Full-process management and control method and system applied to tea processing production line

Abstract

The invention discloses a full-process management and control method and system applied to a tea processing production line, and the method comprises the steps: obtaining production full-process monitoring information, and carrying out the multi-modal feature fusion, and obtaining the multi-modal production full-process monitoring fusion features; establishing a production quality knowledge graph, performing production quality evaluation on the current production line, judging whether the processing quality is unqualified or not, and performing processing technology optimization; analyzing the operation health condition and the residual life of the production equipment of the current production line by using the multi-modal production full-process monitoring fusion features, and generating an operation health condition index and residual life prediction information; and whether the current production line needs processing supervision is analyzed based on the operation health condition index and the residual life prediction information, and a production line management and control scheme is formulated to manage and control the processing production line. The operation condition and health of processing equipment are concerned while tea processing quality management and control are considered, equipment management and control and production quality management and control are achieved, and the comprehensiveness and intelligence of tea processing are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of tea processing production line control, and particularly to a full-process control method and system applied to a tea processing production line. Background Art

[0002] Tea processing is a key link in the combination of traditional agriculture and the food industry, involving multiple processes such as fresh leaf pretreatment, withering, fixation, rolling, fermentation, drying, sorting, and packaging. The complexity of its process and the difficulty of quality control increase significantly with the differences in tea types (such as green tea, black tea, oolong tea). The current tea processing production lines generally have the following technical bottlenecks: First, traditional processing relies on manual experience control, and there is a lack of precise quantification means, especially in the control of key parameters such as withering humidity, fixation temperature, and fermentation duration, resulting in poor quality stability between batches; Second, the data of multiple processes are severely isolated into islands. The equipment in each link operates independently, and there is a lack of real-time correlation analysis of environmental parameters, equipment status, and quality indicators, making it difficult to achieve full-process traceability and collaborative optimization; Third, the lag of quality inspection is obvious. The existing production lines mostly adopt off-line sampling inspection methods and cannot conduct on-line dynamic monitoring of core quality indicators such as tea color, moisture content, and aroma components, and the response efficiency to abnormal working conditions is low.

[0003] With the upgrading of the consumer market's demand for tea standardization and high-endization, the traditional processing mode has been difficult to meet the strict requirements of modern tea enterprises for cost control, energy efficiency optimization, and quality uniformity. Although some enterprises have tried to introduce automated equipment and SCADA systems in recent years, their functions are mostly limited to single-machine control or local data collection, lacking the comprehensive control ability covering the entire "fresh leaf - finished product" chain. At the same time, most tea processing control processes only focus on the output quality, ignoring the health and operating conditions of production equipment, resulting in the phenomenon that the output quality is still unqualified after the production process is adjusted. Therefore, there is an urgent need for a full-process control method for tea processing production lines to improve the intelligence and comprehensiveness of tea processing control. Summary of the Invention

[0004] The present invention overcomes the defects of the prior art and provides a full-process control method and system applied to a tea processing production line, and an important purpose thereof is to improve the comprehensiveness and intelligence of tea processing control.

[0005] To achieve the above object, the first aspect of the present invention provides a full-process control method applied to a tea processing production line, including: Obtaining production full-process monitoring information and performing preprocessing, and performing multi-modal feature fusion according to the preprocessed production full-process monitoring information to obtain multi-modal production full-process monitoring fusion features; Build a production quality knowledge graph, conduct production quality assessment in combination with the multi-modal production full-process monitoring fusion features, and determine whether production quality control is required according to the assessment results; Analyze the production equipment on the current production line using the multi-modal production full-process monitoring fusion features, judge the operating health status of each device on the current production line, and generate an operating health status index; Use the multi-modal production full-process monitoring fusion features to predict the remaining life of the production equipment on the current production line, judge the degradation degree of each device on the current production line, and obtain remaining life prediction information; Analyze whether the current production line needs processing supervision based on the operating health status index and the remaining life prediction information. If processing supervision is required, formulate a production line control plan to control the processing production line.

[0006] In this solution, the production full-process monitoring information is obtained and preprocessed, and multi-modal feature fusion is performed according to the preprocessed production full-process monitoring information to obtain multi-modal production full-process monitoring fusion features, which specifically include: Obtain production full-process monitoring information through the sensor network installed on the tea processing production line. The production full-process monitoring information includes production line equipment operation monitoring information, production line environment monitoring information, and processed product monitoring information; Transmit the obtained production full-process monitoring information to the edge gateway, and preprocess the production full-process monitoring information through the edge computing node. Use the sliding window mechanism to detect abnormal data and use the linear interpolation method for compensation. Perform image enhancement and filtering processing on the processed product monitoring information; Extract features from the preprocessed production full-process monitoring information to obtain single-modal feature extraction information. The single-modal feature extraction information includes production line equipment operation monitoring features, production line environment monitoring features, and processed product monitoring features; Introduce the principal component analysis method to reduce the dimension of the single-modal feature extraction information, calculate the principal component scores corresponding to the extracted single-modal features and judge them against a preset threshold, and define the single-modal features corresponding to the principal component scores greater than the preset threshold as principal component factors; Use the principal component factors for principal component direction projection to obtain a projection scatter plot, and select the single-modal features within a preset selection range in the projection scatter plot to obtain single-modal feature dimension reduction information; Import the single-modal feature dimension reduction information into a pre-trained Bi-LSTM network, calculate the attention scores between different modal features through the attention mechanism, establish an association matrix between temporal features and spatial features for feature fusion, and obtain multi-modal production full-process monitoring fusion features.

[0007] In this solution, establishing a production quality knowledge graph, combining the multi-modal production full-process monitoring fusion features for production quality assessment, and determining whether production quality control is required based on the assessment results specifically includes: Retrieving historical processing instances of different tea categories based on historical data, extracting the final processing quality ratings and production requirements of each historical processing instance, and classifying each historical processing instance into qualified instances and unqualified instances; Respectively extracting the historical processing technology features and historical product production features corresponding to qualified instances and unqualified instances, dividing the historical product production features of each historical processing history through the extracted processing technology features, and generating historical product production feature sets for several different processing stages; Based on the obtained historical product production feature sets for several different processing stages, using the graph theory method to establish a production quality knowledge graph with the tea category - processing stage - processing feature - processing quality as the association path; Obtaining the multi-modal production full-process monitoring fusion features, extracting the tea category of the current processing batch through the multi-modal production full-process monitoring fusion features, generating a retrieval label based on the tea category of the current processing batch, and importing it into the production quality knowledge graph for association path screening; Obtaining several screened association paths, calculating the similarity value with the multi-modal production full-process monitoring fusion features to obtain the similarity value, screening the association path with the highest similarity, and extracting the historical processing quality corresponding to the association path; If the historical processing quality corresponding to the association path is unqualified, indicating that there are production process problems in the current production stage of the current processing batch, then obtain a production regulation strategy from the preset production strategy library for production quality control.

