A control system for adding production auxiliary materials for throat lozenges

Abstract

The present invention discloses a control system for adding auxiliary materials in the production of throat lozenges, belonging to the technical field of data modeling and analysis. It includes a threshold matrix establishment module for constructing a dynamic threshold matrix, a data acquisition module for collecting current production data, a data analysis module for generating a mode selection signal according to the comparison result between the current production data and the dynamic threshold matrix, a mode selection module for activating a corresponding control mode based on the mode selection signal, where the control modes include a normal mode, an adjustment mode, and an exception handling mode, and a control execution module for executing the control instructions corresponding to the control mode and sending the verified control instructions to the auxiliary material addition execution mechanism. Through iterative optimization at the data modeling and algorithm levels, the present invention realizes the intelligent association of multi-source heterogeneous data and the optimization of control strategies.

Description

Technical Field

[0001] The present invention relates to the technical field of data modeling and analysis, and in particular to a control system for adding auxiliary materials in the production of throat lozenges. Background Art

[0002] In the field of intelligent control in the existing food industry production, computer-based data processing methods are widely used in the optimization of process flows. However, current control strategies have significant technical bottlenecks when facing scenarios such as the addition of auxiliary materials for special foods like throat lozenges.

[0003] Currently, traditional methods use linear weighting or static association models to process multi-source heterogeneous data, such as sensor time-series data and process maps, resulting in insufficient cross-modal feature fusion; dynamic threshold setting relies on fixed rules, such as Z-Score statistics, and it is difficult to adapt to the time-varying characteristics of the production line, leading to mode switching delays and parameter oscillations; the simulation verification system ignores equipment dynamic errors, such as servo inertia delay, resulting in deviations between virtual simulation and actual execution; the time-series prediction model has insufficient robustness to non-linear interference, and the fixed time window strategy mismatches with the production rhythm, causing compensation drift; the knowledge transfer across control modes is missing, and parameter inheritance depends on manual configuration, reducing the efficiency of fault tracing. Summary of the Invention

[0004] To solve the above problems, the present invention provides a control system for adding auxiliary materials in the production of throat lozenges, which realizes the intelligent association of multi-source heterogeneous data and the optimization of control strategies through iterative optimization at the data modeling and algorithm levels.

[0005] The above object can be achieved through the following solutions:

[0006] A control system for adding auxiliary materials in the production of throat lozenges includes a threshold matrix establishment module for constructing a dynamic threshold matrix, where the dynamic threshold matrix includes the interval ranges of viscosity reference values, sugar content floating ranges, and temperature and humidity critical points; a data acquisition module for collecting current production data, where the production data includes the measured values of viscosity sensors in the raw material conveying section, temperature and humidity monitoring data of the auxiliary material warehouse, and online detection indicators of finished products; a data analysis module for generating a mode selection signal according to the comparison result between the current production data and the dynamic threshold matrix; a mode selection module for activating the corresponding control mode based on the mode selection signal, where the control mode includes a normal mode, an adjustment mode, and an exception handling mode; and a control execution module for executing the control instructions corresponding to the control mode and sending the verified control instructions to the auxiliary material addition execution mechanism.

[0007] Optionally, the data analysis module includes: a deviation analysis unit for calculating the deviation of each index in the current production data relative to the dynamic threshold matrix; an anomaly judgment unit for triggering the anomaly handling mode when it is detected that the deviation of at least one index exceeds a preset warning value; an adjustment judgment unit for triggering the adjustment mode when the deviation of all indexes is within a preset adjustment range; and a normal judgment unit for maintaining the normal mode when the deviation of all indexes does not exceed the preset allowable fluctuation range.

[0008] Optionally, the adjustment mode includes: retrieving a set of historical parameters matching the current deviation in the historical production batches; calculating a dynamic compensation coefficient for the auxiliary material addition rate based on the set of historical parameters; and superimposing the dynamic compensation coefficient on the current recipe parameters to generate corrected control parameters.

[0009] Optionally, calculating the dynamic compensation coefficient for the auxiliary material addition rate includes: establishing a sliding time window with a length of 10% of the duration of the current production batch; extracting the historical parameter change rate within the sliding time window and calculating the predicted compensation amount; and performing weighted fusion of the predicted compensation amount and the normal compensation amount to output the final dynamic compensation coefficient.

[0010] Optionally, the anomaly handling mode includes: extracting the control parameters of three preset processes in the current production batch; performing multi-dimensional feature matching of the control parameters with a historical fault case library; and generating a processing instruction including the equipment calibration priority and the recipe reset scheme according to the matching result.

[0011] Optionally, performing multi-dimensional feature matching of the control parameters with the historical fault case library includes: constructing a three-dimensional weight matrix of the current anomaly feature, where the three-dimensional weight matrix includes the parameter deviation degree, the equipment aging coefficient, and the environmental interference factor; calculating the feature similarity between the three-dimensional weight matrix and the historical fault cases; and selecting the processing strategy corresponding to the historical case with the highest feature similarity as the basic template.

[0012] Optionally, sending the verified control instruction to the auxiliary material addition actuator includes: creating a digital twin model of the auxiliary material addition process; loading the control instruction and real-time environmental parameters into the digital twin model; obtaining the quality prediction index of the virtual finished product after simulation operation; and confirming the effectiveness of the control instruction when the deviation between the quality prediction index and the preset standard is less than the allowable threshold.

[0013] Optionally, the simulation operation includes: aligning the time series tags of the actual production line and the digital twin model; adding a random disturbance factor to simulate the equipment execution error; and recording the data fluctuation range during the simulation process and generating a safety verification report.

[0014] Optionally, the calculation of the predicted compensation amount includes: obtaining the changing trend of ambient temperature and humidity in the next 15 minutes; predicting the change of auxiliary material demand using a long short-term memory neural network; and correcting the length and weight distribution of the sliding time window based on the prediction results.

