Method and system for quality monitoring in production of soy products

Abstract

The application relates to the technical field of food production quality control, and discloses a quality monitoring method and system for soy product production. The method comprises the following steps: collecting samples of target soy product categories on a soy product production line, recording production batches and key process parameter information of the samples; performing texture testing on the samples to obtain multi-dimensional texture data vectors and normalizing the multi-dimensional texture data vectors, and calling corresponding standard texture data vectors; calculating the dimensional value differences of the two, combining preset weight coefficients to obtain a comprehensive deviation degree; constructing a texture data time sequence of continuous production batches, extracting effective trend change information; when the comprehensive deviation degree exceeds a preset threshold and the trend meets abnormal conditions, determining that the batch is unqualified and alarming, and analyzing deviation dimensions and directions; identifying key mechanical property dimensions, generating process parameter adjustment suggestions, and outputting a quality monitoring report. The application improves the accuracy of soy product quality monitoring and production efficiency.

Description

Technical Field

[0001] This application relates to the field of food production quality control technology, and in particular to a quality monitoring method and system for soy product production. Background Technology

[0002] Soy products are an important plant-based protein food in my country, and the consistency of their texture is the core of their market competitiveness and a key control indicator for industrial-scale production. Currently, quality control in soy product manufacturers relies heavily on sensory tasting by quality inspectors combined with limited instrumental testing. Subjective judgment differences can easily lead to fluctuations in the judgment results for the same batch of products, and comparing the values ​​of a single indicator cannot fully capture the complex characteristics of texture, making it difficult to detect subtle taste deviations that consumers can perceive.

[0003] Meanwhile, the texture of soybean products is determined by multiple mechanical properties, including hardness and elasticity, with complex interrelationships among these properties. Adjustments to process parameters such as grinding, coagulation, and pressing during production simultaneously affect multiple textural dimensions. Existing monitoring methods cannot accurately identify the core dimensions and trends of textural deviations. Process adjustments often rely on trial and error based on experience, which not only slows down production but also results in significant waste of raw materials and energy. Furthermore, traditional monitoring methods lack trend analysis of textural data from continuous production batches, making it impossible to predict quality anomalies in advance. Corrective measures are often only implemented after large quantities of products fail to meet standards, severely impacting production efficiency and product qualification rates.

[0004] Therefore, there is an urgent need for a quality monitoring method for soybean product production that can objectively quantify differences in texture, capture trends in quality changes, and accurately guide process adjustments. Summary of the Invention

[0005] To address the aforementioned technical issues, this application provides a quality monitoring method and system for soybean product production, which enables precise quantitative detection of soybean product quality structure, prediction and traceability of quality anomalies, guides optimization of process parameters, and improves production quality and efficiency.

[0006] Firstly, this application provides a quality control method for soybean product production, the method comprising: Step S1: Collect samples of the target soy product category on the soy product production line and record the production batch information and key process parameters corresponding to the samples; Step S2: Perform texture testing on the sample to obtain a multidimensional texture data vector, normalize the multidimensional texture data vector, and retrieve the standard texture data vector corresponding to the target bean product category from the database; Step S3: Calculate the numerical difference between the multidimensional texture data vector and the standard texture data vector in each dimension, and combine it with the preset weight coefficients to obtain the overall deviation between the current sample and the standard. Step S4: Construct a time series from the multidimensional texture data vector formed by consecutive production batches, including the current production batch, and extract effective trend change information based on the time series; Step S5: When the overall deviation is greater than the preset threshold and the effective trend change information meets the anomaly judgment conditions, the quality of the current production batch sample is determined to be unqualified, a quality anomaly alarm is triggered, and the deviation dimension and deviation direction that caused the quality anomaly are analyzed. Step S6: Based on the deviation dimension and deviation direction, identify the key mechanical property dimensions, generate process parameter adjustment suggestions for the current production batch, and output a quality monitoring report.

[0007] Secondly, this application provides a quality monitoring system for soybean product production, the system comprising: The data acquisition module is used to collect samples of the target soybean product category on the soybean product production line and record the production batch information and key process parameter information corresponding to the samples. The testing module is used to perform texture testing on the samples, obtain multidimensional texture data vectors, normalize the multidimensional texture data vectors, and retrieve standard texture data vectors corresponding to the target bean product category from the database. The calculation module is used to calculate the numerical difference between the multidimensional texture data vector and the standard texture data vector in each dimension, and combine it with preset weight coefficients to obtain the overall deviation of the current sample from the standard. The extraction module is used to construct a time series from the multidimensional texture data vector formed by continuous production batches, including the current production batch, and to extract effective trend change information based on the time series. The determination module is used to determine that the quality of the current production batch sample is unqualified when the overall deviation is greater than a preset threshold and the effective trend change information meets the abnormal determination conditions, trigger a quality abnormality alarm, and analyze the deviation dimension and deviation direction that caused the quality abnormality. The output module is used to identify key mechanical property dimensions based on the deviation dimension and deviation direction, generate process parameter adjustment suggestions for the current production batch, and output a quality monitoring report.

[0008] Compared with the prior art, the beneficial effects of the present invention are at least as follows: This application achieves precise binding of sample texture data with the entire production process data by collecting samples on the soybean product production line and simultaneously recording production batch information and key process parameters, providing a reliable data foundation for accurate tracing of quality anomalies. By using a texture analyzer to obtain mechanical property data of samples in six dimensions—hardness, elasticity, cohesion, chewiness, adhesion, and stickiness—and normalizing the multidimensional texture data vectors, the dimensional differences between different dimensions of data are eliminated, ensuring the scientific rigor and accuracy of subsequent analysis. Furthermore, this application also determines the weight coefficients of each dimension through a dynamic update mechanism, so that the weight coefficients can adapt to the dynamic changes of texture data in the production process. Combined with weighted Euclidean distance calculation, it accurately quantifies the comprehensive deviation of the sample from the standard texture data, avoiding the problem that traditional fixed weights cannot adapt to changes in production data. By constructing a time series of texture data for continuous production batches and extracting effective trend change information, it combines static single-batch texture deviation analysis with dynamic continuous batch trend prediction, realizing dual-condition judgment of quality anomalies, and effectively reducing the risk of misjudgment and omission. Finally, this application also accurately locates the deviation dimension and direction, identifies key mechanical property dimensions by combining weight coefficients, and generates operable adjustment suggestions based on the correspondence between textural properties and production process parameters. This solves the problem of blind trial and error in traditional process adjustments, significantly improves the accuracy, foresight and operability of quality monitoring in soybean product production, ensures product texture consistency, and reduces production trial and error costs and raw material losses. Attached Figure Description

[0009] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 This is a flowchart of a quality monitoring method for soy product production according to an embodiment of this application; Figure 2 This is a schematic diagram of a quality monitoring system for soybean product production according to an embodiment of this application. Detailed Implementation

[0011] This application provides a method and system for quality monitoring in the production of soy products. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0012] For ease of understanding, the specific process of the embodiments of this application is described below. Please refer to [link / reference]. Figure 1 The quality monitoring method for soy product production in this application embodiment includes: Step S1: Collect samples of the target soy product category on the soy product production line and record the production batch information and key process parameters corresponding to the samples.

