Intelligent production equipment control system for seasoning based on multi-source sensor fusion

Abstract

This invention discloses a control system for intelligent seasoning production equipment based on multi-source sensor fusion, comprising: a data identification module for collecting production line operation data, binding process batch identifiers and equipment node identifiers, normalizing the data, and outputting a standardized sensor dataset; a dynamic coupling module for segmenting the standardized sensor dataset by batch and equipment, obtaining the dynamic coupling weight of a single segment based on process deviation using a grey relational model, and generating a dynamic coupling weight set by associating the identifier with the storage address; and a reference acquisition module for constructing historical operating condition feature vectors, matching the dynamic coupling weight set using a cosine similarity algorithm, filtering to obtain a historical optimal strategy reference set, and filtering key control nodes based on the fusion of the dynamic coupling weight set and the historical optimal strategy reference set, combined with a preset response threshold, and retrieving process constraint parameters to generate a key equipment constraint parameter set.

Description

Technical Field

[0001] This invention relates to the field of next-generation information technology, and in particular to a control system for intelligent seasoning production equipment based on multi-source sensor fusion. Background Technology

[0002] With the increasing automation in the food industry, seasoning production is gradually transitioning from manual experience-based control to automated equipment control. In early production models, equipment parameters in each stage operated with fixed preset values, relying solely on manual inspection and adjustment. This resulted in low control precision, slow response, and susceptibility to fluctuations in raw material batches and changes in environmental temperature and humidity, leading to poor quality control stability of the finished product. Subsequently, control systems based on single-equipment closed-loop control were introduced into production scenarios. By collecting operational data from individual devices, local parameter adjustments were achieved, improving equipment stability to some extent. However, the inter-equipment linkage was poor, making it difficult to meet the overall optimization needs of multi-stage coupled processes.

[0003] The control strategies of existing seasoning production control systems rely on historical experience data or fixed algorithms, without fully integrating real-time operating data for dynamic adjustment. This makes it difficult to adapt to batch differences in raw materials and process fluctuations, and is prone to control lag or over-adjustment. Furthermore, the parameter optimization process does not simultaneously consider equipment process constraints, safe operating boundaries, and energy consumption indicators, but only pursues quality or efficiency, which can easily lead to equipment overload or energy waste. As a result, the process coupling relationship between equipment is not fully quantified, making it difficult to accurately assess the impact of single equipment parameter fluctuations on the quality of finished products, and the identification of key control nodes is inaccurate. Therefore, this paper proposes a control system for intelligent seasoning production equipment based on multi-source sensor fusion to solve the above problems. Summary of the Invention

[0004] To overcome the aforementioned technical deficiencies and achieve the above objectives, the present invention proposes the following technical solution: A multi-source sensor fusion-based intelligent seasoning production equipment control system includes: Data Identification Module: Used to collect production line operation data, bind process batch identifiers and equipment node identifiers, normalize the data, and output standardized sensor datasets; Dynamic coupling module: used to segment standardized sensor datasets by batch and device, obtain dynamic coupling weights for each segment based on process deviations using a grey relational model, and generate dynamic coupling weight sets by associating identifiers with storage addresses; Reference acquisition module: used to construct historical operating condition feature vectors, match the dynamic coupling weight set through the cosine similarity algorithm, filter to obtain the historical optimal strategy reference set, and filter key control nodes based on the fusion of the dynamic coupling weight set and the historical optimal strategy reference set, combined with the preset response threshold, and retrieve process constraint parameters to generate key equipment constraint parameter set; Strategy optimization module: retrieves the adjustable range of each device within the key control nodes to obtain the set of adjustable ranges, merges the set of key device constraint parameters with the set of adjustable ranges, and obtains the optimal combination of control parameters for each device through a particle swarm multi-objective optimization algorithm with adaptive adjustment of inertial weights; Control and execution module: Based on the batch identifier parsed by the central control request, the optimal control parameters for a single scenario are retrieved, the corresponding parameters are executed and controlled by the PLC, thus achieving intelligent control.