[0015] Based on the same inventive concept, the present invention also provides a method for controlling the addition of auxiliary materials in the production of throat lozenges, the method comprising: constructing a dynamic threshold matrix, the dynamic threshold matrix comprising a viscosity reference value, a sugar content floating range, and an interval range of temperature and humidity critical points; collecting current production data, the production data comprising viscosity sensor measurement values ​​of the raw material conveying section, auxiliary material bin temperature and humidity monitoring data, and finished product online detection indicators; generating a mode selection signal based on a comparison result between the current production data and the dynamic threshold matrix; activating a corresponding control mode based on the mode selection signal, the control mode comprising a normal mode, an adjustment mode, and an exception handling mode; executing a control instruction corresponding to the control mode, and sending the verified control instruction to an auxiliary material addition actuator.

[0016] Compared with the prior art, the present invention has the following advantages:

[0017] 1. By constructing a dynamic threshold matrix of viscosity, sugar content, temperature and humidity parameters, combined with multi-level deviation analysis, intelligent decision-making from normal mode to abnormal mode is achieved; its advantage lies in overcoming the rigidity of the traditional fixed threshold system, and through dynamic adaptation of process parameters, the compensation strategy is quickly activated when the ambient temperature and humidity suddenly change or the raw material batch fluctuates, thus reducing the deviation of auxiliary material ratio caused by response delay;

[0018] 2. Hypergraph neural network and improved tensor decomposition technology are used to perform multimodal fusion of sensor time series data, process maps and environmental parameters to analyze the implicit correlation rules of cross-domain data. Compared with the traditional static weighted model, this technology expands the feature correlation dimension to three times, and the computational complexity maintains a linear growth. Assisted by the LSTM-enhanced sliding window prediction mechanism, it can predict the trend of auxiliary material demand changes 15 minutes in advance, so that the calculation accuracy of the compensation coefficient is improved, which is especially suitable for the seasonal parameter drift scenario of the production process.

[0019] 3. Based on the incremental digital twin, a bidirectional channel of virtual and real data is constructed to simulate dynamic errors such as equipment inertia delay and mechanical wear before command execution, and generate multi-condition test plans through Monte Carlo perturbation injection; this technology compresses the execution deviation rate between the virtual simulation verification results and the actual production line, avoiding resource waste under the traditional trial and error mechanism; the system uses the timing alignment algorithm and the three-dimensional visualization feedback mechanism to accurately locate potential process defects, greatly shortening the abnormal response time;

[0020] 4. In the abnormal handling mode, a three-dimensional weight matrix is adopted for multi-dimensional feature matching, combined with a case library retrieval technology driven by reinforcement learning, to generate a device calibration priority list and a recipe reset plan; the fault diagnosis efficiency is improved, and the occurrence rate of secondary faults is reduced through a dynamic weight allocation strategy, especially showing significant advantages in complex abnormal scenarios such as mixed temperature runaway and metering pump blockage;

[0021] 5. Design a parameter inheritance engine driven by deep reinforcement learning, establish a weight migration channel between modes through the Q-learning strategy, and ensure that the dynamic compensation coefficient of the adjustment mode can be seamlessly inherited to the abnormal backtracking model; cooperate with a progressive switching strategy and a double-buffer mechanism to eliminate parameter jumps during traditional system switching, reduce the mechanical impact during the transition of control instructions, and extend the service life of the actuator.

[0022] Other features and advantages of the present invention will be described in the following specification, and, in part, will be obvious from the specification, or will be understood by implementing the present invention. The objectives and other advantages of the present invention can be realized and obtained by the structures pointed out in the specification, claims, and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0024] Figure 1 It is a framework diagram of a supplementary material adding control system for throat lozenge production according to an embodiment of the present invention.

[0025] Figure 2 It is a schematic structural diagram of a supplementary material adding control system for throat lozenge production according to an embodiment of the present invention.

[0026] Figure 3 It is a schematic flow diagram of a supplementary material adding control method for throat lozenge production according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0027] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present invention.

[0028] Reference Figure 1

[0029]

[0030]

[0031] A threshold matrix establishment module, configured to construct a dynamic threshold matrix, where the dynamic threshold matrix includes a viscosity reference value, a sugar content floating range, and an interval range of temperature and humidity critical points;

[0031] Specifically, the dynamic threshold matrix is a multi-dimensional parameter set covering the entire production process, including the definition of the viscosity reference value, the sugar content floating range, and the interval of temperature and humidity critical points. The viscosity reference value is the standard reference value of viscosity during the raw material dissolution stage in the production process, and is dynamically adjusted according to the characteristics of raw material batches. The sugar content floating range is the maximum allowable fluctuation range of the sugar content of the finished product, and is related to the percentage deviation from the target sugar degree. The temperature and humidity critical point is the safety and stability boundary value of the auxiliary material storage environment, controlling the maintenance conditions of the active substances of the auxiliary materials.

[0032] Among them, the viscosity reference value is calculated by the median of the viscosity sensor within a sliding time window, the window length L = 60 minutes, and temperature and humidity compensation is considered. For the viscosity reference value , there is

[0033] ,

[0034] In the formula, is the median of the viscosity values in the past 60 minutes, is the temperature and humidity compensation coefficient, is the difference between the current temperature and the reference temperature. The temperature and humidity compensation coefficient has a value range of 0.005 ≤ ≤ 0.02, and is calibrated by the orthogonal experiment method.

[0035] For the floating range , there is

[0036] ,

[0037] In the formula, is the mean value within the statistical period, is the standard deviation within the statistical period, is dynamically adjusted according to the production stage.

