Food ingredient feeding whole process tracing and data association production monitoring method and system

Abstract

The invention provides a food ingredient feeding whole process tracing and data association production monitoring method and system, and relates to the technical field of food safety production and quality control, and the method comprises the steps: automatically synchronizing a production task list from an ERP system; generating a workshop plan according to the production task list and a formula library; in the batching link, weight data are automatically collected through an electronic scale, and batching labels including batch numbers, material codes and supplier information are printed based on the workshop plan; in the feeding link, the ingredient labels are scanned, and checking and mistake proofing are conducted on the station state of the SCADA system; operator, time, equipment and material data of batching and feeding links are automatically collected and bound with corresponding production batches, and a whole-course tracing chain from raw material receiving to finished products is formed. According to the method, a complete traceability chain is formed through whole-process data binding, and the food safety compliance and the production efficiency are improved.

Description

Technical Field

[0001] This invention relates to the field of food safety production and quality control technology, and in particular to a production monitoring method and system for tracing and data association throughout the entire process of food ingredient feeding. Background Technology

[0002] With the rapid development of the food industry, food safety and transparency in the production process have become key concerns. In traditional food production, the ingredient mixing and feeding stages rely heavily on manual operation, leading to problems such as incomplete records, difficulties in data traceability, and high error rates. Production orders are typically transmitted via paper, which is prone to loss and inefficient; weighing data in the ingredient mixing stage relies on manual recording, which is error-prone; and the lack of effective error-proofing mechanisms in the feeding stage can result in material confusion or incorrect feeding order, affecting product quality and production safety. Furthermore, regulatory agencies are increasingly stringent in their requirements for traceability throughout the entire food production process, necessitating companies to establish a complete data chain from raw materials to finished products to meet compliance and customer trust requirements. However, current technologies suffer from low data integration levels between ERP systems, SCADA systems, and production equipment, making real-time data collection and correlation difficult and limiting the level of intelligent production monitoring.

[0003] To address the aforementioned issues, this application proposes a production monitoring method and system for full traceability and data association of food ingredient feeding. Through automated data collection, system integration, and error-proofing verification, a full traceability chain is constructed from production tasks to finished products, thereby improving production efficiency and food safety assurance capabilities. Summary of the Invention

[0004] The purpose of this invention is to provide a production monitoring method and system for tracing and data association of the entire process of food ingredient feeding, so as to solve the problems pointed out in the background art.

[0005] In a first aspect, the production monitoring method for tracing and data association of the entire process of food ingredient feeding provided by embodiments of the present invention includes: Automatically synchronize production task orders from the ERP system; A workshop plan is generated based on the production task order and formula library; In the ingredient preparation process, weight data is automatically collected by electronic scales, and ingredient labels containing batch numbers, material codes and supplier information are printed based on the workshop plan. During the feeding process, the ingredient label is scanned and verified against the workstation status in the SCADA system to prevent errors. The system automatically collects operator, time, equipment, and material data from the batching and feeding processes, and links them to the corresponding production batches to form a complete traceability chain from raw material receipt to finished product.

[0006] Optionally, the formula library is maintained through formula management steps, which include formula entry, version management, process parameter setting, formula download, and formula review.

[0007] Optionally, the ingredient preparation step further includes: Automatically detects whether the weighing data is within the preset error range, and only allows saving and execution of the next weighing if the data is within the error range; When the batching process is finished, the system automatically checks for any missing materials and alerts the operator if any are missing. Supports reprinting of the ingredient labels.

[0008] Optionally, the feeding process further includes: After scanning the ingredient label, it is compared with the materials to be added in the workshop plan, and an error reminder is issued when the materials are inconsistent. Verify the material feeding point according to the workstation type to prevent materials from being fed to the wrong workstation.

[0009] Optional production monitoring methods for tracing and linking the entire process of food ingredient dispensing also include: By entering the production task number or time period, the corresponding ingredient and feeding records can be queried. The records include operation time, product code, raw material code, batch number, supplier and usage information.

[0010] Optional production monitoring methods for tracing and linking the entire process of food ingredient dispensing also include: Initialize inventory material labels and print inventory labels for raw materials; Use a mobile handheld terminal to purchase and receive goods, scan the physical items and enter the batch number and quantity information, and print inventory labels; Perform on-line warehouse receiving, return, and inventory operations, and handle system tail errors caused by weighing errors.

[0011] Optional production monitoring methods for tracing and linking the entire process of food ingredient dispensing also include: Data exchange is achieved through data interfaces with ERP systems, SCADA systems, and electronic scale equipment. The data exchange with the ERP system includes automatic synchronization of production task orders, basic material information and supplier information; Data exchange with the SCADA system includes reading the workstation status; Data exchange with the electronic scale is achieved through serial communication to realize real-time weight acquisition.

[0012] Optional production monitoring methods for tracing and linking the entire process of food ingredient dispensing also include: Security auditing can be achieved by setting user permission controls to access system functions and recording operation logs.

[0013] Optional production monitoring methods for tracing and linking the entire process of food ingredient dispensing also include: Construct a human-machine collaborative production monitoring model; When a user operates the production monitoring model, the user's counterintuitive behaviors are identified and evaluated. These counterintuitive behaviors include: cognitive decay patterns of critical production alarms, spatiotemporal inconsistencies in material delivery sequences, missing opportunity windows for responding to abnormal events, abnormal changes in the information entropy of operating steps, implicit deviations from standard operating procedures, and abnormal distribution of cognitive load in human-computer interaction. When the index of counterintuitive behavior exceeds the first dynamic threshold based on task criticality adaptation, the cognitive-risk dual-track assessment mechanism is activated to generate a real-time intervention capacity window. Based on the real-time intervention capacity window, a first non-intrusive correction strategy is planned, and its expected benefit level is quantified. If the expected benefit exceeds the second dynamic threshold based on trust evolution, then the first seamless correction strategy is implemented for the user. Otherwise, the predictive intervention mode is activated to generate and implement a second, non-intrusive corrective strategy for the user.

[0014] Secondly, the production monitoring system for tracing and data association of the entire food ingredient feeding process provided in this embodiment of the invention includes: The data synchronization module is configured to automatically synchronize production task orders from the ERP system. The plan generation module is configured to generate a workshop plan based on the production task order and the recipe library. The ingredient management module is configured to automatically collect weight data via electronic scales during the ingredient preparation process, and control the label printer to print ingredient labels containing batch numbers, material codes, and supplier information based on the workshop plan. The material feeding verification module is configured to scan the material label using a scanning device during the material feeding process and verify the workstation status against the SCADA system to prevent errors. The data traceability module is configured to automatically collect operator, time, equipment and material data in the batching and feeding process, and bind them to the corresponding production batches to form a full traceability chain from raw material receipt to finished product; The central processing unit is configured to coordinate and control the collaborative work of the data synchronization module, the plan generation module, the batching management module, the material feeding verification module, and the data traceability module.

[0015] The present invention has achieved the following beneficial effects: This invention automates and links the entire process of ingredient preparation and feeding by automatically synchronizing production task sheets with the ERP system, generating workshop plans, and combining electronic scale data collection and label printing. This significantly reduces human error. By scanning ingredient labels and verifying with the SCADA system, it strengthens the error prevention capability of ingredient preparation and ensures production accuracy. The entire process of data binding forms a complete traceability chain, improving food safety compliance and production efficiency.