[0005] The process of outputting a standardized sensor dataset includes: The production line operation data is obtained by collecting operating values ​​from each section using a sensor array and summarizing all collected values. The built-in batch code is retrieved as the process batch identifier, and the inherent hardware number of the equipment body is retrieved as the equipment node identifier. The production line operation data is bound to the process batch identifier and the equipment node identifier one by one through the database field binding technology to obtain the original identifier data. The minimum-maximum normalization algorithm is used to scale the labeled original data to a uniform range, and outliers are removed based on the 3σ gross error to obtain a standardized sensor dataset.

[0006] The process of obtaining the dynamic coupling weight of a single segment includes: The standardized sensor dataset is fragmented and grouped into equipment fragments using process batch identifiers as the primary classification criteria and equipment node identifiers as the secondary classification criteria. The difference between the actual process parameters and the factory standard process parameters within the segmented group data of the equipment is calculated using the difference operation, and all difference values ​​are collected to obtain the segmented process deviation parameters. The segmentation process deviation parameters are imported into a preset grey relational degree calculation model to obtain the dynamic coupling weight of a single segment. Read the preset storage address code of the server disk partition, bind and archive the dynamic coupling weight of a single shard, the corresponding process batch identifier, the equipment node identifier, and the storage address through multi-field association storage in the data table, and integrate the archived content of the entire shard to obtain the dynamic coupling weight set.

[0007] The construction process of the grey relational analysis model includes: The mainstream process parameters of the seasoning production line were selected as the model evaluation index. The measured values ​​of the indexes from multiple batches of historical production were collected, and all measured values ​​were aggregated to obtain the original sample series of the model. Ideal process data is preset as a reference baseline sequence. The absolute difference between the original sample sequence of the model and the reference baseline sequence is calculated. All difference results are collected to obtain the sequence difference matrix. The grey relational coefficient formula is calculated point by point based on the sequence difference matrix to obtain the single-point grey relational coefficient. The average value of all single-point grey relational coefficients corresponding to the same group of indicators is calculated, and the average value calculation formula, sample input port, and result output port are uniformly encapsulated to complete the construction of the grey relational degree calculation model.

[0008] The process of obtaining the historical optimal strategy reference set includes: Retrieve the three types of original records stored in the historical operating condition database: historical sensing, control parameters, and finished product quality inspection. Use the dimensional feature extraction method to extract the key feature components of each batch of data. Summarize the key feature components of all batches to obtain the historical operating condition feature vector. The weight values ​​in the dynamic coupling weight set are organized into real-time feature vectors of the same dimension. The cosine similarity calculation formula is used to substitute the historical working condition feature vectors one by one to complete the numerical calculation and obtain a sequence of similarity results. Based on the sequence of similarity results, historical control records corresponding to those higher than the system's preset matching threshold are selected, and all selected records are merged to obtain a historical optimal strategy reference set.

[0009] The process of obtaining the set of constraint parameters for key equipment includes: Based on the unified device node identifier field in the dynamic coupling weight set and the historical best strategy reference set, all control parameters bound to the identifier in the dynamic coupling weight set and the historical best strategy reference set are collected by filtering by the database primary key and grouped by a single device node identifier to obtain a subset of single device control parameters. Based on the subset of single-device control parameters matched with the finished product inspection results data bound to the same batch, the contribution ratio of parameter fluctuation to quality fluctuation is calculated using the Pearson correlation coefficient and converted into a percentage to obtain the quality influence coefficient of each node. Based on the comparison between the quality impact coefficient of each node and the preset response threshold, the device numbers whose coefficients exceed the threshold are filtered to form a list, thus obtaining the list of key control nodes. Based on the list of key control nodes, the pre-stored upper and lower limit process parameters and safe operation limit parameters of each equipment are read, and all read parameters are summarized to obtain the set of key equipment constraint parameters.

[0010] The process of obtaining the optimal control parameters for a single scenario includes: Based on the equipment's factory calibration and historical process test data, the upper and lower safety limits and dynamic adjustment margins of four parameters—temperature, speed, pressure, and feed rate—are extracted to construct a set of adjustable ranges for the equipment with adjustment margins. The set of key equipment constraint parameters is used as the hard constraint boundary for solving, and the set of adjustable equipment ranges is used as the parameter value boundary. A particle swarm optimization algorithm with adaptive adjustment of inertial weight is adopted, with the product quality control pass rate and unit energy consumption ratio as the optimization objectives. The particle position and velocity are iteratively updated to obtain the optimal control parameters for a single scenario.

