A multi-protocol silicon carbide wafer production data acquisition method and system based on edge computing

Abstract

This invention discloses a multi-protocol silicon carbide wafer production data acquisition method and system based on edge computing. The method includes the following steps: S1: Multi-protocol adaptation and retrofitting of aging equipment. This invention solves the problem of heterogeneity of different equipment protocols by constructing a multi-protocol adaptive parsing module, integrating mainstream industrial protocol drivers and private protocol adaptation interfaces. At the same time, it realizes data acquisition and local preprocessing based on edge computing architecture, combined with dual filtering, dual anomaly identification and missing data compensation schemes, and the IEEE 1588 PTP clock synchronization protocol, which solves the problems of poor data quality and timing disorder in extreme production environments, ensuring data accuracy and continuity. By adding IO acquisition modules and high-precision sensors to aging "dumb" equipment, it fills the data acquisition blind spots and realizes complete data acquisition of all processes and all types of equipment on the production line.

Description

Technical Field

[0001] This invention relates to the field of multi-protocol silicon carbide wafer manufacturing technology, and in particular to a multi-protocol silicon carbide wafer manufacturing data acquisition method and system based on edge computing. Background Technology

[0002] As a core material for third-generation semiconductors, silicon carbide (SiC) wafers involve multiple complex processes such as crystal growth, epitaxy, etching, polishing, and testing. These processes involve various types of equipment, including crystal growth furnaces, epitaxial equipment, etching machines, and testing instruments. Furthermore, different equipment uses various industrial protocols such as Modbus, PROFIBUS, EtherCAT, and OPC UA. Some high-end imported equipment also uses encrypted proprietary protocols, resulting in significant protocol heterogeneity challenges in data acquisition. To address the issues of poor real-time performance and high bandwidth consumption in traditional centralized data acquisition methods, the industry has gradually adopted edge computing architectures to build multi-protocol data acquisition systems. These systems collect data from nearby devices via edge gateways, enabling local preprocessing and real-time transmission. However, in the demanding environment of silicon carbide wafer production, these systems still suffer from significant data quality and acquisition integrity issues, specifically: 1. The silicon carbide production environment presents extreme conditions such as high temperatures (>1000℃), strong electromagnetic interference, dust, and vibration, leading to sensor signal drift, data jumps, and an increase in outliers. Furthermore, factors such as multi-protocol polling, network jitter, and insufficient gateway caching can easily cause data packet loss, timing discrepancies, and duplication. 1. Data: Silicon carbide (SiC) processes have extremely high requirements for the timing accuracy, continuity, and measurement precision of data. Even minor data errors can directly affect wafer yield and defect traceability. 2. Many outdated "dumb" devices on the production line lack communication interfaces and only support analog / switching outputs, resulting in incomplete data acquisition and data blind spots. 3. Existing edge gateways can only perform simple data filtering and cannot accurately reduce noise, correct anomalies, or compensate for missing data specific to the characteristics of SiC production data. This further exacerbates data quality risks, making the acquired data "acquired but unusable" and unable to provide reliable data support for subsequent process optimization, yield improvement, and end-to-end traceability. In response to the above problems, this application proposes a multi-protocol silicon carbide wafer production data acquisition method and system based on edge computing. Summary of the Invention

[0003] Based on the technical problems existing in the background technology, the present invention proposes a multi-protocol silicon carbide wafer production data acquisition method and system based on edge computing.

[0004] This invention proposes a multi-protocol silicon carbide wafer production data acquisition method based on edge computing, comprising the following steps: S1: Multi-protocol Adaptation and Retrofitting of Legacy Equipment: Construct a multi-protocol adaptive parsing module, integrating mainstream industrial protocol parsing drivers and private protocol adaptation interfaces to realize data parsing of multi-protocol equipment; add IO acquisition modules and high-precision sensors to legacy "dumb" equipment to convert analog and switch signals into digital signals and connect them to the edge gateway; and adopt the IEEE1588 PTP clock synchronization protocol to perform unified clock synchronization on each acquisition node, edge gateway, and production line equipment to eliminate measurement delays and ensure data timing consistency. S2: Edge-side data preprocessing: A dual filtering algorithm of Kalman filtering and moving average filtering is used to reduce noise in the collected data. A dual anomaly recognition model combining process threshold and machine learning is constructed to identify and correct abnormal data. Uncorrectable abnormal data is marked and an early warning is triggered. A scheme of cache queue + LSTM neural network prediction is used to compensate for missing data. The preprocessed data is standardized and transformed to construct a unified semantic model and perform lossless compression. S3: Data transmission and redundancy backup: A wired + wireless dual-link redundant transmission scheme is adopted, with priority given to transmission via industrial Ethernet. In case of failure, it automatically switches to the wireless link. The MQTT protocol is used to reduce transmission latency, and a three-level redundancy backup mechanism is constructed, consisting of local edge gateway, edge server cluster, and remote cloud, to achieve secure data storage throughout its entire lifecycle. S4: Data Validation and Dynamic Optimization: A dual validation mechanism is adopted, consisting of local validation at the edge and remote validation in the cloud. When data inconsistency is detected, retransmission and correction are triggered. A data quality assessment system is established to conduct real-time evaluation from four dimensions: data accuracy, integrity, time sequence consistency, and anomaly rate. Based on the evaluation results, the acquisition parameters and model parameters are dynamically optimized. At the same time, the operating status of system components is monitored and faults are investigated.

[0005] Preferably, in S1, the mainstream industrial protocols include Modbus, PROFIBUS, EtherCAT, and OPC UA. The proprietary protocol adapter interface realizes the parsing of encrypted proprietary protocols by reversing protocol parsing and connecting with the manufacturer's authorization. The retrofitting of old equipment adopts anti-interference shielded cable wiring, and the clock synchronization adopts a four-step method to perform round-trip time correction to ensure that the timing error does not exceed 1ms.