[0013] Specifically, in order to achieve quantitative detection of the textural properties of soybean products and accurate monitoring of production quality, it is necessary to obtain multi-dimensional mechanical property data of samples through professional textural testing and construct textural data vectors, so as to lay a core data foundation for subsequent calculation of textural deviation and analysis of production quality deviation.

[0014] Firstly, sample collection of target soy product categories is carried out at the finished product discharge stage of the industrial continuous production line for soy products. The target soy product categories refer to subcategories of soy products with different textural characteristics, such as tofu, dried tofu, bean curd sticks, soy milk, and tofu sheets. The collection operation strictly follows scientific sampling rules adapted to the production characteristics of soy products. These rules take into account the randomness, representativeness, and uniformity of sampling. The sampling node is fixed at the stable production stage after the finished product is continuously discharged to the preset output. The sampling location is selected from multiple angles on the finished product conveying device. For liquid and semi-liquid soy products, they need to be stirred and mixed before sampling. The sampling quantity meets the needs of subsequent textural repetition testing and is controlled within a reasonable range of raw material loss. This ensures that the collected samples can objectively and truthfully reflect the overall textural status of the batch of soy products at the current production stage and avoid the distortion of subsequent quality monitoring results due to sample selection bias.

[0015] While completing sample collection, the industrial data acquisition system and the manufacturing execution system supporting the production line are used to synchronously extract and record the production batch information and key process parameter information corresponding to the sample. The production batch information is exclusive information for uniquely identifying the batch of soybean products, and includes content enabling full production chain traceability such as batch code, production time, production equipment number, and production team. The key process parameter information is the actual process parameter values of core processes that directly affect the texture properties of the batch of soybean products during the production process, specifically covering process data directly related to the formation of soybean product texture, such as the material-to-water ratio and equipment rotation speed in the grinding process, the temperature and heat preservation duration in the heating process, the coagulant addition amount, coagulation temperature and time in the coagulation process, the pressure value and pressing duration in the pressing process, and the temperature and wind speed in the drying process. During the recording process, the physical identifier of each sample is associated and bound with the corresponding production batch information and key process parameter information in a one-to-one manner, so as to establish a unique correspondence between the sample and the production data, and avoid information mismatch and missing. At the same time, all the data that have completed the association and binding are uploaded in real time to the dedicated soybean product production quality monitoring database for unified storage and management, so as to ensure the integrity, accuracy and invocability of the data, enable each soybean product sample to be accurately traced back to the process parameters of its entire production process, and provide complete production data support for process parameter traceability and deviation cause localization in case of abnormal texture quality in the subsequent stage.

[0016] Step S2: performing a texture test on the sample to obtain a multi-dimensional texture data vector, normalizing the multi-dimensional texture data vector, and retrieving the standard texture data vector corresponding to the target soybean product category from the database.

[0017] Step S2 specifically includes: testing each sample of the target soybean product category by using a texture analyzer to obtain mechanical property data of the sample in six dimensions: hardness, springiness, cohesiveness, chewiness, adhesiveness and gumminess, and constructing a multi-dimensional texture data vector of the sample according to the mechanical property data; accessing a pre-established database according to the identifier of the target soybean product category, and extracting the standard texture data vector corresponding to the target soybean product category, wherein the standard texture data vector is determined by performing statistical analysis on historical multi-dimensional texture data of high-quality products, and the standard texture data vector has the same dimension structure as the multi-dimensional texture data vector of the sample.

[0018] Specifically, after the collection of soybean product samples and the recording of production information are completed, a professional texture test is required to obtain multi-dimensional mechanical property data of the samples and construct a standardized texture data vector, and data normalization processing is completed and a matched standard texture data vector is retrieved, so as to provide a standardized and comparable core data basis for subsequent calculation of the comprehensive texture deviation degree.

[0019] First, standardized mechanical property tests were conducted on each sample of the target soy product category using a texture analyzer. The texture analyzer is a professional device capable of quantitatively detecting the mechanical properties of solid and semi-solid foods. By applying a preset force to the sample and recording the force-deformation response curve, precise measurements of the food's texture-related mechanical properties can be achieved. During the test, the sample was placed on the texture analyzer's testing platform. The probe speed and compression distance were adjusted according to the texture characteristics of the soy products to adapt to the texture testing requirements of different soy products. The same testing location for each sample was tested at least three times, and the average of the multiple test results was taken as the final test value to avoid the randomness of a single test. Errors affect data accuracy. The above testing process obtains mechanical property data of the sample in six dimensions: hardness, elasticity, cohesion, chewiness, adhesion, and stickiness. Hardness is the mechanical property of the sample resisting external indentation; elasticity is the ability of the sample to recover its original shape after deformation under external force; cohesion is the binding force between the components inside the sample; chewiness is the energy required to chew the solid sample to a swallowable state; adhesion is the viscous force between the sample and the surface of the detection probe; and stickiness is the energy required to chew the semi-solid sample to a swallowable state. The above six dimensions are the core mechanical indicators characterizing the texture of soy products and can comprehensively reflect the texture and taste characteristics of soy products.

[0020] After obtaining the mechanical property data in six dimensions, a multidimensional texture data vector of the sample is constructed based on the mechanical property data. The multidimensional texture data vector is a vector data form formed by arranging the mechanical property values ​​of multiple dimensions in a fixed dimensional order. It is used to comprehensively characterize the texture properties of the sample in multiple mechanical dimensions. In this application, the mechanical property values ​​of the six dimensions are arranged in a fixed order of hardness, elasticity, cohesion, chewiness, adhesion, and stickiness to construct a six-dimensional texture data vector corresponding to each sample. After completing the construction of the multidimensional texture data vectors of all samples, the multidimensional texture data vectors are normalized. Normalization is a data analysis method that converts texture data with different dimensions and different numerical ranges to the same numerical range. In this application, a linear normalization method is used to convert the mechanical property values ​​of each dimension to the numerical range of 0-1, eliminating the data analysis bias caused by the difference in dimensions and numerical ranges of different dimensions, making the data of each dimension comparable and summable, and laying the data foundation for subsequent weighted calculations.

[0021] After constructing and normalizing the multidimensional texture data vector of the samples, the pre-established soybean product quality monitoring database is accessed based on the identifier of the target soybean product category. The identifier of the target soybean product category is a code or character information used to uniquely distinguish different types of soybean products, forming a one-to-one correspondence with the standard texture data vectors of each category stored in the database. Through this identifier, the standard data of the target category can be accurately retrieved and accessed. The standard texture data vector corresponding to the target soybean product category is extracted from the database. The standard texture data vector is determined by statistical analysis of the historical multidimensional texture data of high-quality products in this category. During the statistical analysis, texture test data of multiple batches of high-quality qualified products in this category are collected. The texture data of each dimension are statistically calculated separately, and the statistical average of the data of each dimension is taken as the value of the corresponding dimension of the standard texture data vector. The standard texture data vector and the multidimensional texture data vector of the samples have the same dimensional structure, that is, they are both six-dimensional vectors, and the arrangement order of each dimension is completely consistent with the mechanical property index, ensuring that the standard texture data vector and the multidimensional texture data vector of the samples are directly comparable.