[0011] The implementation process of the particle swarm optimization algorithm with adaptive adjustment of inertia weights includes: Obtain adjustable boundaries with margin: ; in, The upper limit of the safe adjustable value for the j-th parameter. This is the lower limit of the safety adjustable value for the j-th parameter. For the safety dynamic adjustment margin of the j-th parameter, Indicates the rated upper limit value. Indicates the lower limit value; Based on the standard particle swarm optimization algorithm, the inertia weight is linearly and adaptively adjusted, as expressed by the formula: ; Where t represents the current iteration number, and T represents the preset maximum iteration number. This represents the inertia weight in the t-th iteration. This represents the initial maximum inertia weight. This represents the minimum inertia weight in the later stages of the iteration.

[0012] The process of achieving intelligent control includes: The request message fields issued by the central control are split by TCP segmentation parsing, and the embedded encoded content in the request message fields is extracted to obtain the batch identifier to be retrieved; Based on the batch identifier to be searched, the parameter content corresponding to the optimal control parameter of a single scenario is retrieved to obtain the target control parameter; Based on the target control parameters, the system is converted and sent to the corresponding PLC controller to drive the equipment to perform actions, thereby obtaining the actual control actions on site.

[0013] The present invention has the following beneficial effects: 1. Based on the process deviation and grey relational model, the dynamic coupling relationship between equipment is quantified, and key control nodes that have a significant impact on the quality of finished products are accurately identified, which improves the pertinence and effectiveness of control and reduces ineffective control operations; 2. By integrating real-time dynamic coupling weights and historical optimal control strategies, and using cosine similarity matching to select historical solutions suitable for the operating conditions, a reference basis is provided for real-time control, which improves the adaptability and response speed of the control strategy and reduces the impact of process fluctuations on product quality. 3. Using the set of key equipment constraint parameters as hard boundaries and the set of adjustable equipment ranges as soft boundaries, a particle swarm optimization algorithm with adaptive adjustment of inertial weights is used to solve the optimal combination of control parameters that balances the quality control pass rate and the unit energy consumption ratio. This ensures the safety of equipment operation and achieves synergistic optimization of quality and energy consumption. Attached Figure Description

[0014] Figure 1 This is a system block diagram of the intelligent seasoning production equipment control system based on multi-source sensor fusion proposed in this invention. Detailed Implementation

[0015] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0016] Example 1 Figure 1 As shown, the intelligent seasoning production equipment control system based on multi-source sensor fusion proposed in this invention includes: Data Identification Module: Used to collect production line operation data, bind process batch identifiers and equipment node identifiers, normalize the data, and output standardized sensor datasets; First, deploy sensor arrays to collect production line operation data: In the seasoning production process, temperature, humidity, speed, pressure and flow sensors are set up to form a sensor group in each section, including feeding, mixing, maturation and sterilization, filling and packaging. The real-time operating data of each section is continuously collected using the industrial common timed polling method. All collected data are summarized and stored in the local database to obtain the production line operation data. Then obtain the original identifier data: The batch code built into the production management system is retrieved as the process batch identifier, and the hardware number fixed to the body of each production equipment at the factory is retrieved as the equipment node identifier. Database field binding technology is used to bind each piece of production line operation data with the process batch identifier and the equipment node identifier to obtain the original identifier data. Obtain a standardized sensor dataset: The original identification data is scaled using a min-max normalization algorithm to eliminate differences in the units of measurement between different sensor data. The calculation formula is as follows: ; Where x represents a single original sensor value within the original data. It is the minimum value of the same type of sensor data. The maximum value of the same type of sensor data. The normalized data has a fixed value range of [0,1]. based on The gross error elimination algorithm removes missing and abrupt outlier data to obtain a standardized sensor dataset. For each category of data in the normalized dataset, the following steps are performed: Calculate the arithmetic mean of the normalized data. with standard deviation

[0017] Set the normal data range as Remove all data records that are outside the specified range; All valid data after removing outliers are aggregated to obtain a standardized sensor dataset.

