Chaffing method based on multi-parameter feedback for hot pot seasoning processing parameter adaptive adjustment
By using an adaptive adjustment method based on multi-parameter feedback and flavor transfer kinetics model, the problem of unstable quality in hot pot base production was solved, achieving product standardization and efficient production, and improving finished product consistency and production flexibility.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHONGQING ZHANGBANG FOOD CO LTD
- Filing Date
- 2026-05-13
- Publication Date
- 2026-06-09
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
In the current industrial production of hot pot base, the reliance on manual experience for control leads to unstable product quality, making it difficult to achieve standardization and large-scale production. Furthermore, automated production cannot cope with batch differences in raw materials and environmental fluctuations, resulting in insufficient consistency in the flavor and quality of the finished product.
An adaptive adjustment method for hot pot base processing parameters using multi-parameter feedback is adopted. By loading the flavor feature vector of the target finished product, multi-dimensional parameters are collected in real time, and predictive optimization is performed using a flavor transfer kinetic model to achieve closed-loop adaptive adjustment. Combining mechanistic models and intelligent decision-making, the reliance on human experience is reduced.
It achieves forward-looking global optimization of the hot pot base frying process, ensuring high stability and standardization of product flavor and quality, significantly improving the yield of high-quality products and reducing energy consumption and raw material loss, and supporting rapid adaptation of new formula processes and data asset accumulation.
Smart Images
Figure CN122172601A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of hot pot processing control technology, specifically to an adaptive adjustment method for hot pot base material processing parameters based on multi-parameter feedback. Background Technology
[0002] The current industrial production of hot pot base, especially the core stir-frying process, still heavily relies on experienced chefs who use their senses to control the heat, timing of ingredient addition, and stir-frying rhythm. This manual operation mode has significant drawbacks: first, product quality is highly unstable, greatly influenced by the operator's subjective state, making standardization and large-scale replication difficult; second, valuable technological experience is difficult to digitize and pass on, hindering the sustainable development of the industry. To address this issue, existing technologies have employed automated stir-fryers with program control based on single or a few fixed parameters such as temperature and time. However, these methods lack the flexibility to cope with batch differences in raw materials and environmental fluctuations, essentially remaining "open-loop" control. They cannot respond in real time to the complex physicochemical changes of materials during the stir-frying process, resulting in insufficient consistency in the flavor and quality of the finished product.
[0003] Some improvement solutions attempt to introduce online monitoring methods, such as triggering preset adjustments by detecting temperature or color changes. However, these methods are mostly based on isolated, threshold-based simple feedback, failing to conduct in-depth integrated analysis and collaborative perception of multiple key parameters affecting the final flavor of hot pot base (such as the dynamic evolution of various flavor substances and the rheological properties of materials) from a systems engineering perspective. Their control logic remains localized and lagging, unable to simulate the holistic experience of master chefs who comprehensively consider visual, olfactory, and tactile information for forward-looking judgment and adaptive adjustments. This results in automated production products lagging behind handcrafted products in core indicators such as flavor profile and aroma fullness, and also fails to quickly adapt to the process development needs of new recipe products. Summary of the Invention
[0004] To address the aforementioned technical problems, this paper provides an adaptive adjustment method for hot pot base processing parameters based on multi-parameter feedback. This technical solution solves at least one of the technical problems mentioned in the background section.
[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:
[0006] An adaptive adjustment method for hot pot base processing parameters based on multi-parameter feedback includes:
[0007] Based on the flavor type of the target hot pot base, load the corresponding flavor feature vector of the target finished product, and set the initial processing control parameters based on the basic formula;
[0008] During the stir-frying process, multi-dimensional real-time parameters reflecting the physical state, chemical composition and sensory simulation properties of the base material are collected simultaneously. These multi-dimensional real-time parameters include physical temperature, apparent color, material viscosity and concentration of key volatile flavor substances.
[0009] The multi-dimensional real-time parameters are processed into a current state feature vector and input into a pre-trained flavor transfer dynamics model. The flavor transfer dynamics model, combined with the quantified formula information, predicts the change trajectory of key flavor substance concentrations and comprehensive quality indicators within a preset time period in the future.
[0010] Based on the various future trajectories predicted by the flavor transfer kinetics model, a model predictive control algorithm is adopted. The optimization objective is to be closest to the target finished product flavor feature vector and with the lowest energy consumption. The algorithm continuously calculates and outputs the optimal processing parameter adjustment command within the current control cycle.
[0011] The optimal processing parameter adjustment command is executed to drive the heating, stirring and auxiliary material addition mechanisms to change the frying process. Multi-dimensional real-time parameters reflecting the physical state, chemical composition and sensory simulation attributes of the base material are continuously collected to form a closed-loop adaptive adjustment until the distance between the current state feature vector and the target finished product flavor feature vector is less than a preset threshold, triggering the finished product unloading command.
[0012] Preferably, the steps for constructing the pre-trained flavor transfer dynamics model include:
[0013] In historical frying experiments, complete time-series multi-dimensional real-time parameters, corresponding precise processing control parameter sequences, and expert sensory scores and physicochemical test data of the finished product base were collected simultaneously.
[0014] A model training objective function is constructed. The model training objective function not only requires that the error between the model prediction value and the actual detection data be minimized, but also requires that its prediction law conforms to the embedded mass and heat transfer physicochemical constraints. The constraints include the mass conservation equation and the reaction kinetic relationship based on the Arrhenius equation.
[0015] By employing a physical information neural network structure, the model is trained using collected data and defined constraints. This enables the model to not only fit the data with high accuracy but also learn the inherent physicochemical laws in the base material processing, thus obtaining an interpretable flavor transfer kinetic model.
[0016] Preferably, the process of converting multi-dimensional real-time parameters into a current state feature vector specifically includes:
[0017] The raw sensor signals are preprocessed by filtering, denoising, and standardization.
[0018] From the preprocessed time series data, feature values with process indication significance are extracted. These feature values include temperature change rate, color change rate, viscosity curve inflection point, instantaneous value of specific volatile flavor substance concentration and its first derivative.
[0019] The extracted feature values are combined in sequence to form a comprehensive current state feature vector that characterizes the current frying process.
[0020] Preferably, among the multi-dimensional real-time parameters, the physical temperature is monitored by a distributed temperature sensor array, which includes at least a first, second, and third sensor for monitoring the temperature of the bottom of the pot, the temperature of the center of the material, and the steam temperature at the pot opening.