[0006] Preferably, the specific logical steps of S1 are as follows: S101: Construct a multi-protocol adaptive parsing module, integrating mainstream industrial protocol parsing drivers such as Modbus, PROFIBUS, EtherCAT, and OPC UA. At the same time, set up a private protocol adaptation interface. Obtain private protocol parsing rules by connecting with the manufacturer's authorization through protocol reverse parsing. The multi-protocol adaptive parsing module first scans the protocol frame header information of the access device to identify the protocol type, then matches the corresponding parsing driver and parsing rules, unpacks, decodes and performs preliminary format verification on the protocol data, removes data with incorrect format, and temporarily stores the valid original data in the edge gateway's temporary buffer area to lay the foundation for subsequent processing. S102: For older "dumb" devices without communication interfaces, industrial-grade I / O acquisition modules and high-precision sensors are installed at key monitoring points to collect analog and digital signals from the device. The signal conditioning unit of the I / O acquisition module eliminates signal distortion caused by environmental interference. An A / D converter then converts the analog signals to digital signals, and the digital signals are directly converted to binary digital signals with device identification and temporary timestamps added. Finally, the signals are connected to the edge gateway via anti-interference shielded cables for data aggregation and parsing with mainstream protocol devices. The analog-to-digital conversion uses a standard formula. ,in The actual measured value of the analog signal. Set the minimum input range for the A / D converter. The maximum input range for the A / D converter is n, where n is the number of bits in the A / D converter. The converted digital value; S103: The edge server layer clock module is set as the PTP master clock, and the clock units of each acquisition node, the old equipment upgrade module, the high-precision sensor, and the edge gateway layer are PTP slave clocks. A four-step method is used to collect four key timestamps of master-slave clock interaction, calculate the link transmission delay and clock deviation, and then correct the slave clock time through a calibration formula to achieve clock uniformity across the entire system. The formula used is as follows: The formula for calculating link transmission delay is: ,in, The time when the master clock sends the Sync message. The time when the Sync message is received from the clock. The time when the Delay_Req message is sent from the clock. The time when the master clock receives the Delay_Req message. Link transmission delay between master and slave clocks; The formula for calculating master-slave clock skew is: ,in: Master-slave clock bias > 0 indicates that the clock is ahead. <0 indicates that the clock is too slow, combined with the link transmission delay. Eliminate link interference to ensure that the deviation calculation error is ≤0.5μs; The formula for performing clock calibration is as follows: ,in: To start from the current clock time, To achieve the target time for synchronization with the master clock after calibration, this formula is used to eliminate clock deviations and ensure that the timing error of the entire system is ≤1ms. Finally, the master clock sends a synchronization message every 100ms for real-time calibration and maintenance, and adds a unified timestamp to all raw data in the temporary buffer to ensure data timing consistency.

[0007] Preferably, the specific logical steps of S2 are as follows: S201: Read the raw data processed by S1 from the edge gateway's temporary buffer, filter it by data type, and retain valid time-series datasets. Subsequently, Kalman filtering was used for initial denoising of the dataset, correcting low-frequency interference through prediction and update stages. Next, moving average filtering was used for secondary denoising to remove residual high-frequency interference. Finally, the denoising effect was verified to ensure that the noise amplitude was reduced by more than 30%. The denoised data was then stored in the edge gateway's preprocessing cache. The formulas used for Kalman filtering and moving average filtering are as follows: State prediction formula: ; Covariance prediction formula: ; Kalman gain formula: ; State update formula: ; Covariance update formula: ; Moving average filter formula: ; in Predict the state value at time k. Here is the state transition matrix. For the first The optimal estimate for filtering at time step. For the control matrix, For the first Control the amount at all times. The predicted state covariance at time k. For the first The optimal covariance estimate at any given time. For the transpose of A, The process noise variance is 0.01-0.05. Here, H is the Kalman gain, and H is the observation matrix (set to 1). For the transpose of H, To observe the noise variance (0.05-0.1), This is the original data collected at time k. This represents the optimal estimate of the Kalman filter at time k. It is the identity matrix. The optimal covariance estimate at time k is... This is the final denoised data at time k. The moving average window size is (3-7). The data is the Kalman filtered data at time i. S202: Read the denoising-reduced timing data from the preprocessing buffer after S201. Simultaneously, the threshold database of process parameters for each process step in silicon carbide is retrieved. The first step uses process thresholds to identify explicit anomalies, and then a judgment formula is used. Filter out abnormal data, among which To minimize the threshold of process parameters, a lightweight random forest model is used in the second stage to identify latent anomalies. First, over three months of historical denoised data are retrieved to construct training and testing sets. The Gini coefficient is used as the decision tree node splitting criterion during model training. After training, the model's identification accuracy is ≥98%. The denoised data is then input into the model, and the latent anomaly data is output and labeled as... The two types of abnormal data are combined to form an abnormal dataset. ; S203: For abnormal datasets Distinguish between correctable and uncorrectable anomalies. Correctable anomalies are corrected using a combination of "historical similar data interpolation + process logic correction," with the core formula being: ,in For the corrected data, , For abnormal data adjacent to normal data, For process correction coefficients, uncorrectable anomalies are not corrected; only the anomaly type is marked, and a local audible and visual alarm is triggered. The anomaly information is sent to the operation and maintenance management module to notify staff to troubleshoot the fault. Finally, the corrected data is verified to ensure that it is within the process threshold range and there are no hidden anomalies. The integrated data is then updated to the preprocessing cache. S204: Traverse the dataset integrated by S203, compare it with the unified timestamps of S1, detect missing data locations and record adjacent normal data, temporarily supplement it through the circular cache queue of the edge gateway, and then accurately compensate it through an LSTM neural network. Historical time-series data is retrieved to train the model, minimizing the mean squared error during training. Where N is the number of training sets, For the true value, For the predicted value, the model validation set accuracy is ≥98%. The adjacent normal data is input into the model to obtain the accurate predicted value, which replaces the temporary supplementary data. After verifying the compensation effect, it is updated to the preprocessing cache. S205: Read the complete dataset after compensation by S204, classify and count the differences by data type, unify the units, unify the data format as "parameter name-value-unit-acquisition time-device number-register address", reallocate the register address, unify the alarm threshold and level, and build a unified semantic model; after verifying the standardization effect, store the data in the preprocessing buffer. S206: Read the time-series dataset standardized by S205, and use Delta coding for lossless compression. The core formula is: ,in The difference between adjacent standardized data. , To standardize data for adjacent time points, only the first raw data and subsequent differences are stored. The data packets are encapsulated and compressed, and the data source is labeled. After verifying the compression effect, the data packets are sent to the dual-link transmission module and backed up to the local solid-state drive of the edge gateway.