[0016] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description and the accompanying drawings.

[0017] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0018] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a schematic diagram of a production monitoring method for tracing and data association of the entire food ingredient feeding process in an embodiment of the present invention; Figure 2 This is a schematic diagram of a production monitoring system for tracing and data association of the entire food ingredient feeding process in an embodiment of the present invention. Detailed Implementation

[0019] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0020] Example 1: Figure 1 The flowchart of the production monitoring method for tracing and data association of the entire food ingredient feeding process provided in Example 1 is as follows: Figure 1 As shown, the method includes: Automatically synchronize production task orders from the ERP system; A workshop plan is generated based on the production task order and formula library; In the ingredient preparation process, weight data is automatically collected by electronic scales, and ingredient labels containing batch numbers, material codes and supplier information are printed based on the workshop plan. During the feeding process, the ingredient label is scanned and verified against the workstation status in the SCADA system to prevent errors. The system automatically collects operator, time, equipment, and material data from the batching and feeding processes, and links them to the corresponding production batches to form a complete traceability chain from raw material receipt to finished product.

[0021] This embodiment achieves traceability throughout the entire food production process by automatically synchronizing production task sheets from the ERP system, generating workshop plans, collecting ingredient and feed data, and binding batch information. The production task sheet is a digital task list obtained from the ERP system through a data interface, containing information such as order number, product code, production quantity, and delivery time. It is obtained by periodically pulling the latest task data from the ERP system via a standard API (such as a RESTful interface). The workshop plan is a schedule for guiding ingredient and feed based on the production task sheet and the recipe library, including material codes, usage, and workstation allocation. The recipe library stores product recipe information (such as material ratios and process parameters), and the plan is generated through database queries. During the ingredient mixing process, weight data is collected in real time by electronic scales. The scales transmit weight values ​​via serial communication (RS-232 protocol) with an accuracy of 0.01 kg and a collection frequency of once per second. Ingredient labels are generated by a barcode printer and include a batch number (unique identifier for the material batch, in the format of "date + serial number"), a material code (a unique identifier for the material defined by the ERP system, such as "M001"), and supplier information (supplier codes synchronized from the ERP system). The material feeding process verifies the material feeding status by scanning the ingredient labels (using a 1D or 2D barcode scanner to parse the barcode content) and checking the workstation status in the SCADA system (reading the workstation ID and operating status, such as "standby" or "running," via the Modbus protocol) to prevent feeding errors. Operator, time (accurate to the second), equipment number (uniquely identifying the equipment, such as "SCADA-E01"), and material data are linked to the production batch through a database, forming a traceability chain from raw material receipt to finished product. This chain is stored in a relational database (such as MySQL) to ensure data traceability.

[0022] This invention automates and links the entire process of ingredient preparation and feeding by automatically synchronizing production task sheets with the ERP system, generating workshop plans, and combining electronic scale data collection and label printing. This significantly reduces human error. By scanning ingredient labels and verifying with the SCADA system, it strengthens the error prevention capability of ingredient preparation and ensures production accuracy. The entire process of data binding forms a complete traceability chain, improving food safety compliance and production efficiency.

[0023] Example 2: In Example 2, the formula library is maintained through formula management steps, which include formula entry, version management, process parameter setting, formula download, and formula review.

[0024] The formula library is maintained through five steps: formula entry, version management, process parameter setting, formula download, and formula verification, ensuring the accuracy and consistency of formula data. Formula entry involves inputting product formulas via a human-machine interface (such as a web interface), including material codes, proportions (accurate to 0.1%), and process parameters (such as mixing time, in seconds). This data is stored in the formula database. Version management records formula modification history using version numbers (incrementing from "V1.0"). The system automatically assigns a new version number to each modification and stores the modification time and operator ID. Process parameter setting defines production conditions, such as temperature (range 0-100℃, accuracy 0.1℃) and mixing speed (range 0-1000rpm, accuracy 1rpm). This can be done manually using process parameter templates or imported from historical data. Formula download transmits formulas to the workshop control terminal via a data interface (such as TCP / IP protocol), where the terminal verifies data integrity. Formula verification involves authorized personnel confirming formula accuracy by comparing differences between old and new versions (such as changes in material proportions). The verification result is recorded as "pass" or "fail" and stored in the database.

[0025] Example 2 implements formula entry, version management, process parameter setting, downloading, and verification through formula management steps, which standardizes the formula maintenance process and ensures the accuracy and consistency of formula data; version management effectively tracks formula changes and reduces production risks caused by formula errors; and precise setting of process parameters improves production stability and simplifies formula management operations.

[0026] Example 3: In Example 3, the ingredient preparation step further includes: Automatically detects whether the weighing data is within the preset error range, and only allows saving and execution of the next weighing if the data is within the error range; When the batching process is finished, the system automatically checks for any missing materials and alerts the operator if any are missing. Supports reprinting of the ingredient labels.

[0027] The ingredient batching process improves accuracy by automatically detecting weighing errors, checking for missing materials, and supporting the reprinting of ingredient labels. Weighing data is collected in real-time by an electronic scale via RS-232 protocol with an accuracy of 0.01 kg. The system calculates the error based on the target weight of materials in the formula library (e.g., 500 kg) and a preset error range (±0.5%, or ±2.5 kg). The error range is defined by the formula library and obtained by querying the database. The system only allows data saving and triggers the next weighing if the actual weight is within the error range; otherwise, it displays "out of tolerance." Material omission checks compare the bill of materials (including material codes and quantities) in the workshop plan with the actual weighing records. This is done by querying saved weighing records in the database. If missing materials are found (e.g., the plan requires 3 materials, but only 2 are actually weighed), the system alerts the operator via a pop-up window. Reprinting ingredient labels is performed by a barcode printer. The label content (batch number, material code, supplier information) is extracted from the database, and is triggered by the operator manually selecting the "reprint" function.

[0028] Example 3 significantly improves the accuracy and efficiency of the batching process by automatically detecting weighing data errors, checking for material omissions, and supporting label reprinting. Automatically saving data within the error range reduces manual intervention, the omission reminder function reduces the operational error rate, and the label reprinting function enhances operational flexibility, effectively ensuring the reliability and traceability of the production process.

[0029] Example 4: In Example 4, the feeding process further includes: After scanning the ingredient label, it is compared with the materials to be added in the workshop plan, and an error reminder is issued when the materials are inconsistent. Verify the material feeding point according to the workstation type to prevent materials from being fed to the wrong workstation.

[0030] The material feeding process prevents errors through label comparison and workstation verification. After scanning the ingredient label, the system parses the barcode content (batch number, material code) and queries the database for the list of materials to be fed in the workshop plan (including material code and quantity). This is done by querying the plan table in real time. If the material code is inconsistent (e.g., scanning "M001" but the plan requires "M002"), the system will issue an error alert via the interface or a buzzer. Workstation type verification reads the workstation ID and type (e.g., "mixing workstation" or "packaging workstation") through the SCADA system. This is done by querying the workstation status database via the Modbus protocol. If the material scanned is not suitable for the workstation type (e.g., a mixing workstation requires flour, but the system scans sugar), the system will prompt "workstation error," ensuring that the material is fed to the correct workstation.