[0021] The apparent color is monitored by a hyperspectral camera or an industrial color camera and converted into a standard Lab color space value;
[0022] The viscosity of the material is obtained by an online rotational viscometer or indirectly by means of torque measurement;
[0023] The concentrations of the key volatile flavor compounds are monitored online using a near-infrared spectroscopy probe or a mass spectrometry-type electronic nose.
[0024] Preferably, the step of using the closest possible flavor feature vector of the target finished product and the lowest energy consumption as the optimization objective, and the rolling calculation of the optimal processing parameter adjustment instruction specifically includes:
[0025] Starting with the current state feature vector, and combining multiple different future processing parameter assumption sequences, the flavor transfer kinetics model generates multiple corresponding future quality change trajectories in parallel prediction.
[0026] Calculate the Euclidean distance between the endpoint feature vector of each predicted trajectory and the target finished product flavor feature vector, and simultaneously calculate the estimated total energy consumption for executing the sequence of assumed parameters.
[0027] The weighted summation method or Pareto front method is used to perform multi-objective optimization on the trajectory's endpoint proximity and the estimated total energy consumption, and the predicted trajectory with the highest comprehensive score is selected.
[0028] The first control instruction in the processing parameter hypothesis sequence corresponding to the predicted trajectory with the highest comprehensive score is output as the optimal processing parameter adjustment instruction for the current control cycle.
[0029] Preferably, the method further includes a formulation process adaptive migration step, applied to rapidly generate initial process parameters for a new formulation, the formulation process adaptive migration step including:
[0030] The various ingredients in the new formula are quantified into a multi-dimensional formula vector based on their chemical composition and flavor contribution.
[0031] The formula vector of the new formula is input into the flavor transfer kinetics model, which then matches and interpolates it with a known library of mature formulas. Combined with the learned physicochemical laws, a series of baseline processing parameters suitable for the initial optimization of the new formula is derived.
[0032] Using the aforementioned benchmark processing parameter sequence as the initial setting, a closed-loop adaptive adjustment process is initiated to achieve rapid online fine-tuning and optimization of the preliminary process.
[0033] Preferably, the optimal processing parameter adjustment instruction specifically includes control instructions for at least one of the following actuators:
[0034] Power or temperature adjustment commands for heating components are used to precisely control the heat source input;
[0035] The speed and direction adjustment commands of the stirring device are used to control the heat and mass transfer efficiency and prevent scorching.
[0036] The opening and closing timing and flow command of the automatic dispensing valve for liquid or powdered excipients are used to accurately dispense materials at key process points.
[0037] Preferably, the physical information neural network structure includes a physical calculation module for learning physical law constraints. This module imposes constraints on the output of the intermediate layers of the model, such that the changes in flavor substance concentration predicted by the model satisfy the following relationship:
[0038] The total amount of flavor compounds distributed in the oil phase, aqueous phase and solid phase remains unchanged;
[0039] The rates of the Maillard reaction and caramelization reaction are positively correlated with temperature in a non-linear manner.
[0040] Preferably, the step of introducing active parameter perturbation in the process of simultaneously collecting multi-dimensional real-time parameters reflecting the physical state, chemical composition, and sensory simulation attributes of the base material during the stir-frying process specifically includes:
[0041] Apply a brief, small change to the stirring speed periodically or as needed;
[0042] Simultaneously monitor the dynamic response of viscosity, temperature, and volatile substance signals under this disturbance;
[0043] The characteristics of the dynamic response are used as additional inputs to evaluate the real-time rheological properties of the substrate and the state of the reaction system, thereby enhancing the information dimension of the current state feature vector and the robustness of the control system.
[0044] Preferably, the method further includes a self-learning optimization mechanism, which performs the following steps after each stir-frying process:
[0045] The complete process data chain, including real-time parameters of multiple dimensions of time sequence, the sequence of processing parameter adjustment instructions executed, and the final product quality evaluation, will be stored in the historical process database.
[0046] The expanded historical process database is used periodically to incrementally learn and fine-tune the parameters of the flavor transfer kinetics model and the model prediction controller.
[0047] Based on long-term accumulated finished product quality data and market feedback, the flavor feature vectors of the target finished products corresponding to different flavor types are optimized and calibrated in reverse to continuously track and adapt to changes in consumer preferences.
[0048] Preferably, the method further includes an application calibration step, which is used to adapt to frying equipment of different sizes, specifically including:
[0049] Through a limited number of calibration experiments, a correlation model was established between the critical transfer coefficient and the volume, pot shape, and heating area of the frying equipment.
[0050] The correlation model is used as a compensation factor to scale and adjust the parameters in the flavor transfer kinetics model that are significantly affected by equipment size.
[0051] Applying the adjusted model to new-scale equipment allows the core control strategy to be maintained, enabling rapid replication and consistent scaling up of the process at different production scales.
[0052] Preferably, the method further includes a data sharing mechanism, which interconnects with the enterprise's production execution system and product lifecycle management system, specifically including:
[0053] Receive specific flavor base production orders from MES and automatically load the corresponding target flavor type, formula and process parameter package;
[0054] The parameters of each processing step are linked to the corresponding production batch number to form an immutable digital process file, which is stored in a blockchain or central database.
[0055] Validated and successful process parameter packages for new formulations are automatically pushed to the PLM system, forming a standardized process knowledge base asset for the enterprise to use by the R&D and production departments.
[0056] Compared with the prior art, the beneficial effects of the present invention are as follows:
[0057] This invention, by integrating multi-source sensing, mechanistic models, and intelligent decision-making, transforms the hot pot base preparation process from reactive, single-point control to proactive, global optimization. It reduces absolute reliance on human experience, ensuring high stability and standardization of flavor and quality across different batches through real-time multi-parameter feedback and predictive optimization based on flavor transfer kinetics models. This significantly improves the yield of high-quality products and reduces energy consumption and raw material loss. The invention's cross-recipe process self-generation and self-learning capabilities enable the intelligent transfer of mature processes to new recipes, greatly shortening new product development cycles and continuously accumulating process data assets. This builds an iterative and reusable core process knowledge system for enterprises, achieving flexible and intelligent upgrades in production. Attached Figure Description
[0058] Figure 1 This is a flowchart of the adaptive adjustment method for hot pot base processing parameters based on multi-parameter feedback proposed in Embodiment 1 of this solution;
[0059] Figure 2 This is a flowchart of the active parameter perturbation steps proposed in Embodiment 1 of this scheme;
[0060] Figure 3 This is a flowchart illustrating the method for constructing the flavor transfer kinetics model proposed in Embodiment 1 of this scheme;
[0061] Figure 4 This is a flowchart of the method for calculating and outputting the optimal machining parameter adjustment command within the current control cycle, as proposed in Embodiment 1 of this scheme;
[0062] Figure 5 This is a flowchart of the adaptive migration steps for the formulation process proposed in Embodiment 2 of this scheme. Detailed Implementation
[0063] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.
