Sauce production cooperative control method and system based on edge calculation

Abstract

The invention discloses a sauce production cooperative control method and system based on edge calculation, belongs to the technical field of industrial automation and edge calculation, and aims to solve the problems of control instability and low cloud fault diagnosis efficiency caused by instantaneous sudden change of sensor data in sauce production. The method comprises the steps that an edge control device collects sensing data and generates a dimension-reduced feature abstract copy; performing time sequence comparison on the current abstract and the previous abstract, and when the difference degree exceeds a preset threshold value, judging that the data is abruptly changed; then, abnormal data are prevented from entering a control loop, virtual control data are generated based on the characteristic parameters of the previous normal abstract and output to an execution mechanism, and instant self-healing is achieved; and meanwhile, the mutation event is packaged into a log packet to be asynchronously reported to the cloud. According to the invention, the anti-interference capability and the stability of the control loop are improved, an accurate diagnosis clue is provided for the cloud, and efficient cooperation is realized.

Description

Technical Field

[0001] This invention relates to the fields of industrial automation and edge computing technology, and more specifically, to a collaborative control method and system for sauce production based on edge computing. Background Technology

[0002] In continuous process industries like sauce production, stable control of process parameters such as temperature, viscosity, and pressure is crucial. Currently, a three-layer "cloud-edge-device" architecture is commonly used: field sensors (devices) collect data, edge control devices (edges) perform real-time closed-loop control, and the cloud platform stores data and performs macro-analysis.

[0003] Existing technical solution 1: The edge control device directly uses raw sensor data for PID calculation and outputs control signals. The problem with this solution is that when the sensor is subjected to momentary interference, a brief failure in the transmission link, or abnormal data packets, abrupt erroneous data will be generated. This erroneous data will directly enter the control loop, causing actuators (such as regulating valves and stirring motors) to malfunction, disrupting the stability of the production process and affecting the quality of the sauce. Due to the extremely short control cycle (typically on the order of milliseconds), such momentary errors are difficult to detect and correct in a timely manner.

[0004] Existing technical solution two: The edge control device uploads all raw data to the cloud for data quality analysis and fault diagnosis. The problem with this solution is that the diagnosis is severely delayed. Cloud analysis is time-consuming and cannot provide real-time responses to instantaneous data anomalies at the millisecond / second level. Simultaneously, fault information is buried under massive amounts of normal data, requiring the traversal of large amounts of historical data during troubleshooting, resulting in low efficiency and hindering rapid location and root cause analysis of production anomalies.

[0005] In summary, existing sauce production control systems suffer from technical problems such as unstable edge control loops due to instantaneous changes in sensor data, resulting in low efficiency in cloud-based fault diagnosis and difficulty in location. Summary of the Invention

[0006] The purpose of this invention is to provide a collaborative control method and system for sauce production based on edge computing, so as to solve the technical problems in existing sauce production control, such as the instability of edge control loop caused by instantaneous changes in sensor data, and the resulting low efficiency of cloud fault diagnosis and difficulty in positioning.

[0007] To address the aforementioned technical problems, this invention provides a collaborative control method for sauce production based on edge computing, comprising the following steps: S1: Collect physical quantity sensing data on the production line in real time through the sensor interface, and mark the physical quantity sensing data as the input data at the current time; S2: Call the preset feature extraction algorithm to perform dimensionality reduction encoding on the input data at the current time, generate a copy of the current time feature summary containing trend feature parameters, and write the copy of the current time feature summary into the local cache queue; S3: Read the previous time-lapse feature summary copy stored in the local cache queue, calculate the temporal feature difference between the current time-lapse feature summary copy and the previous time-lapse feature summary copy, and compare the temporal feature difference with a preset mutation threshold; S4: When the time-series feature difference is greater than the preset mutation threshold, a blocking operation is performed to prevent the current input data from entering the downstream control loop; at the same time, the feature model parameters of the previous time-series feature summary copy are called, virtual control data is generated through the reverse reconstruction algorithm, and the virtual control data is used to replace the current input data and output to the actuator. S5: Encapsulate the timestamp of the mutation, the blocked current input data, and the generated virtual control data into an anomaly log package, and send it to the cloud server through an asynchronous channel independent of the downstream control loop.

