

[0028] The quality state evaluation processor includes a weighted fusion operator, a temporal smoothing operator, and a normalization operator. The input of the weighted fusion operator is connected to the output of the feature extractor and the output register of each sub-model. The weighted fusion operator assigns weighting coefficients based on the correlation between the feature quantities of each sensor channel and the output quality index of each sub-model and outputs a fused feature vector. The temporal smoothing operator performs temporal smoothing on the fused feature vector to obtain a smoothed value. The normalization operator normalizes the smoothed value to obtain a quality state index.

[0029] The decision and feedback control module is communicatively connected to the dynamic quality state evaluation engine. The decision and feedback control module includes a deviation calculator, a key adjustment process locator, a key adjustment parameter locator, and an instruction issuer. The deviation calculator calculates the total deviation between the target quality state index and the actual quality state index. The key adjustment process locator determines the deviation component value of each process node based on the total deviation and the parameter transfer register chain, and outputs the target process identifier code corresponding to the process node with the largest deviation component value. The key adjustment parameter locator applies a small perturbation to each input variable in the input register corresponding to the target process and calculates the sensitivity coefficient, and outputs the target parameter identifier code corresponding to the input variable with the largest absolute value of the sensitivity coefficient. The instruction issuer generates an adjustment instruction data frame based on the total deviation, the target process identifier code, and the target parameter identifier code, and sends the adjustment instruction data frame to the execution terminal of the target process.

[0030] The quality data management module is communicatively connected to the decision and feedback control module. The quality data management module includes a batch identifier generator, a quality file storage, and a traceability query processor. The batch identifier generator generates a unique batch identifier code for each processing batch. The quality file storage associates and stores the quality status index, total deviation, target process identifier code, target parameter identifier code, and adjustment instruction data frame of each batch with the batch identifier code to form a batch quality file. The traceability query processor retrieves the corresponding batch quality file based on the input batch identifier code and outputs the traceability query result.

[0031] The quality management terminal is communicatively connected to the dynamic quality status evaluation engine, the decision and feedback control module, and the quality data management module. The quality management terminal includes a display processor, which receives the quality status index, the total deviation, the target process identifier code, the target parameter identifier code, the adjustment instruction data frame, and the traceability query result, and outputs a display signal.

[0032] It should be noted that the food processing process is the steamed bun processing process; the quality status index ranges from 0 to 1, and the closer the quality status index is to 1, the closer the quality status of the corresponding batch of food is to the target quality status; the batch quality file also includes the processing timestamp of the corresponding batch and a summary of the measured process parameters of each process node, which is used for batch-level traceability management of product quality.

[0033] Specifically, the food processing process is a steamed bun product processing process; the quality status index ranges from 0 to 1, and the closer the quality status index is to 1, the closer the quality status of the corresponding batch of food is to the target quality status; the batch identification code includes the batch date code and serial number, and the batch quality file also includes the processing timestamp of the corresponding batch and a summary of the measured process parameters of each process node, which is used for batch-level traceability management of product quality.

[0034] The water addition ratio is the ratio of the amount of water added to the mass of flour.

[0035] The input register of the fermentation sub-model stores the fermentation chamber temperature, fermentation chamber humidity, dough pH value, and fermentation time. The output register of the fermentation sub-model stores the fermentation expansion rate. The fermentation sub-model receives a fermentation mode selection instruction, which corresponds to a full fermentation mode, a semi-fermentation mode, or a no-fermentation mode. The fermentation sub-model configures the input variable set of its input register according to the fermentation mode selection instruction. The full fermentation mode is suitable for yeast-fermented steamed buns, the semi-fermentation mode is suitable for sourdough-fermented steamed buns, and the no-fermentation mode is suitable for unfermented steamed buns. In the no-fermentation mode, the output of the first transmission register is directly connected to the input register of the forming sub-model.

[0036] It should be noted that the humidity sensor is a capacitive humidity-sensitive element, installed at the ventilation openings on the side walls of the fermentation chamber and the proofing room; the pH sensor is a glass electrode pH probe, with its measuring end inserted into the dough sample inside the dough conveying pipe; the dough viscoelasticity sensor is a texture analyzer probe, installed at the dough slice detection station at the exit of the dough mixing process, and performs puncture and tensile tests on the dough slices; the multispectral imaging module includes a white LED light source array and a CMOS image sensor, and the optical sensor outputs RGB three-channel reflectance data and surface texture gradient data; the dew point temperature sensor is arranged in parallel with the wall-mounted thermal resistor on the inner wall of the steaming chamber at the same measurement position to simultaneously acquire steam temperature and dew point temperature.

[0037] The steam supersaturation is calculated from the steam temperature and the dew point temperature, using the following formula:

[0038]

[0039] in For steam supersaturation, For steam temperature, This is the dew point temperature.

[0040] The weighted fusion processor calculates weighting coefficients based on the statistical correlation between the feature quantities of each sensor channel and the output quality indicators of each sub-model. The statistical correlation is measured using the Pearson correlation coefficient. The formula for calculating the Pearson correlation coefficient for each channel is:

[0041]

[0042] in For the first The Pearson correlation coefficient (Cov) between the feature sequence of each sensor channel and the corresponding sub-model output quality index sequence, , Let be the covariance of the two sequences. and These are the standard deviations of the corresponding sequences. This is the sensor channel number.

[0043] The time-series smoothing operator uses an exponentially weighted moving average formula:

[0044]

[0045] in The smoothed value at the current moment. To fuse feature values ​​at the current time, The smoothed value from the previous time step. The smoothing coefficient has a value range of 0 to 1 and is calibrated by fitting historical batch data.

[0046] The normalization operator maps the smoothed value to the interval between 0 and 1, and the calculation formula is as follows:

[0047]

[0048] in The mass state index, For smoothing values, This is the historical minimum. This is the historical maximum value.