[0008] In this solution, analyzing the production equipment on the current production line using the multi-modal production full-process monitoring fusion features, judging the operating health status of each equipment on the current production line, and generating an operating health status index specifically includes: Extracting the historical operating parameters of each production equipment under different processing technologies from the historical production equipment operation library, classifying the operating status and then forming a historical equipment operation data set, where the historical equipment operation data set includes fault historical operating parameters and non-fault historical operating parameters; Constructing a first anomaly detection model based on a convolutional neural network, training the first anomaly detection model through the historical equipment operation data set to construct a first training set, validating it through repeated training and parameter adjustment, and finally outputting a first anomaly detection model that meets the expectations; Extract non-fault historical operation parameters from the historical device operation dataset to construct a second training set, and use SVDD to construct a second anomaly detection model. Import the second training set into the second anomaly detection model for model training; Use the Gaussian kernel function to map the input training data to a high-dimensional space. Solve in the mapped high-dimensional space to obtain the minimum hypersphere boundary containing all non-fault historical operation parameter samples, and then output the trained second anomaly detection model; Cascade the first anomaly detection model and the second anomaly detection model to form a production line anomaly detection model. Obtain the multi-modal production full-process monitoring fusion features and input them into the production line anomaly detection model to judge the operation health status of each device on the current production line; In the first anomaly detection model, extract features based on a preset four-layer convolutional layer to obtain a feature map. Input the obtained feature map into the global average pooling layer and then output the fault probability distribution of each production device at the current moment through the connected fully connected layer; In the second anomaly detection model, generate a real-time feature vector according to the input multi-modal production full-process monitoring fusion features, calculate the Mahalanobis distance of the real-time feature vector in the kernel space. If it is greater than the hypersphere radius, it is marked as an abnormal real-time feature vector, and calculate the relative deviation degree between the abnormal real-time feature vector and the hypersphere boundary; Weightedly fuse the fault probability distribution and the relative deviation degree to generate the operation health status index of each device on the current production line, which is used to characterize the normal operation of the devices for tea processing on the current production line.

[0009] In this solution, the remaining life of the production equipment on the current production line is predicted by using the multi-modal production full-process monitoring fusion features, the degradation degree of each device on the current production line is judged, and the remaining life prediction information is obtained, which specifically includes: Obtain the multi-modal production full-process monitoring fusion features and the historical maintenance records of each device, perform time series processing to generate the time series of the production equipment on the current production line, and input it into the pre-trained LSTM model for remaining life prediction; Capture the device degradation trajectory features through two layers of bidirectional LSTM layers connected to the input layer. In the first layer of bidirectional LSTM, use the forward unit to capture the time series evolution features from the start to the end of the time series, and use the backward unit to retroactively obtain the potential impact of the historical state on the current moment. Concatenate the outputs of the forward unit and the backward unit to output the device degradation trajectory features; Transfer the device degradation trajectory features to the second layer of bidirectional LSTM, introduce the attention mechanism, calculate the attention scores of the device degradation trajectory features at different time steps through the attention mechanism, and weight the device degradation trajectory features through the attention scores; Input the weighted device degradation trajectory features into the fully connected layer, and output the predicted remaining life values of each device on the current production line through the ReLU activation function to obtain the predicted remaining life information.

[0010] In this solution, it is analyzed whether the current production line needs processing supervision based on the operating health status index and the predicted remaining life information. If processing supervision is required, a production line control plan is formulated to control the processing production line, specifically including: Preset four types of processing supervision types: immediate maintenance, continue production, delayed maintenance, and optimize control parameters, and set the corresponding processing supervision determination rules. Obtain the operating health status index and the predicted remaining life information, and judge them against each processing supervision determination rule to obtain the processing supervision determination information; If the processing supervision determination information is to continue production, maintain the current processing production line control parameters for continuous processing; if the processing supervision determination information is one of the other three types of processing supervision types, formulate a production line control plan to control the processing production line; Obtain the operation status data of each production line in the current processing workshop, introduce the sparrow optimization algorithm to optimize the processing control parameters, extract the available status of the tea processing equipment to be controlled through the processing supervision determination information, set the optimization goal according to the available status, and combine the operation status data of each production line in the current processing workshop to initialize the population to obtain the initial sparrow population; Calculate the fitness of each sparrow individual in the initial sparrow population, sort each sparrow individual according to the calculated fitness, select the corresponding number of sparrow individuals as discoverers in combination with the preset discoverer ratio, and update the positions of the remaining sparrow individuals as followers; Set a warning mechanism, randomly select a preset number of individuals in the sparrow population as warning individuals through the warning mechanism. When it is detected that the positions of multiple individuals overlap or the fitness does not change after multiple iterations, the warning mechanism is triggered, and some sparrow individuals are guided by the warning individuals to break away from the area to avoid local optimal solutions; Repeat the iteration until the preset stop rule is met, output the optimal sparrow position, generate the processing control optimization parameters according to the optimal sparrow position, map the processing control optimization parameters to the device control instruction set to generate a production line control plan for processing equipment scheduling and operation parameter regulation.

[0011] The second aspect of the present invention provides a full-process control system applied to a tea processing production line. The system includes: a memory and a processor. The memory contains a full-process control method program for a tea processing production line. When the full-process control method program for a tea processing production line is executed by the processor, the following steps are implemented: Obtain the production whole - process monitoring information and perform pre - processing. Based on the pre - processed production whole - process monitoring information, conduct multi - modal feature fusion to obtain multi - modal production whole - process monitoring fusion features; Build a production quality knowledge graph, combine the multi - modal production whole - process monitoring fusion features to conduct production quality assessment, and determine whether production quality control is required according to the assessment results; Use the multi - modal production whole - process monitoring fusion features to analyze the production equipment of the current production line, judge the operating health status of each equipment on the current production line, and generate an operating health status index; Use the multi - modal production whole - process monitoring fusion features to predict the remaining life of the production equipment on the current production line, judge the degradation degree of each equipment on the current production line, and obtain remaining life prediction information; Based on the operating health status index and the remaining life prediction information, analyze whether the current production line needs processing supervision. If processing supervision is required, formulate a production line control plan to control the processing production line.