[0008] As a further improvement to this technical solution, in step S2, a preset feature extraction algorithm is invoked to perform dimensionality reduction encoding on the input data at the current moment, specifically including: Calculate the first-order difference between the current input data and the previous input data to obtain the slope parameter. ; The amplitude of the input data at the current moment is obtained as the intercept parameter. ; The change slope parameter and the intercept parameter The features are combined and constructed into a feature vector, which serves as a copy of the feature summary at the current time; wherein the storage space occupied by the feature summary copy at the current time is less than the storage space occupied by the input data at the current time.

[0009] As a further improvement to this technical solution, in step S2, before writing the current time feature summary copy into the local cache queue, the method further includes: preprocessing the current time input data, the preprocessing including filtering to remove high-frequency noise; the feature extraction algorithm is executed based on the preprocessed data.

[0010] As a further improvement to this technical solution, in step S4, the feature model parameters of the previous time-lapse feature summary copy are called, and virtual control data is generated through a reverse reconstruction algorithm, specifically including: Parse the previous time step feature summary copy and extract the change slope parameter of the previous time step. Intercept parameter of the previous time step ; Obtain the sampling time interval of the data acquisition ; Using linear extrapolation formula The calculated virtual control data is denoted as... .

[0011] As a further improvement to this technical solution, in step S3, before comparing the temporal feature difference degree with the preset mutation threshold, a step of dynamically configuring the mutation threshold is also included: Receive the production process formula instruction issued by the cloud server, and parse the current process stage identifier, which includes the heating and stirring stage, the constant temperature fermentation stage, or the cooling and filling stage. Query the fluctuation tolerance value corresponding to the process stage identifier in the locally stored threshold mapping table; The fluctuation tolerance value is loaded as the preset mutation threshold; The fluctuation tolerance value corresponding to the heating and stirring stage is greater than the fluctuation tolerance value corresponding to the constant temperature fermentation stage.

[0012] As a further improvement to this technical solution, in step S4, the virtual control data is used to replace the current input data and output to the actuator, specifically including: The virtual control data is converted into analog current signals or bus control commands; The analog current signal or the bus control command is sent to the PID controller; The PID controller adjusts the frequency of the stirring motor's inverter or the opening of the steam pipeline regulating valve based on the received analog current signal or bus control command.

[0013] As a further improvement to this technical solution, the physical quantity sensing data includes the viscosity value, temperature value, liquid level value, or pipeline pressure value of the sauce. The local cache queue is configured as a first-in-first-out queue, and the length of the local cache queue is [missing information]. When a new copy of the feature summary is written, the earliest timestamped copy of the feature summary is automatically discarded.

[0014] As a further improvement to this technical solution, in step S5, the data is sent to the cloud server through an asynchronous channel independent of the downstream control loop, including: The exception log packet is stored in the pending transmission area of ​​the local non-volatile memory; Detect the uplink network idle status of edge control devices; When an uplink network idle is detected, the abnormal log packet is published to the fault diagnosis topic on the cloud server using the MQTT or HTTP protocol. The network priority of the asynchronous channel is lower than the real-time control network priority of the downstream control loop.

[0015] An edge computing-based collaborative control system for sauce production, wherein the edge computing-based collaborative control system for sauce production is used to implement the aforementioned edge computing-based collaborative control method for sauce production, comprising: The data acquisition module collects physical quantity sensing data from the production line in real time through the sensor interface and marks the physical quantity sensing data as the input data at the current moment; The copy generation module is connected to the data acquisition module. It performs dimensionality reduction encoding on the input data at the current moment by calling a preset feature extraction algorithm, generates a current moment feature summary copy containing trend feature parameters, and writes the current moment feature summary copy into a local cache queue. The verification and analysis module is connected to the copy generation module. It reads the previous time-of-flight feature summary copy stored in the local cache queue, calculates the temporal feature difference between the current time-of-flight feature summary copy and the previous time-of-flight feature summary copy, and compares the temporal feature difference with a preset mutation threshold. The self-healing control module, connected to the verification and analysis module, is configured to perform a blocking operation when the temporal feature difference is greater than the preset mutation threshold, prohibiting the current input data from entering the downstream control loop; at the same time, it calls the feature model parameters of the feature summary copy of the previous moment, generates virtual control data through a reverse reconstruction algorithm, and outputs the virtual control data to the actuator instead of the current input data. The cloud collaboration module, connected to the self-healing control module, is used to encapsulate the timestamp of the mutation, the blocked current input data, and the generated virtual control data into an abnormal log package, which is then sent to the cloud server through an asynchronous channel independent of the downstream control loop.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. In this edge computing-based collaborative control method and system for sauce production, by comparing the current data with historical feature summaries in real time, it can identify data mutations in real time and immediately block abnormal data from flowing into the control loop. At the same time, it uses the normal feature parameters from the previous moment to generate smooth virtual control data output, effectively avoiding control command jumps caused by single-point instantaneous interference, and ensuring the instantaneous stability of the core process segment of sauce production (such as constant temperature fermentation).