[0038] A data acquisition module, configured to acquire current production data, where the production data includes the measurement value of the viscosity sensor in the raw material conveying section, the temperature and humidity monitoring data of the auxiliary material warehouse, and the online detection index of the finished product;

[0039] Specifically, the measured value of the viscosity sensor is a physical index for real-time monitoring of the flow resistance of the mixed liquid in the raw material conveying pipeline, reflecting the uniformity and melting state of the raw materials. The temperature and humidity monitoring data of the auxiliary material warehouse are the temperature and humidity parameters of the auxiliary material storage environment collected by distributed sensors, which are used to evaluate the preservation activity of the auxiliary materials. The online inspection indexes of the finished products are the physical and chemical parameters of the finished throat lozenges obtained in real time by a spectrometer or a near-infrared detector, including the sugar content uniformity and the hardness value.

[0040] Among them, the viscosity sensor is of Brookfield DV2T type with a measurement accuracy of ±1%FS; the temperature and humidity sensor is Sensirion SHT35 with a temperature accuracy of ±0.3°C and a humidity accuracy of ±2%RH; the metering pump control resolution reaches 0.01 mL / pulse. The metering pump communicates with the control system through the Modbus RTU protocol, and the instruction format is: {starting address 0x01; function code 0x06; register address 0x2000; flow rate setting value (unit 0.01 mL / s)}, and the CRC check code is generated by the Modbus standard algorithm.

[0041] The data analysis module is used to generate a mode selection signal according to the comparison result between the current production data and the dynamic threshold matrix;

[0042] Specifically, the comparison result is a comprehensive evaluation result formed by the data analysis module through data fusion technology, which performs multi-dimensional matching of the current data with the threshold. The mode selection signal is a set of instructions including mode priorities and response strategies, which is determined by the weight distribution of the deviation degree.

[0043] The mode selection module is used to activate the corresponding control mode based on the mode selection signal, and the control mode includes a normal mode, an adjustment mode, and an exception handling mode;

[0044] Specifically, the normal mode is a standard operation mode when the production parameters are within the threshold range, and directly calls the pre-stored standardized formula parameters. The adjustment mode is a dynamic compensation mode when it is detected that the parameters exceed the reference threshold but do not reach the warning value, triggering historical parameter backtracking and prediction compensation. The exception handling mode is an emergency response mode when it is detected that the key parameters reach the warning threshold, starting fault cause analysis and disposal plan generation.

[0045] The control execution module is used to execute the control instructions corresponding to the control mode and send the verified control instructions to the auxiliary material addition execution mechanism.

[0046] Specifically, the verified control instructions are the double-verification results of digital twin rehearsal and safety verification, ensuring the reliability and executability of the instructions. The auxiliary material addition execution mechanism is a control terminal including a precision metering pump and a dynamic mixing device, which adjusts the auxiliary material ratio and injection rate according to the instructions.

[0047] The present invention constructs a multi-dimensional perception and control model based on a dynamic threshold system, and realizes the active prediction and rapid response of production anomalies through the hierarchical screening and pattern matching of real-time data streams. The system first establishes a dynamically adjusted threshold matrix covering the core parameters of raw material characteristics, environmental conditions, and product quality control, and then triggers the intelligent switching of three modes: normal, adjustment, and anomaly handling through high-frequency data collection and comparison. Closed-loop control is formed among the modes through historical data backtracking, cross-batch parameter sharing, and virtual simulation verification to ensure the accuracy of auxiliary material addition and the stability of the process.

[0048] Optionally, as Figure 2 shown, the data analysis module includes:

[0049] A deviation analysis unit for calculating the deviation of each index in the current production data relative to the dynamic threshold matrix;

[0050] Specifically, the deviation analysis unit is a processing module that calculates the relative deviation amount between each index of the current production data and the corresponding reference value in the dynamic threshold matrix. Its output value quantifying the degree of index fluctuation is the basis for subsequent mode judgment. The deviation uses a percentage difference algorithm to eliminate the influence of inconsistent parameter dimensions.

[0051] An anomaly judgment unit for triggering the anomaly handling mode when it detects that the deviation of at least one index exceeds a preset warning value;

[0052] Specifically, the anomaly judgment unit is a logic controller that marks the anomaly level when the deviation of at least one index exceeds the preset warning value. The warning value is a mandatory intervention boundary comprehensively delimited according to the extreme performance of the equipment and the finished product quality standard. After being triggered, the current production state is immediately locked and the safety protocol is started.

[0053] An adjustment judgment unit for triggering the adjustment mode when the deviation of all indexes is within a preset adjustment range;

[0054] Specifically, the adjustment judgment unit is a parameter correction module activated when the deviation of all indexes is detected to be within the preset adjustment range. The adjustment range is a controllable fluctuation range obtained from the experience of process experts and statistical analysis of historical data, allowing the system to autonomously perform compensation operations without interrupting production.

[0055] A normal judgment unit for maintaining the normal mode when the deviation of all indexes does not exceed the preset allowable fluctuation range.

[0056] Specifically, the conventional judgment unit is a steady-state controller that maintains the basic operation mode. When it detects that all indicators do not exceed the allowable fluctuation range, it continuously calls the standard recipe parameters, and the allowable fluctuation range is the safety margin range of the equipment calibration accuracy.

[0057] Exemplarily, a multi-level deviation judgment system is constructed to achieve precise dynamic mode scheduling. The strong intervention feature of the abnormal judgment unit is used to quickly isolate major risks. The predictive compensation of the adjustment judgment unit is relied on to maintain the steady-state transition of the process. The baseline maintenance ability of the conventional judgment unit is supplemented to optimize resource consumption. Its beneficial effect lies in establishing an intelligent decision-making chain with hierarchical response, realizing progressive handling from minor parameter fluctuations to serious equipment abnormalities, improving the quality risk interception efficiency while ensuring production continuity, and avoiding problems of excessive intervention or response lag in traditional control strategies through the priority scheduling mechanism between modes, forming a closed-loop optimization system that takes into account both production efficiency and safety control.