[0022] By using a texture analyzer to test the mechanical properties of samples in six core texture dimensions, a comprehensive quantitative detection of the texture characteristics of soy products was achieved, overcoming the limitations of traditional human sensory evaluation and single-index detection. Furthermore, the normalization of multi-dimensional texture data vectors eliminated the differences in the dimensions and numerical ranges of data from different dimensions, ensuring the scientific rigor and accuracy of subsequent weighted calculations and deviation analysis.

[0023] Step S3: Calculate the numerical differences between the multidimensional texture data vector and the standard texture data vector in each dimension, and combine them with preset weight coefficients to obtain the overall deviation between the current sample and the standard.

[0024] Step S3 includes: calculating the numerical difference between the multidimensional texture data vector of the sample and the standard texture data vector in each dimension, determining the weight coefficient of each dimension using a dynamic update mechanism, weighting the normalized numerical difference with the updated weight coefficient, and calculating the weighted Euclidean distance based on the weighted difference to obtain the comprehensive deviation between the sample and the standard.

[0025] Specifically, after obtaining and normalizing the sample and standard texture data vectors, the deviation of the sample from the standard texture needs to be quantified by combining dimensional difference calculation with dynamic weight allocation. The calculation of weighted Euclidean distance makes the deviation more in line with the actual taste impact, providing a precise quantitative basis for quality judgment.

[0026] First, for the normalized multidimensional texture data vector of the sample and the standard texture data vector, the numerical differences between the two in six dimensions—hardness, elasticity, cohesion, chewiness, adhesion, and stickiness—are calculated one by one. The numerical difference is the normalized value of each dimension of the sample minus the value of the corresponding dimension of the standard texture data vector. This difference directly represents the deviation of the sample's texture from the standard in a single dimension, providing basic data for subsequent calculation of the overall deviation. Subsequently, a dynamic update mechanism is adopted to determine the weight coefficients of each dimension. The dynamic update mechanism refers to a weight determination method that calibrates and optimizes the initial weight coefficients of each dimension in real time based on the changing trend of texture data from continuous production of soy products. Unlike fixed weights, this method allows the weight coefficients to better reflect the changes in the impact of each dimension on texture during actual production. This mechanism first obtains the time series of texture data from continuous production batches of soy products, extracts the characteristic change patterns of texture data for each dimension, distinguishes between trend changes and random disturbances in texture data, and then calibrates the initial weight coefficients of each dimension online based on the degree of impact of the actual changes in texture data of each dimension on the overall texture. The initial weight coefficients are determined based on the degree of impact of each dimension on the overall taste of soy products. By collecting consumer sensory evaluations and expert tasting data, the correlation coefficients between each dimension and overall taste satisfaction are calculated. After normalizing the correlation coefficients, the initial weights are obtained, ensuring that the sum of all weight coefficients is 1. The higher the degree of impact of a dimension, the larger the initial weight coefficient. After calibration by the dynamic update mechanism, the final updated weight coefficients of each dimension are obtained, which not only retains the core impact of the dimension on taste but also adapts to the data changes during the production process.

[0027] After determining the weight coefficients, the normalized differences between each dimension are weighted and averaged with the updated weight coefficients for the corresponding dimensions. This weighting process involves multiplying the single-dimensional difference by its corresponding weight coefficient to obtain the weighted difference for each dimension. This process is based on the normalized differences, which transform data with different dimensions and numerical ranges into the same numerical interval, completely eliminating calculation biases caused by dimensional differences. This ensures a unified and comparable basis for the differences between dimensions, guaranteeing the scientific validity and accuracy of the weighted calculation. Finally, the weighted Euclidean distance is calculated based on the weighted differences to obtain the overall deviation between the sample and the standard. The weighted Euclidean distance is a distance calculation method that incorporates dimensional weights into the traditional Euclidean distance. First, the weighted differences for each dimension are squared. Then, the squared results for all dimensions are summed. The square root of the sum is then taken to obtain the final weighted Euclidean distance value. This value represents the overall deviation of the sample from the standard. A larger value indicates a higher deviation from the standard texture, and vice versa.

[0028] By combining dynamic weights with normalized weighted Euclidean distance calculation, the overall deviation of the texture of soy products is accurately quantified. This solves the problem that traditional fixed weights cannot adapt to changes in production data and the calculation deviation caused by differences in dimensions. The quantitative results of the deviation are more in line with actual production and taste experience, providing accurate and scientific quantitative basis for subsequent quality anomaly judgment, and greatly improving the accuracy and adaptability of quality monitoring.

[0029] The dynamic update mechanism for determining the weight coefficients of each dimension includes: acquiring a multidimensional texture data vector time series formed by continuous production batches of samples; extracting the spectral features of the texture data in each dimension; distinguishing between trend change components and high-frequency disturbance components; performing morphological operations on the multidimensional texture data vector time series based on the spectral features; and adaptively adjusting the morphological structural elements and length thresholds according to the ratio of low-frequency to high-frequency components in the spectrum to obtain the texture trend sequence after trend extraction; comparing the texture trend sequence with the original texture sequence to obtain the trend offset; and calibrating the weights of each dimension online based on the trend offset.

[0030] Specifically, after calculating the difference between the sample and standard texture data vectors, the weight coefficients of each dimension need to be precisely calibrated through a dynamic update mechanism. This allows the weights to adapt to the texture data change patterns during the production process, making subsequent weighted calculations more consistent with the texture impact characteristics of actual production and improving the accuracy of the comprehensive deviation calculation. Furthermore, this dynamic update mechanism runs online in real time, and the latest weights output are directly used for the comprehensive deviation calculation of the current batch, achieving precise connection between the dynamic weights and the current batch sample calculations, ensuring that the weight coefficients of each batch can adapt to the current production conditions.

[0031] First, a time series of multidimensional texture data vectors formed by samples from continuous batches of soybean products is obtained. This time series is a time-series data set formed by arranging the six-dimensional texture data vectors of continuous batches in chronological order of production time. It serves as the core input for dynamic weight updates and fully reflects the dynamic change trends of each texture dimension (hardness, elasticity, etc.) during the production process, providing real and continuous production data support for subsequent algorithm processing. Based on this time series, spectral analysis was first performed on the texture data of each dimension to extract the spectral features of the texture data of each dimension. The core purpose of spectral analysis is to transform the texture time series data in the time domain to the frequency domain. Through spectral features, trend change components and high-frequency disturbance components can be clearly distinguished. The trend change component corresponds to the low-frequency part in the frequency domain, reflecting the long-term, slow, regular changes in texture data, such as the gradual changes in texture caused by batch differences of raw materials. The high-frequency disturbance component corresponds to the high-frequency part in the frequency domain, representing the short-term, irregular data fluctuations caused by accidental factors in the production process, such as instantaneous voltage fluctuations and uneven mixing of raw materials. Through this frequency domain splitting, the effective signal, i.e., trend change, and the interference signal, i.e., high-frequency disturbance, can be accurately separated, laying the foundation for subsequent trend extraction.