[0018] Dynamic coupling module: used to segment standardized sensor datasets by batch and device, obtain dynamic coupling weights for each segment based on process deviations using a grey relational model, and generate dynamic coupling weight sets by associating identifiers with storage addresses; First, obtain the device fragmentation and grouping data: Generate device fragmented grouped data Using process batch identifier as the primary classification condition and equipment node identifier as the secondary classification condition, the standardized sensor dataset is segmented and aggregated. All sensor data from the same batch and the same equipment are divided into an independent data segment, and the contents of all segments are summarized to obtain the equipment segment grouped data. Specifically, equipment fragmented grouped data refers to the grouped data set obtained by fragmenting and aggregating a standardized sensor dataset based on process batch identifiers and equipment node identifiers as classification conditions; Then define the process deviations associated with the device sharding and grouping data: Segmentation process deviation parameters: refers to the set of differences between the actual process parameters and the factory standard process parameters in the equipment segmentation and grouping data; The difference calculation formula is used to calculate the difference between the actual process parameters and the factory standard process parameters in the equipment segmented and grouped data. The formula is as follows: ; in, The actual operating value of the i-th process parameter is taken from the equipment's segmented and grouped data. The factory standard value of the i-th process parameter is taken from the equipment's factory process document; The process deviation of the i-th process parameter is the difference between the actual value and the standard value. Collect all deviation values ​​to obtain the slicing process deviation parameters; Then, a grey relational degree calculation model is constructed: Obtain the original sample sequence of the model, specifically including: Mainstream process parameters such as temperature, rotation speed, pressure, and material moisture content in the seasoning production process were selected as model evaluation indicators. Multiple batches of historical production data were retrieved, and measured values ​​of each evaluation indicator were collected. All measured values ​​were summarized to obtain the original sample series of the model. Obtain the sequence difference matrix: A reference baseline sequence is formed by setting ideal operating values ​​for each evaluation indicator. This reference baseline sequence consists of the industry's optimal steady-state production parameters. The absolute difference between the original sample sequence of the model and the reference baseline sequence is calculated item by item. The calculation formula is as follows: ; in: To reference the value of the k-th indicator in the benchmark sequence, Let be the value of the k-th indicator in the original sample sequence of the i-th model group. This represents the absolute difference between the two sets of data under the corresponding indicator. Arrange all absolute differences in a row-column grid to obtain the sequence difference matrix; Then, the global maximum and global minimum values ​​within the sequence difference matrix are extracted, and combined with industry-standard resolution coefficients. (With a fixed value of 0.5), the standard grey relational coefficient formula is used for point-by-point calculation, as follows: ; in, The smallest absolute difference in the sequence difference matrix. The global maximum absolute difference of the sequence difference matrix. The resolving factor is used to adjust the distribution range of the correlation coefficient. is the single-point grey relational coefficient corresponding to the kth indicator, with a value range of [0,1]. The arithmetic mean r of all single-point grey relation coefficients corresponding to the same set of indicators is calculated, which is the grey relation degree corresponding to a single set of sample numbers. The average calculation logic, data input port, and result output port are all encapsulated into a program module, thus completing the overall construction of the grey relational degree calculation model. The model can directly access upstream data to complete automatic calculation. Will Organize the sequence into a sequence consistent with the dimensions of the model evaluation index, and input it into the grey relational degree calculation model; The model automatically performs internal calculations and outputs the grey relational degree of the device. ; Grey relational degree After normalization, the weights are mapped to the [0,1] interval to obtain the single-segment dynamic coupling weights of the device. ,Right now: ; In the formula: These represent the minimum and maximum values ​​of the grey relational degree for all devices within the same batch; Each device fragment corresponds to a single fragment dynamic coupling weight. That is, single-slice dynamic coupling weight; Obtain all the dynamic coupling weights of each single shard, as well as the process batch identifier (BID), device node identifier (DID), and the preset storage address code (Addr) of each shard's disk partition. For each device shard, the following information is bound to a single data record using multi-field association storage technology in data tables: Single-slice dynamic coupling weight The batch identifier (BID), the device identifier (DID), and the data storage address (Addr) are also included. All shard binding records are aggregated and archived to form a dynamic coupling weight set, denoted as W; Specifically, each record in W can be quickly retrieved using BID or DID, providing data support for subsequent historical similarity matching and regulation.