[0031] Example 4 achieves precise error prevention in the material feeding process by scanning the ingredient label, comparing it with the workshop plan, and verifying the workstation type. Error alerts when materials are inconsistent and material feeding point verification effectively prevent material confusion and misfeeding, improve the accuracy and safety of the production process, and ensure product quality and production efficiency.

[0032] Example 5: In Example 5, the production monitoring method for tracing and data association of the entire food ingredient feeding process further includes: By entering the production task number or time period, the corresponding ingredient and feeding records can be queried. The records include operation time, product code, raw material code, batch number, supplier and usage information.

[0033] Production process traceability is achieved by querying ingredient and feed records through production task order numbers or time periods. The production task order number is a unique identifier synchronized from the ERP system (e.g., "PO123"), and the time period is the operation time range (e.g., "10:00-12:00"), entered via a web interface. Query results include operation time (accurate to the second, obtained from database timestamps), product code (obtained from the task order), raw material code (obtained from ingredient labels), batch number (obtained from labels), supplier (synchronized from the ERP system), and usage (collected from electronic scales, accuracy 0.01 kg). These are extracted from the database through SQL queries and linked to the production batch table, ingredient table, and feed table to generate complete records.

[0034] Example 5 provides a convenient data traceability function by supporting the query of ingredient feeding records by task number or time period; the records contain key data such as operation time, product code, and raw material information, which facilitates production process analysis and regulatory compliance, and enhances the enterprise's ability to control production data and improve food safety management.

[0035] Example 6: In Example 6, the production monitoring method for tracing and data association of the entire food ingredient feeding process further includes: Initialize inventory material labels and print inventory labels for raw materials; Use a mobile handheld terminal to purchase and receive goods, scan the physical items and enter the batch number and quantity information, and print inventory labels; Perform on-line warehouse receiving, return, and inventory operations, and handle system tail errors caused by weighing errors.

[0036] Inventory management handles material traceability through label initialization, procurement receiving, and line-side warehouse operations. Inventory label initialization generates barcode labels for raw materials, containing the material code, batch number (format "date + serial number"), and supplier. The printer retrieves data from the ERP system. Procurement receiving uses a handheld terminal to scan the physical barcodes, enters the batch number and quantity (accuracy 0.01kg), uploads it to the database via Wi-Fi, and prints the inventory labels. Line-side warehouse operations include receiving (scanning labels to record the quantity received), returns (recording the reason and quantity of returns), and inventory counting (comparing actual inventory with system inventory, calculating the difference (accumulated weighing error, obtained through database comparison)) to ensure accurate inventory data.

[0037] Example 6 optimizes raw material management and inventory counting processes by initializing inventory labels, receiving goods via mobile terminals, and operating at the line-side warehouse; it improves the accuracy of inventory data by scanning and entering batch numbers and quantities and handling weighing errors, simplifies the process of returning materials to the warehouse, and enhances material traceability and production efficiency.

[0038] Example 7: In Example 7, the production monitoring method for tracing and data association of the entire food ingredient feeding process further includes: Data exchange is achieved through data interfaces with ERP systems, SCADA systems, and electronic scale equipment. The data exchange with the ERP system includes automatic synchronization of production task orders, basic material information and supplier information; Data exchange with the SCADA system includes reading the workstation status; Data exchange with the electronic scale is achieved through serial communication to realize real-time weight acquisition.

[0039] Data is exchanged with ERP, SCADA, and electronic scales via data interfaces to ensure system integration. ERP data exchange synchronizes production task orders (including order number and product code), basic material information (material code and name), and supplier information (supplier code and name) via RESTful API, executing once per hour. SCADA data exchange reads workstation status (e.g., "Running" or "Faulty") via Modbus protocol, querying once per second. Electronic scale data is acquired via RS-232 serial communication, with a baud rate of 9600bps, an accuracy of 0.01kg, and a sampling frequency of 1Hz, and the data is stored in the database.

[0040] Example 7 integrates with ERP, SCADA systems and electronic scales through data interfaces, enabling real-time exchange of production tasks, workstation status and weight data; automatically synchronizing material and supplier information simplifies data entry, enhances inter-system collaboration, and improves the real-time performance and data integrity of production monitoring.

[0041] Example 8: In Example 8, the production monitoring method for tracing and data association of the entire food ingredient feeding process further includes: Security auditing can be achieved by setting user permission controls to access system functions and recording operation logs.

[0042] Security auditing is achieved by controlling system access to functions through user permissions and recording operation logs. Access control is based on role assignment (e.g., administrator, operator), and a permission table (containing user ID, role, and function permissions, such as "recipe entry") is stored in a database. This table is retrieved when a user logs in. Operation logs record the user ID, operation time (accurate to the second), operation content (e.g., "modify recipe"), and result, stored in a log database. These logs are retrieved automatically after each operation. Security auditing involves querying the log table to analyze abnormal operations (e.g., unauthorized access).

[0043] Example 8 ensures the security of system function access through user permission control and operation log recording; the security audit function effectively tracks operation behavior, reduces the risk of system misoperation and data leakage, improves the reliability and management standardization of the production monitoring system, and enhances the enterprise's information security protection capabilities.

[0044] Example 9: In Example 9, the production monitoring method for tracing and data association of the entire food ingredient feeding process further includes: Constructing a human-machine collaborative production monitoring model; wherein, the construction of the human-machine collaborative production monitoring model includes: based on the automatically collected operator, time, equipment and material data of the batching and feeding process, and the full traceability chain bound to the production batch, constructing a digital twin monitoring environment covering the entire process from raw material receipt to finished product through a data fusion engine with time-series correlation, using a dynamic adaptive algorithm based on reinforcement learning to optimize the perception-decision-execution closed-loop parameters of the production monitoring model in real time, generating a multi-dimensional visualization view including a production process topology diagram, feeding progress time axis, equipment status monitoring diagram and quality indicator heat map, and introducing a production behavior pattern library based on deep learning, constructing a human-machine collaborative production monitoring model with cognitive ergonomic optimization features through a continuous learning mechanism.