[0012] The present invention discloses a whole - process control method and system applied to a tea processing production line, including: obtaining production whole - process monitoring information and conducting multi - modal feature fusion to obtain multi - modal production whole - process monitoring fusion features; building a production quality knowledge graph, conducting production quality assessment on the current production line, judging whether there are unqualified processing quality phenomena and optimizing the processing technology; using the multi - modal production whole - process monitoring fusion features to analyze the operating health status and remaining life of the production equipment on the current production line, generating an operating health status index and remaining life prediction information; based on the operating health status index and the remaining life prediction information, analyzing whether the current production line needs processing supervision, and formulating a production line control plan to control the processing production line. While taking into account the tea processing quality control, it also pays attention to the operating conditions and health of the processing equipment, realizes equipment control and production quality control, and improves the comprehensiveness and intelligence of tea processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the technical solutions in the embodiments or examples of the present invention, the following will briefly introduce the drawings required for use in the embodiments or example descriptions. Obviously, the following - described drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained according to the drawings shown.

[0014] Figure 1 It is a flowchart of a whole - process control method applied to a tea processing production line provided by an embodiment of the present invention; Figure 2 It is a flowchart of a production equipment control method for a tea processing production line provided by an embodiment of the present invention; Figure 3 A block diagram of a full - process control system applied to a tea processing production line provided by an embodiment of the present invention; The realization of the object of the present invention, functional features and advantages will be further described with reference to the embodiments and the accompanying drawings. Specific embodiments

[0015] In order to more clearly understand the above - mentioned objects, features and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, without conflict, the embodiments of the present application and the features in the embodiments can be combined with each other.

[0016] Many specific details are set forth in the following description in order to fully understand the present invention. However, the present invention can also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited by the specific embodiments disclosed below.

[0017] Figure 1 A flowchart of a full - process control method applied to a tea processing production line provided by an embodiment of the present invention; As Figure 1 shown, the present invention provides a flowchart of a full - process control method applied to a tea processing production line, including: S102, obtaining production full - process monitoring information and performing pre - processing, and performing multi - modal feature fusion according to the pre - processed production full - process monitoring information to obtain multi - modal production full - process monitoring fusion features; S104, establishing a production quality knowledge graph, performing production quality assessment in combination with the multi - modal production full - process monitoring fusion features, and determining whether production quality control is required according to the assessment result; S106, analyzing the production equipment of the current production line by using the multi - modal production full - process monitoring fusion features, judging the operation health status of each equipment of the current production line, and generating an operation health status index; S108, predicting the remaining life of the production equipment of the current production line by using the multi - modal production full - process monitoring fusion features, judging the degradation degree of each equipment of the current production line, and obtaining remaining life prediction information; S110, analyzing whether the current production line needs processing supervision based on the operation health status index and the remaining life prediction information. If processing supervision is required, a production line control plan is formulated to control the processing production line.

[0018] Further, in a preferred embodiment of the present invention, the obtaining production full - process monitoring information and performing pre - processing, and performing multi - modal feature fusion according to the pre - processed production full - process monitoring information to obtain multi - modal production full - process monitoring fusion features specifically includes: The whole production process is monitored through a sensor network installed on the tea processing production line to obtain the whole production process monitoring information, which includes production line equipment operation monitoring information, production line environment monitoring information, and processed product monitoring information; The obtained whole production process monitoring information is transmitted to the edge gateway, and the whole production process monitoring information is preprocessed by the edge computing node. The sliding window mechanism is used for abnormal data detection and the linear interpolation method is used for compensation. Image enhancement and filtering processing are performed on the processed product monitoring information; Feature extraction is performed on the preprocessed whole production process monitoring information to obtain single-modal feature extraction information, which includes production line equipment operation monitoring features, production line environment monitoring features, and processed product monitoring features; The principal component analysis method is introduced to perform feature dimensionality reduction on the single-modal feature extraction information. The principal component scores corresponding to the extracted single-modal features are calculated and judged against a preset threshold. The single-modal features corresponding to the principal component scores greater than the preset threshold are defined as principal component factors; Projection in the principal component direction is performed using the principal component factors to obtain a projection scatter plot. Single-modal features within a preset selection range are selected from the projection scatter plot to obtain single-modal feature dimensionality reduction information; The single-modal feature dimensionality reduction information is imported into a pre-trained Bi-LSTM network. The attention scores between different modal features are calculated through the attention mechanism to establish an association matrix between temporal features and spatial features for feature fusion, obtaining multi-modal whole production process monitoring fusion features.

[0019] It should be noted that first, various types of sensors are deployed at each key process of the production line to collect equipment operation parameters (such as the temperature of the tea-killing machine and the rotation speed of the rolling machine), environmental conditions (temperature, humidity, air flow rate), and processed product characteristics (tea color images, moisture content spectral data), forming a full-dimensional monitoring information flow covering "equipment - environment - product". After the raw data is transmitted to the edge computing node through the edge gateway, the sliding window mechanism is used to dynamically intercept and detect outliers in the time-series data. For the occasional signal loss or noise interference of the sensors, the linear interpolation method is used to smoothly compensate the missing segment; at the same time, for the tea image monitoring data, a combined algorithm of histogram equalization and Gaussian filtering is used to enhance the feature recognition ability, eliminate image blurring caused by uneven illumination or equipment vibration, and ensure the integrity and reliability of the data after preprocessing. In the feature extraction stage of the preprocessed multi-source data, vibration spectrum energy features, motor load fluctuation features, etc. are extracted from the equipment operation monitoring information, temperature and humidity gradient change features, air particulate matter concentration trend features, etc. are extracted from the environmental monitoring information, and single-modal features such as the HSV space distribution feature and texture complexity feature of tea color are extracted from the processed product monitoring information. Subsequently, the principal component analysis method is introduced to reduce the dimension of the high-dimensional features. The correlation between features is calculated through the covariance matrix, and the principal component factors that have a significant impact on the processing quality are selected (such as the principal component of the tea-killing temperature fluctuation and the principal component of the spatial distribution of tea moisture content), and a two-dimensional scatter plot is constructed based on the projection in the principal component direction. Outlier features are removed according to the preset density threshold, and the core feature set representing process stability is retained. Finally, the dimension-reduced single-modal features are input into the pre-trained Bi-LSTM network, and its bidirectional time-series modeling ability is used to capture the evolution law of equipment status and the lag effect of product quality. At the same time, the attention mechanism is used to dynamically calculate the correlation weights between different modal features. For example, the spatial coupling relationship between the temperature sensor data in the tea-killing stage and the tea color features is automatically strengthened, and the redundant association between environmental humidity and equipment vibration signals is weakened, forming multi-dimensional features that integrate time-series dependence and cross-modal interaction characteristics.