[0017] 2. In this edge computing-based collaborative control method and system for sauce production, the edge side not only handles anomalies but also encapsulates detailed anomaly contexts into structured event logs and reports them to the cloud. This provides a high signal-to-noise ratio roadmap for cloud analysis, enabling the cloud to directly locate the precise time and data segment where the problem occurred. This transforms fault analysis from a full data traversal to focused analysis, improving diagnostic efficiency and accuracy.

[0018] 3. In this edge computing-based collaborative control method and system for sauce production, the mutation threshold is dynamically adjusted by receiving formula instructions from the cloud. This allows the sensitivity of data anomaly detection to match the specific production process stage (such as drastic temperature changes and stirring versus stable constant temperature fermentation), avoiding false alarms or missed alarms that may occur when using a fixed threshold, thus making the control strategy more intelligent. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating the overall method of the present invention; Figure 2 This is an overall system block diagram of the present invention. Detailed Implementation

[0020] 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.

[0021] Example 1 In continuous process industries such as sauce production, stable control of process parameters such as temperature, viscosity, and pressure is crucial. Currently, a three-layer architecture of "cloud-edge-device" is commonly used to coordinate the control of sauce production. However, there are technical problems such as the instability of the edge control loop caused by instantaneous changes in sensor data, as well as the resulting low efficiency of cloud fault diagnosis and difficulty in positioning.

[0022] In view of this, please refer to Figure 1 As shown, one of the objectives of this invention is to provide a collaborative control method for sauce production based on edge computing, applied to edge control devices deployed at sauce production sites. The method includes the following steps: S1: Collect physical quantity sensing data on the production line in real time through the sensor interface, and mark the physical quantity sensing data as the input data at the current time; S2: Call the preset feature extraction algorithm to perform dimensionality reduction encoding on the input data at the current time, generate a copy of the current time feature summary containing trend feature parameters, and write the copy of the current time feature summary into the local cache queue; S3: Read the previous time-lapse feature summary copy stored in the local cache queue, calculate the temporal feature difference between the current time-lapse feature summary copy and the previous time-lapse feature summary copy, and compare the temporal feature difference with a preset mutation threshold; S4: When the time-series feature difference is greater than the preset mutation threshold, a blocking operation is performed to prevent the current input data from entering the downstream control loop; at the same time, the feature model parameters of the previous time-series feature summary copy are called, virtual control data is generated through the reverse reconstruction algorithm, and the virtual control data is used to replace the current input data and output to the actuator. S5: Encapsulate the timestamp of the mutation, the blocked current input data, and the generated virtual control data into an anomaly log package, and send it to the cloud server through an asynchronous channel independent of the downstream control loop.