[0032] After extracting spectral features, morphological operations need to be performed. The core is to achieve the connection between "frequency domain analysis to determine parameters and time domain operations to extract trends". The two are not parallel, but progressively connected: spectral analysis is used to determine the key parameters of morphological operations, while morphological operations perform actual data processing in the time domain, making up for the deficiency of spectral analysis that can only decompose signals and cannot extract time domain trends, and finally obtaining an accurate quality trend sequence. Specifically, based on the ratio of low-frequency to high-frequency components obtained from spectral analysis, the structural elements and length thresholds of morphological operations are adaptively adjusted. The morphological structural elements are the basic operational units of morphological operations, employing linear structural elements adapted to the texture time series data, such as line segment structural elements. Their type is determined by the proportion of low-frequency components; the higher the proportion of low-frequency components (i.e., the more pronounced the trend), the longer the linear structural elements are used to better preserve the trend. The length threshold refers to the effective length of the structural elements in morphological operations, directly related to the production batch. The length threshold corresponds to a preset number of consecutive production batches (e.g., 3-5 batches). The higher the proportion of low-frequency components, the larger the length threshold is set to cover more batches; the higher the proportion of high-frequency components, the smaller the length threshold is set to reduce interference. The adaptive adjustment logic for both is as follows: the ratio of low-frequency to high-frequency components is calculated through spectral analysis, and a ratio threshold range is set. Different ratio ranges correspond to different structural element types and length thresholds. This requires no complex mathematical transformations but is based on the correspondence calibrated from production practice data, achieving adaptive parameter matching and avoiding trend extraction deviations caused by fixed parameters.

[0033] A combination of morphological opening and closing operations is employed. Opening operations remove abnormal peaks caused by high-frequency disturbances, while closing operations fill in valleys caused by missing data, resulting in a texture trend sequence after trend extraction. This sequence eliminates interference from high-frequency disturbance components, retaining only the core changing trends of texture data in each dimension. Subsequently, the texture trend sequence is compared dimension-by-dimensionally and batch-by-batch with the original multidimensional texture data vector time series to calculate the trend offset for each dimension. This trend offset is a point-by-point difference, i.e., the difference between the value of the texture trend sequence and the value of the original texture sequence in the same dimension and the same production batch. The point-by-point difference can accurately quantify the deviation of the actual changing trend of texture data in each dimension from the original data in each batch, avoiding the ambiguity of trend offset caused by cumulative differences.

[0034] Finally, the weights of each dimension are calibrated online based on this trend offset. The specific implementation of online calibration is as follows: First, the initial weight coefficients of each dimension are retained. The initial weights are determined based on the degree of influence of each dimension on the overall taste of soy products. The correlation coefficients between each dimension and overall taste satisfaction are calculated through consumer sensory evaluations and expert tasting data, and obtained after normalization to ensure that the sum of all weight coefficients is 1. Then, the absolute values ​​of the point-by-point trend offsets of each dimension are summed to obtain the total offset of each dimension in consecutive batches. The larger the total offset, the greater the impact of that dimension on texture during the production process. The greater the overall change in influence, the greater the change. The initial weights are calibrated based on the total offset. Dimensions with larger total offsets have their weight coefficients adjusted accordingly. If the trend offset is positive and the texture data for that dimension deviates from the standard value, it indicates an increased influence on that dimension, and the weight is appropriately increased. If the trend offset is negative and the texture data for that dimension tends to stabilize, the weight is appropriately decreased. Throughout the calibration process, normalization is used to maintain the sum of all dimension weight coefficients at 1. Calibration calculations are also performed based on the previously completed texture data normalization results to ensure that the weight coefficients for each dimension are adjusted under a unified dimension. The calibrated weight coefficients, as the output of the dynamic update mechanism, are directly used to calculate the overall deviation of the current batch of samples, achieving real-time linkage between dynamic weights and the current batch calculation. This ensures that the weight coefficients not only reflect the actual changes in the production process but also accurately reflect the changes in the influence of each dimension on the overall taste of the soy products, ensuring the accuracy of subsequent overall deviation calculations and providing a more scientific basis for quality judgment.

[0035] The aforementioned dynamic update mechanism achieves precise online calibration of weight coefficients through time-series data analysis and adaptive calculation, solving the problem that fixed weights cannot adapt to the dynamic changes in production texture data. This makes the weights more consistent with the actual texture influence patterns in production, improves the accuracy of comprehensive deviation calculation, and provides a more scientific basis for quality judgment.

[0036] Step S4: Construct a time series from the multidimensional texture data vectors formed by continuous production batches, including the current production batch, and extract effective trend change information based on the time series.

[0037] Step S4 includes: splitting and recombining the multidimensional texture data vectors corresponding to multiple consecutive production batches according to the time sequence of production batches, constructing texture data time series corresponding to each mechanical property dimension, performing denoising and smoothing processing on the texture data time series of each dimension within a preset sliding time window, and performing first-order difference processing or trend fitting processing on the smoothed time series to obtain the trend function of each dimension changing with time; calculating the trend change rate, trend direction and cumulative offset of each dimension within the sliding time window based on the trend function, and comparing the trend change rate with a preset trend change threshold. When the trend change rate is continuously greater than the corresponding preset threshold and the trend direction is consistent with the direction of the comprehensive deviation, it is extracted as valid trend change information.

[0038] Specifically, after calculating the overall deviation between the sample and the standard texture, it is necessary to construct a time series of texture data from consecutive batches to mine the changing patterns, extract effective information that reflects production trends, and combine static deviation with dynamic trends in quality anomaly judgment to improve the foresight and accuracy of monitoring.

[0039] First, according to the chronological order of production batches, the multidimensional texture data vectors corresponding to multiple consecutive production batches are split and recombined according to their dimensions, constructing texture data time series for each mechanical property dimension. The texture data time series refers to the time-series data set formed by arranging the texture data of consecutive production batches under the same mechanical property dimension in chronological order of production time. In this application, independent texture data time series are constructed for six mechanical property dimensions: hardness, elasticity, cohesion, chewiness, adhesion, and stickiness, ensuring that the trend changes of each dimension can be analyzed independently. Then, within a preset sliding time window, the texture data time series of each dimension are subjected to denoising and smoothing processing. The sliding time window is a fixed-length data analysis window divided according to production batches, which can be updated by sliding with the generation of new production batches, ensuring the timeliness and continuity of the analyzed data. Denoising processing is used to eliminate abnormal fluctuations in texture data caused by accidental factors during the production process, while smoothing processing is used to weaken local fluctuations in the data and highlight the overall trend of data change. After these two processing steps, texture data time series characteristics that more closely reflect actual production can be obtained. After data processing, trend fitting is performed on the smoothed time series to obtain trend functions for each dimension as a function of production batches. The trend fitting process uses the least squares method to fit the time series data into a continuous function with production batches as independent variables and texture data as dependent variables. This function can be linear or nonlinear depending on the data characteristics, accurately representing the overall pattern of texture data changes over time.