[0008] Preferably, the specific logical steps of S3 are as follows: S301: Start the dual-link transmission module, initialize industrial Ethernet and 5G / industrial WiFi; configure the MQTT protocol, and calculate the link switching threshold. ,in To switch thresholds, =50ms, verify link and protocol configuration to ensure normal standby; S302: Reads the compressed data packet from S2 and adds a transmission identifier. It prioritizes transmission via Industrial Ethernet using the MQTT protocol. The transmission delay formula is... ,in For the total delay, For transmission delay, In the event of a link interruption, the system will switch to a wireless link within 1 second. The system detects the signal, receives confirmation, and then completes the transmission. S303: Construct a three-level redundancy backup system. The first level is a local backup for the edge gateway, and the formula used is: ,in Available for at least 7 days, the secondary backup is a RAID5 edge server cluster, and the recovery formula is: Level 3 is a cloud-based off-site backup, and its encryption verification formula is: This ensures that the data is not tampered with and that execution is synchronized at all three levels. S304: Verify the transmission and backup results. The transmission success rate formula is: Verify the consistency of the hash values ​​of the three-level backup data and record the logs in the operation and maintenance module.

[0009] Preferably, the execution logic steps of S4 are as follows: S401: Perform local verification on the edge side. The edge server reads the compressed data transmitted by S3 and the local backup data, verifies the data integrity and format standardization, compares the hash values ​​of the data packets before and after transmission to confirm that the data has not been tampered with, and then performs remote verification on the cloud. The cloud platform reads the core data uploaded by the edge server and performs a three-way comparison with the edge server backup data and the edge gateway local backup data to verify the data consistency. If data inconsistency, missing data or tampering is found, the data retransmission mechanism is immediately triggered, instructing the edge gateway to retransmit the corresponding data. If the retransmission fails, an operation and maintenance warning is triggered, and inconsistent data is corrected to ensure that the data is accurate. S402: Establish an evaluation system covering four dimensions: data accuracy, integrity, time series consistency, and anomaly rate, and clarify the evaluation standards for each dimension: data accuracy error ≤ ±0.5%, integrity ≥ 99.5%, time series consistency error ≤ 1ms, and anomaly rate ≤ 0.5%. Collect data for each dimension in real time, dynamically evaluate the pre-processed and backed-up data, generate real-time data quality reports, and clarify the compliance status and existing problems for each dimension. S403: Based on the data quality assessment report, optimize system parameters accordingly: If accuracy is not up to standard, adjust the Q and R parameters of the dual filtering and the moving average window size in S201; if the anomaly rate is too high, optimize the parameters of the random forest anomaly identification model and the process threshold in S202; if timing consistency is not up to standard, adjust the calibration cycle of clock synchronization in S103; if integrity is not up to standard, optimize the dual-link transmission switching threshold in S3 and the missing data compensation model parameters in S204. The optimization cycle is 2-3 days to ensure that data quality continues to improve after parameter adjustments. S404: Real-time monitoring of the operational status of the acquisition layer, edge gateway layer, edge server layer, cloud platform layer, and transmission links, including device operating parameters, link transmission status, and component load; if a component failure is detected, the location and type of the failure are immediately marked, triggering the corresponding warning, and the failure information is pushed to the operation and maintenance management module. Operation and maintenance personnel can troubleshoot the problem according to the failure prompts, repair the failure in a timely manner, ensure the normal operation of all system components, and ensure that the entire process of data acquisition, transmission, and processing is uninterrupted.

[0010] This invention also proposes a multi-protocol silicon carbide wafer production data acquisition system based on edge computing, including an acquisition layer, an edge gateway layer, an edge server layer, a cloud platform layer, and an operation and maintenance management module; The acquisition layer includes a multi-protocol acquisition node, an old equipment renovation module, a high-precision sensor group, and a clock synchronization module. The multi-protocol acquisition node is used to connect various protocol devices on the production line and collect data. The old equipment renovation module is used to renovate old "dumb" devices and realize the conversion of analog and digital signals. The high-precision sensor group is used to collect environmental and production status parameters and adopts a high-temperature and electromagnetic interference resistant design. The clock synchronization module is used to realize the unified clock synchronization of all components. The edge gateway layer includes a multi-protocol adaptive parsing module, a data preprocessing module, a cache queue module, a dual-link transmission module, and a local backup module. The multi-protocol adaptive parsing module is used for parsing and converting multi-protocol data, the data preprocessing module is used for data noise reduction, anomaly handling, missing data compensation, and standardization conversion, the cache queue module is used for caching collected data, the dual-link transmission module is used for implementing dual-link redundant transmission, and the local backup module is used for local data storage. The edge server layer includes a data central management module, a redundancy backup module, a data verification module, and a dynamic optimization module. The data central management module is used for centralized data management and classified storage, the redundancy backup module is used for data cluster backup, the data verification module is used for dual data verification, and the dynamic optimization module is used to optimize the acquisition parameters and model parameters and monitor the system operating status. The cloud platform layer includes a data storage module, a data quality assessment module, and a data application interface module. The data storage module is used for off-site encrypted backup of core data, the data quality assessment module is used for real-time data quality assessment and report generation, and the data application interface module is used to interface with MES and APC systems to realize closed-loop data application. The operation and maintenance management module is used to remotely monitor the status of various system components, locate faults, perform batch operation and maintenance, and upgrade configurations.