[0058] Optionally, the adjustment mode includes:

[0059] Retrieve the historical parameter set that matches the current deviation in the historical production batches;

[0060] Specifically, the historical parameter set is a group of past production record data stored in the system database. Its sampling standard is the working condition characteristics that match the current deviation threshold interval. The dynamic time warping algorithm is used in the matching process to align the time series differences between batches. The selected parameter set needs to meet the similarity requirements of the three-axis fluctuation trends of temperature, humidity, and viscosity, and the Pearson correlation coefficient is used to verify the correlation strength between the historical data and the current working conditions.

[0061] Based on the historical parameter set, calculate the dynamic compensation coefficient of the auxiliary material addition rate;

[0062] Specifically, the dynamic compensation coefficient is a dimensionless index that characterizes the adjustment amplitude of the auxiliary material addition rate. In its calculation process, the sliding window average method is used to process the additive delivery records in the historical parameter set, and the compensation weight value is calculated in combination with the real-time throughput of the current production line. Finally, the time response curve of the compensation coefficient is optimized through the fuzzy logic inference engine.

[0063] Superimpose the dynamic compensation coefficient on the current recipe parameters to generate the corrected control parameters.

[0064] Specifically, the corrected control parameters are a new instruction set that has been iteratively optimized. In its generation process, the benchmark framework of the original recipe parameters is retained and the compensation coefficient is superimposed to form a composite control instruction. The superimposition algorithm uses the half-life decay model to ensure the smooth transition of the old and new parameters. The instruction queue uses the double-buffer mechanism to avoid mechanical shocks to the production line caused by parameter jumps.

[0065] For example, the intelligent matching and compensation calculation of historical working conditions can be used to predict production fluctuations, and refined dynamic adjustments can be made while maintaining the original formula framework, which can effectively balance the relationship between process stability and production flexibility. Its beneficial effect is to reduce the trial and error cost of the parameter setting process through historical data reuse, capture the short-term fluctuation characteristics of process parameters using sliding window technology, and improve the applicability of compensation coefficients by combining the reasoning ability of fuzzy logic engines. Finally, the control instructions can be switched without disturbance through a double buffer mechanism, which can improve the consistency of finished product quality while extending the service life of the equipment, forming an intelligent control paradigm that is driven by historical experience and coordinated with real-time data feedback.

[0066] Optionally, the dynamic compensation coefficient for calculating the auxiliary material addition rate includes:

[0067] Establish a sliding time window, the length of which is 10% of the duration of the current production batch;

[0068] Specifically, the sliding time window is a data collection period whose length is automatically adjusted according to the current production progress. The window length is set to 10% of the length of the current production batch to ensure that the sampling cycle is updated synchronously with the process rhythm. The sliding step size of the window is configured as the minimum time unit of the production rhythm to achieve continuous coverage of data fragments. The data in the window uses timestamp alignment technology to eliminate the time accumulation error of the production equipment.

[0069] Extracting the historical parameter change rate within the sliding time window and calculating the predicted compensation amount;

[0070] Specifically, the historical parameter change rate is the fluctuation gradient indicator of the auxiliary material addition rate within the statistical window. The calculation process uses the least squares method to fit the parameter change trend line and quantify the change rate through differential operation. The predicted compensation amount generation method introduces an adaptive Kalman filter algorithm to reduce the noise of the original fluctuation data, and finally the theoretical compensation demand within the next three production cycles is deduced through the gray prediction model.

[0071] Among them, historical parameter matching uses the dynamic time warping algorithm (DTW) to calculate the similarity between historical batches and current data sequences. ,have

[0072] ,

[0073] In the formula, The first data points, is the first A number of data points. Historical parameters with a similarity greater than 0.8 are included in the candidate set. Linear interpolation is performed on the historical batch data sequence, and the unified sampling frequency is 1 Hz; when matching, the maximum allowable time stretching ratio is ±20%, and sequences outside this range are considered irrelevant sequences.

[0074] The predicted compensation amount and the conventional compensation amount are weighted and fused to output the final dynamic compensation coefficient.

[0075] Specifically, the conventional compensation amount is the basic adjustment amount reference value generated according to the equipment calibration manual, and its weight distribution coefficient is dynamically adjusted through the real-time monitoring values of the temperature and humidity in the production environment. The weighted fusion algorithm uses an improved entropy weight method to calculate the contribution degree of each compensation component, and at the same time introduces the process optimization rules in the expert experience library to constrain the reasonable range of the fusion result, and finally generates a dynamic compensation coefficient that takes into account both prediction accuracy and operation stability.

[0076] Among them, the weighting logic is the fusion weight for the conventional compensation amount and the predicted compensation amount , there is

[0077] ,

[0078] In the formula, is the environmental humidity change rate, which represents the change rate of the current environmental humidity and is used to dynamically adjust the compensation strategy;

[0079] For the final dynamic compensation coefficient , there is

[0080] ,

[0081] In the formula, is the predicted compensation amount, is the conventional compensation amount.

[0082] Exemplarily, through the sliding window algorithm, the time characteristics of historical data are innovatively incorporated into the compensation calculation process, the trend changes of process parameters are captured in the dynamic time dimension, and the accurate prediction of the compensation amount is achieved by combining an advanced prediction model and a robust weighting strategy. Its beneficial effect is to break through the static limitations of traditional compensation methods, establish a synergistic mechanism of time-varying parameter analysis and adaptive weight adjustment, effectively reduce the adjustment lag effect caused by equipment response delay, improve the predictability and compliance of the auxiliary material addition rate control, ensure the robust adaptability of product quality to production environment fluctuations, and form an intelligent compensation technology system with both timeliness and reliability.

[0083] Optionally, the abnormal handling mode includes:

[0084] Extract the control parameters of three preset processes in the current production batch;

[0085] Specifically, the three preset processes are the control variable sets of the mixing temperature control process in the raw material pretreatment stage, the auxiliary material quantitative addition process, and the semi-finished product forming process. The selection basis is the key quality control nodes determined by process stability analysis. The parameter extraction range covers the real-time recorded values of equipment operation parameters and the process state marking values, and the complete data sequence within the last two production cycles is obtained through the backtracking interface of the time series database.