[0045] In this step, the human-machine collaborative production monitoring model is a digital system integrating data fusion, dynamic optimization, and visualization functions, aiming to optimize production monitoring efficiency through real-time data analysis and human-machine interaction. The model is based on automatically collected data from the batching and feeding stages (including operator ID, operation time, equipment number, material code, batch number, usage, etc., collected via RS-232 serial communication of electronic scales, with an accuracy of 0.01kg and a frequency of 1Hz; the SCADA system obtains workstation status via Modbus protocol, with a frequency of 1Hz; the ERP system synchronizes task orders and material information via RESTful API, once per hour) and a full-process traceability chain bound to production batches (stored in a MySQL database, generated by querying the batch table, batching table, and feeding table via SQL). The data fusion engine employs a time-series correlation algorithm, specifically a time-series analysis algorithm based on timestamps (accurate to the second), combined with sliding window technology (window size set to 60 seconds, step size 10 seconds), aligning multi-source data (such as weight, workstation status, operator behavior) along the time axis to generate a unified data stream. The data fusion engine is built on the Apache Kafka stream processing platform. It pulls data in real-time from databases and device interfaces via KafkaConnect, configures topics to store different data sources (such as "weightdata" and "stationstatus"), and uses Kafka Streams for distributed stream processing to merge data streams and generate a time-series dataset. The digital twin monitoring environment is built on the Industrial Internet of Things (IIoT) framework using the ThingWorx platform. Specific steps include: defining a data model (containing entities such as equipment, workstations, and materials); importing the fused data stream into ThingWorx via REST API; generating a virtualized production environment; mapping physical equipment (such as scales and mixers) to digital entities; and updating status in real-time (such as equipment operation and material usage). The dynamic adaptive algorithm uses the Q-learning algorithm from reinforcement learning. The state space is defined as a production state vector (including workstation status, material usage, and operation latency), the action space is for adjusting monitoring parameters (such as alarm thresholds and intervention frequency), and the reward function comprehensively considers production efficiency (number of orders completed per minute), quality stability (error rate, target <0.5%), and resource consumption (electricity, unit: kWh). The Q-learning algorithm initializes the Q-table through offline training (using historical production data, approximately 100,000 records), updates the Q-value in real time (learning rate α=0.1, discount factor γ=0.9), and optimizes the closed-loop parameters of perception-decision-execution (such as alarm trigger time and intervention priority).The multi-dimensional visualization view is generated using the D3.js library and includes a production process topology diagram (based on a directed acyclic graph, with nodes representing workstations and edges representing material flows), a material feeding progress timeline (in minutes, showing the material feeding completion rate), an equipment status monitoring diagram (using color coding to indicate operating / fault status), and a quality indicator heatmap (using error rate and usage deviation as coordinates, with color depth representing risk level). The production behavior pattern library is built based on a deep learning recurrent neural network (RNN), specifically an LSTM model. The input is a sequence of operator behavior (operation time, action type, material code, etc.), and the output is a behavior pattern classification (e.g., normal, abnormal). The training data consists of historical operation logs (approximately 500,000 entries), trained using the TensorFlow framework. The model structure includes a 3-layer LSTM (128 units per layer) and a 1-layer fully connected layer, with ReLU activation function and Adam optimizer (learning rate 0.001). The continuous learning mechanism updates the LSTM model through online learning, incrementally pulling new logs from the database daily (approximately 1000 logs / day) and updating weights using mini-batch (batch size 32) to maintain the model's adaptability to new behaviors. Cognitive ergonomics optimization is implemented through the human-computer interaction interface (web interface, based on the React framework). The interface layout is optimized according to Fitts' Law (e.g., button size 20px, spacing 10px) to reduce the cognitive load on operators (target response time <0.5 seconds).

[0046] When a user operates the production monitoring model, the system identifies and evaluates the user's counterintuitive behavior. This identification and evaluation includes: extracting behavioral features based on multi-source heterogeneous data; capturing potential abnormal patterns in user operations using a spatiotemporal graph convolutional network; constructing a multi-dimensional quantitative evaluation system for counterintuitive behavior; and identifying and evaluating the user's counterintuitive behavior based on the potential abnormal patterns and the multi-dimensional quantitative evaluation system. The counterintuitive behaviors include: cognitive decay patterns of critical production alarms, spatiotemporal inconsistencies in material delivery sequences, missing opportunity windows for responding to abnormal events, abnormal changes in the information entropy of operational steps, implicit deviations from standard operating procedures, and abnormal distribution of cognitive load in human-computer interaction.

[0047] In this step, counterintuitive behavior refers to operator deviations from standard operating procedures (SOPs) or abnormal operating patterns during production, which may lead to production errors or safety risks. Behavioral feature extraction is based on multi-source heterogeneous data, including operator input logs (collected via a web interface, recording key presses, mouse clicks, scan times, etc.), equipment status (obtained by the SCADA system via Modbus protocol, 1Hz frequency), and material data (collected by electronic scales, accuracy 0.01kg). A spatiotemporal graph convolutional network (ST-GCN) is used to capture potential abnormal patterns. The ST-GCN model construction steps include: defining a spatiotemporal graph, where nodes represent operator actions (clicks, scans), equipment status, and material data; edges represent temporal associations (time difference < 5 seconds) and spatial associations (same workstation ID); and feature vectors include operation time, material code, workstation status, etc. The model structure includes a 3-layer GCN (64 units per layer) and a 2-layer LSTM (128 units per layer). The input is a sequence of behaviors within a time window (30 seconds), and the output is the anomaly probability (0-1). The threshold for determining the anomaly probability is determined using the ROC curve on the historical validation set. The initial threshold is set to 0.7, meaning that when the probability value is greater than or equal to 0.7, the operation is considered a potential anomalous pattern. To further adapt to changes in the production environment, the system introduces a dynamic threshold adjustment mechanism: the ROC curve is recalculated quarterly based on the operation data of the past quarter, and the threshold is updated according to the principle of maximizing the Youden exponent, ensuring that the accuracy and recall of the model are balanced. ST-GCN is implemented using the PyTorch framework, with training data consisting of historical operation logs and device status (approximately 200,000 entries), the loss function being cross-entropy, and the optimizer being Adam (learning rate 0.001). The multi-dimensional quantitative evaluation system includes six indicators: cognitive decay (ignoring alarm frequency, calculated as the number of times there is no response within 5 seconds after an alarm is triggered, target <1 time / hour), spatiotemporal inconsistency (material feeding sequence does not match the plan, calculated by comparing the edit distance between the actual material feeding sequence and the workshop plan), missing opportunity window (failure to respond to anomalies within the specified time, calculated as the number of timeout responses, target <1 time / shift), information entropy anomaly (entropy value of the operation sequence, calculated based on Shannon entropy, normal range 0.5-1.5), latent deviation (number of deviations from SOP, calculated as the number of differences between the operation log and the SOP template, target <2 times / shift), and cognitive load anomaly (operational delay, calculated as the average response time of clicks or scans, target <1 second). The evaluation system extracts behavioral data from the database through SQL queries, combines it with Python scripts to calculate the values ​​of each indicator, and generates a comprehensive anomaly index (weighted average, with weights dynamically adjusted according to the criticality of the task).

[0048] The criticality of the task is quantitatively categorized based on the food safety risk level and process stability requirements defined in the standard operating procedures: Key tasks: These are ingredient addition processes, critical control point operations, or ingredient preparation processes that have a decisive impact on the flavor and texture of the finished product. Their risk level is marked as "High." The weight of these tasks in the evaluation system is set at 0.4.

[0049] Non-critical tasks: These involve ordinary raw material input or auxiliary processes, and their risk level is marked as "medium" or "low." The weight of these tasks in the evaluation system is set to 0.2. Task criticality information is obtained from the risk level field attached to the production task sheet in the ERP system and correlated with operational behavior data during the data fusion phase.

[0050] When the index of counterintuitive behavior exceeds a first dynamic threshold based on task criticality adaptation, a cognitive-risk dual-track assessment mechanism is activated to generate a real-time intervention capacity window. This activation includes: dynamically assessing the user's cognitive processing capacity in the current production environment using a context-aware cognitive load model; identifying potential risk points where the user's intention deviates from the optimal path using a Bayesian inference-based operation path prediction engine; and establishing a dynamic mapping relationship between cognitive capacity (represented by cognitive processing capacity) and production risk values ​​(represented by potential risk points) to generate a real-time intervention capacity window with spatiotemporal constraints.