[0064] Example 1:
[0065] Reference Figure 1 As shown, this embodiment proposes an adaptive adjustment method for hot pot base processing parameters based on multi-parameter feedback, including:
[0066] Based on the flavor type of the target hot pot base, the corresponding flavor feature vector of the target finished product is loaded, and based on the basic formula, initial processing control parameters are set. Specifically, this is achieved through the following steps:
[0067] A digital flavor database containing multiple standard flavor types, such as spicy butter, tomato soup, and mushroom aroma, is pre-built. Each flavor type corresponds to a multi-dimensional target finished product flavor feature vector. This vector is generated by instrumental sensory analysis of the standard finished product of this type, such as electronic tongue, electronic nose, colorimeter, texture analyzer and physicochemical detection, such as GC-MS spectra of capsaicin and volatile aroma components, and by integrating expert sensory evaluation scores.
[0068] Once the operator selects the target flavor type on the control interface, the system automatically loads the corresponding target feature vector from the database.
[0069] It also retrieves the basic formula quantification list corresponding to the flavor type and the set initial processing control parameters, including but not limited to: feeding oil temperature, heating power curve, stirring speed and the planned addition time of auxiliary materials, so as to provide an initial process setting for subsequent adaptive adjustment.
[0070] During the stir-frying process, multi-dimensional real-time parameters reflecting the physical state, chemical composition, and sensory simulation properties of the base material are collected simultaneously. These multi-dimensional real-time parameters include physical temperature, apparent color, material viscosity, and concentration of key volatile flavor compounds. Physical temperature is monitored using a distributed temperature sensor array, which includes at least three sensors monitoring the temperature of the bottom of the pot, the center temperature of the material, and the steam temperature at the pot opening. Apparent color is monitored using a hyperspectral camera or an industrial color camera and converted to standard Lab color space values. Material viscosity is obtained through an online rotational viscometer or an indirect method based on torque measurement. The concentration of key volatile flavor compounds is monitored online using a near-infrared spectral probe or a mass spectrometer-type electronic nose.
[0071] Specifically, the acquisition process of the above parameters also includes an active parameter perturbation step, as described above. Figure 2 As shown, the steps of the active parameter perturbation process include:
[0072] Apply a brief, small change to the stirring speed periodically or as needed;
[0073] Simultaneously monitor the dynamic response of viscosity, temperature, and volatile substance signals under this disturbance;
[0074] By using the characteristics of the dynamic response as an additional input, the real-time rheological properties of the substrate and the state of the reaction system are evaluated, thereby enhancing the information dimension of the current state feature vector and the robustness of the control system.
[0075] In the specific processing, the aforementioned multi-dimensional parameter acquisition and active parameter perturbation work in tandem. After the stir-frying process enters a steady state, or when the system detects that a key parameter has entered a plateau period, the controller will autonomously trigger an active parameter perturbation according to a preset cycle, set to 5 minutes in some preferred embodiments. The stirring speed is set to increase by a step value from the current steady-state value within 0.5 seconds. In some preferred embodiments, the step value is set to 15% of the steady-state stirring speed, and this higher speed is maintained for 2-3 seconds. Subsequently, the speed decreases by a step value from the original steady-state value within 0.5 seconds, maintained for 2-3 seconds, and finally, the speed smoothly returns to the original steady-state speed within 1 second. One perturbation step is completed.
[0076] During a perturbation step, real-time readings from the viscosity sensor, temperatures at various points in the distributed temperature sensor array, and the intensity of specific spectral absorption peaks from the near-infrared spectral probe are simultaneously acquired at a higher frequency. In some preferred embodiments, the sampling frequency during the perturbation is set to 10 times the conventional sampling frequency. These high-density dynamic data streams are recorded as time series. After the perturbation ends, the system immediately performs feature extraction on each signal sequence, calculates the instantaneous rate of change of the viscosity signal during the acceleration and deceleration phases, analyzes the maximum fluctuation amplitude of the pot center temperature during the perturbation, and calculates the difference in the spectral signals of key volatile substances before and after the perturbation to obtain dynamic response characteristic values.
[0077] Subsequently, the dynamic response feature values are used as a new set of feature dimensions and combined with conventionally collected static parameters, such as current absolute temperature, color Lab value, and viscosity steady-state value, to jointly construct an enhanced current state feature vector with richer information dimensions and better reflecting the dynamic characteristics of mass and heat transfer inside the material, thereby providing deeper working condition information for subsequent intelligent decision-making.
[0078] Multidimensional real-time parameters are processed into current state feature vectors and input into a pre-trained flavor transfer dynamics model. The flavor transfer dynamics model, combined with the quantified formula information, predicts the change trajectory of key flavor substance concentrations and comprehensive quality indicators within a preset time period in the future.
[0079] The process of converting multi-dimensional real-time parameters into a current state feature vector specifically includes:
[0080] The raw sensor signals are preprocessed by filtering, denoising, and standardization.
[0081] From the preprocessed time series data, feature values with process indication significance are extracted. These feature values include temperature change rate, color change rate, viscosity curve inflection point, instantaneous value of specific volatile flavor substance concentration and its first derivative.
[0082] The extracted feature values are combined in sequence to form a comprehensive current state feature vector that characterizes the current frying process.
[0083] Specifically, refer to Figure 3 As shown, the method for constructing the flavor transfer kinetics model is as follows;
[0084] In historical frying experiments, complete time-series multi-dimensional real-time parameters, corresponding precise processing control parameter sequences, and expert sensory scores and physicochemical test data of the finished product base were collected simultaneously.