[0086] Perform multi-dimensional feature matching between the control parameters and the historical failure case database;

[0087] Specifically, the historical failure case database is a process exception event database stored in a structured manner. Each case record contains failure characterization parameters, failure cause codes, and disposal plan indexes. The multi-dimensional feature matching uses an improved cosine similarity algorithm to compare the dimension space distance between the current parameter feature vector and the feature matrix of the case database, and optimizes the efficiency and accuracy of similarity calculation through the principal component analysis method.

[0088] Generate a processing instruction containing the equipment calibration priority and the formula reset plan according to the matching result.

[0089] Specifically, the equipment calibration priority is a sorted list generated according to the failure impact degree and repair urgency. The sorting weight is dynamically determined by the product of the equipment downtime cost coefficient and the failure propagation risk value. The formula reset plan is an optimized ratio parameter set output by an adaptive strategy generation module based on the failure type. Its generation process integrates expert experience rules and optimization suggestions output by a reinforcement learning model.

[0090] Exemplarily, this solution realizes the efficient traceability and intelligent disposal decision-making of production anomalies. By focusing on the multi-dimensional failure feature matching of key processes, it accurately locks the root cause of the problem, and combines the deep learning of historical cases to establish a fast failure mode recognition mechanism. Its beneficial effects are to shorten the abnormal response time while improving the scientificity and reliability of the disposal plan, optimize the allocation efficiency of maintenance resources through the calibration priority sorting mechanism, reduce the quality loss in batch production by means of the dynamic generation ability of the formula reset plan, and form an intelligent optimization system for closed-loop management of failures.

[0091] Optionally, the performing multi-dimensional feature matching between the control parameters and the historical failure case database includes:

[0092] Construct a three-dimensional weight matrix of the current abnormal features, and the three-dimensional weight matrix includes the parameter deviation degree, the equipment aging coefficient, and the environmental interference factor;

[0093] Specifically, the three-dimensional weight matrix is a three-dimensional evaluation model integrating the parameter deviation degree, equipment aging coefficient, and environmental interference factor. The parameter deviation degree is the percentage deviation of the real-time monitoring value from the standard threshold. The equipment aging coefficient is the performance decay index calculated based on the cumulative operation duration and maintenance records of the equipment. The environmental interference factor is the normalized value of the temperature, humidity, and air pressure data in the production workshop. In the matrix construction process, the analytic hierarchy process is used to determine the weight distribution ratio of each dimension, and the robustness of the model is verified through orthogonal experiments.

[0094] Calculate the feature similarity between the three-dimensional weight matrix and historical failure cases;

[0095] Specifically, the feature similarity calculation uses an improved cosine similarity algorithm to quantify the matching score of the standardized matrix data. The improvement lies in applying a double weight factor to the parameter deviation degree dimension and performing Gaussian kernel function smoothing on the environmental interference factor. In the data dimensionality reduction stage, the principal component analysis method is used to extract the top three principal component vectors of influence as the core index set for similarity calculation.

[0096] Among them, the parameter deviation degree is , where is the current parameter value, representing the current actual operating parameters of the equipment or system, is the standard parameter value, representing the parameters of the equipment or system during design or normal operation;

[0097] The equipment aging coefficient is , where is the cumulative operating hours, is the number of historical failures, , ; The parameters , are obtained by fitting the equipment failure rate curve based on the Weibull distribution model. When the cumulative operating time is greater than 2000 hours, the aging coefficient A increases according to an exponential law.

[0098] The feature similarity is , where is the environmental interference factor.

[0099] Select the processing strategy corresponding to the historical case with the highest feature similarity as the basic template.

[0100] Specifically, the basic template is the standard disposal plan framework for screening the best matching cases from the historical case library. The selection process introduces a dual optimization mechanism of case effectiveness index and disposal timeliness score. In the disposal strategy generation stage, the priority is adjusted and the parameters are fine-tuned based on the actual working conditions of the current production line on the basis of the template. Finally, a personalized disposal plan is output through the weight decision tree.

[0101] Exemplarily, the device state environmental factors and process parameters are innovatively combined through a three-dimensional weight matrix, breaking through the limitations of single-parameter comparison in traditional fault analysis. An intelligent feature space matching algorithm is used to improve the accuracy of similar case retrieval and the adaptability of processing decisions. The beneficial effects are as follows: a fault diagnosis system with multi-dimensional feature fusion is established to achieve rapid analysis of complex abnormal scenarios. The processing weight of key influencing factors is enhanced through weight allocation optimization. Combining the continuous learning ability of the case library forms a dynamically optimized disposal strategy library, significantly improving equipment maintenance efficiency and reducing the probability of secondary faults.

[0102] Optionally, the sending the verified control instruction to the auxiliary material adding actuator includes:

[0103] Creating a digital twin model of the auxiliary material adding process;

[0104] Specifically, the digital twin model is a virtual simulation environment that integrates the operating rules of the physical production line and real-time sensor data. Its modeling basis includes the three-dimensional structure topology data of the equipment, the dynamic parameters of the transmission system, and the rheological characteristic curves of the materials. Through the industrial Internet of Things interface, the second-level data synchronization update with the physical production line is realized. The physical engine built into the model supports the accurate simulation of the gradual change process of the auxiliary material mixing state.

[0105] Loading the control instruction and real-time environmental parameters in the digital twin model;

[0106] Specifically, the real-time environmental parameters are a multi-dimensional data set of temperature, humidity, air pressure, and light collected by distributed sensors in the workshop. The loading process uses a dynamic resource allocation algorithm to match the timestamps and spatial coordinate systems of the physical production line and the virtual model. The loading strategy of the control instruction injects simulation triggers according to the execution queue order of the equipment response priority.