[0020] Furthermore, in a preferred embodiment of the present invention, the establishment of the production quality knowledge graph, the production quality assessment is carried out in combination with the multi-modal production full-process monitoring fusion features, and according to whether production quality control is required based on the assessment results, specifically includes: Retrieve historical processing instances of different tea categories based on historical data, extract the final processing quality ratings and production requirements of each historical processing instance, and divide each historical processing instance into qualified instances and unqualified instances; Respectively extract the historical processing process features and historical product production features corresponding to the qualified instances and unqualified instances, and divide the historical product production features of each historical processing history into processing stage subsets through the extracted processing process features, generating several historical product production feature sets of different processing stages; Based on the obtained historical product production feature sets at several different processing stages, a production quality knowledge spectrum graph is established using the graph theory method with the tea category - processing stage - processing feature - processing quality as the association path. Obtain the multi-modal production full-process monitoring fusion features, extract the tea category of the current processing batch through the multi-modal production full-process monitoring fusion features, generate a retrieval label according to the tea category of the current processing batch, and import it into the production quality knowledge graph for association path screening. Obtain several screened association paths, calculate the similarity value with the multi-modal production full-process monitoring fusion features to obtain the similarity value, and screen the association path with the highest similarity, and extract the historical processing quality corresponding to the association path. If the historical processing quality corresponding to the association path is unqualified, it means that there are production process problems in the current production stage of the current processing batch, and then obtain the production regulation strategy from the preset production strategy library for production quality control.

[0021] It should be noted that through the construction of a deep interaction mechanism between the historical process knowledge base and real-time production data, early warning and active regulation of production quality anomalies are realized. First, the historical processing data of multi-category tea is structured, and the instances are divided into two categories: qualified and unqualified according to the physical and chemical indexes of the finished products and the sensory evaluation results, and the process parameters (such as withering duration gradient, fixing temperature curve) and production features (raw material batch attributes, production equipment operation parameters) recorded during the processing are extracted. Through the time series segmentation algorithm, the full-cycle data of each instance is segmented into stages to form a set of feature segments for key processes such as withering, rolling, and fermentation. Based on the processing stage feature set, using the multi-attribute association modeling method in graph theory, with the tea category as the root node, the processing stage as the intermediate node, the process feature as the edge attribute, and the quality rating as the leaf node, a production quality knowledge graph with weight transfer characteristics is constructed. In the real-time monitoring scenario, the current processed tea category is extracted through multi-modal feature fusion, and a retrieval label containing the category identifier and process status is generated. After mapping this label to the knowledge graph, the label propagation algorithm is used to screen out the subgraph of the association path that matches the current production context. Subsequently, the dynamic time warping algorithm is used to calculate the morphological similarity between the real-time process feature curve and the historical path feature curve, and the reference path with the highest matching degree is screened out. When the matching path points to a historical unqualified instance, combined with the current process positioning (such as in the middle of fermentation) and abnormal feature performance (such as the temperature rise rate deviation), the corresponding regulation measures (such as adjusting the ventilation volume of the fermentation room, inserting an intermediate quality inspection link) are called from the strategy library, and at the same time, the process parameter compensation algorithm is triggered to prospectively correct the set value of the subsequent process, forming a closed-loop control loop of "data comparison - knowledge reasoning - strategy execution" to effectively block the spread of quality defects in the production line.

[0022] Further, in a preferred embodiment of the present invention, analyzing the production equipment of the current production line by using the multi-modal full-process monitoring fusion features of production, and judging the operation health status of each equipment on the current production line to generate an operation health status index, specifically including: Extract the historical operation parameters of each production equipment under different processing technologies from the historical production equipment operation library, classify the operation status and then form a historical equipment operation data set, where the historical equipment operation data set includes fault historical operation parameters and non-fault historical operation parameters; Construct a first anomaly detection model based on a convolutional neural network, train the first anomaly detection model through the historical equipment operation data set to construct a first training set, verify it through repeated training and parameter adjustment, and finally output a first anomaly detection model that meets the expectations; Extract non-fault historical operation parameters from the historical equipment operation data set to construct a second training set, use SVDD to construct a second anomaly detection model, and import the second training set into the second anomaly detection model for model training; Use the Gaussian kernel function to map the input training data to a high-dimensional space, solve in the mapped high-dimensional space to obtain the minimum hypersphere boundary containing all non-fault historical operation parameter samples, and then output the trained second anomaly detection model; Cascade the first anomaly detection model and the second anomaly detection model to form a production line anomaly detection model, obtain the multi-modal full-process monitoring fusion features of production and input them into the production line anomaly detection model to judge the operation health status of each equipment on the current production line; In the first anomaly detection model, extract features based on a preset four-layer convolutional layer to obtain a feature map, input the obtained feature map into the global average pooling layer, and then output the fault probability distribution of each production equipment at the current moment through the connected fully connected layer; In the second anomaly detection model, generate a real-time feature vector according to the input multi-modal full-process monitoring fusion features of production, calculate the Mahalanobis distance of the real-time feature vector in the kernel space, if it is greater than the radius of the hypersphere, it is marked as an abnormal real-time feature vector, and calculate the relative deviation degree of the abnormal real-time feature vector from the hypersphere boundary; Fuse the fault probability distribution and the relative deviation degree through weighting to generate the operation health status index of each equipment on the current production line, which is used to characterize the normal operation of the equipment for tea processing on the current production line.