[0023] The implementation process of this invention is described in detail below with reference to specific steps: Considering that physical parameters such as viscosity, temperature, liquid level, and pipeline pressure directly reflect the production status during sauce production, real-time acquisition of these parameters is the foundation for subsequent control. Edge devices collect data on-site, avoiding the uploading of large amounts of raw data to the cloud and reducing network transmission pressure. Therefore, the purpose of step S1 is to collect and label physical quantity sensor data in real time, as specifically implemented as follows: The edge control device is equipped with multiple sensor interfaces, including a 4-20mA analog signal interface, an RS485 bus interface, and an Ethernet interface, which are respectively connected to the viscosity sensor, temperature sensor, liquid level sensor, and pressure sensor of the sauce production line; The physical quantity sensing data includes the viscosity, temperature, liquid level, and pipeline pressure of the sauce, which are collected by the corresponding viscosity sensor, temperature sensor, liquid level sensor, and pressure sensor, respectively. At the same time, the sauce production site is characterized by high temperature, high viscosity, and corrosiveness, so high temperature resistant (0-150℃) and corrosion resistant sensors must be selected, such as rotary viscosity sensors and armored thermocouple temperature sensors, to ensure the stability and accuracy of data acquisition. The sensor collects data at a preset sampling frequency (in this embodiment, the sampling frequency is set to 10Hz. The basis for determining this sampling frequency is that the change cycle of the sauce production process parameters is usually greater than 0.1s, and a 10Hz sampling frequency can completely capture the parameter change trend while avoiding excessive computational load on the edge device caused by high-frequency sampling). After receiving the data through the sensor interface, the edge control device adds a collection timestamp to each set of data, marking it as the input data at the current time. ; Through the above steps, real-time process parameters of the production line are obtained, providing raw data for subsequent feature extraction and anomaly detection; local data collection at the edge reduces the amount of data transmitted to the cloud and lowers network bandwidth usage.

[0024] Considering that the sensor may be subject to electromagnetic interference from production line motors, frequency converters, and other equipment during data acquisition, generating high-frequency noise (interference frequency is usually above 50Hz), this noise will be superimposed on the effective signal, causing distortion of parameter change trends and affecting the accuracy of subsequent feature extraction. Therefore, the purpose of step S2 is to perform feature extraction and caching on the data acquired in step S1; the specific implementation method is as follows: The first step is preprocessing: During data acquisition, sensors are subject to electromagnetic interference from production line motors, frequency converters, and other equipment, generating high-frequency noise (interference frequencies are typically above 50Hz). This noise is superimposed on the valid signal, distorting parameter trends and affecting the accuracy of subsequent feature extraction. Therefore, data filtering is necessary before initiating the process. This operation filters the input data at the current moment before calling the feature extraction algorithm. Low-pass filtering is performed. In this embodiment, a moving average filtering algorithm is used, with the filtering window size set to 5 sampling points (the 5 sampling points are determined as follows: the sampling frequency in this embodiment is 10Hz, the single sampling period is 0.1s, and the window time corresponding to 5 sampling points is 0.5s; the electromagnetic interference on the production line is a 50Hz high-frequency signal, and the 0.5s window time can cover at least 25 interference cycles. The moving average can effectively cancel out the positive and negative fluctuations of high-frequency interference). The calculation formula is as follows: ,in, The current time-time input data is filtered; high-frequency noise in the input data is removed through preprocessing operations to improve data quality and ensure the reliability of subsequent feature extraction results. The second step is feature extraction and encoding: The raw sensor data is large in volume, and direct storage and computation would consume the limited storage resources of edge devices. Generating feature summary copies through dimensionality reduction encoding can significantly reduce data storage volume while preserving the core trend features of the data, facilitating subsequent time-series comparisons. The specific process is as follows: The feature extraction algorithm pre-installed in the edge control device is invoked to process the filtered input data at the current time. Processing: First, calculate the first-order difference between the current input data and the previous input data to obtain the slope parameter. The calculation formula is: ,in, The sampling time interval for data acquisition (in this embodiment) =0.1s, determined by a 10Hz sampling frequency); Secondly, the amplitude of the input data at the current moment is obtained as the intercept parameter. ,Right now ; Next, the slope parameter will be changed. and intercept parameter Combined to construct a feature vector ( , ), which serves as a copy of the feature summary at the current moment.