[0085] Specifically, before building the model, a large number of frying experiments covering different formulas and different process conditions need to be conducted in a controlled experimental or production environment in order to build a high-quality dataset.
[0086] In each frying experiment, multi-dimensional real-time parameters (T(t), C(t), μ(t), S(t)) and precise processing control parameter sequences (P(t)) were synchronously recorded at a fixed frequency, where T represents temperature array data, C represents color value, μ represents viscosity, S represents near-infrared or mass spectrometry signal, and P represents control commands. All data were strictly timestamped and aligned.
[0087] After stir-frying, standardized samples of the finished base are taken. On one hand, instrumental physicochemical testing is performed, such as using high-performance liquid chromatography (HPLC) to determine the content of capsaicin and thiazolyl amide, and using gas chromatography-mass spectrometry (GC-MS) to analyze the spectral composition of volatile flavor compounds, forming a quantitative index vector. On the other hand, a sensory evaluation panel conducted blind tastings, scoring dimensions such as spiciness, richness, and burnt flavor to form a sensory score vector. ;
[0088] The basic recipe used in each experiment was quantified and encoded. The recipe was converted into a vector, whose elements included, but were not limited to: butter ratio, total chili pepper content and corresponding Scoville index estimate, total Sichuan peppercorn content, moisture content, and total sugar content. This constituted the prior knowledge input for the model.
[0089] Construct a model training objective function. The model training objective function not only requires that the error between the model prediction value and the actual detection data be minimized, but also requires that its prediction law conforms to the embedded mass and heat transfer physicochemical constraints, including the mass conservation equation and the reaction kinetic relationship based on the Arrhenius equation.
[0090] By employing a physical information neural network structure, the model is trained using collected data and defined constraints. This enables the model to not only fit the data with high accuracy but also learn the inherent physicochemical laws in the base material processing, thus obtaining an interpretable flavor transfer kinetic model.
[0091] The model training employs a physical information neural network structure. This structure not only takes the aforementioned time-series data, control parameters, and recipe vectors as input, but its loss function and network structure are also designed to adhere to fundamental physicochemical laws. Specifically:
[0092] The physical information neural network structure includes an input module, a feature fusion and processing module, and an output module;
[0093] The input module includes:
[0094] The state feature input layer is used to receive the enhanced state feature vector F(t) at time t;
[0095] The control sequence input layer is used to receive the hypothetical control parameter sequence P(t:t+Δt) for a future period starting from time t.
[0096] The recipe prior input layer is used to receive the recipe vector;
[0097] The feature fusion and processing module is connected to the input module. The feature fusion and processing module includes:
[0098] The splicing and fusion layer is used to splice the enhanced state feature vector F(t) at time t from the state feature input layer, the hypothetical control parameter sequence P(t:t+Δt) from the control sequence input layer for a future period starting from time t, and the recipe vector from the recipe prior input layer to form a comprehensive feature vector that includes the current state, future control strategy, and material properties. The comprehensive feature vector is in the form of [F(t), P(t:t+Δt), recipe vector];
[0099] The physical information processing layer consists of multiple fully connected layers that perform nonlinear transformations to learn complex mapping relationships. The network structure design of the physical information processing layer includes a physical variable calculation branch, which is used to calculate intermediate variables directly related to physical laws from the features of the physical information processing layer. In this scheme, two design ideas for the physical variable calculation branch are given: one is to calculate the oil phase content, water phase content and solid phase content of key flavor substances, and the other is to calculate the instantaneous rate of key chemical reactions.
[0100] The output module is connected to the feature fusion and processing module. Based on the comprehensive feature vector formed by the splicing and fusion layer, which includes the current state, future control strategy and material properties, the output module outputs the trajectory of the key quality indicators Y(t+1), Y(t+2), ... Y(t+Δt) in the future time interval Δt, which is predicted by the complex mapping relationship learned by multiple fully connected layers of the physical information processing layer. Y may include the predicted capsaicin concentration, color value, comprehensive flavor score, etc.
[0101] Construct a loss function that embeds physical constraints, and the model's total loss function. It consists of two parts:
[0102] ;
[0103] in, The mean squared error between predicted and actual detected values is used to ensure the model fits the training data. The physical constraint loss forces the model to learn physical laws;
[0104] This embodiment provides two specific examples of physical constraint losses. The first is a constraint based on mass conservation, where, for key flavor compounds, it is assumed that their total amount remains constant during cooking, distributing only among the oil, aqueous, and solid phases. The physical constraint loss is defined as:
[0105] ;
[0106] in, It is a collection of key flavor compounds. For the i-th key flavor compound, Let i be the initial total amount of the i-th key flavor compound. Let be the oil phase content of the i-th key flavor compound. The aqueous phase content of the i-th key flavor compound. For the solid phase content of the i-th key flavor compound, in this embodiment of physical constraint loss, the physical variable calculation branch is designed to calculate the oil phase content, aqueous phase content and solid phase content of the key flavor compound.
[0107] Secondly, there is the kinetic constraint based on the Arrhenius equation. For key chemical reactions such as the Maillard reaction, the rate constant should follow an Arrhenius relationship with temperature. In the model, constraints can be imposed on the predicted reaction rate versus temperature relationship, and the physical constraint loss is defined as:
[0108] ;
[0109] It is a collection of key chemical reactions. For the j-th key chemical reaction, To provide the model's predicted rate for the j-th key chemical reaction, As the pre-factor, For activation energy, Let be the ideal gas constant. Given the measured absolute temperature, in this embodiment with physical constraint loss, the physical variable calculation branch is designed to calculate the instantaneous rate of the key chemical reaction.
[0110] The prepared dataset is then sliced according to time windows and organized into a large number of input-output sample pairs. Training is performed to learn the complex mapping relationships of the physical information processing layer, aiming to minimize the loss function, thus obtaining a flavor transfer dynamics model. During real-time control, the system inputs the current state feature vector F(t), the recipe vector R, and multiple different future control hypothesis sequences {P1, P2, ...} into the model. The model runs in parallel, quickly predicting the future quality trajectory under each set of hypothetical control parameters.