[0023] It should be noted that for the full - process control of the tea processing production line, only focusing on the quality changes of tea during processing is not enough to comprehensively improve the tea processing quality. The normal and healthy operation of the production equipment on the line is also a major factor affecting tea quality. First, extract the operation parameters of each device under different processing technologies (such as green tea fixation, black tea fermentation) from the historical production database, including time - series data such as vibration spectra, temperature curves, current waveforms, etc. Classify the data into two categories: faulty and non - faulty through expert annotation and automatic clustering algorithms. Faulty data covers scenarios such as equipment degradation and sudden failures (such as a sharp increase in the high - frequency component of vibration before a bearing jams), and non - faulty data includes the states during normal working conditions and planned maintenance periods. In the data pre - processing stage, sliding window segmentation, missing value imputation, and standardization processing are adopted to construct a historical equipment operation dataset containing multi - dimensional features. Among them, the vibration signal is decomposed by wavelet packet to extract energy entropy, the temperature data generates heat - map features through spatial interpolation, and the current signal calculates electrical indicators such as harmonic distortion rate and power factor. During this process, aiming at the physical property differences of different equipment types (such as fixators, rolling machines, dryers), the sliding window adaptive segmentation technology is used to slice the continuous operation data into working conditions, ensuring that each data sample corresponds to the steady - state or transient behavior of the equipment under a specific process stage.

[0024] Based on this dataset, first, a first anomaly detection model is constructed using a convolutional neural network (CNN). A four-layer deep architecture is designed: in the first layer, a one-dimensional convolutional kernel with a width of 5 is used to extract local features from the original signal, and the ReLU activation function is used to enhance the non-linear expression ability; in the second layer, a max-pooling operation is performed to compress the feature dimension and retain the significant vibration patterns; in the third layer, dilated convolutions are used to expand the receptive field and capture the degradation trends during the long-term operation of the equipment; in the fourth layer, a channel attention mechanism is introduced to dynamically weight the importance of different sensor signals. At the end of the model, global average pooling is used to replace the traditional fully connected layer, effectively suppressing overfitting while generating the fault probability distribution vectors of each subsystem of the equipment. During the training process, a cross-entropy loss function with weight adjustment is adopted, focusing on strengthening the learning ability for low-frequency but highly harmful fault modes (such as bearing fractures), and transfer learning technology is used to transfer general equipment fault knowledge to the model fine-tuning of specific tea processing equipment. A second anomaly detection model constructed using support vector data description (SVDD) is also adopted. Non-fault samples are screened from the historical dataset to construct a second training set, and the input features are mapped to a high-dimensional space through a Gaussian kernel function to solve the boundary of the smallest hypersphere containing all normal samples. Specifically, for the equipment operation feature vector (such as the three-dimensional combination of vibration energy entropy, temperature gradient, and harmonic distortion rate), the kernel function parameter controls the distribution tightness of the samples in the high-dimensional space, and the optimization goal is to minimize the hypersphere radius and constrain most samples to be inside the sphere. After training, after the real-time feature vector is mapped by the same kernel function, the Mahalanobis distance between it and the center of the sphere is calculated. If it exceeds the radius, it is determined as an anomaly, and the degree of anomaly is quantified by the relative deviation (deviation distance / boundary radius). It not only retains the global anomaly detection ability but also takes into account the local characteristics at different operation stages.

[0025] Subsequently, in the model cascading stage, the fault probability distribution output by the CNN and the anomaly deviation calculated by the SVDD are adaptively weighted and fused. The weight coefficients are dynamically adjusted according to the device type: for devices with clear known fault modes (such as the heating tube of the tea withering machine), a higher weight is given to the CNN (such as 0.7); for complex systems (such as the temperature and humidity joint control module of the fermentation chamber), the anomaly detection ability of the SVDD is emphasized (weight 0.6). The fusion formula is: Health Index = 1 - (α × Fault Probability + β × Deviation), where α + β = 1 and α, β ∈ [0, 1]. When the health index is lower than the threshold (such as 0.5), an early warning is triggered to guide the maintenance personnel to conduct targeted inspections. For example, when the health index of a certain dryer drops to 0.55, the CNN indicates that the fault probability of the fan is 0.68, and the SVDD shows that the current harmonic deviation is 0.8. It will be prioritized to recommend checking the bearing and motor insulation status of the fan, forming a closed-loop management from anomaly detection to root cause location. This model uses a dual-channel detection mechanism, which not only uses the CNN to identify known fault modes but also uses the SVDD to capture unknown anomalies, achieving full-coverage monitoring of the equipment health status, ensuring the stability of continuous production in tea processing, and effectively preventing the occurrence of major equipment accidents.

[0026] Further, in a preferred embodiment of the present invention, the remaining life of the production equipment on the current production line is predicted by using the multi-modal production full-process monitoring fusion features, and the degradation degree of each device on the current production line is judged to obtain the remaining life prediction information, which specifically includes: Obtain the multi-modal production full-process monitoring fusion features and the historical maintenance records of each device, perform time series processing to generate the time series of the production equipment on the current production line, and input it into the pre-trained LSTM model for remaining life prediction; Capture the device degradation trajectory features through two layers of bidirectional LSTM layers connected to the input layer. In the first layer of bidirectional LSTM, use the forward unit to capture the time series evolution features from the start point to the end point of the time series, and use the backward unit to retroactively obtain the potential impact of the historical state on the current moment, and splice the outputs of the forward unit and the backward unit to output the device degradation trajectory features; Transfer the device degradation trajectory features to the second layer of bidirectional LSTM, introduce the attention mechanism, calculate the attention scores of the device degradation trajectory features at different time steps through the attention mechanism, and weight the device degradation trajectory features through the attention scores; Input the weighted device degradation trajectory features into the fully connected layer, and output the remaining life prediction values of each device on the current production line through the ReLU activation function to obtain the remaining life prediction information.

[0027] It should be noted that the multi-modal production full-process monitoring fusion features and the historical maintenance records of each device are fused. The continuous data is cut into fixed-length time series segments through the sliding window technique, and each time step contains a multi-dimensional feature vector at the current moment. Missing values are processed by combining linear interpolation and neighboring filling. At the same time, the heterogeneous data is normalized and encoded to form a standardized three-dimensional input tensor. The preprocessed time series is input into a pre-trained bidirectional LSTM network for degradation trajectory modeling. Among them, the Huber loss function is used in the training stage to balance the sensitivity of MSE to outliers and the robustness of MAE. The optimizer selects Nadam to adaptively adjust the learning rate, and the initial learning rate is set to 0.001, which decays by 20% every 50 epochs until the validation set loss converges.