[0049] As an optional implementation, the multimodal sensor array includes a temperature sensor, a humidity sensor, a pH sensor, a dough viscoelasticity sensor, an optical sensor, and a dew point temperature sensor.

[0050] The temperature sensor includes an insertable thermocouple probe and a wall-mounted resistance thermometer. The insertable thermocouple probe is installed inside the temperature measuring sleeve of the mixing shaft in the dough mixing chamber and is in direct contact with the dough. The wall-mounted resistance thermometer is installed on the inner wall of the fermentation chamber, the inner wall of the proofing chamber, and the inner wall of the steaming chamber.

[0051] The humidity sensor is a capacitive humidity sensitive element, which is installed at the ventilation opening on the side wall of the fermentation chamber and the ventilation opening on the side wall of the proofing room.

[0052] The pH sensor is a glass electrode pH probe, and the measuring end of the glass electrode pH probe is inserted into the dough sample inside the dough delivery pipe;

[0053] The dough viscoelasticity sensor is a texture analyzer probe, which is installed at the dough slice detection station at the exit of the dough kneading process. The dough viscoelasticity sensor performs puncture and tensile tests on the dough slices.

[0054] The optical sensor is a multispectral imaging module, which is installed at the bread surface scanning station at the exit of the forming process. The multispectral imaging module includes a white LED light source array and a CMOS image sensor. The optical sensor outputs RGB three-channel reflectivity data and surface texture gradient data.

[0055] The dew point temperature sensor is installed on the inner wall of the steaming chamber. The output terminal of the dew point temperature sensor is connected to the input register of the steaming sub-model. The dew point temperature sensor and the wall-mounted thermal resistor on the inner wall of the steaming chamber are arranged in parallel at the same measurement position to synchronously acquire the steam temperature and dew point temperature.

[0056] As an optional implementation, three temperature sensors are arranged at each process node, and the three temperature sensors are distributed in a triangle in the space of the corresponding process. The measured temperature value of each process node is the arithmetic mean of the temperature values ​​measured by the three temperature sensors. The triangular arrangement makes the three temperature measurement points evenly distributed in the process space, eliminating the problem of insufficient representativeness caused by local temperature unevenness in single-point measurement. The arithmetic mean of the three temperature values ​​is taken to suppress the measurement error introduced by the drift of a single sensor.

[0057] It should be noted that the triangular arrangement ensures that the three temperature measuring points are evenly distributed within the process space, eliminating the problem of insufficient representativeness caused by local temperature unevenness in single-point measurements; the arithmetic mean of the three temperature values ​​is taken to suppress the measurement error introduced by the drift of a single sensor.

[0058] As an optional implementation, the input register of the dough kneading sub-model stores the dough temperature, water ratio, and kneading time, and the output register of the dough kneading sub-model stores the dough viscoelasticity index; the input register of the shaping sub-model stores the dough division mass and shaping pressure, and the output register of the shaping sub-model stores the dough regularity; the input register of the proofing sub-model stores the proofing room temperature, proofing room humidity, and proofing time, and the output register of the proofing sub-model stores the proofing expansion rate; the input register of the steaming sub-model stores the steam temperature, steam pressure, and steam supersaturation, and the output register of the steaming sub-model stores the specific volume.

[0059] Specifically, the mapping relationship between each of the sub-models is as follows:

[0060]

[0061] in For the first Quality indicators for each process For the first The first process input register stores the... One measured process parameter, For the first The number of measured process parameters in the input register of each process step For the first The constant term of each sub-model For the first The first process The regression coefficients corresponding to each measured process parameter For process sequence number, The process input parameter number is used. The constant term and the regression coefficient are fitted and calibrated by the least squares method from historical batch production data. After calibration, they are written into the corresponding sub-model for online calculation and calling.

[0062] The formula for calculating the viscoelasticity index of the dough is as follows:

[0063]

[0064] in The viscoelasticity index of dough. This refers to the peak puncture force during the puncture test. This represents the fracture displacement during the tensile test.

[0065] The formula for calculating the fermentation expansion rate is as follows:

[0066]

[0067] in This refers to the fermentation expansion rate. This represents the volume of the dough at the point where fermentation ceases. This represents the volume of the dough at the initial stage of fermentation.

[0068] The formula for calculating the regularity of the steamed bun dough is:

[0069]

[0070] in For the regularity of the steamed bun dough, This represents the intersection area of ​​the actual outline region of the steamed bun and the standard circular outline region. The area is the standard circular outline.

[0071] The formula for calculating the fermentation expansion rate is as follows:

[0072]

[0073] in For the fermentation expansion rate, This refers to the volume of the dough at the end of the proofing process. This refers to the volume of the dough at the initial stage of proofing.

[0074] The formula for calculating the specific volume is:

[0075]

[0076] in For volume, This represents the volume of the food after steaming. For the quality of food.

[0077] As an optional implementation, the weighted fusion processor includes a correlation calculation circuit, a weighting coefficient allocation circuit, and a fusion vector synthesis circuit. The input of the correlation calculation circuit is connected to the output of the feature extractor and the output register of each sub-model. The correlation calculation circuit outputs a correlation metric between the feature quantities of each sensor channel and the output quality index of each sub-model. The input of the weighting coefficient allocation circuit is connected to the output of the correlation calculation circuit. The weighting coefficient allocation circuit allocates weighting coefficients to each channel according to the magnitude of the correlation metric, with channels having higher correlation metric values ​​receiving higher weighting coefficients. The input of the fusion vector synthesis circuit is connected to the output of the weighting coefficient allocation circuit and the output of the feature extractor. The fusion vector synthesis circuit multiplies each channel feature quantity by its corresponding weighting coefficient and then sums the results to output a fused feature vector.