[0040] When calculating the rate of change, trend direction, and cumulative offset of each dimension within a sliding time window based on a trend function, the rate of change is the first derivative of the trend function, representing the rate of change of texture data with production batches within the sliding time window. For linear trend functions, the rate of change is the slope of the function; the larger the absolute value of the slope, the faster the rate of change of texture data. For nonlinear trend functions, the average first derivative value within the sliding time window is calculated as the rate of change within that window, thus unifying the calculation standard for the rate of change of different types of trend functions. The trend direction is determined by the sign of the rate of change. If the rate of change is positive, it indicates that within the sliding time window, the texture data of the corresponding dimension is changing at a relatively high rate. Texture data shows an upward trend as production batches progress. If the trend change rate is negative, it indicates that the texture data of the corresponding dimension shows a downward trend as production batches progress. The trend direction directly reflects the change direction of texture data, providing a core basis for the extraction of effective trend change information. The calculation of cumulative offset uses the texture data of the first production batch within the sliding time window as the benchmark value. The fitted value of texture data of the corresponding dimension of each production batch within the window is calculated by using the trend function. Then, the difference between each fitted value and the benchmark value is accumulated. The accumulated value is the cumulative offset within the sliding time window. This cumulative offset can be quantitatively characterized as the total degree of deviation of texture data from the initial benchmark within the entire time window.

[0041] The trend change rate is then compared with a preset trend change threshold. The preset trend change threshold is a critical value determined based on historical data of soybean product production, which can define whether the trend change of texture data is abnormal. This threshold is set separately for each dimension, matching the degree of influence of each dimension on the texture of soybean products. The higher the degree of influence of a dimension, the more stringent the trend change threshold setting. At the same time, the relevant data is normalized before the threshold is set to eliminate the difference in the dimensions of the data of each dimension and ensure the scientific nature of the threshold comparison. When the trend change rate is continuously greater than the corresponding preset threshold and the trend direction is consistent with the direction of the comprehensive deviation, the trend change information is extracted as valid trend change information. The direction of the comprehensive deviation is the deviation direction of the texture data of each dimension of the current production batch relative to the standard texture data. Only when the dynamic trend change and the static batch deviation form a positive correlation are they judged as valid trend change information with actual quality monitoring significance.

[0042] By constructing time-series data and extracting effective trend change information, the static single-batch texture deviation analysis is combined with the dynamic continuous batch trend analysis. This solves the problem that traditional quality monitoring can only determine whether a single batch is qualified and cannot predict the trend of quality changes. It can identify abnormal texture trends in the production process in advance, making process adjustments more forward-looking, significantly reducing the risk of large-scale quality failure, and improving the comprehensiveness and predictability of quality monitoring in soy product production.

[0043] Step S5: When the overall deviation is greater than the preset threshold and the effective trend change information meets the anomaly judgment conditions, the quality of the current production batch sample is determined to be unqualified, a quality anomaly alarm is triggered, and the deviation dimension and deviation direction that caused the quality anomaly are analyzed.

[0044] In step S5, determining that the current sample quality is unqualified includes: based on the overall deviation being greater than a preset threshold, determining whether the effective trend change information meets the anomaly judgment conditions. The anomaly judgment conditions include: within a preset sliding time window, the trend change rate of at least one mechanical characteristic dimension is greater than the corresponding trend change threshold for a consecutive preset number of batches, and the trend direction is consistent with the direction of the numerical difference corresponding to the dimension. When the overall deviation is greater than the preset threshold and the anomaly judgment conditions are met, the current sample quality is determined to be unqualified, and the corresponding unqualified judgment result is generated.

[0045] Specifically, after extracting the trend change information of texture data, quality judgment needs to be carried out by combining the comprehensive deviation of the sample with the effective trend change information. This dual-condition verification achieves accurate judgment of the production batch quality, avoiding misjudgments caused by single-dimensional judgment and improving the rigor and accuracy of quality monitoring. When the comprehensive deviation exceeds a preset threshold and the effective trend change information meets the anomaly judgment conditions, the current production batch sample is judged to be substandard, triggering a quality anomaly alarm. When analyzing the deviation dimension and direction that caused the quality anomaly, the current sample quality is first judged to be substandard. This judgment is based on dual conditions: the comprehensive deviation must exceed the preset threshold, and the effective trend change information must also meet the anomaly judgment conditions. The preset threshold is a critical value for the comprehensive deviation determined based on statistical analysis of historical qualified data from soy product production. This threshold is set in conjunction with the impact of texture data from each dimension on the overall taste, and data normalization was performed before setting to eliminate judgment bias caused by differences in the dimensions of different dimensions, ensuring the scientific nature of the threshold judgment.

[0046] If the overall deviation exceeds a preset threshold, further judgment is made as to whether the effective trend change information meets the anomaly judgment conditions. The anomaly judgment conditions are judgment rules set based on the trend changes of continuous production batch texture data. Specifically, within a preset sliding time window, the trend change rate of at least one mechanical property dimension is greater than the corresponding trend change threshold for a consecutive preset number of batches, and the trend direction is consistent with the direction of the numerical difference corresponding to the dimension. The preset number of batches is determined according to the batch scale and quality control precision of industrialized production of soy products. The trend change threshold is a critical value set separately for each dimension. This threshold is statistically derived based on the historical trend change patterns of texture data of each dimension. The higher the degree of influence on the overall taste, the more stringent the trend change threshold is set. The direction of the numerical difference is the deviation direction of the texture data of each dimension of the current production batch sample relative to the standard texture data. That is, the sample value is higher than the standard value as a positive deviation, and the sample value is lower than the standard value as a negative deviation.

[0047] During the judgment process, the trend change rate of each mechanical property dimension within the preset sliding time window is first checked to confirm whether there is a dimension where the number of consecutive preset batches exceeds the corresponding trend change threshold. Then, the consistency between the trend direction and the direction of the numerical difference of the dimension is checked. When at least one dimension meets both of the above requirements, the effective trend change information is judged to meet the abnormal judgment conditions. When the overall deviation is greater than the preset threshold and the effective trend change information meets the abnormal judgment conditions, the quality of the current production batch sample is judged to be unqualified, and the corresponding unqualified judgment result is generated. This result includes the current production batch information, the overall deviation value, the specific dimension that triggered the abnormal judgment, and the trend change information, ensuring the integrity and traceability of the judgment result. Upon determining that the current production batch sample is substandard, a quality anomaly alarm is immediately triggered. This alarm is an early warning mechanism of the soybean product production quality monitoring system, which can provide early warnings through audible and visual prompts and system pop-ups at the production end, promptly reminding production control personnel to handle the quality anomaly. Simultaneously, it initiates the deviation dimension and deviation direction analysis process that led to the quality anomaly. This analysis process is based on the multidimensional texture data vector of the current production batch sample and the standard texture data vector, calculating the numerical difference between the two dimension by dimension. Combining this with the preset allowable fluctuation threshold for each dimension, the dimension whose absolute value of the numerical difference exceeds the corresponding allowable fluctuation threshold is identified as the deviation dimension. Then, based on the positive or negative value of the numerical difference, the specific deviation direction of each deviation dimension is determined. A positive numerical difference indicates that the dimension is too high, and a negative numerical difference indicates that the dimension is too low. Finally, all deviation dimensions and their corresponding deviation directions are integrated to form the basic analysis data for the cause of the quality anomaly, providing precise directional guidance for subsequent process parameter adjustments.

[0048] By employing dual criteria, the quality of soy product production batches can be accurately determined, solving the problem of misjudgment and omission that is prone to occur in traditional quality monitoring with a single judgment dimension. Combining the judgment method of static deviation degree and dynamic trend change makes the identification of quality anomalies more accurate. At the same time, real-time quality anomaly alarms and deviation dimension analysis can quickly locate problems and improve the response efficiency of production quality control.