[0028] In the degradation trajectory modeling of the bidirectional LSTM network, the first layer of the bidirectional LSTM contains two processing channels, forward and backward: the forward unit analyzes the data frame by frame from the start point (t = 0) to the end point (t = T) of the time series, capturing the cumulative effect of device degradation features (such as the progressive upward trend of vibration energy entropy); the backward unit processes the sequence in reverse (from t = T to t = 0), revealing the potential influence mode of historical states on the current moment (such as the lag effect of temperature anomalies three weeks ago on the current bearing wear rate). The outputs of the forward and backward units at each time step (each 128-dimensional) are concatenated to form a 256-dimensional fused feature vector, representing the degradation state of the device at the corresponding time point. The second layer of the bidirectional LSTM receives the output sequence of the first layer and introduces an attention mechanism to enhance the weights of key time steps. After the forward and backward units of this layer (each 64-dimensional) further extract high-order time series features, a 64-dimensional time series feature matrix is output. The attention module calculates the importance scores of each time step through a trainable weight matrix: first, the feature matrix is input into a fully connected layer to be mapped into a 128-dimensional query vector, and the similarity score is calculated with a learnable context vector. After Softmax normalization, the attention weight distribution (the sum is 1) is obtained. The context vector after weighted summation focuses on the feature mutations in the degradation acceleration stage and suppresses noise interference. Finally, the attention-weighted feature vector is input into a fully connected layer for remaining useful life prediction. The fully connected layer contains 128 hidden units, and a non-linear transformation is introduced through the ReLU activation function to filter out negative value interference (such as invalid negative remaining useful life predictions). The output layer uses a linear activation function to generate the regression prediction value of the remaining useful life.

[0029] Figure 2 It is a flowchart of a production equipment control method for a tea processing production line provided by an embodiment of the present invention; As Figure 2 shown, the present invention provides a flowchart of a production equipment control method for a tea processing production line, including: S202, Preset four types of processing supervision types: immediate maintenance, continuous production, delayed maintenance, and optimization control parameters, and set corresponding processing supervision judgment rules. Obtain the operation health status index and remaining life prediction information, and judge with each processing supervision judgment rule to obtain the processing supervision judgment information; S204, If the processing supervision judgment information is continuous production, then maintain the current processing line control parameters for continuous processing; if the processing supervision judgment information is the other three types of processing supervision types, then formulate a production line control plan for production line control; S206, Obtain the operation status data of each production line in the current processing workshop, introduce the sparrow optimization algorithm for optimizing processing control parameters, extract the available status of the tea processing equipment to be controlled through the processing supervision judgment information, set the optimization goal according to the available status, and combine the operation status data of each production line in the current processing workshop to perform population initialization to obtain the initial sparrow population; S208, Calculate the fitness of each sparrow individual in the initial sparrow population, sort each sparrow individual according to the calculated fitness, select the corresponding number of sparrow individuals as discoverers in combination with the preset discoverer ratio, and update the positions of the remaining sparrow individuals as followers; S210, Set a warning mechanism, randomly select a preset number of individuals in the sparrow population as warning individuals through the warning mechanism. When it is detected that the positions of multiple individuals overlap or the fitness does not change after multiple iterations, the warning mechanism is triggered, and some sparrow individuals are guided by the warning individuals to break away from the area to avoid local optimal solutions; S212, Repeat the iteration until the preset stop rule is met and output the optimal sparrow position. Generate the processing control optimization parameters according to the optimal sparrow position, map the processing control optimization parameters to the equipment control instruction set to generate a production line control plan for processing equipment scheduling and operation parameter regulation.

[0030] It should be noted that four types of processing supervision types are preset (immediate maintenance, continuous production, delayed maintenance, and optimized control parameters), and decision rules are set based on the equipment operation health index and the predicted remaining life value. The health index is generated by fusing the CNN failure probability and the SVDD abnormal deviation degree, and the remaining life is predicted by the LSTM model. The combination of the two triggers the supervision decision: if the health index is higher than the safety threshold and the remaining life is sufficient (such as the health index ≥ 0.8 and RUL > 30 days), it is determined as continuous production or optimized control parameters based on the tea production quality; if the health index is lower than the threshold but the remaining life allows short-term operation (such as the health index 0.6 - 0.8 and RUL > 7 days), delayed maintenance is triggered; when the health index drops suddenly or the RUL is approaching failure (such as the health index < 0.5 and RUL < 3 days), immediate maintenance is required. When it is determined that maintenance or parameter optimization needs to be performed, the sparrow optimization algorithm is started to generate the production line control plan. First, the real-time data of the workshop is collected (such as equipment load rate, energy consumption, and priority of batches to be processed), the available status of the equipment to be controlled is extracted, and the optimization goal is set. The definition method of the available status is as follows: if the processing supervision type is immediate maintenance or delayed maintenance, the corresponding tea processing equipment is marked as unavailable equipment; if the processing supervision type is optimized control parameters, the corresponding tea processing equipment is marked as available equipment; The optimization objective is set as a multi-objective combination: minimizing the equipment failure risk while ensuring production capacity and balancing the loads of each production line. When initializing the sparrow population, each sparrow represents a set of control parameter combinations, and the parameter value range is restricted by the physical limits of the equipment and process standards. The fitness function is designed with comprehensive multi-dimensional indicators: the single-equipment failure risk (converted from the health index and the predicted RUL value), the production line load balance degree, and the order delay penalty. After completing the population initialization, calculate the fitness of each sparrow, sort them in ascending order (the lower the fitness, the better the solution), and select the top 30% of the individuals as discoverers, who are responsible for exploring high-quality areas; the remaining 70% are followers, who move closer to the positions of the discoverers. When updating the positions, the discoverers adjust the parameter combinations according to the direction of the current optimal solution. The warning mechanism prevents premature convergence by dynamically monitoring the population diversity. After every 5 iterations, randomly select 10% of the individuals as warning monitors. If it is detected that more than 20% of the individuals gather in the same parameter area or the optimal fitness has not been updated continuously for 10 times, the warning operation is triggered. The warning monitors guide some sparrows to jump out of the current area and adopt the Levy flight strategy for long-distance jumps to break through the local optimal trap. At the same time, apply Gaussian perturbation to the stagnant individuals, add random noise to their current parameters, and activate the search activity. The iterative optimization continues until the stop rule is met (the maximum number of iterations is 200 or the fitness improvement rate < 0.1%), and output the global optimal sparrow position (i.e., the best parameter combination). Map this position to the equipment control instruction set, convert it into instructions executable by the PLC, and send them to the corresponding equipment controller. At the same time, generate scheduling instructions, such as enabling a standby dryer to share the load, or adjusting the batch priority to match the production capacity rhythm after parameter changes, to drive the production line to perform adaptive regulation and ensure the dynamic balance of process compliance and equipment safety.