[0025] It should be noted that in this embodiment, the original input data is a 16-bit precision value, and each set of data occupies 2 bytes of storage space, while the feature summary copy contains only two floating-point parameters and occupies 8 bytes of storage space. When 10 sets of data are collected, the original data occupies 20 bytes, while the feature summary copy occupies only 8 bytes, thus achieving data dimensionality reduction. Through the above steps, the generated feature summary copy retains the data's trend characteristics while significantly reducing storage usage; it also reduces the computational and storage load on edge devices and improves data processing efficiency. The third step is to write to the local cache queue: To perform a time-series comparison with the feature summary copy for the next time step, a temporary storage of the current time step's feature summary copy is required. A first-in-first-out (FIFO) queue structure is used to automatically manage historical data and avoid data redundancy. Therefore, the edge control device has a built-in FIFO cache queue with a queue length of... Set to 10 (queue length) The basis for setting it to 10 is as follows: This invention only needs a copy of the feature summary from the previous moment for difference calculation. Setting the queue length to 10 can store the historical data of the most recent 1 second (10Hz sampling frequency). Even if a single queue read failure occurs, the parameters of the previous moment can still be retrieved from the historical data, improving the system's fault tolerance. The generated copy of the feature summary at the current moment is written to the tail of the queue. When the queue length reaches 10, the feature summary copy with the earliest timestamp at the head of the queue is automatically discarded. Through the above steps, the orderly storage and management of feature summary copies are achieved, providing data support for subsequent time-series feature difference calculation; expired data is automatically discarded to avoid cache queue overflow.

[0026] Considering that determining whether the process parameters at the current moment have undergone a sudden change requires comparison with the parameter characteristics at the previous moment, it is necessary to read a copy of the feature summary from the previous moment from the cache queue. Therefore, the purpose of step S3 is to calculate the temporal feature difference and compare the threshold; the specific implementation method is as follows: First, read the feature summary copy from the previous time step: To determine whether the process parameters at the current time step have undergone a sudden change, it is necessary to compare them with the parameter characteristics at the previous time step. Therefore, it is necessary to read the feature summary copy from the previous time step from the cache queue. The specific process is as follows: Read the previous element from the tail of the first-in-first-out cache queue, which is the feature summary copy from the previous time step. , ),in, The slope parameter is the change in the previous time step. This is the intercept parameter from the previous time step; Secondly, the temporal feature difference degree is calculated: by calculating the degree of difference between the current time and the feature summary copy of the previous time, the change range of process parameters can be quantified, and it can be determined whether a sudden change has occurred; this embodiment uses the Euclidean distance algorithm to calculate the temporal feature difference degree. The calculation formula is: ,in, , The parameters of the current time feature summary copy are used to quantify the magnitude of parameter changes in a numerical manner, providing a precise basis for mutation judgment; Then, the mutation threshold is dynamically configured: different stages of sauce production have different tolerances for parameter fluctuations. For example, the temperature rises rapidly during the heating and stirring stage, resulting in larger parameter fluctuations and higher tolerance; the isothermal fermentation stage requires stable parameters and has lower tolerance. Using a fixed threshold can lead to false positives or false negatives; therefore, the threshold needs to be dynamically configured. The specific implementation process is as follows: The edge control device receives the production process formula instruction from the cloud server via Ethernet. The instruction contains the current process stage identifier, which includes the heating and stirring stage, the constant temperature fermentation stage, and the cooling and filling stage. The edge control device has a locally stored threshold mapping table. This table stores the fluctuation tolerance values ​​corresponding to each process stage, determined through a limited number of production experiments. Specifically, parameter data for each process stage of 100 batches of sauce production were collected, and the maximum time-series characteristic difference under normal production conditions was statistically analyzed. To avoid misjudgment, a 20% increase was added to the normal maximum difference value as the abrupt change threshold for each stage. The final determined value is: the fluctuation tolerance value for the heating and stirring stage is 0.8 (Heating and stirring stage: the sauce is in a flowing state, the heating rate is fast, and the parameter changes drastically; the normal maximum difference value is 0.67, which is set to 0 after a 20% increase). 0.8, which can cover normal process fluctuations and avoid accidental triggering of blocking operations), the fluctuation tolerance value corresponding to the constant temperature fermentation stage is 0.3 (constant temperature fermentation stage: the sauce is in a steady state of fermentation, the parameters need to be kept constant, the normal maximum difference is 0.25, and after a 20% increase, it is set to 0.3, which can accurately identify abnormal mutations and avoid parameter fluctuations affecting product quality), the fluctuation tolerance value corresponding to the cooling and filling stage is 0.5 (cooling and filling stage: the sauce temperature slowly decreases and the viscosity gradually increases, the fluctuation degree is between the first two stages, the normal maximum difference is 0.42, and after a 20% increase, it is set to 0.5, which takes into account both production efficiency and quality control). After parsing the process stage identifier, the edge control device queries the corresponding fluctuation tolerance value from the threshold mapping table and loads this value as the preset mutation threshold. ; Through the above steps, the threshold is accurately adapted to the production needs of different process stages, which greatly improves the accuracy of mutation identification and avoids quality problems such as misjudgment, shutdown or missed judgment caused by fixed threshold.