[0049] In step S5, the analysis of the deviation dimensions and deviation directions that lead to quality abnormalities includes: for the numerical differences of each dimension of the multidimensional texture data vector of the sample and the standard texture data vector, the absolute value of the numerical difference of each dimension is compared with the corresponding allowable fluctuation threshold, and the dimension with the absolute value greater than the allowable fluctuation threshold is determined as the deviation dimension. For each deviation dimension, the deviation direction is determined according to the sign of the numerical difference.

[0050] Specifically, after determining that the quality of the current production batch sample is unqualified and triggering an alarm, it is necessary to accurately locate the core texture dimension that caused the quality abnormality and the specific deviation direction. By comparing and analyzing the dimension difference with the allowable fluctuation threshold, a clear and accurate directional guide is provided for subsequent process parameter adjustments, ensuring the pertinence and effectiveness of quality rectification.

[0051] First, the normalized sample multidimensional texture data vector and the standard texture data vector corresponding to the target bean product category are retrieved. Based on the two, the numerical differences of each dimension are recalculated to ensure the accuracy of the difference data. The numerical difference is the sample texture data of each dimension minus the value of the corresponding dimension of the standard texture data vector. This difference directly reflects the deviation of the single-dimensional texture data from the standard. Since the normalization process has been completed in the early stage, the differences of each dimension have eliminated the differences in units and numerical ranges, and have a unified basis for comparison.

[0052] Subsequently, the allowable fluctuation threshold for each dimension in the standard texture data vector is obtained. This allowable fluctuation threshold is a value determined based on statistical analysis of historical texture data from high-quality products in the target soy product category. It represents the reasonable range within which the texture data of each dimension deviates from the standard value. This threshold corresponds one-to-one with the texture dimension; the higher the dimension's impact on the overall taste of the soy product, the stricter its allowable fluctuation threshold. This threshold is also based on normalized data, maintaining consistency with the numerical differences across data dimensions. After obtaining the differences and thresholds, the absolute value of the numerical difference for each dimension is compared one by one with the corresponding allowable fluctuation threshold. The calculation of the absolute value eliminates the directional influence of the positive or negative sign of the difference, retaining only the quantitative degree of deviation. Through this comparison, dimensions with absolute values ​​greater than the corresponding allowable fluctuation threshold are identified as deviation dimensions. These deviation dimensions are the core mechanical property dimensions that directly cause abnormal texture quality in soy products and are the dimensions that need to be focused on in subsequent process parameter adjustments. After identifying all deviation dimensions, for each deviation dimension, the specific direction of deviation is determined based on the sign of its original numerical difference. The sign of the numerical difference directly reflects the direction of deviation. If the difference is positive, it indicates that the sample texture data of the deviation dimension is higher than the standard texture data, and it is determined to be a deviation direction where the dimension is too high. If the difference is negative, it indicates that the sample texture data of the deviation dimension is lower than the standard texture data, and it is determined to be a deviation direction where the dimension is too low. All deviation dimensions and their corresponding deviation directions together constitute the core cause analysis results of quality anomalies, providing direct data support for subsequent identification of key mechanical property dimensions and generation of process parameter adjustment suggestions.

[0053] By scientifically comparing the numerical difference with the allowable fluctuation threshold, the deviation dimension and direction of quality anomalies are accurately located, solving the problem that traditional quality monitoring cannot accurately identify the core dimension of quality deviation. This provides a clear direction for process parameter adjustment, greatly improves the efficiency and accuracy of quality rectification, and effectively reduces the trial and error cost of process adjustment.

[0054] Step S6: Based on the deviation dimension and deviation direction, identify the key mechanical property dimensions, generate process parameter adjustment suggestions for the current production batch, and output a quality monitoring report.

[0055] In step S6, identifying key mechanical property dimensions includes: sorting the deviation dimensions according to the deviation dimension and deviation direction, and combining the weight coefficient of each deviation dimension in the overall deviation degree, and identifying the top preset number of mechanical property dimensions with the highest weight coefficient as key mechanical property dimensions.

[0056] Specifically, after completing the precise analysis of the dimensions and directions of quality anomalies, it is necessary to further identify the mechanical property dimensions that have a key impact on production quality. Combined with weighting coefficients, the core direction of process adjustments is clarified, and then targeted process parameter adjustment suggestions are generated, forming a complete quality monitoring report to guide production optimization. The first step is to identify the key mechanical property dimensions. This identification process is based on the deviation dimension, deviation direction, and the weighting coefficient of each deviation dimension in the overall deviation degree. The weighting coefficient is a value determined after calibration through a dynamic update mechanism. This coefficient is set according to the degree of influence of each mechanical property dimension on the overall taste and texture of the soy products; the higher the degree of influence, the larger the weighting coefficient. Furthermore, the sum of the weighting coefficients of all dimensions is normalized to 1, eliminating the influence of dimensional differences on the weighting ratio and ensuring the scientific nature of the coefficient comparison. During the identification process, all identified deviation dimensions and their corresponding weight coefficients are retrieved first. Combined with the analysis results of the deviation direction, all deviation dimensions are sorted in descending order using the weight coefficients as the core indicator. The sorting process is based solely on the weight coefficient values; the larger the weight coefficient value, the higher the ranking. Then, a preset quantity is set according to the precision requirements of quality control in soy product production. This preset quantity is the number of key dimensions selected based on actual production needs, representing the number of dimensions that require key process adjustments. From the sorted deviation dimensions, the mechanical characteristic dimensions with the highest weight coefficients are selected and identified as key mechanical characteristic dimensions. These dimensions are the core influencing dimensions that cause abnormal quality in the current batch of soy products and are also the focus of subsequent process parameter adjustments.

[0057] After identifying the key mechanical property dimensions, based on the analysis results of deviation dimensions, deviation directions, and key mechanical property dimensions, process parameter adjustment suggestions are generated for the current production batch. First, a correspondence between textural properties and production process parameters is established. This correspondence is determined through statistical analysis of historical process data and textural testing data from soybean product production. The mapping relationship between each mechanical property dimension and the process parameters of core production processes such as grinding, heating, coagulation, pressing, and drying is clarified. Then, for each deviation dimension, the corresponding mechanical property value adjustment direction is determined based on its deviation direction. If the deviation direction is too high, the mechanical property value of that dimension needs to be reduced; if the deviation direction is too low, the mechanical property value of that dimension needs to be increased. Combining the correspondence between textural properties and production process parameters, the adjustment direction of mechanical property values ​​is mapped to the adjustment direction of specific production process parameters. Simultaneously, the adjustment priority of each process parameter is determined based on the weight coefficient of each deviation dimension. The larger the weight coefficient, the higher the adjustment priority of the process parameter corresponding to the deviation dimension. Specific adjustment guidance suggestions are formulated for the process parameters corresponding to the key mechanical property dimensions, clarifying the adjustment trend and range of the process parameters to ensure the operability of the suggestions. Finally, based on all the above quality monitoring and analysis results, a complete quality monitoring report is output. This report includes current production batch information, quality anomaly judgment results, deviation dimensions and deviation directions analysis, key mechanical property dimension identification results, specific process parameter adjustment suggestions and adjustment priorities, etc. The report content is accurate, logically clear, and provides specific suggestions, which can directly guide production personnel to carry out process parameter adjustments and production quality rectification, realizing the transformation of quality monitoring results into production optimization guidance.