[0111] Based on the flavor transfer kinetics model predicting multiple future trajectories, a model predictive control algorithm is employed. With the optimization objective of approximating the target finished product's flavor feature vector while minimizing energy consumption, the algorithm continuously calculates and outputs the optimal processing parameter adjustment command for the current control cycle. Figure 4 As shown, this step specifically includes:
[0112] Starting with the current state feature vector, and combining multiple different sequences of future processing parameter hypotheses, the flavor transfer kinetics model generates multiple corresponding future quality change trajectories in parallel prediction. The specific steps are as follows:
[0113] At time k, the current state feature vector Assuming an initial state, and setting the prediction time domain to N future times, generate multiple different sequences of hypothetical future processing parameters. , ,in, The current state feature vector These are the initial processing parameters in the initial state. The feature vector of the current state The initial state is the assumed value of the future processing parameters at time n in the future processing parameter hypothesis sequence, where N is the final time. Specifically, the future processing parameter hypothesis sequence... The generation method adopts a traversal combination approach, that is, to obtain all future processing parameters that meet the processing requirements at future moments, and to perform traversal combination to obtain multiple different future processing parameter hypothesis sequences.
[0114] The generation process of the above-mentioned future processing parameter hypothesis sequence is illustrated by a simulation example:
[0115] Suppose we need to generate a sequence of future processing parameters for the next three time points. The initial processing parameter is A. The future processing parameters that meet the processing requirements at time 1 are B1 and B2. The future processing parameters that meet the processing requirements at time 2 are C1, C2, and C3. The future processing parameters that meet the processing requirements at time 3 are D1 and D2. Then the generated sequences of future processing parameters are: [A, B1, C1, D1], [A, B1, C1, D2], [A, B1, C2, D1], [A, B1, C2, D2], [A, B1, C3, D1], [A, B1, C3, D2], [A, B2, C1, D1], [A, B2, C1, D2], [A, B2, C2, D1], [A, B2, C2, D2], [A, B2, C3, D1], [A, B2, C3, D2].
[0116] Multiple different sequences of future processing parameters The inputs are fed in parallel into a pre-trained flavor transport dynamics model F, and the model outputs a corresponding future state trajectory for each input sequence. , ,in, The feature vector of the current state Let the quality vector be the initial state. The feature vector of the current state Given the initial state, the future processing parameters are assumed to be the quality vector at time n in the sequence;
[0117] Calculate the Euclidean distance between the endpoint feature vector of each predicted trajectory and the target finished product flavor feature vector, and simultaneously calculate the estimated total energy consumption for executing the sequence of assumed parameters.
[0118] The method for calculating Euclidean distance is as follows:
[0119] ;
[0120] The Euclidean distance between the endpoint feature vector and the target finished product flavor feature vector is given. The set of element types in the feature vector. Let u be the type of element in the feature vector. The feature vector of the current state Given the initial state, the future processing parameters are assumed to be the value of the u-th element in the quality vector at the final time in the sequence. Let u be the value of the type of element u in the target finished product flavor feature vector;
[0121] The formula for assessing total energy consumption is:
[0122] ;
[0123] This is the predicted and assessed value of total energy consumption. Let n be the heating power at time n in the sequence. Let n be the stirring speed at the nth time in the sequence. The time interval between adjacent moments in the sequence. , The conversion factor is determined by the energy efficiency attribute of the frying equipment itself;
[0124] The weighted summation method or Pareto front method is used to perform multi-objective optimization on the trajectory's endpoint proximity and the estimated total energy consumption, and the predicted trajectory with the highest comprehensive score is selected.
[0125] Among them, the weighted summation method is to normalize the calculated Euclidean distance and total energy consumption, and then perform weighted summation to transform multiple objectives into a single objective for optimization. The trajectory with the smallest weighted summation value is selected and recorded as the predicted trajectory with the highest comprehensive score.
[0126] The Pareto front method selects the minimum Euclidean distance and the minimum total energy consumption to obtain the ideal target solution. Then, it analyzes the Euclidean distance and total energy consumption of all predicted trajectories to obtain at least one Pareto front solution. The method for determining the Pareto front solution is to determine whether there are other predicted trajectories whose Euclidean distance and total energy consumption are both less than those of the current predicted trajectory. If so, the Euclidean distance and total energy consumption of the current predicted trajectory do not constitute a Pareto front solution; otherwise, the Euclidean distance and total energy consumption of the current predicted trajectory constitute a Pareto front solution. After normalizing the Euclidean distance and total energy consumption, the Euclidean distance between each Pareto front solution and the ideal target solution is calculated. The Pareto front solution corresponding to the minimum Euclidean distance of the ideal target solution is selected as the optimal solution. The trajectory corresponding to the optimal solution is recorded as the predicted trajectory with the highest comprehensive score.
[0127] The first control command in the processing parameter hypothesis sequence corresponding to the predicted trajectory with the highest comprehensive score is output as the optimal processing parameter adjustment command for the current control cycle.
[0128] The entire optimal control sequence is not executed for the selected optimal trajectory. Instead, it only takes control commands. The optimal adjustment command for the current control cycle is issued to the actuator. At the start of the next control cycle, the system reads the new actual state measurements to update the initial state, and then repeats the above steps, re-predicting and optimizing at the new time starting point. Through this rolling approach, the system can continuously utilize the latest feedback information to overcome model errors and external disturbances, achieving stable and adaptive closed-loop control.
[0129] The system executes the optimal processing parameter adjustment command, drives the heating, stirring and auxiliary material addition mechanisms to change the frying process, and continuously collects multi-dimensional real-time parameters reflecting the physical state, chemical composition and sensory simulation attributes of the base material to form a closed-loop adaptive adjustment until the distance between the current state feature vector and the target finished product flavor feature vector is less than the preset threshold, triggering the finished product unloading command.
[0130] The optimal processing parameter adjustment instruction specifically includes control instructions for at least one of the following actuators:
[0131] Power or temperature adjustment commands for heating components are used to precisely control the heat source input;
[0132] The speed and direction adjustment commands of the stirring device are used to control the heat and mass transfer efficiency and prevent scorching.
[0133] The opening and closing timing and flow command of the automatic dispensing valve for liquid or powdered excipients are used to accurately dispense materials at key process points.
[0134] The adaptive adjustment method for hot pot base processing parameters based on multi-parameter feedback proposed in this embodiment, by integrating multi-source sensing, mechanistic models, and intelligent decision-making, achieves a shift from lagging, single-point control to forward-looking, global optimization of the hot pot base cooking process. This reduces absolute reliance on human experience, and through real-time multi-parameter feedback and predictive optimization based on flavor transfer kinetics models, ensures high stability and standardization of flavor and quality across different batches of products, significantly improving the yield of high-quality products while reducing energy consumption and raw material loss.