[0027] Finally, threshold comparison: The calculated time-series feature difference is compared with the mutation threshold to determine whether a sudden change has occurred in the current process parameters. Specifically, the time-series feature difference is... With mutation threshold Perform a size comparison, if > If the parameter changes abruptly, then it is determined that a sudden change has occurred; if ≤ If the parameter is within the normal fluctuation range, the input data at the current moment will directly enter the downstream control loop.

[0028] The purpose of step S4 is to generate blocking and virtual control data during mutation, and the specific process is as follows: When parameters change abruptly, the current input data becomes abnormal (e.g., a sudden pressure change due to sensor malfunction or pipe blockage). If this data enters the downstream control loop, it can cause malfunctions in the actuators (e.g., an abnormally large opening of the steam regulating valve or a sudden increase in the speed of the stirring motor), leading to uneven mixing of the sauce, temperature runaway, and other quality problems. Therefore, it is necessary to block abnormal data. In view of this, the edge control device has a built-in control loop switch. When a parameter change is detected, the control switch is turned off, preventing the current input data from entering the downstream control loop, thus performing a blocking operation to avoid malfunctions in the actuators caused by abnormal data and preventing product quality fluctuations from the source. If the control command output is directly interrupted after blocking abnormal data, the production line will stop, and the loss from a single batch of downtime can reach tens of thousands of yuan. Generating virtual control data based on the normal characteristic parameters from the previous moment can maintain the continuity of the production process until the fault is cleared. Therefore, the feature model parameters of the feature summary copy from the previous moment are called (…). , The virtual control data is generated through a reverse reconstruction algorithm. The specific steps are as follows: First, parse the feature summary copy from the previous time step and extract the slope parameter. and intercept parameter ; Then, obtain the sampling time interval for data acquisition. In this embodiment =0.1s (determined by a 10Hz sampling frequency); Finally, the virtual control data is calculated using a linear extrapolation formula, which is: ,in, This is virtual control data. It should be further explained that the parameter changes in each process stage of sauce production show a linear trend in a short time (0.1s). The virtual data is generated by using a linear extrapolation formula, which can not only keep the parameter trend consistent with the previous moment, but also has the advantages of simple calculation and strong real-time performance, making it suitable for the rapid processing needs of edge devices. The steps of generating virtual control data through the above reverse reconstruction algorithm are used to generate virtual data based on historical normal parameters, maintain the continuous operation of the production line, and avoid downtime losses; the linear extrapolation algorithm has low computational load and ensures the real-time generation of virtual data; Actuators (such as frequency converters for stirring motors and regulating valves for steam pipelines) typically receive 4-20mA analog current signals or Modbus bus control commands. Virtual control data needs to be converted into standard signals or commands to control the actuators. Therefore, edge control devices have built-in signal conversion modules to convert virtual control data... It converts the signal into a 4-20mA analog current signal or Modbus bus control command, and sends the converted analog current signal or bus control command to the PID controller. The PID controller adjusts the frequency of the frequency converter of the stirring motor or the opening of the steam pipeline regulating valve according to the received signal or command, thereby completing the purpose of virtual data output to the actuator. Through the above steps, seamless integration of virtual data with existing actuators is achieved without modifying existing production line equipment, reducing system transformation costs; and the stable operation of actuators is maintained, ensuring the continuity of sauce production.

[0029] To facilitate subsequent analysis of the causes of parameter mutations and optimization of production processes by the cloud server, it is necessary to completely encapsulate the mutation-related information. Therefore, the purpose of step S5 is to encapsulate the exception log package and upload it asynchronously. The specific process is as follows: Input the timestamp of the mutation and the current time when the blockage occurred into the data. Generated virtual control data The logs are packaged into anomaly log packages according to a preset format. The log package format is: [timestamp][input data][virtual data], thereby completely recording mutation-related information and providing comprehensive data support for cloud-based fault diagnosis and process optimization. Uploading log packets via the control loop's network channel will consume bandwidth for real-time control, affecting the transmission efficiency of control commands. Uploading via an independent asynchronous channel can avoid impacting the control loop. The specific process is as follows: First, the exception log packet is stored in the send area of ​​the edge control device's local non-volatile memory (such as an SD card); Then, the edge control device has a built-in network detection module to detect the idle status of the uplink network in real time; When an uplink network idle is detected, the abnormal log packet is published to the fault diagnosis topic on the cloud server using the MQTT protocol. At the same time, the network priority of the asynchronous channel is set lower than the real-time control network priority of the downstream control loop to ensure that the transmission of control commands is not affected.