[0051] In this step, the cognitive-risk dual-track assessment mechanism generates a real-time intervention capacity window by evaluating operator cognitive load and production risk. This window is defined as the allowable intervention time and resource range (time window 10-60 seconds, resources include prompts, interface adjustments, etc.). The cognitive load model is adapted from the NASA-TLX scale. Inputs include operator behavior data (click frequency, scan interval), task complexity (number of task steps, obtained from the shop floor plan), and environmental factors (workstation status, alarm frequency). Outputs include a cognitive load index (0-100, 0.7 times) and an intervention capacity window (time range 10-60 seconds, resource allocation priority 1-5). The regression model is trained in Scikit-learn using historical intervention records (approximately 50,000 records) as training data. Window generation is achieved by extracting behavioral and risk data in real-time via SQL queries, combined with the regression model to predict the time and resource range.

[0052] Based on the real-time intervention capacity window, a first seamless correction strategy is planned, and its expected benefit level is quantified. This planning and quantification includes: based on the boundary constraints of the real-time intervention capacity window, selecting a subset of decision information matching the window capacity from the user's real-time production information flow using information entropy optimization theory; combining a behavior correction strategy library with transfer learning capabilities to generate a personalized first seamless correction strategy based on the decision information subset; and using a multi-objective optimization algorithm with the real-time intervention capacity window as a constraint to quantify the expected benefit level of the first seamless correction strategy. The calculation of the expected benefit level comprehensively considers the degree of improvement in production efficiency, the degree of improvement in quality stability, the degree of optimization in resource consumption, and the degree of enhancement in safety compliance, with the weights of each indicator dynamically adjusted according to the remaining capacity of the real-time intervention capacity window.

[0053] In this step, the first seamless correction strategy is based on seamless intervention measures generated by the real-time intervention capacity window (such as adjusting interface prompts and optimizing task order) to minimize operator interference. Information entropy optimization theory is used to filter a subset of decision information. Specifically, it calculates the entropy value of the production information flow (operation logs, equipment status, material data) (based on Shannon entropy, ranging from 0 to 2), and uses a greedy algorithm to filter the subset with the lowest entropy value (the subset size is 50% of the window capacity, approximately 5-10 pieces of information). The algorithm is implemented in Python, with the input being the real-time data stream and the output being the filtered information subset. The behavior correction strategy library is based on transfer learning, using a pre-trained LSTM model (the same as the production behavior pattern library). Weights are adjusted through transfer learning, with the input being a subset of decision information and the output being a correction strategy (such as the prompt "Check material M001"). The multi-objective optimization algorithm is NSGA-II, with objectives including production efficiency (order completion rate, target >95%) and quality stability (efficiency weight 0.4 when error rate is 50%; safety weight 0.4 when error rate is <50%), and the revenue value range is 0-100. The quantification and normalization methods for each revenue indicator are as follows: Production efficiency improvement: Calculated as the difference between the expected order completion rate after implementing the strategy and the current average completion rate, and normalized to the interval [0, 100] (e.g., 5 points for every 0.1% difference).

[0054] Quality stability improvement: Calculated as the percentage reduction in the expected quality error rate (such as feed weight deviation) and normalized to the [0, 100] interval (e.g., 10 points for each 0.05% reduction).

[0055] Resource consumption optimization degree: Calculated as the percentage reduction in expected unit product energy consumption (kWh / kg) and normalized to the [0, 100] interval.

[0056] Safety compliance enhancement: Calculated as the reduction in the number of expected SOP deviations, and normalized to the range [0, 100].

[0057] The expected improvement in production efficiency and quality stability, among other indicators, are estimated based on the corrective strategies output from the behavior correction strategy library and by combining the average effect data of similar strategies implemented in similar production scenarios in the historical database. For example, historical data shows that strategies that prompt material inspection can reduce the weight error rate by an average of 0.05%-0.1%. This estimate serves as input to a multi-objective optimization algorithm to calculate the expected benefit level.

[0058] The expected return is the weighted sum of the four normalized indicators mentioned above, with the weights dynamically adjusted based on the remaining capacity of the real-time intervention capacity window: when the remaining capacity is >50%, efficiency and quality have higher weights (e.g., 0.3 each); when the remaining capacity is ≤50%, safety and resource consumption have higher weights (e.g., 0.3 each).

[0059] If the expected benefit exceeds the second dynamic threshold based on trust evolution, then the first seamless correction strategy is implemented for the user. Otherwise, a predictive intervention mode is activated, generating and implementing a second seamless corrective strategy for the user. Activating the predictive intervention mode and generating and implementing the second seamless corrective strategy includes: generating a sequence of future user behaviors based on a behavioral evolution prediction model, according to the production information sequence, user operations, and intent paths; determining the latest intervention point when the production risk is about to exceed the tolerance threshold; generating the second seamless corrective strategy through a strategy optimization engine; and implementing the second seamless corrective strategy for the user.

[0060] The expected benefit level is compared with a second dynamic threshold (based on trust evolution, ranging from 50-80, calculated through historical intervention success rates; the threshold is 60 when the success rate is >80%). If the threshold is exceeded, the first non-intrusive corrective strategy is implemented, either by pushing a prompt through the web interface (e.g., a pop-up displaying "Please scan material M001") or adjusting the task order (updating the workshop plan via API). If the threshold is not exceeded, a predictive intervention mode is activated. The behavior evolution prediction model is based on LSTM, with inputs being production information sequences (operation logs, equipment status) and intent paths (Bayesian inference results), and outputting future behavior sequences (predicting operations in the next 30 seconds). The model is implemented in TensorFlow, with training data consisting of historical operation sequences (approximately 200,000 records). The latest intervention time is determined through time series analysis, calculating the moment when the risk exceeds the tolerance (anomaly index >0.8). The algorithm is implemented in Python, using Pandas to process the time series. The strategy optimization engine, based on a genetic algorithm (population size 50, 50 iterations), generates a second non-intrusive corrective strategy (e.g., adjusting alarm frequency), implemented through the web interface. The second dynamic threshold is based on a trust evolution model, which uses the number of consecutive successful interventions as the core indicator. The initial threshold is set to 50. Each successful intervention (i.e., the system detects a decrease in the anomaly index after intervention) increases trust and raises the threshold accordingly (e.g., by 5 units, up to a maximum of 80); each failure decreases trust and lowers the threshold accordingly (e.g., by 10 units). Through this mechanism, the system can adaptively adjust the trigger threshold of the intervention strategy to match the operator's actual performance and reliability.

[0061] The following is a specific implementation example of Example 9: Example 9 describes a production monitoring method for tracing and data association throughout the entire process of food ingredient feeding in a food processing enterprise (such as a biscuit factory). This method deploys a human-machine collaborative production monitoring model to optimize the production process and improve the accuracy and efficiency of operator behavior management. Biscuit production involves various raw materials (such as flour, sugar, and butter). The ingredient feeding and dispensing processes must strictly adhere to the formula and SOPs to ensure food safety and production efficiency. The system is deployed on the enterprise's private cloud server (based on Ubuntu 20.04, equipped with 32GB of memory and an 8-core CPU). The database uses MySQL to store ingredient feeding, dispensing, and batch data. The web interface is developed based on the React framework. The device interface connects to electronic scales (accuracy 0.01kg) and a SCADA system (collecting workstation status) via an industrial gateway (supporting Modbus and RS-232 protocols). Production orders are synchronized hourly from the SAP ERP system via a RESTful API, including the order number (e.g., "PO2023001"), product code (e.g., "BISCUIT01"), and production quantity (e.g., 1000kg). The recipe library stores cookie recipes (such as 50% flour, 20% sugar, and 30% butter), which can be entered and maintained through a web interface.