[0078] The timing smoothing unit includes a delay register and a smoothing coefficient multiplier. The delay register stores the smoothed value of the previous time step. The smoothing coefficient multiplier multiplies the current time step's fused feature value with the smoothing coefficient and adds it to the complement of the previous time step's smoothed value stored in the delay register to output the current time step's smoothed value. The normalization unit includes an extreme value memory and a divider. The extreme value memory stores the historical minimum value and the historical maximum value. The divider subtracts the historical minimum value from the current smoothed value and divides it by the difference between the historical maximum value and the historical minimum value to output the normalized mass state index.

[0079] The formula for allocating the weighting coefficients is as follows:

[0080]

[0081] in For the first The weighting coefficients for each channel, It is a natural constant. For the first Pearson correlation coefficient for each channel, To find the summation index for all channels, This represents the total number of sensor channels.

[0082] It should be noted that the correlation metric is the Pearson correlation coefficient, the first... The formula for calculating the Pearson correlation coefficient for each channel is:

[0083]

[0084] in For the first The Pearson correlation coefficient (Cov) between the feature sequence of each sensor channel and the corresponding sub-model output quality index sequence, , ) represents the covariance of the two sequences. and These are the standard deviations of the corresponding sequences. This is the sensor channel number.

[0085] The formula for allocating the weighting coefficients is as follows:

[0086]

[0087] in For the first The weighting coefficients for each channel, It is a natural constant. For the first Pearson correlation coefficient for each channel, To find the summation index for all channels, The total number of sensor channels; the timing smoothing formula is:

[0088]

[0089] in The smoothed value at the current moment. To fuse feature values ​​at the current time, The smoothed value from the previous time step. The smoothing coefficient ranges from 0 to 1 and is determined by fitting historical batch data; the normalization formula is:

[0090]

[0091] in The mass state index, For smoothing values, This is the historical minimum. This is the historical maximum value.

[0092] As an optional implementation, the deviation calculator includes a subtractor, whose first input receives the target quality state index, second input receives the actual quality state index, and outputs the total deviation. The key adjustment process locator includes a path coefficient memory, a multiplier, an accumulator, and a first comparator. The path coefficient memory stores the standardized path coefficients of each sub-model. The multiplier multiplies the total deviation by the standardized path coefficients. The accumulator normalizes all multiplication results to obtain the deviation component values ​​of each process node. The first comparator compares the deviation component values ​​of each process node and outputs the target process identifier code corresponding to the process node with the largest deviation component value. The key adjustment parameter locator includes a disturbance generator, a differential calculator, and a second comparator. The disturbance generator applies a small disturbance to each input variable of the target process. The differential calculator calculates the ratio of the change in the sub-model output to the small disturbance to obtain a sensitivity coefficient. The second comparator compares the absolute values ​​of the sensitivity coefficients of each input variable and outputs the target parameter identifier code corresponding to the input variable with the largest absolute value. The command issuer includes a direction determiner and a data frame encapsulator. The direction determiner determines the adjustment direction based on the sign of the total deviation. The data frame encapsulator encapsulates the target device address, parameter identifier code, adjustment direction, and adjustment amount into an industrial communication protocol data frame.

[0093] Specifically, the standardized path coefficients are the standardized regression coefficients of the linear regression equation between the quality indicators of each sub-model and the sensory scores of the food, obtained by fitting historical batch production data; the formula for calculating the deviation component value is:

[0094]

[0095] in For the first Deviation component values ​​at each process node This is the total deviation. For the first Standardized path coefficients for each process, The summation sequence number is used to iterate through all processes, and the total number of processes is five. The small disturbance amount is calibrated according to the difference approximation accuracy of the difference calculator and is smaller than the actual adjustment step size of the corresponding input variable, so as to ensure the difference approximation accuracy without interfering with normal production.

[0096] The formula for calculating the sensitivity coefficient is as follows:

[0097]

[0098] in For the first The sensitivity coefficient of each input variable. This represents the change in the sub-model output after applying a small perturbation. For small perturbations, The input variable number is used; the direction determiner determines the adjustment direction as increasing when the total deviation is positive, and determines the adjustment direction as decreasing when the total deviation is negative; the CRC16 checksum of the industrial communication protocol data frame is the remainder obtained by performing a modulo-2 division operation between the data frame byte sequence and the generator polynomial.

[0099] It should be noted that the standardized path coefficients are the standardized regression coefficients of the linear regression equation between the quality indicators of each sub-model and the sensory scores of the food, obtained by fitting historical batch production data; the formula for calculating the deviation component value is:

[0100]

[0101] in For the first Deviation component values ​​at each process node This is the total deviation. For the first Standardized path coefficients for each process, To iterate through the summation sequence number of all processes, there are a total of five processes; the small disturbance amount is calibrated according to the differential approximation accuracy of the differential calculator and is smaller than the actual adjustment step size of the corresponding input variable, so as to ensure the differential approximation accuracy without interfering with normal production; the formula for calculating the sensitivity coefficient is:

[0102]

[0103] in For the first The sensitivity coefficient of each input variable. This represents the change in the sub-model output after applying a small perturbation. For small perturbations, The input variable number is used; the direction determiner determines the adjustment direction as increasing when the total deviation is positive, and determines the adjustment direction as decreasing when the total deviation is negative; the CRC16 checksum of the industrial communication protocol data frame is the remainder obtained by performing a modulo-2 division operation between the data frame byte sequence and the generator polynomial.

[0104] As an optional implementation, the dynamic quality state evaluation engine also has a built-in quality early warning module, which includes a threshold memory, a multi-level comparator, and a signal output device; the threshold memory includes a first threshold register, a second threshold register, a third threshold register, and a fourth threshold register.