[0019] Reference acquisition module: used to construct historical operating condition feature vectors, match the dynamic coupling weight set through the cosine similarity algorithm, filter to obtain the historical optimal strategy reference set, and filter key control nodes based on the fusion of the dynamic coupling weight set and the historical optimal strategy reference set, combined with the preset response threshold, and retrieve process constraint parameters to generate key equipment constraint parameter set; First, obtain the historical operating condition feature vector: Retrieve three types of original records stored in the historical operating condition database: historical sensor data, historical equipment control parameters, and finished product quality inspection data. Use the dimensional feature extraction method to extract the key feature components of each batch of production data. Arrange the feature components in a fixed order to form a single batch historical operating condition feature vector. Summarize all single batch historical operating condition feature vectors to obtain the complete historical operating condition feature vector. Obtain the sequence of similarity results: The weight values ​​within the dynamic coupling weight set W are organized into a real-time feature vector A with the same dimension as the historical feature vector. Any historical feature vector within the historical operating condition feature vector is taken as B, and the cosine similarity formula is used to perform the calculation step by step. The formula is as follows: ; Where A is the real-time feature vector and B is the historical feature vector. For vector dot product, For the real-time feature vector magnitude, The length of the historical feature vector. Let be the cosine similarity between the two sets of vectors, with values ​​ranging from [-1, 1]. The calculation is completed by traversing the historical working condition feature vectors, resulting in a sequence of similarity results; Set the system's preset matching threshold In this embodiment The value is set to 0.8, and each similarity value in each sequence of similarity results is compared with the matching threshold. Compare and filter those with similarity scores greater than the matching threshold. The corresponding historical control records are merged, and all filtered historical control records are combined to obtain the historical optimal strategy reference set.

[0020] Obtain a subset of single-device control parameters Both the dynamic coupling weight set and the historical optimal strategy reference set are equipped with a unified device node identifier field. Using database primary key filtering technology, the device node identifier is used as the retrieval primary key to collect all control parameters that are centrally bound to the same device node identifier in the two types of datasets, thus obtaining a subset of single device control parameters. Then obtain the quality impact coefficient of each node: A subset of single-equipment control parameters is correlated and matched with the finished product quality inspection results data bound to the same batch. The Pearson correlation coefficient is used to calculate the linear correlation between equipment parameter fluctuations and finished product quality fluctuations. The calculated Pearson correlation coefficient is then used to... The contribution percentage is converted into a percentage form to obtain the quality impact coefficient of each node. The Pearson correlation coefficient ranges from -1 to 1, with a larger absolute value indicating a higher degree of correlation between the two sets of data. Obtain a list of key regulatory nodes: Set preset response threshold In this embodiment, The quality impact coefficient of each node is compared with the response threshold. Item by item was compared, and those with coefficient values ​​greater than the response threshold were selected. The equipment numbers are combined to form a list, resulting in a list of key control nodes; Obtain the set of constraint parameters for key equipment: Based on the list of key control nodes, the upper limit of the corresponding equipment process parameters, the lower limit of the process parameters, and the equipment safe operation limit parameters are read from the pre-stored system database. All read parameters are summarized to obtain the set of key equipment constraint parameters. Specifically, the upper limit of the corresponding equipment process parameters, the lower limit of the process parameters, and the limit parameters for safe operation of the equipment are obtained directly through historical data statistics.