[0062] During the production process, the system pulls data in real time from electronic scales (collecting flour weight 500.02kg), SCADA system (workstation ID "MIX01", status "Running"), and ERP system (material code "M001", supplier "SUP001") via KafkaConnect. It then uses KafkaStreams to fuse the data and generate a time-series dataset (timestamps accurate to the second). A digital twin monitoring environment is built using the ThingWorx platform, mapping physical equipment (such as mixer "EQUIP01") to digital entities, displaying equipment status and material usage in real time. The Q-learning algorithm optimizes monitoring parameters (alarm threshold adjusted from 5 seconds to 3 seconds), and an LSTM model (3 layers, 128 units) analyzes operator behavior (scanning labels, clicking confirmation) to generate a behavior pattern classification (normal probability 0.95). The multi-dimensional visualization view, generated using D3.js, displays the production process topology (workstations “MIX01” to “PACK01”), the material feeding progress timeline (80% completion), the equipment status monitoring chart (green indicates operation), and the quality indicator heatmap (error rate 0.3%). The ST-GCN model detected that the operator ignored the alarm during material feeding (cognitive decay, 2 times / hour), calculating an anomaly index of 0.75. The cognitive load model (random forest, 100 trees) assessed the operator load at 45, Bayesian inference predicted an intention deviation probability of 0.2, and generated an intervention capacity window (30 seconds, resource priority 3). Information entropy optimization filtered the decision subset (containing material code “M001” and workstation status), and the NSGA-II algorithm generated the first non-intrusive correction strategy (prompting “Please check flour feeding amount”), with an expected benefit level of 85 (efficiency improvement of 0.3%, error rate reduction of 0.2%). Because the benefit exceeded the threshold of 60, the system pushed a prompt through the web interface, and the error rate dropped to 0.1% after the operator responded. When an operator is detected to have made three consecutive incorrect feeding sequences (anomaly index of 0.85), the profit drops to 50, which is below the threshold. A predictive intervention mode is then activated. LSTM predicts the operator's behavior over the next 30 seconds (potentially missing sugar), and a genetic algorithm generates a second, non-intrusive correction strategy (adjusting the alarm frequency to once every 10 seconds). After implementation via the interface, the feeding accuracy improves to 98%. Ultimately, the system achieves full traceability (from flour receipt to finished biscuit product), with an error rate controlled within 0.2%, production efficiency increased by 15%, and significant reductions in operational errors and safety risks.

[0063] Example 9, by constructing a human-machine collaborative production monitoring model and combining digital twin, reinforcement learning, and deep learning technologies, achieves precise traceability and data association throughout the entire food ingredient feeding process; spatiotemporal graph convolutional networks and multi-dimensional quantitative evaluation systems effectively identify counterintuitive behaviors and reduce operational error rates; a cognition-risk dual-track evaluation mechanism and a seamless correction strategy optimize human-machine interaction efficiency and reduce cognitive load; multi-dimensional visualization views enhance production transparency and facilitate real-time monitoring by managers; and a continuous learning mechanism ensures the model adapts to dynamic production environments, significantly improving production efficiency, quality stability, and safety compliance, providing strong technical support for food safety and production management.

[0064] Example 10: In Example 10, implementing the first or second seamless error correction strategy for the user includes: employing a context-aware strategy delivery mechanism to achieve smooth strategy implantation through implicit interaction technology; wherein, the implicit interaction technology includes dynamic optimization of interface elements based on visual perception, information density adjustment based on auditory perception, and automated correction based on workflow perception; constructing a strategy execution engine with feedback learning capabilities to evaluate the error correction effect in real time and dynamically adjust strategy parameters; and feeding the evaluation results back to the user profile and production behavior pattern library through a digital twin evaluation system to form a complete closed loop of continuous optimization.

[0065] The working principle of Example 10 is achieved through the following four technical steps: 1. Context-aware policy delivery mechanism Context-aware strategy delivery mechanism refers to dynamically adjusting the push method of corrective strategies based on the production environment, operator behavior, and equipment status to ensure that strategies are delivered to operators at the appropriate time and in the appropriate form, reducing interference and improving acceptance. Context-aware input data includes production environment data (workstation status, equipment operating parameters, collected from the SCADA system via the Modbus protocol at a frequency of 1Hz, with statuses including "Running," "Standby," and "Fault"), operator behavior data (operation time, click frequency, and scan interval, collected through web interface logs, with timestamps accurate to the second and click frequency in times / second), and task criticality (risk level field obtained from production task orders in the ERP system, categorized as "High," "Medium," and "Low," with high-risk tasks involving allergens or critical control points). The data is acquired as follows: the SCADA system connects to the industrial gateway via the Modbus TCP protocol (port 502) to read the workstation ID (e.g., “MIX01”) and status in real time; the web interface, based on the React framework, records operator interaction logs (e.g., timestamps of clicking the “Confirm” button and material code “M001”) and stores them in a MySQL database; the ERP system provides task order data via a RESTful API (synchronized hourly, URL format “ / api / tasks”), including order number (e.g., “PO2023001”), product code (e.g., “BISCUIT01”) and risk level. The context-aware model is built based on the Conditional Random Field (CRF) algorithm. The CRF model uses the production environment, behavioral data, and task criticality as feature vectors, and outputs a strategy delivery method (such as pop-up prompts, interface highlighting, or sound alerts). The model is implemented using the Python CRFsuite library. The training data consists of historical production and operation logs (approximately 100,000 entries, including environmental status, operation sequences, and delivery results). Features include timestamps, workstation status, and click frequency. The label is the success rate of the delivery method (0-1, success is defined as operator response time < 2 seconds). The CRF model training steps include: extracting feature vectors (e.g., workstation status = "Running", click frequency = 2 times / second, task risk = "High"), optimizing model parameters using maximum likelihood estimation, storing the trained model as a binary file, and calling it via API during real-time inference (response time < 0.1 seconds). The delivery mechanism selects the strategy form based on the CRF output. For example, high-risk tasks prioritize pop-up prompts (interface element size 20px, red background), while low-risk tasks use interface highlighting (thickened border 2px).