[0058] By accurately identifying key mechanical property dimensions through weighted coefficients and generating targeted adjustment suggestions based on the correspondence between texture and process, this solves the problem of traditional process adjustments lacking a clear core direction and involving blind trial and error. This makes process optimization more targeted, significantly improves the efficiency of production quality rectification, and reduces the waste of raw materials and energy.

[0059] Step S6 further includes: finding the pre-established correspondence between the texture properties and the production process parameters based on the mechanical property index corresponding to the deviation dimension, determining the production process parameters associated with the mechanical property index, generating the corresponding production process parameter adjustment direction based on the deviation direction, determining the adjustment priority of each process parameter by combining the weight coefficient of each deviation dimension in the overall deviation degree, and outputting the corresponding process parameter adjustment suggestions.

[0060] Specifically, after identifying the key mechanical property dimensions, it is necessary to rely on the correspondence between the texture properties and the production process parameters to transform the deviation adjustment requirements of the texture dimensions into specific process parameter adjustment directions, and combine the weighting coefficients to clarify the adjustment priority, thereby generating process parameter adjustment suggestions that can directly guide production.

[0061] First, based on the mechanical property indicators corresponding to the determined deviation dimensions, the pre-established correspondence between textural properties and production process parameters is found. This correspondence is established based on the historical textural testing data and process parameter data of soybean product production, after statistical analysis and process verification. Specifically, historical process parameter data of grinding, heating, coagulation, pressing, drying and other processes in soybean product production are collected, as well as textural testing data such as hardness, elasticity, cohesiveness, chewiness, adhesion and stickiness of corresponding batches of products. Correlation analysis and regression fitting are performed on the above data to determine the direction, magnitude and significance of the influence of each process parameter on each textural property indicator. Then, the statistically obtained correlation is calibrated, corrected and verified through multiple sets of process verification experiments to form a stable and reliable textural property-production process parameter mapping relationship. This mapping relationship is stored in the system database in the form of a pre-set mapping library or mapping table, which is represented by the corresponding association between each mechanical property indicator and one or more production process parameters, clarifying the influence trend of process parameter changes on textural indicators, and providing a direct basis for generating the direction of subsequent process parameter adjustments.

[0062] Based on this correspondence, the production process parameters directly related to the mechanical property indicators corresponding to each deviation dimension are determined from the process parameters of the entire soybean product production process. This clarifies the types of process parameters that have a core impact on the mechanical property indicator, achieving precise conversion of the textural property dimension into production process parameters. Subsequently, based on the deviation direction of each deviation dimension, corresponding adjustment directions for the production process parameters are generated. If the deviation direction of the deviation dimension is too high (i.e., the sample textural data is higher than the standard textural data), an instruction is generated to adjust the corresponding process parameter in the direction of reducing the mechanical property indicator. If the deviation direction of the deviation dimension is too low (i.e., the sample textural data is lower than the standard textural data), an instruction is generated to adjust the corresponding process parameter in the direction of increasing the mechanical property indicator, ensuring a high degree of matching between the adjustment direction of the process parameters and the correction requirements of the textural dimension.

[0063] Next, the adjustment priority of each process parameter is determined by combining the weight coefficients of each deviation dimension in the overall deviation degree. The weight coefficients of each deviation dimension are values ​​that reflect the degree of influence of that dimension on the overall texture of the soy products after being calibrated by a dynamic update mechanism, and have been normalized to ensure uniformity of dimensions. The larger the weight coefficient of a deviation dimension, the higher the importance of its corresponding process parameter in production quality rectification, and the higher its adjustment priority. For deviation dimensions with the same weight coefficient, they can be further ranked according to their degree of deviation to ensure that the setting of adjustment priorities is scientific and reasonable. Finally, based on the determined adjustment direction and priority of the process parameters, corresponding process parameter adjustment suggestions are output. These suggestions clarify the specific adjustment trend of each process parameter, and, in conjunction with the production process specifications of the target soy product category, provide a reasonable adjustment range for the process parameters to ensure that the suggestions are operable and can directly guide production personnel to optimize and adjust the process parameters. At the same time, the process parameter adjustment suggestions are integrated with the previous quality anomaly judgment results, deviation dimension analysis results, and key mechanical property dimension identification results and included in the final quality monitoring report.

[0064] The above operations have enabled the precise transformation of texture deviation problems into process parameter adjustment schemes, solving the problem of the disconnect between texture problems and process adjustments in traditional soy product production, and the lack of clear basis for adjustments. Combined with the priority setting of weight coefficients, the process adjustments are more targeted, which greatly improves the efficiency of production quality rectification and reduces the trial and error costs of process adjustments.

[0065] The above describes a quality monitoring method for soybean product production in an embodiment of this application. The following describes a quality monitoring system for soybean product production in an embodiment of this application. Please refer to [link / reference]. Figure 2 A quality monitoring system for soy product production, as described in this application embodiment, includes: The data acquisition module is used to collect samples of the target soy product category on the soy product production line and record the corresponding production batch information and key process parameters of the samples. The testing module is used to perform texture testing on samples, obtain multidimensional texture data vectors, normalize the multidimensional texture data vectors, and retrieve standard texture data vectors corresponding to the target bean product category from the database. The calculation module is used to calculate the numerical difference between the multidimensional texture data vector and the standard texture data vector in each dimension, and combine it with preset weight coefficients to obtain the overall deviation of the current sample from the standard. The extraction module is used to construct a time series from the multidimensional texture data vectors formed by continuous production batches, including the current production batch, and to extract effective trend change information based on the time series. The judgment module is used to determine that the quality of the current production batch sample is unqualified when the overall deviation is greater than the preset threshold and the effective trend change information meets the abnormal judgment conditions, trigger a quality abnormality alarm, and analyze the deviation dimension and deviation direction that caused the quality abnormality. The output module is used to identify key mechanical property dimensions based on the deviation dimension and direction, generate process parameter adjustment suggestions for the current production batch, and output a quality monitoring report.

[0066] In summary, the quality monitoring method and system for soybean product production provided in this application achieves precise binding of texture data with the production process by collecting samples and associating them with production batches and process parameters. It acquires and normalizes six-dimensional mechanical property data using a texture analyzer, and calculates weighted Euclidean distance using dynamically updated weighting coefficients, accurately quantifying the comprehensive deviation of samples from standards. By constructing continuous batch texture time series and extracting effective trend change information, it combines static deviation analysis with dynamic trend prediction, achieving dual-condition judgment of quality anomalies and effectively avoiding misjudgments and omissions caused by single-dimensional judgments. After determining non-compliance, it accurately locates the core dimension and direction of texture deviation by comparing the deviation dimension with the allowable fluctuation threshold, and then identifies key mechanical property dimensions by combining weighting coefficients. Based on the correspondence between texture and process, it generates actionable process adjustment suggestions, ultimately outputting a complete quality monitoring report. This solution solves the problems of subjective monitoring and blind process adjustments in traditional methods, improves the accuracy and foresight of quality monitoring, reduces production trial-and-error costs, and ensures the consistency of soybean product texture and production efficiency.