[0021] Strategy optimization module: retrieves the adjustable range of each device within the key control nodes to obtain the set of adjustable ranges, merges the set of key device constraint parameters with the set of adjustable ranges, and obtains the optimal combination of control parameters for each device through a particle swarm multi-objective optimization algorithm with adaptive adjustment of inertial weights; Retrieve the equipment's factory calibration documents and records of all previous process tests, and extract the rated upper limits of four parameters: temperature, rotational speed, pressure, and feed rate. Rated lower limit value and the safety dynamic adjustment margin determined by the equipment factory test. ; Specifically, safety dynamic adjustment margin This refers to the range of parameter fluctuations that can be safely adjusted temporarily and for a short period, beyond the upper and lower limits of the rated process parameters specified at the factory. This range is determined by the redundancy of the equipment's mechanical structure, drive system, and control system. During factory testing, without triggering hardware protection or causing irreversible damage, the maximum percentage by which each process parameter can safely deviate from its rated upper and lower limits is considered safe. That is, a maximum deviation of 10% from the rated value is allowed; Then obtain the adjustable boundary with margin: ; in, The upper limit of the safe adjustable value for the j-th parameter. This is the lower limit of the safety adjustable value for the j-th parameter. This represents the safety dynamic adjustment margin for the j-th parameter; Then, based on the standard particle swarm optimization algorithm, the inertia weights are linearly and adaptively adjusted to balance global exploration and local convergence, avoiding the drawback of fixed weights easily getting trapped in local optima. The steps are as follows: In the standard particle swarm optimization algorithm, the inertia weight is a fixed constant. This scheme improves it to an adaptive weight that decreases linearly with the number of iterations, as expressed by the formula: ; Where t represents the current iteration number, and T represents the preset maximum iteration number. This represents the inertia weight in the t-th iteration. This represents the initial maximum inertia weight. This represents the minimum inertia weight in the later stages of the iteration; Furthermore, in the early stages of iteration Larger size, strong global particle exploration capability, later iterations The particle size is relatively small, and the local development capability of the particles is strong. This formula can balance the global search and local convergence performance. Then, based on the update process of the standard particle swarm optimization algorithm, and with the yield rate and unit energy consumption ratio as the optimization objectives, an optimization objective function is constructed: ; in, Indicates the finished product qualification rate. Indicates the percentage of energy consumption per unit. The adaptive weights decrease linearly with the number of iterations. Furthermore, the particle swarm update rule is expressed as: Speed ​​updates: ; Location update: ; in, Let be the particle velocity in the t-th iteration. Let be the particle position in the t-th iteration. The learning factor is taken from an industry-standard value. , A random number in the interval [0,1]. This represents the optimal position in the history of an individual particle. The optimal position for the entire group; With the yield rate and unit energy consumption ratio as optimization objectives, the particle position and velocity are iteratively updated: Within the constraint boundaries of the key equipment constraint parameter set and the equipment adjustable range set, the initial positions of particles are randomly generated. With initial velocity Obtain the objective function value for each particle. Update the individual's optimal position. with the global optimal position ; Calculate the inertia weight based on the number of iterations t. Update particle velocity and update particle positions. If the value exceeds the adjustable range set, then the boundary value is taken. The iterations are repeated until the maximum number of iterations T is reached. This refers to the optimal control parameters for a single scenario; Furthermore, the optimal control parameters for a single scenario It is a set of key process parameters that correspond one-to-one with the seasoning production equipment, specifically including temperature parameters, rotation speed parameters, pressure parameters, and feed rate parameters. Each parameter value simultaneously satisfies the hard constraint boundary of the key equipment constraint parameter set and the safe adjustable boundary of the equipment adjustable range set.

[0022] Control and execution module: Based on the batch identifier parsed by the central control request, it retrieves the optimal control parameters for a single scenario, calls up the corresponding parameters, and implements control via PLC to achieve intelligent control; The existing TCP segmentation and parsing technology is used to split the fields of the request message sent by the central control system, and the encoded character ID encapsulated inside the message is extracted. This encoded character is a batch-specific identifier, and the batch identifier to be retrieved is obtained. Using the batch identifier to be searched as the search condition, the dynamic coupling weight set and the optimal control parameter for a single scenario are searched respectively, and the corresponding parameter content of the batch is extracted to obtain the target control parameter. Based on the PLC national standard engineering unit conversion rules, the target control parameters are converted to a new format. Let the converted parameters be... Through the industrial bus protocol The data is sent to the corresponding PLC controller to drive the production equipment to complete its operation, thereby obtaining the actual on-site control actions and completing the control of the intelligent seasoning production equipment.