[0135] Example 2:
[0136] This embodiment, based on Embodiment 1, further introduces a formulation process adaptive migration step, applied to rapidly generate initial process parameters for new formulations, referring to... Figure 5 As shown, the adaptive migration steps of the formulation process include:
[0137] The various ingredients in the new formula are quantified into a multi-dimensional formula vector based on their chemical composition and flavor contribution.
[0138] The formula vector of the new formula is input into the flavor transfer kinetics model, which then matches and interpolates it with a known library of mature formulas. Combined with the learned physicochemical laws, a series of baseline processing parameters suitable for the initial optimization of the new formula is derived.
[0139] Using the baseline processing parameter sequence as the initial setting, a closed-loop adaptive adjustment process is initiated to achieve rapid online fine-tuning and optimization of the preliminary process.
[0140] When a new recipe is introduced, the system parses it into structured data. Then, based on the raw material database, each raw material in the recipe is converted into a multi-dimensional digital recipe vector according to its feeding ratio. The dimensions of the digital recipe vector are preset. In some preferred embodiments, the dimensions of the digital recipe vector are [total oil content, capsaicin equivalent concentration, salinomycin concentration, total sugar content, reducing sugar ratio, estimated initial concentration of characteristic aroma components, moisture content, average particle size, salt concentration]. This transforms the unstructured recipe text into a standardized mathematical expression that the model can understand and calculate.
[0141] This solution incorporates a mature formula-high-quality process database, which is built through a self-learning optimization mechanism. The specific process is as follows:
[0142] The complete process data chain, including real-time parameters of multiple dimensions of time sequence, the sequence of processing parameter adjustment instructions executed, and the final product quality evaluation, will be stored in the historical process database.
[0143] The expanded historical process database is used regularly to incrementally learn and fine-tune the parameters of the flavor transfer kinetics model and the model prediction controller.
[0144] Based on long-term accumulated finished product quality data and market feedback, the flavor feature vectors of target finished products corresponding to different flavor types are optimized and calibrated in reverse to continuously track and adapt to changes in consumer preferences.
[0145] The Euclidean distance between the digital recipe vector of the new recipe and all mature recipe vectors in the database is calculated, and at least one mature recipe vector similar to the digital recipe vector of the new recipe is selected based on the Euclidean distance.
[0146] Using the process notification sequence of each similar mature recipe vector as a candidate initial control sequence, and the new recipe as model input, the model is used to quickly simulate and predict the theoretical endpoint quality after executing the candidate initial control sequence.
[0147] The process of selecting the mature formula vector that is closest to the theoretical endpoint quality and the target quality of the new formula is used as the basic template;
[0148] Based on the basic template, with the target quality of the new formula as the optimization target, the multi-dimensional real-time parameters reflecting the physical state, chemical composition and sensory simulation properties of the base material are collected simultaneously during the frying process in Example 1. The multi-dimensional real-time parameters include physical temperature, apparent color, material viscosity and concentration of key volatile flavor substances.
[0149] Multidimensional real-time parameters are processed into current state feature vectors and input into a pre-trained flavor transfer dynamics model. The flavor transfer dynamics model, combined with the quantified formula information, predicts the change trajectory of key flavor substance concentrations and comprehensive quality indicators within a preset time period in the future.
[0150] Based on the prediction of multiple future trajectories by the flavor transfer kinetics model, a model predictive control algorithm is adopted. The optimization objective is to be closest to the flavor feature vector of the target finished product and with the lowest energy consumption. The algorithm continuously calculates and outputs the optimal processing parameter adjustment command within the current control cycle.
[0151] The system executes the optimal processing parameter adjustment command, drives the heating, stirring and auxiliary material addition mechanisms to change the frying process, and continuously collects multi-dimensional real-time parameters reflecting the physical state, chemical composition and sensory simulation attributes of the base material to form a closed-loop adaptive adjustment until the distance between the current state feature vector and the target finished product flavor feature vector is less than a preset threshold, triggering the finished product unloading command frying step, and adding the process parameters after each frying to the historical process database, and performing further optimization calculations in the subsequent new formula frying process.
[0152] To further achieve data interoperability across the entire enterprise process chain and optimize the aforementioned adaptive migration steps of the formulation process, this embodiment also proposes a data sharing mechanism. This mechanism interconnects with the enterprise's Manufacturing Execution System (MES) and Product Lifecycle Management System (PWMS), specifically including:
[0153] The system receives specific flavor base production orders from the MES (Management Execution System), automatically loads the corresponding target flavor type, formula, and process parameter package, and connects to the enterprise's MES system via a standard API or middleware. Both parties define a unified data exchange protocol, and the host computer receives production orders from the MES. Based on the product code and formula version number in the order, the system automatically retrieves and loads the corresponding verified process parameter package from its own process knowledge base or PLM (Production Management System). This parameter package is a structured file containing a complete target finished product flavor feature vector, a quantified formula vector, and a sequence of baseline processing parameters for initial settings. The system is then ready; after operator confirmation, production can begin, achieving seamless automated integration from order to production.
[0154] By binding the end-to-end parameters of each processing step with the corresponding production batch number, an immutable digital process file is created and stored in a blockchain or central database. During production, the adaptive adjustment system uses the production batch number issued by the MES as a unique identifier. All real-time collected multi-dimensional parameters, intermediate state feature vectors predicted by the model, actual control command sequences output by MPC rolling optimization, and the final system-determined unloading trigger signal are strictly aligned and packaged in terms of timestamps. This associated data is then further correlated with the online or offline quality inspection results of the final product, such as rapid testing of physicochemical indicators and sampling sensory scores, to form a complete and traceable digital process file. To ensure data immutability and enhance the credibility of quality traceability, the system writes key summary information of each batch of products' digital process file, such as batch number, hash value, and timestamp, into the enterprise's internal permissioned blockchain nodes. The original complete data can be stored in a highly available centralized database. The blockchain's notarization ensures the authenticity and non-repudiation of the process file, providing a reliable data foundation for quality auditing and root cause analysis.