[0105] The first threshold register stores the total deviation threshold obtained by inverse operation of the regression equation from the first threshold value of the sensory rating. The second, third, and fourth threshold registers respectively store the total deviation threshold obtained by inverse operation of the regression equation from the second, third, and fourth threshold values ​​of the sensory rating. The first, second, third, and fourth threshold values ​​correspond to the boundaries of four levels of sensory quality: excellent, good, average, and poor, respectively, and are determined by the statistical distribution of historical batch sensory rating data. The regression equation is a univariate linear regression equation.

[0106]

[0107] in Rate the senses. This is the total deviation. For regression coefficients, The intercept term is the regression coefficient, and the intercept term is determined by fitting historical production data using the least squares method; the first threshold... The second threshold The third threshold The fourth threshold ,in , , and The first boundary value, the second boundary value, the third boundary value, and the fourth boundary value are, in sequence.

[0108] The multi-level comparator includes a first comparator stage, a second comparator stage, a third comparator stage, and a fourth comparator stage. The first comparator stage compares the total deviation with a first threshold, the second comparator stage compares the total deviation with a second threshold, the third comparator stage compares the total deviation with a third threshold, and the fourth comparator stage compares the total deviation with a fourth threshold. The signal output device outputs a level 1 normal signal, a level 2 attention signal, a level 3 warning signal, a level 4 emergency signal, or a level 5 shutdown signal based on the four-level comparison results. The level 2 attention signal triggers a fine-tuning command, the level 3 warning signal triggers a normal adjustment command, the level 4 emergency signal triggers a large-amplitude adjustment command, and the level 5 shutdown signal triggers an emergency shutdown command. The adjustment amount of the fine-tuning command is less than the adjustment amount of the normal adjustment command, and the adjustment amount of the normal adjustment command is less than the adjustment amount of the large-amplitude adjustment command. The adjustment amounts of the fine-tuning command, the normal adjustment command, and the large-amplitude adjustment command are all proportional to the total deviation, and the proportionality coefficient is calibrated by the device adjustment range of the execution terminal.

[0109] It should be noted that the first threshold register stores the total deviation threshold obtained by inverse operation of the regression equation from the first threshold value of the sensory rating; the second threshold register, the third threshold register, and the fourth threshold register respectively store the total deviation threshold obtained by inverse operation of the regression equation from the second, third, and fourth threshold values ​​of the sensory rating; the first, second, third, and fourth threshold values ​​correspond to the boundaries of the four levels of sensory quality—excellent, good, average, and poor—in turn, and are determined by the statistical distribution of historical batch sensory rating data; the regression equation is a univariate linear regression equation.

[0110]

[0111] in Rate the senses. This is the total deviation. For regression coefficients, The intercept term is the regression coefficient, and the intercept term is determined by fitting historical production data using the least squares method; the first threshold... The second threshold The third threshold The fourth threshold ,in , , and The first boundary value, the second boundary value, the third boundary value, and the fourth boundary value are, in sequence; the adjustment amounts of the fine-tuning command, the normal adjustment command, and the large-amplitude adjustment command are all proportional to the total deviation, and the proportional coefficient is calibrated by the device adjustment range of the execution terminal.

[0112] As an optional implementation, the data acquisition and transmission module further includes an edge computing node, an industrial communication gateway, and a data security unit. The edge computing node is deployed in each processing workshop and includes a buffer, a statistical distribution analyzer, and a feature compressor. The buffer temporarily stores the raw data output by the multimodal sensor array locally. The statistical distribution analyzer calculates the mean and standard deviation of the sampled data of each sensor channel and removes sampling points that deviate from the mean by more than three standard deviations. The feature compressor calculates the mean, standard deviation, maximum value, minimum value, and interquartile range of the data of each channel within a fixed time window and outputs a 5-dimensional compressed feature vector. The industrial communication gateway transmits the compressed feature vector to a cloud server via an industrial Ethernet network. The data security unit includes an encryption processor that performs encryption operations on the compressed feature vector before transmission.

[0113] Specifically, the elimination rule of the statistical distribution analyzer is: when the sampling point satisfy They were removed at that time, among which This is the mean of the sampled data for this channel. The standard deviation of the sampled data for this channel is given, and the rejection rule is determined based on the statistical characteristics of the normal distribution; the interquartile range is calculated using the following formula:

[0114]

[0115] in Interquartile range, It is the 75th percentile. It is the 25th percentile; the encryption operation uses the AES symmetric encryption algorithm.

[0116] It should be noted that the rejection rule of the statistical distribution analyzer is: when the sampling point satisfy They were removed at that time, among which This is the mean of the sampled data for this channel. The standard deviation of the sampled data for this channel is given, and the rejection rule is determined based on the statistical characteristics of the normal distribution; the interquartile range is calculated using the following formula:

[0117]

[0118] in Interquartile range, It is the 75th percentile. It is the 25th percentile; the encryption operation uses the AES symmetric encryption algorithm.

[0119] As an optional implementation, the dynamic quality assessment engine also incorporates a model self-learning module. This module includes a sample recorder, a partitioner, an error calculator, and a coefficient adjustment circuit. The sample recorder includes an input feature memory and a label memory. The input feature memory records the fused feature vector obtained after the measured process parameters of each process node in the entire food supply chain for each batch are processed by the quality assessment processor. The label memory records the sensory evaluation score of the corresponding batch. The partitioner includes a time window counter, which divides the samples into a training set and a validation set according to the sample collection time sequence. The collection time of the training set samples is earlier than that of the validation set samples. The error calculator includes a subtractor and a square multiplier. The subtractor calculates the difference between the predicted sensory score and the sensory score in the label memory. The predicted sensory score is obtained by calibrating and mapping the quality state index output by the quality state evaluation processor. The square multiplier sums the squares of the differences of each sample and divides them by the number of samples to obtain the mean square error value. The coefficient adjustment circuit includes a sign determiner and a step multiplier. The sign determiner determines the direction of increase or decrease of the mean square error value. The step multiplier determines the adjustment direction and adjustment step size of the weighting coefficients based on the output of the sign determiner. When the mean square error value increases, the step multiplier adjusts the weighting coefficients in the opposite direction. When the mean square error value decreases, the step multiplier adjusts the weighting coefficients in the same direction. The error calculator also includes a square root operator. The square root operator takes the square root of the mean square error value on the validation set to obtain the root mean square error value. When the root mean square error value exceeds the historical best root mean square error value, the model retraining process is triggered.