[0011] Preferably, the data preprocessing module includes a dual filtering unit, an anomaly identification and correction unit, a missing data compensation unit, and a data standardization unit. The anomaly identification and correction unit has a built-in process threshold library and a lightweight machine learning anomaly identification model, and the missing data compensation unit has a built-in LSTM neural network prediction model.

[0012] Compared with existing technologies, the beneficial effects of this invention are: 1. By constructing a dual noise reduction algorithm combining Kalman filtering and moving average filtering, a dual anomaly identification model combining process threshold and machine learning, and a missing data compensation scheme of cache queue + LSTM neural network, we achieve accurate data noise reduction, anomaly correction and missing data compensation. At the same time, we adopt the IEEE 1588 PTP clock synchronization protocol to ensure the consistency of data timing throughout the system, significantly reducing data errors. This ensures that the data accuracy, timing consistency and integrity meet the stringent requirements of silicon carbide process, avoids the impact of small data errors on wafer yield and defect traceability, and solves the problem of "acquired but difficult to use" caused by data quality risks in the existing system. 2. By adding IO acquisition modules and high-precision sensors to old "dumb equipment", analog and switch signals are converted into digital signals. Combined with the compatibility of multi-protocol adaptive parsing modules, data can be uniformly accessed between old "dumb equipment" and mainstream protocol equipment and high-end imported encrypted private protocol equipment. This breaks the data acquisition blind spot and covers all processes and all types of equipment, such as crystal growth, epitaxy, and etching, ensuring the integrity of the acquisition range and providing comprehensive data support for the optimization and traceability of the entire process. 3. Based on the edge computing architecture, a multi-protocol adaptive parsing module is built, integrating mainstream industrial protocol parsing drivers and private protocol adaptation interfaces to achieve unified parsing and local acquisition of multi-protocol data. Through local preprocessing on the edge side and a "wired + wireless" dual-link redundant transmission scheme, combined with the MQTT protocol, transmission latency is reduced. At the same time, a three-level redundant backup mechanism is built to ensure data security. This not only solves the data acquisition problem caused by protocol heterogeneity, but also improves the real-time performance and reliability of data acquisition and transmission, reduces cloud transmission bandwidth consumption, and solves the inherent drawbacks of traditional centralized acquisition.

[0013] This invention overcomes the challenge of protocol heterogeneity across different devices by constructing a multi-protocol adaptive parsing module that integrates mainstream industrial protocol drivers and proprietary protocol adapter interfaces. Simultaneously, it achieves local data acquisition and preprocessing based on an edge computing architecture. Combined with dual filtering, dual anomaly identification, and missing data compensation schemes, along with the IEEE 1588 PTP clock synchronization protocol, it solves problems such as poor data quality and timing discrepancies in extreme production environments, ensuring data accuracy and continuity. By adding IO acquisition modules and high-precision sensors to older "dumb" devices, it fills data acquisition blind spots, achieving complete data acquisition across all processes and types of equipment on the production line. Furthermore, by employing dual-link redundant transmission and a three-level redundant backup mechanism, it improves the real-time performance and reliability of data transmission and storage, reduces bandwidth consumption, and ultimately solves the problem of "acquired but unusable" data. This provides comprehensive and reliable data support for process optimization, yield improvement, and full-process traceability in silicon carbide wafer production. Attached Figure Description

[0014] Figure 1This is a flowchart of a multi-protocol silicon carbide wafer production data acquisition method based on edge computing proposed in this invention; Figure 2 This is a block diagram of a multi-protocol silicon carbide wafer production data acquisition system based on edge computing proposed in this invention. Detailed Implementation

[0015] The present invention will be further explained below with reference to specific embodiments.