[0067] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0068] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0069] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A quality control method for soybean product production, characterized in that, The method includes: Step S1: Collect samples of the target soy product category on the soy product production line and record the production batch information and key process parameters corresponding to the samples; Step S2: Perform texture testing on the sample to obtain a multidimensional texture data vector, normalize the multidimensional texture data vector, and retrieve the standard texture data vector corresponding to the target bean product category from the database; Step S3: Calculate the numerical difference between the multidimensional texture data vector and the standard texture data vector in each dimension, and combine it with the preset weight coefficients to obtain the overall deviation between the current sample and the standard. Step S4: Construct a time series from the multidimensional texture data vector formed by consecutive production batches, including the current production batch, and extract effective trend change information based on the time series; Step S5: When the overall deviation is greater than the preset threshold and the effective trend change information meets the anomaly judgment conditions, the quality of the current production batch sample is determined to be unqualified, a quality anomaly alarm is triggered, and the deviation dimension and deviation direction that caused the quality anomaly are analyzed. Step S6: Based on the deviation dimension and deviation direction, identify the key mechanical property dimensions, generate process parameter adjustment suggestions for the current production batch, and output a quality monitoring report.

2. The quality control method for soybean product production according to claim 1, characterized in that, Step S2 includes: For each sample of the target bean product category, a texture analyzer was used to test the mechanical properties of the sample in six dimensions: hardness, elasticity, cohesion, chewiness, adhesion and stickiness. Based on the mechanical property data, a multidimensional texture data vector of the sample was constructed. Based on the identifier of the target soy product category, a pre-established database is accessed to extract the standard texture data vector corresponding to the target soy product category. The standard texture data vector is determined by statistical analysis of historical multidimensional texture data of high-quality products, and the standard texture data vector has the same dimensional structure as the multidimensional texture data vector of the sample.

3. The quality monitoring method for soybean product production according to claim 1, characterized in that, Step S3 includes: For the multidimensional texture data vector of the sample and the standard texture data vector, the numerical difference between the two in each dimension is calculated. A dynamic update mechanism is used to determine the weight coefficient of each dimension. The normalized numerical difference is weighted with the updated weight coefficient, and the weighted Euclidean distance is calculated based on the weighted difference to obtain the comprehensive deviation between the sample and the standard.

4. The quality control method for soybean product production according to claim 3, characterized in that, The weight coefficients for each dimension are determined using a dynamic update mechanism, including: Obtain the time series of multidimensional texture data vectors formed by samples from continuous production batches, extract the spectral features of texture data in each dimension, and distinguish between trend change components and high-frequency disturbance components. Based on the spectral features, morphological operations are performed on the multidimensional texture data vector time series, and the morphological structural elements and length thresholds are adaptively adjusted according to the ratio of low-frequency components to high-frequency components in the spectrum to obtain the texture trend sequence after trend extraction. The texture trend sequence is compared with the original texture sequence to obtain the trend offset, and the weights of each dimension are calibrated online based on the trend offset.

5. The quality control method for soybean product production according to claim 1, characterized in that, Step S4 includes: According to the time sequence of production batches, the multidimensional texture data vectors corresponding to multiple consecutive production batches are split and recombined according to dimensions to construct texture data time series corresponding to each mechanical property dimension. Within a preset sliding time window, the texture data time series of each dimension are denoised and smoothed. The smoothed time series are then subjected to first-order difference processing or trend fitting processing to obtain the trend function of each dimension changing with time. Based on the trend function, the trend change rate, trend direction and cumulative offset of each dimension within the sliding time window are calculated, and the trend change rate is compared with a preset trend change threshold. When the trend change rate is continuously greater than the corresponding preset threshold and the trend direction is consistent with the direction of the comprehensive deviation, it is extracted as valid trend change information.

6. The quality control method for soybean product production according to claim 1, characterized in that, Step S5, determining that the current sample quality is unqualified, includes: Based on the overall deviation being greater than a preset threshold, it is determined whether the effective trend change information meets the anomaly judgment conditions. The anomaly judgment conditions include: within a preset sliding time window, the trend change rate of at least one mechanical characteristic dimension is greater than the corresponding trend change threshold for a consecutive preset number of batches, and the trend direction is consistent with the direction of the numerical difference corresponding to the dimension. When the overall deviation is greater than the preset threshold and meets the anomaly judgment conditions, the current sample quality is determined to be unqualified, and a corresponding unqualified judgment result is generated.

7. The quality control method for soybean product production according to claim 6, characterized in that, Step S5 analyzes the deviation dimensions and directions that lead to quality anomalies, including: For the numerical differences of each dimension between the multidimensional texture data vector of the sample and the standard texture data vector, the absolute value of the numerical difference of each dimension is compared with the corresponding allowable fluctuation threshold. The dimension with an absolute value greater than the allowable fluctuation threshold is determined as the deviation dimension. For each deviation dimension, the deviation direction is determined according to the sign of the numerical difference.

8. The quality control method for soybean product production according to claim 1, characterized in that, Step S6 identifies key mechanical property dimensions, including: Based on the deviation dimensions and deviation directions, and combined with the weight coefficients of each deviation dimension in the overall deviation degree, the deviation dimensions are sorted, and the mechanical characteristic dimensions with the highest weight coefficients are identified as key mechanical characteristic dimensions.

9. The quality control method for soybean product production according to claim 8, characterized in that, Step S6 also includes: Based on the mechanical property index corresponding to the deviation dimension, the pre-established correspondence between the texture properties and the production process parameters is found, the production process parameters associated with the mechanical property index are determined, and the corresponding production process parameter adjustment direction is generated according to the deviation direction. The adjustment priority of each process parameter is determined by combining the weight coefficient of each deviation dimension in the overall deviation degree, and the corresponding process parameter adjustment suggestions are output.

10. A quality monitoring system for soybean product production, used to implement the quality monitoring method for soybean product production as described in any one of claims 1-9, characterized in that, The system includes: The data acquisition module is used to collect samples of the target soybean product category on the soybean product production line and record the production batch information and key process parameter information corresponding to the samples. The testing module is used to perform texture testing on the samples, obtain multidimensional texture data vectors, normalize the multidimensional texture data vectors, and retrieve standard texture data vectors corresponding to the target bean product category from the database. The calculation module is used to calculate the numerical difference between the multidimensional texture data vector and the standard texture data vector in each dimension, and combine it with preset weight coefficients to obtain the overall deviation of the current sample from the standard. The extraction module is used to construct a time series from the multidimensional texture data vector formed by continuous production batches, including the current production batch, and to extract effective trend change information based on the time series. The determination module is used to determine that the quality of the current production batch sample is unqualified when the overall deviation is greater than a preset threshold and the effective trend change information meets the abnormal determination conditions, trigger a quality abnormality alarm, and analyze the deviation dimension and deviation direction that caused the quality abnormality. The output module is used to identify key mechanical property dimensions based on the deviation dimension and deviation direction, generate process parameter adjustment suggestions for the current production batch, and output a quality monitoring report.

Images