[0120] Specifically, the measured process parameters across the entire process chain include time-series data on temperature, humidity, pH value, dough viscoelasticity, surface reflectivity, and steam supersaturation at each process node; the sensory evaluation score is obtained by taking the arithmetic mean of the scores given by the evaluation panel to the finished food product in four dimensions: appearance, aroma, taste, and internal structure; the calibration mapping is the mapping relationship between the quality state index and the sensory score, determined by fitting historical batch data; the formula for calculating the mean square error is:

[0121]

[0122] in Mean square error, For the sample size, The sample number. For the first The rating residuals for each sample are calculated using the following formula:

[0123]

[0124] in For the first The actual sensory scores of each sample For the first The predicted sensory score for each sample; the adjustment formula for the coefficient adjustment circuit is:

[0125]

[0126] in These are the updated weighting coefficients. These are the weighting coefficients before the update. To adjust the step size, For symbolic functions, The adjustment step size is the change in mean square error between two consecutive verifications, and is calibrated based on the convergence of mean square error during historical adjustments. After the model retraining process is triggered, the parameters of the time-series smoothing operator are frozen, and the weighting coefficients and the historical extreme value parameters of the normalization operator are retrained.

[0127] It should be noted that the measured process parameters across the entire process chain include time-series data on temperature, humidity, pH value, dough viscoelasticity, surface reflectance, and steam supersaturation at each process node; the sensory evaluation score is obtained by taking the arithmetic mean of the scores given by the evaluation panel to the finished food product in four dimensions: appearance, aroma, taste, and internal structure; the calibration mapping is the mapping relationship between the quality state index and the sensory score, determined by fitting historical batch data; the formula for calculating the mean square error is:

[0128]

[0129] in Mean square error, For the sample size, The sample number. For the first The rating residuals for each sample are calculated using the following formula:

[0130]

[0131] in For the first The actual sensory scores of each sample For the first The predicted sensory score for each sample; the adjustment formula for the coefficient adjustment circuit is:

[0132]

[0133] in These are the updated weighting coefficients. These are the weighting coefficients before the update. To adjust the step size, For symbolic functions, The adjustment step size is the change in mean square error between two consecutive verifications, and is calibrated based on the convergence of mean square error during historical adjustments. After the model retraining process is triggered, the parameters of the time-series smoothing operator are frozen, and the weighting coefficients and the historical extreme value parameters of the normalization operator are retrained.

[0134] As an optional implementation, a dynamic quality management method for a food processing process is applied to the food processing process, which includes dough mixing, fermentation, shaping, proofing, and steaming processes, and includes the following steps:

[0135] Step S1: At the nodes of the dough mixing process, fermentation process, shaping process, proofing process and steaming process, the dough temperature, ambient humidity, dough pH value, dough viscoelasticity, surface optical properties and steam dew point temperature are collected by a multimodal sensor array to obtain the measured process parameter sequence.

[0136] Step S2: Perform analog-to-digital conversion on the measured process parameter sequence, align the sampling data of each sensor channel according to a unified timestamp, perform statistical distribution analysis on the sampling data of each channel and remove sampling points that deviate from the channel mean by more than three standard deviations, and extract the mean, standard deviation, maximum value, minimum value and interquartile range of each channel data within the sliding time window to obtain standardized time series data;

[0137] Step S3: Construct a quality state mapping model library, which includes a dough kneading sub-model, a fermentation sub-model, a shaping sub-model, a proofing sub-model, and a steaming sub-model. Each sub-model takes the measured process parameters of the corresponding process as input and the quality index of that process as output. The input register of the dough kneading sub-model stores the dough temperature, water ratio, and kneading time, and the output register of the dough kneading sub-model stores the dough viscoelasticity index. The fermentation sub-model switches between full fermentation mode, semi-fermentation mode, or no fermentation mode according to the fermentation mode selection command. The input register of the steaming sub-model stores the steam temperature, steam pressure, and steam supersaturation. The sub-models are linked through a parameter transfer register chain, and the quality index of the previous process is used as the input of the sub-model of the next process for calculation through the parameter transfer register chain.

[0138] Step S4: The standardized time series data is processed by the quality state evaluation processor. The quality state evaluation processor includes a weighted fusion processor, a time series smoothing processor, and a normalization processor. The weighted fusion processor outputs a fused feature vector, the time series smoothing processor outputs a smoothed value, and the normalization processor outputs a quality state index.

[0139] Step S5: The total deviation is obtained by comparing the target quality state index with the actual quality state index through the deviation calculator. The deviation component value of each process node is determined by the key adjustment process locator based on the reverse transmission path of the parameter transfer register chain and the target process identification code is output. The sensitivity coefficient is calculated and the target parameter identification code is output after applying a small disturbance to the input variable of the target process through the key adjustment parameter locator. The adjustment instruction data frame is generated by the instruction sender and sent to the execution terminal.

[0140] Step S6: Generate a unique batch identifier code for the current processing batch through the quality data management module, associate and store the quality status index, total deviation, target process identifier code, target parameter identifier code and adjustment instruction data frame of the current batch with the batch identifier code to form a batch quality file, and retrieve the corresponding batch quality file according to the input batch identifier code and output the traceability query result.