[0016] Example Reference Figure 1 This embodiment proposes a multi-protocol silicon carbide wafer production data acquisition method based on edge computing, including the following steps: S1: Multi-protocol Adaptation and Retrofitting of Legacy Equipment: Construct a multi-protocol adaptive parsing module, integrating mainstream industrial protocol parsing drivers and private protocol adaptation interfaces to realize data parsing of multi-protocol equipment; add IO acquisition modules and high-precision sensors to legacy "dumb" equipment to convert analog and switch signals into digital signals and connect them to the edge gateway; and adopt the IEEE1588 PTP clock synchronization protocol to perform unified clock synchronization on each acquisition node, edge gateway, and production line equipment to eliminate measurement delays and ensure data timing consistency. The mainstream industrial protocols include Modbus, PROFIBUS, EtherCAT, and OPC UA. The proprietary protocol adapter interface realizes the parsing of encrypted proprietary protocols by reversing protocol parsing and connecting with manufacturer authorization. Anti-interference shielded cable wiring is used for the renovation of old equipment. Clock synchronization adopts a four-step method to correct round-trip time and ensure that the timing error does not exceed 1ms. The specific logical steps are as follows: S101: Construct a multi-protocol adaptive parsing module, integrating mainstream industrial protocol parsing drivers such as Modbus, PROFIBUS, EtherCAT, and OPC UA. At the same time, set up a private protocol adaptation interface. Obtain private protocol parsing rules by connecting with the manufacturer's authorization through protocol reverse parsing. The multi-protocol adaptive parsing module first scans the protocol frame header information of the access device to identify the protocol type, then matches the corresponding parsing driver and parsing rules, unpacks, decodes and performs preliminary format verification on the protocol data, removes data with incorrect format, and temporarily stores the valid original data in the edge gateway's temporary buffer area to lay the foundation for subsequent processing. S102: For older "dumb" devices without communication interfaces, industrial-grade I / O acquisition modules and high-precision sensors are installed at key monitoring points to collect analog and digital signals from the device. The signal conditioning unit of the I / O acquisition module eliminates signal distortion caused by environmental interference. An A / D converter then converts the analog signals to digital signals, and the digital signals are directly converted to binary digital signals with device identification and temporary timestamps added. Finally, the signals are connected to the edge gateway via anti-interference shielded cables for data aggregation and parsing with mainstream protocol devices. The analog-to-digital conversion uses a standard formula. ,in The actual measured value of the analog signal. Set the minimum input range for the A / D converter. The maximum input range for the A / D converter is n, where n is the number of bits in the A / D converter. The converted digital value; S103: The edge server layer clock module is set as the PTP master clock, and the clock units of each acquisition node, the old equipment upgrade module, the high-precision sensor, and the edge gateway layer are PTP slave clocks. A four-step method is used to collect four key timestamps of master-slave clock interaction, calculate the link transmission delay and clock deviation, and then correct the slave clock time through a calibration formula to achieve clock uniformity across the entire system. The formula used is as follows: The formula for calculating link transmission delay is: ,in, The time when the master clock sends the Sync message. The time when the Sync message is received from the clock. The time when the Delay_Req message is sent from the clock. The time when the master clock receives the Delay_Req message. Link transmission delay between master and slave clocks; The formula for calculating master-slave clock skew is: ,in: Master-slave clock bias > 0 indicates that the clock is ahead. <0 indicates that the clock is too slow, combined with the link transmission delay. Eliminate link interference to ensure that the deviation calculation error is ≤0.5μs; The formula for performing clock calibration is as follows: ,in: To start from the current clock time, To achieve the target time synchronized with the master clock after calibration, this formula is used to eliminate clock deviation and ensure that the timing error of the entire system is ≤1ms. Finally, the master clock sends a synchronization message every 100ms for real-time calibration and maintenance, and adds a unified timestamp to all raw data in the temporary buffer to ensure data timing consistency. S2: Edge-side data preprocessing: A dual filtering algorithm of Kalman filtering and moving average filtering is used to reduce noise in the collected data. A dual anomaly recognition model combining process threshold and machine learning is constructed to identify and correct abnormal data. Uncorrectable abnormal data is marked and an early warning is triggered. A scheme of cache queue + LSTM neural network prediction is used to compensate for missing data. The preprocessed data is standardized and transformed to construct a unified semantic model and perform lossless compression. The specific logical steps are as follows: S201: Read the raw data processed by S1 from the edge gateway's temporary buffer, filter it by data type, and retain valid time-series datasets. Subsequently, Kalman filtering was used for initial denoising of the dataset, correcting low-frequency interference through prediction and update stages. Next, moving average filtering was used for secondary denoising to remove residual high-frequency interference. Finally, the denoising effect was verified to ensure that the noise amplitude was reduced by more than 30%. The denoised data was then stored in the edge gateway's preprocessing cache. The formulas used for Kalman filtering and moving average filtering are as follows: State prediction formula: ; Covariance prediction formula: ; Kalman gain formula: ; State update formula: ; Covariance update formula: ; Moving average filter formula: ; in Predict the state value at time k. Here is the state transition matrix. For the first The optimal estimate for filtering at time step. For the control matrix, For the first Control the amount at all times. The predicted state covariance at time k. For the first The optimal covariance estimate at any given time. For the transpose of A, The process noise variance is 0.01-0.05. Here, H is the Kalman gain, and H is the observation matrix (set to 1). For the transpose of H, To observe the noise variance (0.05-0.1), This is the original data collected at time k. This represents the optimal estimate of the Kalman filter at time k. It is the identity matrix. The optimal covariance estimate at time k is... This is the final denoised data at time k. The moving average window size is (3-7). The data is the Kalman filtered data at time i. S202: Read the denoising-reduced timing data from the preprocessing buffer after S201. Simultaneously, the threshold database of process parameters for each process step in silicon carbide is retrieved. The first step uses process thresholds to identify explicit anomalies, and then a judgment formula is used. Filter out abnormal data, among which To minimize the threshold of process parameters, a lightweight random forest model is used in the second stage to identify latent anomalies. First, over three months of historical denoised data are retrieved to construct training and testing sets. The Gini coefficient is used as the decision tree node splitting criterion during model training. After training, the model's identification accuracy is ≥98%. The denoised data is then input into the model, and the latent anomaly data is output and labeled as... The two types of abnormal data are combined to form an abnormal dataset. ; S203: For abnormal datasets Distinguish between correctable and uncorrectable anomalies. Correctable anomalies are corrected using a combination of "historical similar data interpolation + process logic correction," with the core formula being: ,in For the corrected data, , For abnormal data adjacent to normal data, For process correction coefficients, uncorrectable anomalies are not corrected; only the anomaly type is marked, and a local audible and visual alarm is triggered. The anomaly information is sent to the operation and maintenance management module to notify staff to troubleshoot the fault. Finally, the corrected data is verified to ensure that it is within the process threshold range and there are no hidden anomalies. The integrated data is then updated to the preprocessing cache. S204: Traverse the dataset integrated by S203, compare it with the unified timestamps of S1, detect missing data locations and record adjacent normal data, temporarily supplement it through the circular cache queue of the edge gateway, and then accurately compensate it through an LSTM neural network. Historical time-series data is retrieved to train the model, minimizing the mean squared error during training. Where N is the number of training sets, For the true value, For the predicted value, the model validation set accuracy is ≥98%. The adjacent normal data is input into the model to obtain the accurate predicted value, which replaces the temporary supplementary data. After verifying the compensation effect, it is updated to the preprocessing cache. S205: Read the complete dataset after compensation by S204, classify and count the differences by data type, unify the units, unify the data format as "parameter name-value-unit-acquisition time-device number-register address", reallocate the register address, unify the alarm threshold and level, and build a unified semantic model; after verifying the standardization effect, store the data in the preprocessing buffer. S206: Read the time-series dataset standardized by S205, and use Delta coding for lossless compression. The core formula is: ,in The difference between adjacent standardized data. , To standardize data for adjacent time points, only the first raw data and subsequent differences are stored. The data packets are encapsulated and compressed, and the data source is labeled. After verifying the compression effect, the data packets are sent to the dual-link transmission module and backed up to the local solid-state drive of the edge gateway. S3: Data transmission and redundancy backup: A wired + wireless dual-link redundant transmission scheme is adopted, with priority given to transmission via industrial Ethernet. In case of failure, it automatically switches to the wireless link. The MQTT protocol is used to reduce transmission latency, and a three-level redundancy backup mechanism is constructed, consisting of local edge gateway, edge server cluster, and remote cloud, to achieve secure data storage throughout its entire lifecycle. The specific logical steps are as follows: S301: Start the dual-link transmission module, initialize industrial Ethernet and 5G / industrial WiFi; configure the MQTT protocol, and calculate the link switching threshold. ,in To switch thresholds, =50ms, verify link and protocol configuration to ensure normal standby; S302: Reads the compressed data packet from S2 and adds a transmission identifier. It prioritizes transmission via Industrial Ethernet using the MQTT protocol. The transmission delay formula is... ,in For the total delay, For transmission delay, In the event of a link interruption, the system will switch to a wireless link within 1 second. The system detects the signal, receives confirmation, and then completes the transmission. S303: Construct a three-level redundancy backup system. The first level is a local backup for the edge gateway, and the formula used is: ,in Available for at least 7 days, the secondary backup is a RAID5 edge server cluster, and the recovery formula is: Level 3 is a cloud-based off-site backup, and its encryption verification formula is: This ensures that the data is not tampered with and that execution is synchronized at all three levels. S304: Verify the transmission and backup results. The transmission success rate formula is: Verify the consistency of hash values ​​of the three-level backup data and record the logs in the operation and maintenance module; S4: Data Validation and Dynamic Optimization: A dual validation mechanism is adopted, consisting of local validation at the edge and remote validation in the cloud. When data inconsistency is detected, retransmission and correction are triggered. A data quality assessment system is established to evaluate the data in real time from four dimensions: data accuracy, integrity, time sequence consistency, and anomaly rate. Based on the evaluation results, the acquisition parameters and model parameters are dynamically optimized. At the same time, the operating status of system components is monitored and faults are investigated. Its operational logic steps are as follows: S401: Perform local verification on the edge side. The edge server reads the compressed data transmitted by S3 and the local backup data, verifies the data integrity and format standardization, compares the hash values ​​of the data packets before and after transmission to confirm that the data has not been tampered with, and then performs remote verification on the cloud. The cloud platform reads the core data uploaded by the edge server and performs a three-way comparison with the edge server backup data and the edge gateway local backup data to verify the data consistency. If data inconsistency, missing data or tampering is found, the data retransmission mechanism is immediately triggered, instructing the edge gateway to retransmit the corresponding data. If the retransmission fails, an operation and maintenance warning is triggered, and inconsistent data is corrected to ensure that the data is accurate. S402: Establish an evaluation system covering four dimensions: data accuracy, integrity, time series consistency, and anomaly rate, and clarify the evaluation standards for each dimension: data accuracy error ≤ ±0.5%, integrity ≥ 99.5%, time series consistency error ≤ 1ms, and anomaly rate ≤ 0.5%. Collect data for each dimension in real time, dynamically evaluate the pre-processed and backed-up data, generate real-time data quality reports, and clarify the compliance status and existing problems for each dimension. S403: Based on the data quality assessment report, optimize system parameters accordingly: If accuracy is not up to standard, adjust the Q and R parameters of the dual filtering and the moving average window size in S201; if the anomaly rate is too high, optimize the parameters of the random forest anomaly identification model and the process threshold in S202; if timing consistency is not up to standard, adjust the calibration cycle of clock synchronization in S103; if integrity is not up to standard, optimize the dual-link transmission switching threshold in S3 and the missing data compensation model parameters in S204. The optimization cycle is 2-3 days to ensure that data quality continues to improve after parameter adjustments. S404: Real-time monitoring of the operational status of the acquisition layer, edge gateway layer, edge server layer, cloud platform layer, and transmission links, including device operating parameters, link transmission status, and component load; if a component failure is detected, the location and type of the failure are immediately marked, triggering the corresponding warning, and the failure information is pushed to the operation and maintenance management module. Operation and maintenance personnel can troubleshoot the problem according to the failure prompts, repair the failure in a timely manner, ensure the normal operation of all system components, and ensure that the entire process of data acquisition, transmission, and processing is uninterrupted.