[0066] 2. Strategy Embedding of Implicit Interaction Technology Implicit interaction technology seamlessly integrates corrective strategies through visual, auditory, and workflow awareness, ensuring operators receive guidance without significant interference. Visually-aware dynamic optimization of interface elements guides operator behavior by adjusting the color, size, and position of web interface elements (such as buttons and text boxes). Parameters include color (RGB values, e.g., red #FF0000), size (pixels, range 10-30px), and position (screen coordinates, unit: pixels). These parameters are dynamically calculated based on the context-aware model output. For example, buttons for high-risk tasks are enlarged to 25px, while those for low-risk tasks are adjusted to 15px. Coordinates are optimized using Fitts' Law (target click time style={{backgroundColor: '#FF0000', width: '25px'}}), and the optimization effect is verified through A / B testing (1000 user interactions) (target click-through rate > 95%). Auditory perception information density adjustment is achieved by adjusting the frequency (range 500-2000Hz, obtained by mapping according to task urgency, with high frequency 1500Hz for urgent tasks and low frequency 800Hz for ordinary tasks) and duration (range 0.5-2 seconds, determined by the context-aware model output). This is implemented by generating prompts via the WebAudio API, with audio files pre-stored on the server (WAV format, size <100KB), and invoked via JavaScript (e.g., audio.play()). Workflow-aware automated correction optimizes operational processes by adjusting task sequence or workstation allocation. Parameters include task sequence (material delivery list, obtained from the shop floor plan, formatted as material code sequences such as ["M001", "M002"]) and workstation allocation (workstation ID, obtained from the SCADA system). This is obtained by querying the shop floor plan table (containing material codes and delivery order) via SQL and combining it with the SCADA system workstation status (read via Modbus protocol). The automated correction algorithm is based on a rule engine (Drools framework). The rule is defined as "If material M001 needs to be placed at workstation MIX01 but is scanned to PACK01, then adjust the task order and push a prompt." The rule base is stored in the database and loaded into the Drools engine in real time (loading time <0.2 seconds). The integration of implicit interaction technology is achieved through a web interface. React components listen to context-aware output and dynamically call visual, auditory, and workflow adjustment functions to ensure the strategy is seamlessly implemented (user-perceived interference rate <5%, obtained through a questionnaire survey of 100 operators).

[0067] 3. A policy execution engine with feedback learning capabilities. The strategy execution engine optimizes the accuracy and efficiency of intervention by evaluating the effectiveness of corrective strategies in real time and dynamically adjusting parameters. The engine's input data includes strategy execution results (operator response time, in seconds, collected via web interface logs; a success indicator, 1 for success, 0 for failure, obtained by comparing actual and expected operations) and production metrics (production efficiency, order completion rate, target >95%; quality error rate, percentage, target 80%). These metrics are calculated by querying log tables using SQL and stored in a database for subsequent analysis. The engine uses a random forest regression model to dynamically adjust strategy parameters. This model is implemented using the Scikit-learn library, and the training data consists of approximately 50,000 historical strategy execution records. Features include strategy type, initial parameters, context state (e.g., operator load, task risk), and labels indicating parameter adjustments after successful strategy execution (e.g., optimal increment for pop-up window size). The model is initialized through offline training and incrementally updated weekly based on newly added execution records.

[0068] 4. Closed-loop optimization of the digital twin evaluation system The digital twin evaluation system forms a closed loop for continuous optimization by feeding back strategy execution results to user profiles and a production behavior pattern library. User profiles include operator ID, operating habits (click frequency, response time), preferences (preference for visual or auditory cues, statistically derived from historical interaction data), and skill level (based on SOP execution accuracy, ranging from 0-100%, calculated from operation logs). The production behavior pattern library is based on an LSTM model (3 layers, 128 units per layer, training data of 500,000 operation logs), with input being operation sequences (timestamp, action type, material code) and output being behavior pattern classification (normal, abnormal). Data acquisition methods are as follows: user profiles are generated by querying log tables via SQL (e.g., operator ID "USER001" has an average response time of 1.5 seconds, a preference for pop-up prompts, and an accuracy of 90%); the behavior pattern library is incrementally updated daily using the TensorFlow framework (1000 new log entries added, batch size 32). The digital twin evaluation system, based on the ThingWorx platform, includes the following steps: importing strategy execution results (response time, success flags) into ThingWorx via REST API to update the status of digital twin entities (operators, equipment, workstations); analyzing strategy effectiveness by querying user profiles and behavior pattern libraries using SQL queries (e.g., pop-up prompt success rate of 85%); updating behavior classifications online using an LSTM model (anomaly probability threshold of 0.7, determined based on ROC curves); and updating user profiles to reflect operational habits (e.g., reducing response time from 1.5 seconds to 1.2 seconds). Closed-loop optimization uses a genetic algorithm (population size 50, 50 iterations) to adjust strategy parameters, aiming to maximize the success rate and minimize the interference rate (target <5%). Parameters include pop-up size and prompt audio frequency. The genetic algorithm is implemented in Python, taking strategy effectiveness data as input and outputting optimized parameter values ​​(e.g., pop-up size 28px), updating the strategy execution engine and web interface via API.

[0069] The fitness function of the genetic algorithm is defined as: Fitness = 0.7 * SuccessRate - 0.3 * InterferenceRate. This function guides the search direction, ensuring a high success rate while strictly suppressing user interference.

[0070] The definitions and acquisition methods of each parameter are as follows: Success Rate: Calculated as N_success / N_total, normalized to the range [0, 1]. N_success represents the number of successes, judged by the following criteria: within 5 seconds of strategy implementation, the operator completes the expected corrective action and the relevant anomaly index decreases by more than 20%; N_total represents the total number of strategy implementations within the same evaluation period (e.g., one production shift). Data is obtained by querying the strategy_logs table using SQL.

[0071] Interference Rate: Calculated as N_complaint / N_total, normalized to the range [0, 1]. N_complaint represents the number of complaints, collected through the system's built-in lightweight feedback mechanism (such as the "Yes / No" button for "Was this prompt annoying?"). Data is obtained by querying the feedback_logs and strategy_logs tables using SQL.

[0072] The weighting coefficients of 0.7 and 0.3 were determined based on historical data A / B testing to prioritize production efficiency while keeping user interference at an acceptable level (target value <5%).

[0073] The following is a specific implementation example of Example 10: In Example 10 of a food processing enterprise (such as a biscuit factory), a production monitoring method for the entire process of food ingredient feeding and traceability and data association is presented. This method utilizes a context-aware strategy delivery mechanism and implicit interaction technology to smoothly implement the first or second non-intrusive corrective strategies. Furthermore, it optimizes production monitoring efficiency through feedback learning and a digital twin evaluation system. Biscuit production involves various raw materials (such as flour, sugar, and butter), and the ingredient feeding and feeding processes must strictly adhere to standard operating procedures (SOPs) to ensure food safety and production efficiency. The system is deployed on a private cloud server (based on Ubuntu 20.04, equipped with 32GB of memory and an 8-core CPU). The database uses MySQL to store ingredient feeding, feeding, and operation log data. The web interface is developed based on the React framework. The device interface connects to an electronic scale (accuracy 0.01kg, data acquisition frequency 1Hz) and a SCADA system (collecting workstation status, frequency 1Hz) via an industrial gateway (supporting Modbus and RS-232 protocols). Production task orders are retrieved from the SAP ERP system via a RESTful API (URL " / api / tasks", synchronized hourly), containing order number "PO2023001", product code "BISCUIT01", production quantity 1000kg, and risk level "high" (involving allergenic flour). A recipe library stores biscuit recipes (50% flour, 20% sugar, 30% butter), entered and maintained via a web interface. A context-aware model (CRF algorithm, trained on 100,000 log entries) outputs a red pop-up notification (25px in size, #FF0000 background) based on the workstation status collected by the SCADA system (workstation "MIX01", status "running"), operator behavior recorded by the web interface (operator ID "USER001", click frequency 2 times / second, scanned material "M001" timestamp 2023-10-10 10:00:01), and the task criticality (risk level "high") provided by the ERP system. Implicit interaction technology dynamically adjusts the interface using the React framework (the "Confirm" button is enlarged to 25px, and its border is thickened by 2px), plays a 1500Hz notification sound (lasting 1 second) using the WebAudio API, and adjusts the task order using the Drools rule engine (prioritizing the feeding of material "M001," with the rule loaded from the database and executed in 0.15 seconds). When an operator ignores an alarm (response time > 5 seconds, anomaly index 0.75), the strategy execution engine (random forest model, 100 trees) collects response data (response time 5.2 seconds, correction failed), predicts parameter adjustment values ​​(enlarging the pop-up window to 30px), updates the web interface via API, and executes the new strategy (displaying a 30px pop-up window prompting "Please check flour feeding amount"), increasing the success rate to 82%.The digital twin assessment system updates operator profiles (“USER001” response time reduced to 1.3 seconds, preference pop-up prompts, accuracy 92%) and behavioral pattern library (LSTM model updated, anomaly probability reduced to 0.65) through the ThingWorx platform. The genetic algorithm optimizes the pop-up window size to 28px, achieving an intervention success rate of 85% and reducing the interference rate to 4%. Ultimately, the system controls the error rate within 0.2% in the material feeding stage, increasing production efficiency by 12% and reducing operator cognitive load by 15% (response time reduced from 1.5 seconds to 1.2 seconds), significantly improving the accuracy and safety of the production process.