[0141] Step S7: The quality management terminal receives the quality status index, the total deviation, the target process identifier code, the target parameter identifier code, the adjustment instruction data frame, and the traceability query result through its display processor and outputs a display signal.

[0142] In step S1, the sampling frequency of the multimodal sensor array is set according to the rate of physicochemical change of each process, and the sampling frequency is higher for processes with faster physicochemical change rates. In step S2, the three standard deviations are determined based on the statistical characteristics of the normal distribution. In step S7, the display processor shows the time-series change curve of the quality state index in a line graph, shows the distribution of the deviation component values ​​of the total deviation at each process node in a bar graph, and lists the specific contents of the target process, target parameters, adjustment amount and batch quality file in a table.

[0143] It should be noted that the sampling frequency of the multimodal sensing array in step S1 is set according to the rate of physicochemical change of each process, with a higher sampling frequency corresponding to the process with a faster rate of physicochemical change; the formula for calculating the steam supersaturation in step S3 is:

[0144]

[0145] in For steam supersaturation, For steam temperature, The dew point temperature; in step S4, the weighted fusion calculator assigns weighting coefficients based on the Pearson correlation coefficient; in step S5, the formula for calculating the deviation component value is:

[0146]

[0147] The formula for calculating the sensitivity coefficient is as follows:

[0148]

[0149] In step S7, the display processor displays the time-series change curve of the quality state index using a line graph, displays the distribution of the deviation component values ​​of the total deviation at each process node using a bar graph, and lists the specific contents of the target process, target parameters, adjustment amount, and batch quality file in tabular form.

[0150] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A dynamic quality monitoring system for a food processing process, wherein the food processing process includes dough mixing, fermentation, shaping, proofing, and steaming, characterized in that, include: A multimodal sensor array is distributed at the nodes of the dough mixing, fermentation, shaping, proofing and steaming processes to collect measured process parameters at each process node. A data acquisition and transmission module is communicatively connected to the multimodal sensor array. The data acquisition and transmission module includes an analog-to-digital converter, a timestamp aligner, an outlier remover, and a feature extractor. The analog-to-digital converter converts the measured process parameters into digital sampling sequences. The timestamp aligner aligns the digital sampling sequences of each sensor channel according to a unified timestamp to obtain time-series aligned data. The outlier removes outlier sampling points from the time-series aligned data to obtain cleaned data. The feature extractor extracts statistical features from the cleaned data to obtain standardized time-series data. A dynamic quality state evaluation engine is communicatively connected to the data acquisition and transmission module. The dynamic quality state evaluation engine has a built-in quality state mapping model library and a quality state evaluation processor. The quality state mapping model library includes a dough kneading sub-model, a fermentation sub-model, a shaping sub-model, a proofing sub-model, and a steaming sub-model that correspond one-to-one with the dough kneading process, fermentation process, shaping process, proofing process, and steaming process. Each sub-model includes an input register and an output register. The input register stores the measured process parameters of the corresponding process, and the output register stores the quality indicators of the corresponding process. The sub-models are connected through a parameter transfer register chain, which sequentially connects the output register of the previous process sub-model to the input register of the next process sub-model. The quality index of the previous process is used as the input of the next process sub-model for calculation through the parameter transfer register chain. The quality state evaluation processor includes a weighted fusion operator, a temporal smoothing operator, and a normalization operator. The input of the weighted fusion operator is connected to the output of the feature extractor and the output register of each sub-model. The weighted fusion operator assigns weighting coefficients based on the correlation between the feature quantities of each sensor channel and the output quality index of each sub-model and outputs a fused feature vector. The temporal smoothing operator performs temporal smoothing on the fused feature vector to obtain a smoothed value. The normalization operator normalizes the smoothed value to obtain a quality state index. The decision and feedback control module is communicatively connected to the dynamic quality state evaluation engine. The decision and feedback control module includes a deviation calculator, a key adjustment process locator, a key adjustment parameter locator, and an instruction issuer. The deviation calculator calculates the total deviation between the target quality state index and the actual quality state index. The key adjustment process locator determines the deviation component value of each process node based on the total deviation and outputs the target process identifier code corresponding to the process node with the largest deviation component value. The key adjustment parameter locator outputs the target parameter identifier code based on the sensitivity coefficient of each input variable of the target process. The instruction issuer generates adjustment instruction data frames and sends them to the execution terminal of the target process. The quality data management module is communicatively connected to the decision and feedback control module. The quality data management module includes a batch identifier generator, a quality file storage, and a traceability query processor. The batch identifier generator generates a unique batch identifier code for each processing batch. The quality file storage associates and stores the quality status index, total deviation, target process identifier code, target parameter identifier code, and adjustment instruction data frame of each batch with the batch identifier code to form a batch quality file. The traceability query processor retrieves the corresponding batch quality file based on the input batch identifier code and outputs the traceability query result. The quality management terminal is communicatively connected to the dynamic quality status evaluation engine, the decision and feedback control module, and the quality data management module. The quality management terminal includes a display processor, which receives the quality status index, the total deviation, the target process identifier code, the target parameter identifier code, the adjustment instruction data frame, and the traceability query result, and outputs a display signal.

2. The dynamic quality status monitoring system for food processing according to claim 1, characterized in that, The multimodal sensor array includes a temperature sensor, a humidity sensor, a pH sensor, a dough viscoelasticity sensor, an optical sensor, and a dew point temperature sensor. The temperature sensor includes an insertable thermocouple probe and a wall-mounted resistance thermometer. The insertable thermocouple probe is installed inside the temperature measuring sleeve of the mixing shaft in the dough mixing chamber and is in direct contact with the dough. The wall-mounted resistance thermometer is installed on the inner wall of the fermentation chamber, the inner wall of the proofing chamber, and the inner wall of the steaming chamber. The optical sensor is a multispectral imaging module installed at the surface scanning station of the steamed bun at the exit of the forming process; The dew point temperature sensor is installed on the inner wall of the steaming chamber, and the output terminal of the dew point temperature sensor is connected to the input register of the steaming sub-model.