[0107] Obtaining the quality prediction index of the virtual finished product after simulation operation;

[0108] Specifically, the quality prediction index of the virtual finished product includes the calculation results of physical and chemical parameters such as sugar content uniformity, hardness value, and dissolution rate. Its generation process simulates the formation process of the sugar granule microstructure by the discrete element method and calculates the macroscopic physical properties by the finite element algorithm. The prediction index is displayed in the form of a heat map through a visualization interface to show the spatial distribution characteristics.

[0109] When the deviation between the quality prediction index and the preset standard is less than the allowable threshold, confirm that the control instruction is valid.

[0110] Specifically, the allowable threshold is a fluctuation tolerance interval dynamically adjusted based on the product acceptance specification. The deviation judgment uses a fuzzy logic algorithm to comprehensively evaluate the cumulative effect of multiple indicators. The confirmation process includes an automatic verification module of the cross-professional expert knowledge base, and automatically triggers a manual review request for critical state cases.

[0111] Exemplarily, the pre-verification of the control strategy is realized by reconstructing the digital mirror body of the actual production line in the virtual space, and a real-time mapping channel from the physical world to the digital space is constructed. Its beneficial effects are as follows: relying on high-precision simulation to expose potential production process defects in advance, reducing the physical trial-and-error cost through the quantitative analysis of quality prediction indicators, using intelligent threshold judgment technology to form a multi-level safety verification mechanism for control instructions, enhancing the traceability of the quality control process while improving production safety, and forming a closed-loop optimization control system of virtual-real collaboration.

[0112] Optionally, the simulation operation includes:

[0113] Aligning the timing tags of the actual production line and the digital twin model;

[0114] Specifically, the timing tag is a synchronization identifier that accurately marks the device action time point. The alignment operation is realized through the combined action of a hardware clock synchronization module and a software delay compensation algorithm. Specifically, the network time protocol is used to check the clock reference of the physical production line controller and the simulation server, and the slot jitter offset in data transmission is repaired through an interpolation algorithm, ultimately ensuring that the operation rhythm of the virtual model is exactly the same as the actual production.

[0115] Among them, the alignment of the timing tags realizes time synchronization through the industrial Internet of Things OPC UA protocol; first, the physical device controller sends a time stamp every 200 ms ; then, the simulation server calculates the clock deviation , where is the time stamp of the simulation server itself; when the clock deviation is greater than 10 ms, a time interpolation correction is triggered.

[0116] Adding a random disturbance factor to simulate the device execution error;

[0117] Specifically, the random disturbance factor is a virtual interference quantity that simulates the accuracy attenuation and mechanical wear of the device. Its generation is based on the failure probability distribution in the device historical maintenance record. The disturbance amplitude is dynamically adjusted according to the servo mechanism positioning accuracy level. The injection method uses the Monte Carlo method to superimpose normal distribution random noise in the control instruction, and at the same time introduces a non-linear interference model to reproduce the mechanical transmission characteristic deviations such as gear clearance.

[0118] Recording the data fluctuation range of the simulation process and generating a safety verification report.

[0119] Specifically, the data fluctuation range is the maximum and minimum values ​​and standard deviation statistics of key parameters during the simulation operation. The recording process uses the sliding window analysis method to capture transient abnormal signals. The safety verification report includes the statistical curve of the equipment action exceeding the limit frequency and the parameter out-of-bounds heat map. The report generation module integrates the fuzzy comprehensive evaluation algorithm and the risk matrix template to automatically divide the risk level and generate a list of disposal recommendations.

[0120] Among them, the threshold setting rule for the deviation of the quality prediction index is: when the production batch is greater than 1000 pieces: max(0.5%, 0.1% of the total cost of raw materials); the Monte Carlo perturbation factor injection formula is: , where is a normally distributed random number.

[0121] For example, a digital twin test environment close to reality is built through high-precision timing calibration and controllable disturbance injection technology, and potential operational risks are deeply explored using multi-dimensional data analysis methods. The beneficial effect is that it breaks through the limitations of idealized assumptions in traditional simulation tests, improves the credibility and engineering practical value of test results by simulating real equipment errors, and provides accurate decision-making basis for process optimization in combination with an automated risk quantification assessment system, forming a complete digital verification closed loop covering test design execution and analysis, significantly enhancing the fault tolerance and safety protection level of the control system.

[0122] Optionally, the calculation of the predicted compensation amount includes:

[0123] Get the ambient temperature and humidity change trend for the next 15 minutes;

[0124] Specifically, the changing trend of ambient temperature and humidity is the short-term climate change data predicted by the fusion of the workshop's external meteorological station and the indoor microenvironment sensor. The prediction algorithm uses the time series decomposition method to separate the seasonal and random components and then uses the ARIMA model to deduce the future change curve. The data collection cycle is set to update every 30 seconds to capture the transient characteristics of rapid disturbances.

[0125] Long short-term memory neural network is used to predict changes in auxiliary material demand;

[0126] Specifically, the long short-term memory neural network is a deep time series prediction model equipped with a forget gate and a memory gate structure. Its input features include the temperature and humidity correlation factors of the historical consumption rate of auxiliary materials and the equipment operation status encoding. The attention mechanism is introduced into the model training process to strengthen the feature learning of key time nodes. The prediction results are decomposed into a dual-channel output of the demand trend component and the periodic fluctuation component.

[0127] Among them, the structure of the long short-term memory neural network can be that the input layer has 6 nodes, namely the historical values of temperature, humidity, and viscosity; each of the two-layer LSTM has 64 units. The training data is 50,000 sets of production records containing different seasons / raw material batches. The loss function is SmoothL1Loss, and the learning rate is 0.001, that is, the Adam optimizer. The RobustScaler method is used for input feature normalization.

[0128] Modify the length and weight distribution of the sliding time window based on the prediction results.