[0155] Validated successful process parameter packages for new formulations are automatically pushed to the PLM system, forming a standardized process knowledge base asset for the enterprise, which can be used by R&D and production departments. When a batch of production is completed and the finished product of that batch is determined by the quality system to meet the standards, the system automatically triggers the knowledge capture process. The system marks the corresponding quantitative formulation vector and the final optimized and stable actual control instruction sequence of that batch as a new successful process parameter package. Through the integration interface with the enterprise PLM system, the newly generated successful process parameter package, along with its associated quantitative formulation vector and quality evaluation results, is pushed as a new process instance into the standardized process knowledge base of the PLM system.
[0156] The cross-formula process self-generation and self-learning capability proposed in this embodiment can intelligently migrate mature processes to new formulations, greatly shorten the new product development cycle, and continuously accumulate process data assets, thus building an iterative and reusable core process knowledge system for enterprises and realizing the flexible and intelligent upgrading of production.
[0157] Example 3:
[0158] Based on Example 1, this embodiment proposes an application calibration mechanism to adapt to frying equipment of different sizes, specifically including:
[0159] Through a limited number of calibration experiments, a correlation model was established between the critical transfer coefficient and the volume, pot shape, and heating area of the frying equipment.
[0160] The correlation model was used as a compensation factor to scale and adjust the parameters in the flavor transfer kinetics model that are significantly affected by equipment size.
[0161] Apply the adjusted model to the equipment with a new scale, so that the core control strategy can be maintained, and rapid replication and consistent amplification of the process under different production scales can be achieved.
[0162] First, select a piece of equipment as the reference equipment, on which the flavor transfer kinetic model has been fully trained and verified. Then, design and execute a set of calibration experiments with a limited number of times for the target scaled-up equipment. The specific experimental steps are as follows:
[0163] On the reference equipment and the scaled-up equipment, use exactly the same formula and initial raw materials. On the reference equipment, run its optimized standard process. On the scaled-up equipment, start with the preliminary process parameters scaled according to traditional experience, such as equal power-volume ratio, but allow the adaptive adjustment steps in Example 1 to run, in order to record the sequence of control parameters actually dynamically adjusted by the system to achieve the same or approximate finished product flavor target as the reference equipment;
[0164] On both pieces of equipment, synchronously collect complete data during the frying process. Focus on analyzing the dynamic parameters that can directly reflect the transfer process, such as: standardized heating curve, moisture evaporation rate, dissolution kinetic curve of specific flavor substances, such as capsaicin. From these data, extract quantifiable key transfer coefficients, and based on the similarity principle, establish an association model between the key transfer coefficients and the characteristic dimensions of the equipment. After obtaining the scale association model, use it as a compensation factor to specifically adjust the internal parameters in the flavor transfer kinetic model applied to the scaled-up equipment that are significantly affected by the equipment scale. Deploy the flavor transfer kinetic model after the above parameter scaling adjustment to the adaptive adjustment system supporting the target scaled-up equipment.
[0165] In summary, the advantages of the present invention are as follows: It solves the problem of quality standardization in the industrial production of hot pot bases, and ensures the ultimate stability and consistency of product quality through forward-looking optimization instead of lagging adjustment; the flavor transfer kinetic model realizes process understanding, combined with the formula process migration and the equipment scale adaptive mechanism, which can greatly shorten the process development cycle of new products and new production lines, reduce the trial-and-error cost, and form digital process knowledge assets that can be precipitated and reused, achieving the leap from automated production to intelligent innovation for enterprises.
[0166] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art of this industry should understand that the present invention is not limited by the above embodiments. What is described in the above embodiments and the specification is only the principle of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed. The scope of protection required by the present invention is defined by the appended claims and their equivalents.
Claims
1. A method for adaptive adjustment of hot pot base processing parameters based on multi-parameter feedback, characterized in that, include: Based on the flavor type of the target hot pot base, load the corresponding flavor feature vector of the target finished product, and set the initial processing control parameters based on the basic formula; During the stir-frying process, multi-dimensional real-time parameters, including physical temperature, apparent color, material viscosity, and concentration of key volatile flavor compounds, are collected simultaneously. The multi-dimensional real-time parameters are processed into current state feature vectors and input into a pre-trained flavor transfer dynamics model. The flavor transfer dynamics model, combined with the quantified formula information, predicts the change trajectory of key flavor substance concentrations and comprehensive quality indicators within a preset time period. The flavor transfer dynamics model adopts a physical information neural network structure, which is designed to obey constraints based on mass conservation or Arrhenius relation constraints. Based on the various future trajectories predicted by the flavor transfer kinetics model, a model predictive control algorithm is adopted. The optimization objective is to be closest to the target finished product flavor feature vector and with the lowest energy consumption. The algorithm continuously calculates and outputs the optimal processing parameter adjustment command within the current control cycle. The optimal processing parameter adjustment command is executed to drive the heating, stirring and auxiliary material addition mechanisms to change the frying process. Multi-dimensional real-time parameters reflecting the physical state, chemical composition and sensory simulation attributes of the base material are continuously collected to form a closed-loop adaptive adjustment until the distance between the current state feature vector and the target finished product flavor feature vector is less than a preset threshold, triggering the finished product unloading command.
2. The adaptive adjustment method for hot pot base processing parameters based on multi-parameter feedback according to claim 1, characterized in that, The steps for constructing the pre-trained flavor transport dynamics model include: In historical frying experiments, complete time-series multi-dimensional real-time parameters, corresponding precise processing control parameter sequences, and expert sensory scores and physicochemical test data of the finished product base were collected simultaneously. A model training objective function is constructed. The model training objective function not only requires that the error between the model prediction value and the actual detection data be minimized, but also requires that its prediction law conforms to the embedded mass and heat transfer physicochemical constraints. The constraints include the mass conservation equation and the reaction kinetic relationship based on the Arrhenius equation. By employing a physical information neural network structure, the model is trained using collected data and defined constraints. This enables the model to not only fit the data with high accuracy but also learn the inherent physicochemical laws in the base material processing, thus obtaining an interpretable flavor transfer kinetic model.
3. The adaptive adjustment method for hot pot base processing parameters based on multi-parameter feedback according to claim 2, characterized in that, The process of converting multi-dimensional real-time parameters into a current state feature vector specifically includes: The raw sensor signals are preprocessed by filtering, denoising, and standardization. From the preprocessed time series data, feature values with process indication significance are extracted. These feature values include temperature change rate, color change rate, viscosity curve inflection point, instantaneous value of specific volatile flavor substance concentration and its first derivative. The extracted feature values are combined in sequence to form a comprehensive current state feature vector that characterizes the current frying process.