[0031] Figure 3 A full-process control system 3 for a tea processing production line provided by an embodiment of the present invention includes: a memory 31 and a processor 32. The memory 31 contains a full-process control method program for a tea processing production line. When the full-process control method program for a tea processing production line is executed by the processor 32, the following steps are implemented: Obtain the production full-process monitoring information and perform preprocessing, and perform multi-modal feature fusion based on the preprocessed production full-process monitoring information to obtain multi-modal production full-process monitoring fusion features; Establish a production quality knowledge graph, combine the multi-modal production full-process monitoring fusion features to perform production quality assessment, and determine whether production quality control is required according to the assessment results; Use the multi-modal production full-process monitoring fusion features to analyze the production equipment of the current production line, judge the operating health status of each device on the current production line, and generate an operating health status index; Predict the remaining useful life of the production equipment on the current production line by using the multi-modal full-process monitoring fusion features of production, judge the degradation degree of each equipment on the current production line, and obtain the remaining useful life prediction information; Analyze whether the current production line needs processing supervision based on the operating health status index and the remaining useful life prediction information. If processing supervision is required, formulate a production line control plan to control the processing production line.

[0032] In several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are only illustrative. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined, or can be integrated into another system, or some features can be ignored, or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed with each other can be through some interfaces. The indirect coupling or communication connection of devices or units can be electrical, mechanical, or other forms.

[0033] The units described above as separate components may or may not be physically separated. The components shown as units may or may not be physical units; they can be located in one place or distributed to multiple network units; some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0034] In addition, each functional unit in the embodiments of the present invention can be all integrated in a processing unit, or each unit can be separately used as a unit, or two or more units can be integrated in a unit; the above integrated unit can be implemented in the form of hardware, or in the form of a combination of hardware and software functional units.

[0035] Those of ordinary skill in the art can understand that all or part of the steps of implementing the above method embodiments can be completed by hardware related to program instructions. The foregoing program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps including the above method embodiments; and the foregoing storage medium includes: various media that can store program codes such as removable storage devices, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disks, or optical discs.

[0036] Alternatively, if the above-integrated units of the present invention are implemented in the form of software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiments of the present invention, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as removable storage devices, ROM, RAM, magnetic disks, or optical discs.

[0037] As described above, the above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or substitutions, which should all be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. A full-process control method applied to a tea processing production line, characterized in that, Including: Obtain the production full-process monitoring information and perform preprocessing, and perform multimodal feature fusion based on the preprocessed production full-process monitoring information to obtain multimodal production full-process monitoring fusion features; Build a production quality knowledge graph, combine the multimodal production full-process monitoring fusion features to perform production quality assessment, and determine whether production quality control is required according to the assessment results; Use the multimodal production full-process monitoring fusion features to analyze the production equipment of the current production line, judge the operating health status of each equipment on the current production line, and generate an operating health status index; Use the multimodal production full-process monitoring fusion features to predict the remaining life of the production equipment of the current production line, judge the degradation degree of each equipment on the current production line, and obtain remaining life prediction information; Analyze whether the current production line needs processing supervision based on the operating health status index and the remaining life prediction information. If processing supervision is required, formulate a production line control plan to control the processing production line.

2. The full-process control method applied to a tea processing production line according to claim 1, characterized in that The obtaining of the production full-process monitoring information and performing preprocessing, and performing multimodal feature fusion based on the preprocessed production full-process monitoring information to obtain multimodal production full-process monitoring fusion features specifically includes: Obtain the production full-process monitoring information through the sensor network installed on the tea processing production line. The production full-process monitoring information includes production line equipment operation monitoring information, production line environment monitoring information, and processed product monitoring information; Transmit the obtained production full-process monitoring information to the edge gateway, and perform preprocessing on the production full-process monitoring information through the edge computing node. Use the sliding window mechanism to detect abnormal data and use the linear interpolation method for compensation, and perform image enhancement and filtering processing on the processed product monitoring information; Extract features from the preprocessed production full-process monitoring information to obtain unimodal feature extraction information. The unimodal feature extraction information includes production line equipment operation monitoring features, production line environment monitoring features, and processed product monitoring features; Introduce the principal component analysis method to perform feature dimensionality reduction on the unimodal feature extraction information, calculate the principal component scores corresponding to the extracted unimodal features and judge them with a preset threshold, and define the unimodal features corresponding to the principal component scores greater than the preset threshold as principal component factors; Use the principal component factors to perform principal component direction projection to obtain a projection scatter plot, and select the unimodal features within a preset selection range in the projection scatter plot to obtain unimodal feature dimensionality reduction information; Import the unimodal feature dimensionality reduction information into a pre-trained Bi-LSTM network, calculate the attention scores between different modal features through the attention mechanism, establish a correlation matrix between temporal features and spatial features for feature fusion, and obtain multimodal production full-process monitoring fusion features.

3. A full-process control method applied to a tea processing production line according to claim 1, characterized in that The building of the production quality knowledge graph, combining the multimodal production full-process monitoring fusion features to perform production quality assessment, and determining whether production quality control is required according to the assessment results specifically includes: Retrieve historical processing instances of different tea categories based on historical data retrieval, extract the final processing quality ratings and production requirements of each historical processing instance, and divide each historical processing instance into qualified instances and unqualified instances; Extract the historical processing technology characteristics and historical product production characteristics corresponding to qualified instances and unqualified instances respectively. Divide the historical product production characteristics of each historical processing history through the extracted processing technology characteristics to generate historical product production characteristic sets in several different processing stages; Based on the obtained historical product production characteristic sets in several different processing stages, establish a production quality knowledge spectrum diagram using the graph theory method with the tea category - processing stage - processing characteristics - processing quality as the association path; Obtain the multi-modal production full-process monitoring fusion characteristics, extract the tea category of the current processing batch through the multi-modal production full-process monitoring fusion characteristics, generate a retrieval label according to the tea category of the current processing batch, and import it into the production quality knowledge graph for association path screening; Obtain several screened association paths, calculate the similarity value with the multi-modal production full-process monitoring fusion characteristics to obtain the similarity value, screen the association path with the highest similarity, and extract the historical processing quality corresponding to the corresponding association path; If the historical processing quality corresponding to the corresponding association path is unqualified, it means that there are production process problems in the current generation stage of the current processing batch, and then obtain the production regulation strategy from the preset production strategy library for production quality control.