[0030] Through the above steps, the real-time performance of control commands is guaranteed without consuming real-time control network bandwidth; non-volatile memory caches log packets to avoid data loss due to network interruptions; the MQTT protocol is lightweight and has low bandwidth characteristics, making it suitable for communication between edge devices and the cloud.

[0031] Example 2 To implement the edge computing-based collaborative control method for sauce production described in Example 1, please refer to... Figure 2As shown, the purpose of Embodiment 2 is to provide a sauce production collaborative control system based on edge computing, which includes: The data acquisition module collects physical quantity sensing data from the production line in real time through the sensor interface and marks the physical quantity sensing data as the input data at the current moment; The copy generation module is connected to the data acquisition module. It performs dimensionality reduction encoding on the input data at the current moment by calling a preset feature extraction algorithm, generates a current moment feature summary copy containing trend feature parameters, and writes the current moment feature summary copy into a local cache queue. The verification and analysis module is connected to the copy generation module. It reads the previous time-of-flight feature summary copy stored in the local cache queue, calculates the temporal feature difference between the current time-of-flight feature summary copy and the previous time-of-flight feature summary copy, and compares the temporal feature difference with a preset mutation threshold. The self-healing control module, connected to the verification and analysis module, is configured to perform a blocking operation when the temporal feature difference is greater than the preset mutation threshold, prohibiting the current input data from entering the downstream control loop; at the same time, it calls the feature model parameters of the feature summary copy of the previous moment, generates virtual control data through a reverse reconstruction algorithm, and outputs the virtual control data to the actuator instead of the current input data. The cloud collaboration module, connected to the self-healing control module, is used to encapsulate the timestamp of the mutation, the blocked current input data, and the generated virtual control data into an abnormal log package, which is then sent to the cloud server through an asynchronous channel independent of the downstream control loop.

[0032] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A collaborative control method for sauce production based on edge computing, characterized in that, The method, applied to edge control devices deployed at a sauce production site, includes the following steps: S1: Collect physical quantity sensing data on the production line in real time through the sensor interface, and mark the physical quantity sensing data as the input data at the current time; S2: Call the preset feature extraction algorithm to perform dimensionality reduction encoding on the input data at the current time, generate a copy of the current time feature summary containing trend feature parameters, and write the copy of the current time feature summary into the local cache queue; S3: Read the previous time-lapse feature summary copy stored in the local cache queue, calculate the temporal feature difference between the current time-lapse feature summary copy and the previous time-lapse feature summary copy, and compare the temporal feature difference with a preset mutation threshold; S4: When the time-series feature difference is greater than the preset mutation threshold, a blocking operation is performed to prevent the current input data from entering the downstream control loop; at the same time, the feature model parameters of the previous time-series feature summary copy are called, virtual control data is generated through the reverse reconstruction algorithm, and the virtual control data is used to replace the current input data and output to the actuator. S5: Encapsulate the timestamp of the mutation, the blocked current input data, and the generated virtual control data into an anomaly log package, and send it to the cloud server through an asynchronous channel independent of the downstream control loop.

2. The edge computing-based collaborative control method for sauce production according to claim 1, characterized in that, In step S2, a preset feature extraction algorithm is invoked to perform dimensionality reduction encoding on the input data at the current time, specifically including: Calculate the first-order difference between the current input data and the previous input data to obtain the slope parameter. ; The amplitude of the input data at the current moment is obtained as the intercept parameter. ; The change slope parameter and the intercept parameter The features are combined and constructed into a feature vector, which serves as a copy of the feature summary at the current time; wherein the storage space occupied by the feature summary copy at the current time is less than the storage space occupied by the input data at the current time.