4. The adaptive adjustment method for hot pot base processing parameters based on multi-parameter feedback according to claim 3, characterized in that, Among the multi-dimensional real-time parameters, the physical temperature is monitored by a distributed temperature sensor array, which includes at least a first, second and third sensor for monitoring the temperature of the bottom of the pot, the temperature of the center of the material, and the steam temperature at the mouth of the pot. The apparent color is monitored by a hyperspectral camera or an industrial color camera and converted into a standard Lab color space value; The viscosity of the material is obtained by an online rotational viscometer or indirectly by means of torque measurement; The concentrations of the key volatile flavor compounds are monitored online using a near-infrared spectroscopy probe or a mass spectrometry-type electronic nose.
5. The adaptive adjustment method for hot pot base processing parameters based on multi-parameter feedback according to claim 4, characterized in that, The specific instructions for adjusting the optimal processing parameters, which aim to achieve the closest flavor feature vector to the target finished product while minimizing energy consumption, include: Starting with the current state feature vector, and combining multiple different future processing parameter assumption sequences, the flavor transfer kinetics model generates multiple corresponding future quality change trajectories in parallel prediction. Calculate the Euclidean distance between the endpoint feature vector of each predicted trajectory and the target finished product flavor feature vector, and simultaneously calculate the estimated total energy consumption for executing the sequence of assumed parameters. The weighted summation method or Pareto front method is used to perform multi-objective optimization on the trajectory's endpoint proximity and the estimated total energy consumption, and the predicted trajectory with the highest comprehensive score is selected. The first control instruction in the processing parameter hypothesis sequence corresponding to the predicted trajectory with the highest comprehensive score is output as the optimal processing parameter adjustment instruction for the current control cycle.
6. The method for adaptive adjustment of hot pot base processing parameters based on multi-parameter feedback according to any one of claims 1-5, characterized in that, It also includes a formulation process adaptive migration step, applied to quickly generate initial process parameters for new formulations, the formulation process adaptive migration step including: The various ingredients in the new formula are quantified into a multi-dimensional formula vector based on their chemical composition and flavor contribution. The formula vector of the new formula is input into the flavor transfer kinetics model, which then matches and interpolates it with a known library of mature formulas. Combined with the learned physicochemical laws, a series of baseline processing parameters suitable for the initial optimization of the new formula is derived. Using the aforementioned benchmark processing parameter sequence as the initial setting, a closed-loop adaptive adjustment process is initiated to achieve rapid online fine-tuning and optimization of the preliminary process.
7. The adaptive adjustment method for hot pot base processing parameters based on multi-parameter feedback according to claim 1, characterized in that, The optimal processing parameter adjustment command specifically includes control commands for at least one of the following actuators: Power or temperature adjustment commands for heating components are used to precisely control the heat source input; The speed and direction adjustment commands of the stirring device are used to control the heat and mass transfer efficiency and prevent scorching. The opening and closing timing and flow command of the automatic dispensing valve for liquid or powdered excipients are used to accurately dispense materials at key process points.
8. The adaptive adjustment method for hot pot base processing parameters based on multi-parameter feedback according to claim 2, characterized in that, The physical information neural network structure includes a physical calculation module for learning physical constraints. This module imposes constraints on the output of the intermediate layers of the model, such that the changes in flavor substance concentrations predicted by the model satisfy the following relationship: The total amount of flavor compounds distributed in the oil phase, aqueous phase and solid phase remains unchanged; The rates of the Maillard reaction and caramelization reaction are positively correlated with temperature in a non-linear manner.
9. The method for adaptive adjustment of hot pot base processing parameters based on multi-parameter feedback according to claim 1, characterized in that, The step of introducing active parameter perturbation during the stir-frying process, which involves simultaneously collecting multi-dimensional real-time parameters reflecting the physical state, chemical composition, and sensory simulation properties of the base material, specifically includes: Apply a brief, small change to the stirring speed periodically or as needed; Simultaneously monitor the dynamic response of viscosity, temperature, and volatile substance signals under this disturbance; The characteristics of the dynamic response are used as additional inputs to evaluate the real-time rheological properties of the substrate and the state of the reaction system, thereby enhancing the information dimension of the current state feature vector and the robustness of the control system.
10. The method for adaptive adjustment of hot pot base processing parameters based on multi-parameter feedback according to claim 1, characterized in that, It also includes a self-learning optimization mechanism, which performs the following steps after each cooking process: The complete process data chain, including real-time parameters of multiple dimensions of time sequence, the sequence of processing parameter adjustment instructions executed, and the final product quality evaluation, will be stored in the historical process database. The expanded historical process database is used periodically to incrementally learn and fine-tune the parameters of the flavor transfer kinetics model and the model prediction controller. Based on long-term accumulated finished product quality data and market feedback, the flavor feature vectors of the target finished products corresponding to different flavor types are optimized and calibrated in reverse to continuously track and adapt to changes in consumer preferences.
11. The method for adaptive adjustment of hot pot base processing parameters based on multi-parameter feedback according to claim 1, characterized in that, It also includes an application calibration step, which is used to adapt to frying equipment of different sizes, specifically including: Through a limited number of calibration experiments, a correlation model was established between the critical transfer coefficient and the volume, pot shape, and heating area of the frying equipment. The correlation model is used as a compensation factor to scale and adjust the parameters in the flavor transfer kinetics model that are significantly affected by equipment size. Applying the adjusted model to new-scale equipment allows the core control strategy to be maintained, enabling rapid replication and consistent scaling up of the process at different production scales.
12. The method for adaptive adjustment of hot pot base processing parameters based on multi-parameter feedback according to claim 1, characterized in that, It also includes a data sharing mechanism that interconnects with the enterprise's manufacturing execution system and product lifecycle management system, specifically including: Receive specific flavor base production orders from MES and automatically load the corresponding target flavor type, formula and process parameter package; The parameters of each processing step are linked to the corresponding production batch number to form an immutable digital process file, which is stored in a blockchain or central database. Validated and successful process parameter packages for new formulations are automatically pushed to the PLM system, forming a standardized process knowledge base asset for the enterprise to use by the R&D and production departments.