4. A full-process control method applied to a tea processing production line according to claim 1, characterized in that, Analyze the production equipment of the current production line using the multi-modal production full-process monitoring fusion characteristics, judge the operation health status of each equipment on the current production line, and generate an operation health status index, specifically including: Extract the historical operation parameters of each production equipment under different processing technologies from the historical production equipment operation library, classify the operation status and then form a historical equipment operation data set. The historical equipment operation data set includes fault historical operation parameters and non-fault historical operation parameters; Construct a first anomaly detection model based on a convolutional neural network, train the first anomaly detection model through the historical equipment operation data set to construct a first training set, verify it through repeated training and parameter adjustment, and finally output a first anomaly detection model that meets the expectations; Extract non-fault historical operation parameters from the historical equipment operation data set to construct a second training set, use SVDD to construct a second anomaly detection model, and import the second training set into the second anomaly detection model for model training; Use the Gaussian kernel function to map the input training data to a high-dimensional space, solve in the mapped high-dimensional space to obtain the minimum hypersphere boundary containing all non-fault historical operation parameter samples, and then output the trained second anomaly detection model; Cascade the first anomaly detection model and the second anomaly detection model to form a production line anomaly detection model, obtain the multi-modal production full-process monitoring fusion characteristics and input them into the production line anomaly detection model to judge the operation health status of each equipment on the current production line; In the first anomaly detection model, feature extraction is performed based on a preset four-layer convolutional layer to obtain a feature map. The obtained feature map is input into a global average pooling layer and then output through a connected fully connected layer as the failure probability distribution of each production device at the current moment; In the second anomaly detection model, a real-time feature vector is generated according to the input multi-modal production full-process monitoring fusion features. The Mahalanobis distance of the real-time feature vector in the kernel space is calculated. If it is greater than the radius of the hypersphere, it is labeled as an abnormal real-time feature vector, and the relative deviation degree between the abnormal real-time feature vector and the hypersphere boundary is calculated; The failure probability distribution and the relative deviation degree are weighted and fused to generate the operation health status index of each device on the current production line, which is used to characterize the normal operation of the devices for tea processing on the current production line.

5. The full-process control method applied to a tea processing production line according to claim 1, wherein, Using the multi-modal production full-process monitoring fusion features to predict the remaining life of the production devices on the current production line, judge the degradation degree of each device on the current production line, and obtain the remaining life prediction information, specifically including: Obtain the multi-modal production full-process monitoring fusion features and the historical maintenance records of each device, perform time series processing to generate the time series of the production devices on the current production line, and input it into a pre-trained LSTM model for remaining life prediction; Capture the device degradation trajectory features through two layers of bidirectional LSTM layers connected to the input layer. In the first layer of bidirectional LSTM, use the forward unit to capture the time series evolution features from the start point to the end point of the time series, and use the backward unit to retroactively obtain the potential impact of the historical state on the current moment. Concatenate the outputs of the forward unit and the backward unit to output the device degradation trajectory features; Transfer the device degradation trajectory features to the second layer of bidirectional LSTM, introduce the attention mechanism, calculate the attention scores of the device degradation trajectory features at different time steps through the attention mechanism, and weight the device degradation trajectory features through the attention scores; Input the weighted device degradation trajectory features into the fully connected layer, and output the remaining life prediction values of each device on the current production line through the ReLU activation function to obtain the remaining life prediction information.

6. A full-process control method applied to a tea processing production line according to claim 1, characterized in that, Based on the operation health status index and the remaining life prediction information, analyze whether the current production line needs processing supervision. If processing supervision is required, formulate a production line control plan to control the processing production line, specifically including: Preset four types of processing supervision types: immediate maintenance, continue production, delayed maintenance, and optimize control parameters, and set the corresponding processing supervision judgment rules. Obtain the operation health status index and the remaining life prediction information, and judge with each processing supervision judgment rule to obtain the processing supervision judgment information; If the processing supervision judgment information is to continue production, maintain the control parameters of the current processing production line for continuous processing; if the processing supervision judgment information is the other three types of processing supervision types, formulate a production line control plan to control the processing production line; Obtain the operation status data of each production line in the current processing workshop, introduce the sparrow optimization algorithm to optimize the processing control parameters, extract the available status of the tea processing equipment to be controlled through the processing supervision judgment information, set the optimization goal according to the available status, and perform population initialization in combination with the operation status data of each production line in the current processing workshop to obtain the initial sparrow population; Calculate the fitness of each sparrow individual in the initial sparrow population, sort each sparrow individual according to the calculated fitness, select the corresponding number of sparrow individuals as discoverers in combination with the preset discoverer ratio, and update the positions of the remaining sparrow individuals as followers; Set a warning mechanism, randomly select a preset number of individuals in the sparrow population as warning individuals through the warning mechanism. When it is detected that the positions of multiple individuals overlap or the fitness does not change after multiple iterations, the warning mechanism is triggered, and some sparrow individuals are guided by the warning individuals to break away from the area to avoid local optimal solutions; Repeat the iteration until the optimal sparrow position is output according to the preset stop rule, generate the processing control optimization parameters according to the optimal sparrow position, map the processing control optimization parameters to the equipment control instruction set to generate a production line control plan for processing equipment scheduling and operation parameter regulation.

7. A full-process control system applied to a tea processing production line, characterized in that, The system includes: a memory and a processor. The memory contains a full-process control method program for the tea processing production line. When the full-process control method program for the tea processing production line is executed by the processor, the following steps are implemented: Obtain the production full-process monitoring information and perform preprocessing, and perform multi-modal feature fusion according to the preprocessed production full-process monitoring information to obtain the multi-modal production full-process monitoring fusion feature; Establish a production quality knowledge graph, combine the multi-modal production full-process monitoring fusion feature to conduct production quality assessment, and determine whether production quality control is required according to the assessment result; Analyze the production equipment of the current production line by using the multi-modal production full-process monitoring fusion feature, judge the operation health status of each equipment on the current production line, and generate an operation health status index; Predict the remaining life of the production equipment of the current production line by using the multi-modal production full-process monitoring fusion feature, judge the degradation degree of each equipment on the current production line, and obtain the remaining life prediction information; Analyze whether the current production line needs processing supervision based on the operation health status index and the remaining life prediction information. If processing supervision is required, formulate a production line control plan to control the processing production line.

Images