3. The dynamic quality status monitoring system for food processing according to claim 2, characterized in that, The number of temperature sensors at each process node is three. The three temperature sensors are distributed in a triangle in the space of the corresponding process. The actual temperature measurement value at each process node is the arithmetic mean of the temperature measurement values ​​of the three temperature sensors.

4. The dynamic quality status monitoring system for food processing according to claim 1, characterized in that, The input register of the dough kneading model stores the dough temperature, water ratio, and kneading time, while the output register of the dough kneading model stores the viscoelasticity index of the dough. The input register of the fermentation sub-model stores the fermentation chamber temperature, fermentation chamber humidity, dough pH value, and fermentation time. The output register of the fermentation sub-model stores the fermentation expansion rate. The fermentation sub-model receives a fermentation mode selection instruction, which corresponds to a full fermentation mode, a semi-fermentation mode, or a no-fermentation mode. The fermentation sub-model configures the input variable set of its input register according to the fermentation mode selection instruction. The input register of the forming sub-model stores the dough division quality and forming pressure, and the output register of the forming sub-model stores the dough regularity. The input register of the proofing sub-model stores the proofing room temperature, proofing room humidity, and proofing time, while the output register of the proofing sub-model stores the proofing expansion rate. The input register of the steaming sub-model stores steam temperature, steam pressure, and steam supersaturation, while the output register of the steaming sub-model stores specific volume. The parameter transfer register chain includes a first transfer register, a second transfer register, a third transfer register, and a fourth transfer register. The input of the first transfer register is connected to the output register of the dough-making sub-model, and the output is connected to the input register of the fermentation sub-model. The input of the second transfer register is connected to the output register of the fermentation sub-model, and the output is connected to the input register of the shaping sub-model. The input of the third transfer register is connected to the output register of the shaping sub-model, and the output is connected to the input register of the proofing sub-model. The input of the fourth transfer register is connected to the output register of the proofing sub-model, and the output is connected to the input register of the steaming sub-model.

5. The dynamic quality status monitoring system for food processing according to claim 4, characterized in that, The weighted fusion processor includes a correlation calculation circuit, a weighting coefficient allocation circuit, and a fusion vector synthesis circuit. The input terminal of the correlation calculation circuit is connected to the output terminal of the feature extractor and the output register of each sub-model. The correlation calculation circuit outputs the correlation metric between the feature quantity of each sensor channel and the output quality index of each sub-model. The input of the weighting coefficient allocation circuit is connected to the output of the correlation calculation circuit, and the weighting coefficient allocation circuit allocates the weighting coefficients of each channel according to the correlation metric. The input of the fusion vector synthesis circuit is connected to the output of the weighting coefficient allocation circuit and the output of the feature extractor. The fusion vector synthesis circuit multiplies the feature values ​​of each channel by the corresponding weighting coefficients and then sums them to output a fusion feature vector. The timing smoothing operator includes a delay register and a smoothing coefficient multiplier. The delay register stores the smoothing value of the previous time step. The smoothing coefficient multiplier multiplies the current time step's fused feature value with the smoothing coefficient and adds it to the complement of the previous time step's smoothing value stored in the delay register to output the current time step's smoothing value. The normalization operator includes an extreme value memory and a divider. The extreme value memory stores historical minimum and historical maximum values. The divider subtracts the historical minimum value from the current smoothed value and divides it by the difference between the historical maximum value and the historical minimum value, outputting the normalized mass state index.

6. The dynamic quality status monitoring system for food processing according to claim 5, characterized in that, The deviation calculator includes a subtractor, the first input of which receives the target mass state index, the second input of which receives the actual mass state index, and the output of which outputs the total deviation. The key adjustment process locator includes a path coefficient memory, a multiplier, an accumulator, and a first comparator. The path coefficient memory stores the standardized path coefficients of each sub-model. The multiplier multiplies the total deviation by the standardized path coefficients. The accumulator normalizes all multiplication results to obtain the deviation component values ​​of each process node. The first comparator compares the deviation component values ​​of each process node and outputs the target process identifier code corresponding to the process node with the largest deviation component value. The key adjustment parameter locator includes a disturbance generator, a difference calculator, and a second comparator. The disturbance generator applies a small disturbance to each input variable of the target process. The difference calculator calculates the ratio of the change in the sub-model output to the small disturbance to obtain a sensitivity coefficient. The second comparator compares the absolute values ​​of the sensitivity coefficients of each input variable and outputs the target parameter identification code corresponding to the input variable with the largest absolute value. The instruction transmitter includes a direction determiner and a data frame encapsulator. The direction determiner determines the adjustment direction based on the sign of the total deviation. The data frame encapsulator encapsulates the target device address, parameter identification code, adjustment direction, and adjustment amount into an industrial communication protocol data frame.

7. The dynamic quality status monitoring system for food processing according to claim 6, characterized in that, The dynamic quality state evaluation engine also has a built-in quality early warning module, which includes a threshold memory, a multi-level comparator, and a signal output device. The threshold memory includes a first threshold register, a second threshold register, a third threshold register, and a fourth threshold register; The multi-stage comparator includes a first comparator stage, a second comparator stage, a third comparator stage, and a fourth comparator stage. The first comparator stage compares the total deviation with a first threshold, the second comparator stage compares the total deviation with a second threshold, the third comparator stage compares the total deviation with a third threshold, and the fourth comparator stage compares the total deviation with a fourth threshold. The signal output device outputs a Level 1 normal signal, a Level 2 attention signal, a Level 3 warning signal, a Level 4 emergency signal, or a Level 5 shutdown signal based on the four-level comparison results. The Level 2 attention signal triggers a fine-tuning command, the Level 3 warning signal triggers a normal adjustment command, the Level 4 emergency signal triggers a large-amplitude adjustment command, and the Level 5 shutdown signal triggers an emergency shutdown command. The adjustment amount of the fine-tuning command is less than the adjustment amount of the normal adjustment command, and the adjustment amount of the normal adjustment command is less than the adjustment amount of the large-amplitude adjustment command.