[0017] Reference Figure 2 This embodiment proposes a multi-protocol silicon carbide wafer production data acquisition system based on edge computing, including an acquisition layer, an edge gateway layer, an edge server layer, a cloud platform layer, and an operation and maintenance management module; The acquisition layer includes multi-protocol acquisition nodes, an old equipment renovation module, a high-precision sensor group, and a clock synchronization module. The multi-protocol acquisition nodes are used to connect various protocol devices on the production line and collect data. The old equipment renovation module is used to renovate old "dumb" devices and realize the conversion between analog and digital signals. The high-precision sensor group is used to collect environmental and production status parameters and adopts a high-temperature and electromagnetic interference resistant design. The clock synchronization module is used to realize the unified clock synchronization of all components. The edge gateway layer includes a multi-protocol adaptive parsing module, a data preprocessing module, a cache queue module, a dual-link transmission module, and a local backup module. The multi-protocol adaptive parsing module is used for parsing and converting multi-protocol data, the data preprocessing module is used for data noise reduction, anomaly handling, missing data compensation, and standardization conversion, the cache queue module is used for caching collected data, the dual-link transmission module is used for implementing dual-link redundant transmission, and the local backup module is used for local data storage. The data preprocessing module includes a dual filtering unit, an anomaly identification and correction unit, a missing data compensation unit, and a data standardization unit. The anomaly identification and correction unit has a built-in process threshold library and a lightweight machine learning anomaly identification model, and the missing data compensation unit has a built-in LSTM neural network prediction model. The edge server layer includes a data central management module, a redundancy backup module, a data verification module, and a dynamic optimization module. The data central management module is used for centralized data management and classified storage, the redundancy backup module is used for data cluster backup, the data verification module is used for dual data verification, and the dynamic optimization module is used to optimize acquisition parameters and model parameters and monitor the system operating status. The cloud platform layer includes a data storage module, a data quality assessment module, and a data application interface module. The data storage module is used for off-site encrypted backup of core data, the data quality assessment module is used for real-time data quality assessment and report generation, and the data application interface module is used to interface with MES and APC systems to achieve closed-loop data application. The operation and maintenance management module is used for remote status monitoring, fault location, batch operation and maintenance, and configuration upgrades of various system components; This embodiment solves the problem of heterogeneity of protocols across different devices by constructing a multi-protocol adaptive parsing module that integrates mainstream industrial protocol drivers and proprietary protocol adapter interfaces. Simultaneously, it achieves local data acquisition and preprocessing based on an edge computing architecture. Combined with dual filtering, dual anomaly identification, and missing data compensation schemes, along with the IEEE 1588 PTP clock synchronization protocol, it addresses issues such as poor data quality and timing discrepancies in extreme production environments, ensuring data accuracy and continuity. By adding IO acquisition modules and high-precision sensors to older "dumb" devices, it fills data acquisition blind spots, achieving complete data acquisition across all processes and types of equipment on the production line. Furthermore, the adoption of dual-link redundant transmission and a three-level redundant backup mechanism improves the real-time performance and reliability of data transmission and storage, reduces bandwidth consumption, and ultimately solves the problem of "acquired but unusable" data. This provides comprehensive and reliable data support for process optimization, yield improvement, and full-process traceability in silicon carbide wafer production.