[0023] In the application, several formulas are calculated by removing dimensions and taking their numerical values. The formulas are established by collecting a large amount of data and simulating the most recent real situation. Some coefficients or weights in the formulas are set by those skilled in the art according to the actual situation, so they will not be elaborated here.

[0024] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0025] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A control system for intelligent seasoning production equipment based on multi-source sensor fusion, characterized in that, include: Data Identification Module: Used to collect production line operation data, bind process batch identifiers and equipment node identifiers, normalize the data, and output standardized sensor datasets; Dynamic coupling module: used to segment standardized sensor datasets by batch and device, obtain dynamic coupling weights for each segment based on process deviations using a grey relational model, and generate dynamic coupling weight sets by associating identifiers with storage addresses; Reference acquisition module: used to construct historical operating condition feature vectors, match the dynamic coupling weight set through the cosine similarity algorithm, filter to obtain the historical optimal strategy reference set, and filter key control nodes based on the fusion of the dynamic coupling weight set and the historical optimal strategy reference set, combined with the preset response threshold, and retrieve process constraint parameters to generate key equipment constraint parameter set; Strategy optimization module: retrieves the adjustable range of each device within the key control nodes to obtain the set of adjustable ranges, merges the set of key device constraint parameters with the set of adjustable ranges, and obtains the optimal combination of control parameters for each device through a particle swarm multi-objective optimization algorithm with adaptive adjustment of inertial weights; Control and execution module: Based on the batch identifier parsed by the central control request, the optimal control parameters for a single scenario are retrieved, the corresponding parameters are executed and controlled by the PLC, thus achieving intelligent control.

2. The intelligent seasoning production equipment control system based on multi-source sensor fusion according to claim 1, characterized in that, The process of outputting a standardized sensor dataset includes: The production line operation data is obtained by collecting operating values ​​from each section using a sensor array and summarizing all collected values. The built-in batch code is retrieved as the process batch identifier, and the inherent hardware number of the equipment body is retrieved as the equipment node identifier. The production line operation data is bound to the process batch identifier and the equipment node identifier one by one through the database field binding technology to obtain the original identifier data. The minimum-maximum normalization algorithm is used to scale the labeled original data to a uniform range, and outliers are removed based on the 3σ gross error to obtain a standardized sensor dataset.

3. The intelligent seasoning production equipment control system based on multi-source sensor fusion according to claim 2, characterized in that, The process of obtaining the dynamic coupling weight of a single segment includes: The standardized sensor dataset is fragmented and grouped into equipment fragments using process batch identifiers as the primary classification criteria and equipment node identifiers as the secondary classification criteria. The difference between the actual process parameters and the factory standard process parameters within the segmented group data of the equipment is calculated using the difference operation, and all difference values ​​are collected to obtain the segmented process deviation parameters. The segmentation process deviation parameters are imported into a preset grey relational degree calculation model to obtain the dynamic coupling weight of a single segment. Read the preset storage address code of the server disk partition, bind and archive the dynamic coupling weight of a single shard, the corresponding process batch identifier, the equipment node identifier, and the storage address through multi-field association storage in the data table, and integrate the archived content of the entire shard to obtain the dynamic coupling weight set.

4. The intelligent seasoning production equipment control system based on multi-source sensor fusion according to claim 3, characterized in that, The construction process of the grey relational analysis model includes: The mainstream process parameters of the seasoning production line were selected as the model evaluation index. The measured values ​​of the indexes from multiple batches of historical production were collected, and all measured values ​​were aggregated to obtain the original sample series of the model. Ideal process data is preset as a reference baseline sequence. The absolute difference between the original sample sequence of the model and the reference baseline sequence is calculated. All difference results are collected to obtain the sequence difference matrix. The grey relational coefficient formula is calculated point by point based on the sequence difference matrix to obtain the single-point grey relational coefficient. The average value of all single-point grey relational coefficients corresponding to the same group of indicators is calculated, and the average value calculation formula, sample input port, and result output port are uniformly encapsulated to complete the construction of the grey relational degree calculation model.