8. The dynamic quality status monitoring system for food processing according to claim 1, characterized in that, The data acquisition and transmission module also includes an edge computing node, an industrial communication gateway, and a data security unit; The edge computing nodes are deployed in each processing workshop. Each edge computing node includes a buffer, a statistical distribution analyzer, and a feature compressor. The buffer temporarily stores the raw data output by the multimodal sensor array locally. The statistical distribution analyzer calculates the mean and standard deviation of the sampled data for each sensor channel and removes sampled points that deviate from the mean by more than three standard deviations. The feature compressor calculates the mean, standard deviation, maximum value, minimum value, and interquartile range of the data for each channel within a fixed time window and outputs a 5-dimensional compressed feature vector. The industrial communication gateway transmits the compressed feature vector to the cloud server via industrial Ethernet. The data security unit includes an encryption processor that performs encryption operations on the compressed feature vector before transmission.

9. The dynamic quality status monitoring system for food processing according to claim 8, characterized in that, The dynamic quality state evaluation engine also has a built-in model self-learning module, which includes a sample recorder, a divider, an error calculator, and a coefficient adjustment circuit. The sample recorder includes an input feature memory and a tag memory. The input feature memory records the fused feature vector obtained by the quality status evaluation processor after the measured process parameters of each process node in the entire chain of each batch of food are processed. The tag memory records the sensory evaluation score of the corresponding batch. The divider includes a time window counter, which divides the samples into a training set and a validation set according to the time sequence of sample collection. The collection time of the training set samples is earlier than that of the validation set samples. The error calculator includes a subtractor and a square multiplier. The subtractor calculates the difference between the predicted sensory score and the sensory score in the label memory. The predicted sensory score is obtained by calibration mapping from the quality state index output by the quality state evaluation processor. The square multiplier sums the squares of the differences of each sample and divides them by the number of samples to obtain the mean square error value. The coefficient adjustment circuit includes a sign detector and a step multiplier. The sign detector determines the direction of increase or decrease of the mean square error value. The step multiplier determines the adjustment direction and adjustment step size of the weighting coefficients based on the output of the sign detector. When the mean square error value increases, the step multiplier adjusts the weighting coefficients in the opposite direction. When the mean square error value decreases, the step multiplier adjusts the weighting coefficients in the same direction. The error calculator also includes a square root operator, which takes the square root of the mean square error value on the validation set to obtain the root mean square error value. When the root mean square error value exceeds the historical best root mean square error value, the model retraining process is triggered.

10. A method for dynamic quality status management in food processing, applied to a dynamic quality status monitoring system for food processing as described in any one of claims 1-9, characterized in that, Includes the following steps: Step S1: At the nodes of the dough mixing process, fermentation process, shaping process, proofing process and steaming process, the dough temperature, ambient humidity, dough pH value, dough viscoelasticity, surface optical properties and steam dew point temperature are collected by a multimodal sensor array to obtain the measured process parameter sequence. Step S2: Perform analog-to-digital conversion on the measured process parameter sequence, align the sampling data of each sensor channel according to a unified timestamp, perform statistical distribution analysis on the sampling data of each channel and remove sampling points that deviate from the channel mean by more than three standard deviations, and extract the mean, standard deviation, maximum value, minimum value and interquartile range of each channel data within the sliding time window to obtain standardized time series data; Step S3: Construct a quality state mapping model library, which includes a dough-making sub-model, a fermentation sub-model, a shaping sub-model, a proofing sub-model, and a steaming sub-model. Each sub-model takes the measured process parameters of the corresponding process as input and the quality index of the process as output. The input register of the dough kneading sub-model stores the dough temperature, water ratio, and kneading time, while the output register stores the dough viscoelasticity index. The fermentation sub-model switches between full fermentation, semi-fermentation, and no fermentation modes according to the fermentation mode selection command. The input register of the steaming sub-model stores the steam temperature, steam pressure, and steam supersaturation. The sub-models are linked through a parameter transfer register chain, with the quality indicators of the previous process serving as inputs for the next process sub-model in the calculation. Step S4: The standardized time series data is processed by the quality state evaluation processor. The quality state evaluation processor includes a weighted fusion processor, a time series smoothing processor, and a normalization processor. The weighted fusion processor outputs a fused feature vector, the time series smoothing processor outputs a smoothed value, and the normalization processor outputs a quality state index. Step S5: The total deviation is obtained by comparing the target quality state index with the actual quality state index through the deviation calculator. The deviation component value of each process node is determined by the key adjustment process locator based on the reverse transmission path of the parameter transfer register chain and the target process identification code is output. The sensitivity coefficient is calculated and the target parameter identification code is output after applying a small disturbance to the input variable of the target process through the key adjustment parameter locator. The adjustment instruction data frame is generated by the instruction sender and sent to the execution terminal. Step S6: Generate a unique batch identifier code for the current processing batch through the quality data management module, associate and store the quality status index, total deviation, target process identifier code, target parameter identifier code and adjustment instruction data frame of the current batch with the batch identifier code to form a batch quality file, and retrieve the corresponding batch quality file according to the input batch identifier code and output the traceability query result. Step S7: The quality management terminal receives the quality status index, the total deviation, the target process identifier code, the target parameter identifier code, the adjustment instruction data frame, and the traceability query result through its display processor and outputs a display signal.

Images