[0018] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A multi-protocol silicon carbide wafer production data acquisition method based on edge computing, characterized in that, Includes the following steps: S1: Multi-protocol Adaptation and Retrofitting of Legacy Equipment: Construct a multi-protocol adaptive parsing module, integrating mainstream industrial protocol parsing drivers and private protocol adaptation interfaces to realize data parsing of multi-protocol equipment; add IO acquisition modules and high-precision sensors to legacy "dumb" equipment to convert analog and switch signals into digital signals and connect them to the edge gateway; and adopt the IEEE1588 PTP clock synchronization protocol to perform unified clock synchronization on each acquisition node, edge gateway, and production line equipment to eliminate measurement delays and ensure data timing consistency. S2: Edge-side data preprocessing: A dual filtering algorithm of Kalman filtering and moving average filtering is used to reduce noise in the collected data. A dual anomaly recognition model combining process threshold and machine learning is constructed to identify and correct abnormal data. Uncorrectable abnormal data is marked and an early warning is triggered. A scheme of cache queue + LSTM neural network prediction is used to compensate for missing data. The preprocessed data is standardized and transformed to construct a unified semantic model and perform lossless compression. S3: Data transmission and redundancy backup: A wired + wireless dual-link redundant transmission scheme is adopted, with priority given to transmission via industrial Ethernet. In case of failure, it automatically switches to the wireless link. The MQTT protocol is used to reduce transmission latency, and a three-level redundancy backup mechanism is constructed, consisting of local edge gateway, edge server cluster, and remote cloud, to achieve secure data storage throughout its entire lifecycle. S4: Data Validation and Dynamic Optimization: A dual validation mechanism is adopted, consisting of local validation at the edge and remote validation in the cloud. When data inconsistency is detected, retransmission and correction are triggered. A data quality assessment system is established to conduct real-time evaluation from four dimensions: data accuracy, integrity, time sequence consistency, and anomaly rate. Based on the evaluation results, the acquisition parameters and model parameters are dynamically optimized. At the same time, the operating status of system components is monitored and faults are investigated.

2. The multi-protocol silicon carbide wafer production data acquisition method based on edge computing according to claim 1, characterized in that, In S1, mainstream industrial protocols include Modbus, PROFIBUS, EtherCAT, and OPC UA. The proprietary protocol adapter interface realizes the parsing of encrypted proprietary protocols by reversing protocol parsing and connecting with manufacturer authorization. Anti-interference shielded cable wiring is used for the renovation of old equipment. Clock synchronization adopts a four-step method for round-trip time correction to ensure that the timing error does not exceed 1ms.