[0129] Specifically, the sliding window length correction algorithm dynamically adjusts the data interception range according to the gradient change of the auxiliary material demand trend component. The window is shortened in the steep rise stage to enhance sensitivity, and the window is extended in the stable stage to improve stability. The weight distribution is updated by assigning different time decay factors to historical data according to the amplitude characteristics of the periodic fluctuation component, and the weight allocation ratio is optimized through the matrix confusion algorithm.

[0130] Exemplarily, a refined auxiliary material demand prediction mechanism is established by integrating meteorological prediction technology and deep time series models, and the intelligent adaptation of the data analysis model is realized by combining the dynamic self-adjustment strategy of sliding window parameters. The beneficial effect is to break through the applicability limitation of the traditional fixed window analysis mode to dynamic process scenarios, accurately capture the coupling relationship between complex environmental variables and auxiliary material demands by using the non-linear mapping ability of the neural network, effectively balance the sensitivity and stability of the prediction model through real-time optimization of window parameters, form an intelligent control system with strong climate adaptability, and significantly improve the active adjustment ability of the production system to cope with environmental fluctuations.

[0131] Example 1:

[0132] To verify the implementation effect of the present invention, the present invention is applied to the throat lozenge production line under a certain food group. This production line mainly produces functional throat lozenges containing herbal extracts, and the addition ratios of auxiliary materials such as honey, menthol, and Chinese herbal medicine extracts need to be accurately controlled. The traditional production method relies on manual experience to adjust parameters and cannot respond to raw material fluctuations and environmental changes in real time, resulting in problems such as uneven sugar content of the finished product and deviation of the effective ingredient content.

[0133] In this embodiment, a dynamic threshold matrix is constructed through the threshold matrix establishment module. The viscosity reference value is set to 4500 cP (25 °C), the sugar content floating range is ±2%, and the temperature and humidity critical points are 25 °C ± 3 °C and 45% RH ± 5% respectively. The data acquisition module integrates a high-precision viscosity sensor, an infrared saccharimeter, and a temperature and humidity detector to collect the viscosity parameters of the raw material mixing tank, the environmental data of the auxiliary material warehouse, and the sugar content of the finished product in the forming section in real time.

[0134] When the ambient temperature is suddenly detected to rise to 28°C (exceeding the baseline threshold by 5%), the data analysis module triggers the adjustment mode. First, historical parameter matching is performed, retrieving historical data of similar temperature rise scenarios in the past 30 batches, and screening out the 5 groups of samples with the highest auxiliary material compensation amounts. Then, dynamic compensation calculation is carried out. Based on the sliding window algorithm with a window length of 10% of the current batch production duration, the increment of the honey addition rate is calculated, predicting the temperature rise trend in the next 15 minutes through LSTM network prediction, thereby generating a compensation coefficient of 0.15. Next, instruction pre-verification is carried out. The corrected instruction is loaded into the digital twin model, and the simulation operation shows that the sugar content fluctuation of the finished product drops from ±3% to ±1.2%, confirming the effectiveness of the instruction. Finally, smooth switching execution is carried out. The control execution module adopts a progressive transition strategy, where the inertia compensation factor is 0.8, and the metering pump rate is gradually adjusted within 10 seconds to avoid mechanical shock.

[0135] Two production lines with the same production capacity are selected for comparison. The experimental group applies the present invention, and the control group uses traditional PID control; continuous production is carried out for 100 batches, 2 hours per batch. The experimental group and the control group use the same raw materials, and the key indicators monitored include: the qualified rate of the finished product sugar degree, with the target value of 75%-80%; the metering error of auxiliary material addition, less than or equal to ±1.5%; the abnormal response time, which is the time from detecting parameter overstep to executing adjustment.

[0136] The data shows that the qualified rate of the finished product sugar degree in the experimental group reaches 98.3%, and that in the control group is 84.7%. During the afternoon period with frequent ambient temperature fluctuations, the experimental group autonomously compensates 26 times through the adjustment mode, and the standard deviation of the sugar degree is 0.8%, significantly lower than 2.1% of the control group; when the viscosity sensor detects that the raw material viscosity deviates by 8%: the experimental group triggers the quality traceback mode, locates the problem of over-limit mixing temperature in the previous process within 10 seconds, generates an equipment calibration instruction, and the abnormal response time is 45 seconds; the control group relies on manual troubleshooting, with an average time-consuming of 15 minutes. The average waste rate of auxiliary materials in the experimental group is reduced to 0.7%, a 72% reduction compared to 2.4% of the control group.

[0137] Table 1 Comparison Table of Blood Pressure Fluctuation Data of Hypertension Patients

[0138] Index Experimental group Control group Qualified rate of sugar content 98.3% 84.7% Measurement error (standard deviation) 0.9% 2.3% Abnormal response time ≤60 seconds 8 - 20 minutes Waste rate of auxiliary materials 0.7% 2.4%

[0139] As can be seen from the data in Table 1 above, the control system of the present invention significantly improves the quality control accuracy of throat lozenge production through the dynamic threshold matrix and multi-mode collaborative mechanism, realizing full-process automation from real-time monitoring, intelligent decision-making to closed-loop verification. Compared with traditional methods, its adaptability and abnormal response efficiency in complex working conditions are significantly superior, effectively reducing raw material losses and ensuring product consistency, meeting the strict production requirements of functional foods.

[0140] Based on the same inventive concept, such as Figure 3As shown, the present invention also provides a method for controlling the addition of adjuvants in throat lozenge production, and the method includes:

[0141] Construct a dynamic threshold matrix, which includes the viscosity reference value, the sugar content floating range, and the range of temperature and humidity critical points;

[0142] Collect current production data, which includes the measured value of the viscosity sensor in the raw material conveying section, the temperature and humidity monitoring data of the adjuvant warehouse, and the online detection indicators of the finished product;

[0143] Generate a mode selection signal according to the comparison result between the current production data and the dynamic threshold matrix;

[0144] Activate the corresponding control mode based on the mode selection signal, and the control mode includes a normal mode, an adjustment mode, and an exception handling mode;

[0145] Execute the control instructions corresponding to the control mode, and send the verified control instructions to the adjuvant addition execution mechanism.