5. The intelligent seasoning production equipment control system based on multi-source sensor fusion according to claim 4, characterized in that, The process of obtaining the historical optimal strategy reference set includes: Retrieve the three types of original records stored in the historical operating condition database: historical sensing, control parameters, and finished product quality inspection. Use the dimensional feature extraction method to extract the key feature components of each batch of data. Summarize the key feature components of all batches to obtain the historical operating condition feature vector. The weight values ​​in the dynamic coupling weight set are organized into real-time feature vectors of the same dimension. The cosine similarity calculation formula is used to substitute the historical working condition feature vectors one by one to complete the numerical calculation and obtain a sequence of similarity results. Based on the sequence of similarity results, historical control records corresponding to those higher than the system's preset matching threshold are selected, and all selected records are merged to obtain a historical optimal strategy reference set.

6. The intelligent seasoning production equipment control system based on multi-source sensor fusion according to claim 5, characterized in that, The process of obtaining the set of constraint parameters for key equipment includes: Based on the unified device node identifier field in the dynamic coupling weight set and the historical best strategy reference set, all control parameters bound to the identifier in the dynamic coupling weight set and the historical best strategy reference set are collected by filtering by the database primary key and grouped by a single device node identifier to obtain a subset of single device control parameters. Based on the subset of single-device control parameters matched with the finished product inspection results data bound to the same batch, the contribution ratio of parameter fluctuation to quality fluctuation is calculated using the Pearson correlation coefficient and converted into a percentage to obtain the quality influence coefficient of each node. Based on the comparison between the quality impact coefficient of each node and the preset response threshold, the device numbers whose coefficients exceed the threshold are filtered to form a list, thus obtaining the list of key control nodes. Based on the list of key control nodes, the pre-stored upper and lower limit process parameters and safe operation limit parameters of each equipment are read, and all read parameters are summarized to obtain the set of key equipment constraint parameters.

7. The intelligent seasoning production equipment control system based on multi-source sensor fusion according to claim 5, characterized in that, The process of obtaining the optimal control parameters for a single scenario includes: Based on the equipment's factory calibration and historical process test data, the upper and lower safety limits and dynamic adjustment margins of four parameters—temperature, speed, pressure, and feed rate—are extracted to construct a set of adjustable ranges for the equipment with adjustment margins. The set of key equipment constraint parameters is used as the hard constraint boundary for solving, and the set of adjustable equipment ranges is used as the parameter value boundary. A particle swarm optimization algorithm with adaptive adjustment of inertial weight is adopted, with the product quality control pass rate and unit energy consumption ratio as the optimization objectives. The particle position and velocity are iteratively updated to obtain the optimal control parameters for a single scenario.

8. The intelligent seasoning production equipment control system based on multi-source sensor fusion according to claim 7, characterized in that, The implementation process of the particle swarm optimization algorithm with adaptive adjustment of inertia weights includes: Obtain adjustable boundaries with margin: ; in, The upper limit of the safe adjustable value for the j-th parameter. This is the lower limit of the safety adjustable value for the j-th parameter. For the safety dynamic adjustment margin of the j-th parameter, Indicates the rated upper limit value. Indicates the lower limit value; Based on the standard particle swarm optimization algorithm, the inertia weight is linearly and adaptively adjusted, as expressed by the formula: ； Where t represents the current iteration number, and T represents the preset maximum iteration number. This represents the inertia weight in the t-th iteration. This represents the initial maximum inertia weight. This represents the minimum inertia weight in the later stages of the iteration.

9. The intelligent seasoning production equipment control system based on multi-source sensor fusion according to claim 7, characterized in that, The process of achieving intelligent control includes: The request message fields issued by the central control are split by TCP segmentation parsing, and the embedded encoded content in the request message fields is extracted to obtain the batch identifier to be retrieved; Based on the batch identifier to be searched, the parameter content corresponding to the optimal control parameter of a single scenario is retrieved to obtain the target control parameter; Based on the target control parameters, the system is converted and sent to the corresponding PLC controller to drive the equipment to perform actions, thereby obtaining the actual control actions on site.

Images