[0074] Example 10 achieves seamless implantation of corrective strategies through a context-aware strategy delivery mechanism and implicit interaction technology, significantly reducing operator interference; the digital twin evaluation system forms a closed-loop optimization mechanism through continuous updates of user profiles and behavior pattern libraries, ensuring that the system adapts to dynamic production environments; the overall effect is improved production efficiency, enhanced quality stability, and improved safety compliance, providing strong technical support for accurate traceability and efficient monitoring of the entire food ingredient feeding process.

[0075] Example 11: Figure 2 This is a schematic diagram of the production monitoring system for tracing and linking the entire process of food ingredient feeding provided in Example 11, as shown below. Figure 2 As shown, the system includes: The data synchronization module is configured to automatically synchronize production task orders from the ERP system. The plan generation module is configured to generate a workshop plan based on the production task order and the recipe library. The ingredient management module is configured to automatically collect weight data via electronic scales during the ingredient preparation process, and control the label printer to print ingredient labels containing batch numbers, material codes, and supplier information based on the workshop plan. The material feeding verification module is configured to scan the material label using a scanning device during the material feeding process and verify the workstation status against the SCADA system to prevent errors. The data traceability module is configured to automatically collect operator, time, equipment and material data in the batching and feeding process, and bind them to the corresponding production batches to form a full traceability chain from raw material receipt to finished product; The central processing unit is configured to coordinate and control the collaborative work of the data synchronization module, the plan generation module, the batching management module, the material feeding verification module, and the data traceability module.

[0076] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A production monitoring method for tracing and data association throughout the entire process of food ingredient dispensing, characterized in that, include: Automatically synchronize production task orders from the ERP system; A workshop plan is generated based on the production task order and formula library; In the ingredient preparation process, weight data is automatically collected by electronic scales, and ingredient labels containing batch numbers, material codes and supplier information are printed based on the workshop plan. During the feeding process, the ingredient label is scanned and verified against the workstation status in the SCADA system to prevent errors. The system automatically collects operator, time, equipment, and material data from the batching and feeding processes, and links them to the corresponding production batches to form a complete traceability chain from raw material receipt to finished product.

2. The production monitoring method for full-process traceability and data association of food ingredient feeding according to claim 1, characterized in that, The formula library is maintained through formula management steps, which include formula entry, version management, process parameter setting, formula download, and formula review.

3. The production monitoring method for full traceability and data association of food ingredient feeding process according to claim 1, characterized in that, The ingredient preparation process also includes: Automatically detects whether the weighing data is within the preset error range, and only allows saving and execution of the next weighing if the data is within the error range; When the batching process is finished, the system automatically checks for any missing materials and alerts the operator if any are missing. Supports reprinting of the ingredient labels.

4. The production monitoring method for full traceability and data association of food ingredient feeding process according to claim 1, characterized in that, The feeding process also includes: After scanning the ingredient label, it is compared with the materials to be added in the workshop plan, and an error reminder is issued when the materials are inconsistent. Verify the material feeding point according to the workstation type to prevent materials from being fed to the wrong workstation.

5. The production monitoring method for full-process traceability and data association of food ingredient feeding according to claim 1, characterized in that, Also includes: By entering the production task number or time period, the corresponding ingredient and feeding records can be queried. The records include operation time, product code, raw material code, batch number, supplier and usage information.

6. The production monitoring method for full traceability and data association of food ingredient feeding process according to claim 1, characterized in that, Also includes: Initialize inventory material labels and print inventory labels for raw materials; Use a mobile handheld terminal to purchase and receive goods, scan the physical items and enter the batch number and quantity information, and print inventory labels; Perform on-line warehouse receiving, return, and inventory operations, and handle system tail errors caused by weighing errors.

7. The production monitoring method for full traceability and data association of food ingredient feeding process according to claim 1, characterized in that, Also includes: Data exchange is achieved through data interfaces with ERP systems, SCADA systems, and electronic scale equipment. The data exchange with the ERP system includes automatic synchronization of production task orders, basic material information and supplier information; Data exchange with the SCADA system includes reading the workstation status; Data exchange with the electronic scale is achieved through serial communication to realize real-time weight acquisition.

8. The production monitoring method for full traceability and data association of food ingredient feeding process according to claim 1, characterized in that, Also includes: Security auditing can be achieved by setting user permission controls to access system functions and recording operation logs.

9. The production monitoring method for full traceability and data association of food ingredient feeding process according to claim 1, characterized in that, Also includes: Construct a human-machine collaborative production monitoring model; When a user operates the production monitoring model, the user's counterintuitive behaviors are identified and evaluated. These counterintuitive behaviors include: cognitive decay patterns of critical production alarms, spatiotemporal inconsistencies in material delivery sequences, missing opportunity windows for responding to abnormal events, abnormal changes in the information entropy of operating steps, implicit deviations from standard operating procedures, and abnormal distribution of cognitive load in human-computer interaction. When the index of counterintuitive behavior exceeds the first dynamic threshold based on task criticality adaptation, the cognitive-risk dual-track assessment mechanism is activated to generate a real-time intervention capacity window. Based on the real-time intervention capacity window, a first non-intrusive correction strategy is planned, and its expected benefit level is quantified. If the expected benefit exceeds the second dynamic threshold based on trust evolution, then the first seamless correction strategy is implemented for the user. Otherwise, the predictive intervention mode is activated to generate and implement a second, non-intrusive corrective strategy for the user.

10. A production monitoring system for the entire process of food ingredient dispensing with traceability and data association, characterized in that: include: The data synchronization module is configured to automatically synchronize production task orders from the ERP system. The plan generation module is configured to generate a workshop plan based on the production task order and the recipe library. The ingredient management module is configured to automatically collect weight data via electronic scales during the ingredient preparation process, and control the label printer to print ingredient labels containing batch numbers, material codes, and supplier information based on the workshop plan. The material feeding verification module is configured to scan the material label using a scanning device during the material feeding process and verify the workstation status against the SCADA system to prevent errors. The data traceability module is configured to automatically collect operator, time, equipment and material data in the batching and feeding process, and bind them to the corresponding production batches to form a full traceability chain from raw material receipt to finished product; The central processing unit is configured to coordinate and control the collaborative work of the data synchronization module, the plan generation module, the batching management module, the material feeding verification module, and the data traceability module.

Images