3. The edge computing-based collaborative control method for sauce production according to claim 1, characterized in that: In step S2, before writing the current time feature summary copy into the local cache queue, the method further includes: preprocessing the current time input data, the preprocessing including filtering to remove high-frequency noise; the feature extraction algorithm is executed based on the preprocessed data.

4. The edge computing-based collaborative control method for sauce production according to claim 1, characterized in that, In step S4, the feature model parameters of the previous time-lapse feature summary copy are called, and virtual control data is generated through a reverse reconstruction algorithm, specifically including: Parse the previous time step feature summary copy and extract the change slope parameter of the previous time step. Intercept parameter of the previous time step ; Obtain the sampling time interval of the data acquisition ; Using linear extrapolation formula The calculated virtual control data is denoted as... .

5. The edge computing-based collaborative control method for sauce production according to claim 1, characterized in that, In step S3, before comparing the temporal feature difference with the preset mutation threshold, the step of dynamically configuring the mutation threshold is also included: Receive the production process formula instruction issued by the cloud server, and parse the current process stage identifier, which includes the heating and stirring stage, the constant temperature fermentation stage, or the cooling and filling stage. Query the fluctuation tolerance value corresponding to the process stage identifier in the locally stored threshold mapping table; The fluctuation tolerance value is loaded as the preset mutation threshold; The fluctuation tolerance value corresponding to the heating and stirring stage is greater than the fluctuation tolerance value corresponding to the constant temperature fermentation stage.

6. The edge computing-based collaborative control method for sauce production according to claim 1, characterized in that, In step S4, the virtual control data is used to replace the current input data and output to the actuator, specifically including: The virtual control data is converted into analog current signals or bus control commands; The analog current signal or the bus control command is sent to the PID controller; The PID controller adjusts the frequency of the stirring motor's inverter or the opening of the steam pipeline regulating valve based on the received analog current signal or bus control command.

7. The edge computing-based collaborative control method for sauce production according to claim 1, characterized in that, The physical quantity sensing data includes the viscosity value, temperature value, liquid level value, or pipeline pressure value of the sauce. The local cache queue is configured as a first-in-first-out queue, and the length of the local cache queue is [missing information]. When a new copy of the feature summary is written, the earliest timestamped copy of the feature summary is automatically discarded.

8. The edge computing-based collaborative control method for sauce production according to claim 1, characterized in that, In step S5, the data is sent to the cloud server via an asynchronous channel independent of the downstream control loop, including: The exception log packet is stored in the pending transmission area of ​​the local non-volatile memory; Detect the uplink network idle status of edge control devices; When an uplink network idle is detected, the abnormal log packet is published to the fault diagnosis topic on the cloud server using the MQTT or HTTP protocol. The network priority of the asynchronous channel is lower than the real-time control network priority of the downstream control loop.

9. A collaborative control system for sauce production based on edge computing, wherein the collaborative control system for sauce production based on edge computing is used to implement the collaborative control method for sauce production based on edge computing according to any one of claims 1 to 8, characterized in that, include: The data acquisition module collects physical quantity sensing data from the production line in real time through the sensor interface and marks the physical quantity sensing data as the input data at the current moment; The copy generation module is connected to the data acquisition module. It performs dimensionality reduction encoding on the input data at the current moment by calling a preset feature extraction algorithm, generates a current moment feature summary copy containing trend feature parameters, and writes the current moment feature summary copy into a local cache queue. The verification and analysis module is connected to the copy generation module. It reads the previous time-of-flight feature summary copy stored in the local cache queue, calculates the temporal feature difference between the current time-of-flight feature summary copy and the previous time-of-flight feature summary copy, and compares the temporal feature difference with a preset mutation threshold. The self-healing control module, connected to the verification and analysis module, is configured to perform a blocking operation when the temporal feature difference is greater than the preset mutation threshold, prohibiting the current input data from entering the downstream control loop; at the same time, it calls the feature model parameters of the feature summary copy of the previous moment, generates virtual control data through a reverse reconstruction algorithm, and outputs the virtual control data to the actuator instead of the current input data. The cloud collaboration module, connected to the self-healing control module, is used to encapsulate the timestamp of the mutation, the blocked current input data, and the generated virtual control data into an abnormal log package, which is then sent to the cloud server through an asynchronous channel independent of the downstream control loop.

Images