[0146] It should be noted that the electrical connections between the above-mentioned various units do not necessarily mean direct connections of the lines. Indirect connection methods, as long as they can achieve the purpose of the present invention, are applicable to the embodiments of the present invention. The above are only exemplary embodiments of the present invention, and the scope of the present invention cannot be limited by this.

[0147] That is, any equivalent changes and modifications made in accordance with the teachings of the present invention still fall within the scope covered by the present invention. After considering the specification and the disclosure of the practical truth, those skilled in the art will easily think of other implementation schemes of the present invention. This application aims to cover any variations, uses, or adaptive changes of the present invention, and these variations, uses, or adaptive changes follow the general principles of the present invention and include the common general knowledge or conventional technical means in the technical field not recorded in the present invention.

Claims

1. A control system for adding auxiliary materials to throat lozenge production, characterized in that: The system comprises: A threshold matrix building module is used to build a dynamic threshold matrix, wherein the dynamic threshold matrix includes a viscosity reference value, a sugar content floating range, and an interval range of temperature and humidity critical points; A data acquisition module is used to collect current production data, including viscosity sensor measurement values ​​of the raw material conveying section, temperature and humidity monitoring data of the auxiliary material bin, and online detection indicators of finished products; A data analysis module, configured to generate a mode selection signal according to a comparison result between the current production data and the dynamic threshold matrix; A mode selection module, configured to activate a corresponding control mode based on the mode selection signal, wherein the control mode includes a normal mode, an adjustment mode, and an abnormality handling mode; A control execution module, used for executing the control instructions corresponding to the control mode, and sending the verified control instructions to the auxiliary material adding execution mechanism; The adjustment modes include: Retrieve a set of historical parameters from historical production batches that match the current deviation; Calculating a dynamic compensation coefficient of the auxiliary material addition rate based on the historical parameter set; Adding the dynamic compensation coefficient to the current recipe parameter to generate a corrected control parameter; The exception handling modes include: Extract the control parameters of the three processes preset for the current production batch; Perform multi-dimensional feature matching on the control parameters and the historical fault case library; Generate processing instructions containing device calibration priorities and recipe reset solutions based on the matching results.

2. A control system for adding auxiliary materials to throat lozenge production according to claim 1, characterized in that: The data analysis module includes: A deviation analysis unit, used to calculate the deviation of each indicator in the current production data relative to the dynamic threshold matrix; An abnormality judgment unit, used to trigger the abnormality handling mode when it is detected that the deviation of at least one indicator exceeds a preset warning value; An adjustment judgment unit, configured to trigger the adjustment mode when the deviation of all indicators is within a preset adjustment interval; The normal judgment unit is used to maintain the normal mode when the deviation of all indicators does not exceed the preset allowable fluctuation range.

3. A control system for adding auxiliary materials to throat lozenge production according to claim 1, characterized in that: The dynamic compensation coefficient for calculating the auxiliary material addition rate includes: Establish a sliding time window, the length of which is 10% of the duration of the current production batch; Extracting the historical parameter change rate within the sliding time window and calculating the predicted compensation amount; The predicted compensation amount is weightedly fused with the conventional compensation amount to output a final dynamic compensation coefficient.

4. A control system for adding auxiliary materials to throat lozenge production according to claim 1, characterized in that: The multi-dimensional feature matching of the control parameters with the historical fault case library includes: Constructing a three-dimensional weight matrix of the current abnormal characteristics, wherein the three-dimensional weight matrix includes parameter deviation, equipment aging coefficient and environmental interference factor; Calculating feature similarity between the three-dimensional weight matrix and historical fault cases; The processing strategy corresponding to the historical case with the highest feature similarity is selected as the basic template.

5. A control system for adding auxiliary materials to throat lozenge production according to claim 1, characterized in that: The step of sending the verified control instruction to the auxiliary material adding actuator comprises: Create a digital twin model of the excipient addition process; Loading the control instructions and real-time environmental parameters into the digital twin model; After the simulation runs, quality prediction indicators of the virtual finished product are obtained; When the deviation between the quality prediction index and the preset standard is less than an allowable threshold, the control instruction is confirmed to be valid.

6. A control system for adding auxiliary materials to throat lozenge production according to claim 5, characterized in that: The simulation operation includes: Align the timing labels of the actual production line with the digital twin model; Add random disturbance factors to simulate equipment execution errors; Record the data fluctuation range of the simulation process and generate a safety verification report.

7. A control system for adding auxiliary materials to throat lozenge production according to claim 3, characterized in that: The calculation of the predicted compensation amount includes: Get the ambient temperature and humidity change trend for the next 15 minutes; Long short-term memory neural network is used to predict changes in auxiliary material demand; The length and weight distribution of the sliding time window are modified based on the prediction results.

8. A method for controlling the addition of auxiliary materials in the production of throat lozenges, using a control system for adding auxiliary materials in the production of throat lozenges as claimed in any one of claims 1 to 7, characterized in that: The method comprises: Constructing a dynamic threshold matrix, wherein the dynamic threshold matrix includes a viscosity reference value, a sugar content floating range, and an interval range of a temperature and humidity critical point; Collect current production data, including viscosity sensor measurement values ​​of the raw material conveying section, temperature and humidity monitoring data of the auxiliary material bin, and online detection indicators of finished products; generating a mode selection signal according to a comparison result between the current production data and the dynamic threshold matrix; activating a corresponding control mode based on the mode selection signal, the control mode including a normal mode, an adjustment mode and an abnormality handling mode; Execute the control instructions corresponding to the control mode, and send the verified control instructions to the auxiliary material adding execution mechanism.

Images