3. The method according to claim 2, wherein, The specific logical steps of S1 are as follows: S101: Construct a multi-protocol adaptive parsing module, integrating mainstream industrial protocol parsing drivers such as Modbus, PROFIBUS, EtherCAT, and OPC UA. At the same time, set up a private protocol adaptation interface. Obtain private protocol parsing rules by connecting with the manufacturer's authorization through protocol reverse parsing. The multi-protocol adaptive parsing module first scans the protocol frame header information of the access device to identify the protocol type, then matches the corresponding parsing driver and parsing rules, unpacks, decodes and performs preliminary format verification on the protocol data, removes data with incorrect format, and temporarily stores the valid original data in the edge gateway's temporary buffer area to lay the foundation for subsequent processing. S102: For older "dumb" devices without communication interfaces, industrial-grade I / O acquisition modules and high-precision sensors are installed at key monitoring points to collect analog and digital signals from the device. The signal conditioning unit of the I / O acquisition module eliminates signal distortion caused by environmental interference. An A / D converter then converts the analog signals to digital signals, and the digital signals are directly converted to binary digital signals with device identification and temporary timestamps added. Finally, the data is connected to the edge gateway via an anti-interference shielded cable for data aggregation and parsing with mainstream protocol devices. The analog-to-digital conversion uses a standard formula. ,in The actual measured value of the analog signal. Set the minimum input range for the A / D converter. The maximum input range for the A / D converter is n, where n is the number of bits in the A / D converter. The converted digital value; S103: The edge server layer clock module is set as the PTP master clock, and the clock units of each acquisition node, the old equipment upgrade module, the high-precision sensor, and the edge gateway layer are PTP slave clocks. A four-step method is used to collect four key timestamps of master-slave clock interaction, calculate the link transmission delay and clock deviation, and then correct the slave clock time through a calibration formula to achieve clock uniformity across the entire system. The formula used is as follows: The formula for calculating link transmission delay is: ,in, The time when the master clock sends the Sync message. The time when the Sync message is received from the clock. The time when the Delay_Req message is sent from the clock. The time when the master clock receives the Delay_Req message. Link transmission delay between master and slave clocks; The formula for calculating master-slave clock skew is: ,in: Master-slave clock bias > 0 indicates that the clock is ahead. <0 indicates that the clock is too slow, combined with the link transmission delay. Eliminate link interference to ensure that the deviation calculation error is ≤0.5μs; The formula for performing clock calibration is as follows: ,in: To start from the current clock time, To achieve the target time for synchronization with the master clock after calibration, this formula is used to eliminate clock deviations and ensure that the timing error of the entire system is ≤1ms. Finally, the master clock sends a synchronization message every 100ms for real-time calibration and maintenance, and adds a unified timestamp to all raw data in the temporary buffer to ensure data timing consistency.

4. The method for acquiring multi-protocol silicon carbide wafer production data based on edge computing according to claim 3, characterized in that, The specific logical steps of S2 are as follows: S201: Read the raw data processed by S1 from the edge gateway's temporary buffer, filter it by data type, and retain valid time-series datasets. Subsequently, Kalman filtering was used for initial denoising of the dataset, correcting low-frequency interference through prediction and update stages. Next, moving average filtering was used for secondary denoising to remove residual high-frequency interference. Finally, the denoising effect was verified to ensure that the noise amplitude was reduced by more than 30%. The denoised data was then stored in the edge gateway's preprocessing cache. The formulas used for Kalman filtering and moving average filtering are as follows: State prediction formula: ; Covariance prediction formula: ; Kalman gain formula: ; State update formula: ; Covariance update formula: ; Moving average filter formula: ; in Predict the state value at time k. Here is the state transition matrix. For the first The optimal estimate for filtering at time step. For the control matrix, For the first Control the amount at all times. The predicted state covariance at time k. For the first The optimal covariance estimate at any given time. For the transpose of A, The process noise variance is 0.01-0.

05. Here, H is the Kalman gain, and H is the observation matrix (set to 1). For the transpose of H, To observe the noise variance (0.05-0.1), This is the original data collected at time k. This represents the optimal estimate of the Kalman filter at time k. It is the identity matrix. The optimal covariance estimate at time k is... This is the final denoised data at time k. The moving average window size is (3-7). The data is the Kalman filtered data at time i. S202: Read the denoising-reduced timing data from the preprocessing buffer after S201. Simultaneously, the threshold database of process parameters for each process step in silicon carbide is retrieved. The first step uses process thresholds to identify explicit anomalies, and then a judgment formula is used. Filter out abnormal data, among which To minimize the threshold of process parameters, a lightweight random forest model is used in the second stage to identify latent anomalies. First, over three months of historical denoised data are retrieved to construct training and testing sets. The Gini coefficient is used as the decision tree node splitting criterion during model training. After training, the model's identification accuracy is ≥98%. The denoised data is then input into the model, and the latent anomaly data is output and labeled as... The two types of abnormal data are combined to form an abnormal dataset. ; S203: For abnormal datasets Distinguish between correctable and uncorrectable anomalies. Correctable anomalies are corrected using "historical similar data interpolation + process logic correction," with the core formula being: ,in For the corrected data, , For abnormal data adjacent to normal data, For process correction coefficients, uncorrectable anomalies are not corrected; only the anomaly type is marked, and a local audible and visual alarm is triggered. The anomaly information is sent to the operation and maintenance management module to notify staff to troubleshoot the fault. Finally, the corrected data is verified to ensure that it is within the process threshold range and there are no hidden anomalies. The integrated data is then updated to the preprocessing cache. S204: Traverse the dataset integrated by S203, compare it with the unified timestamps of S1, detect missing data locations and record adjacent normal data, temporarily supplement it through the circular cache queue of the edge gateway, and then accurately compensate it through an LSTM neural network. Historical time-series data is retrieved to train the model, minimizing the mean squared error during training. Where N is the number of training sets, For the true value, For the predicted value, the model validation set accuracy is ≥98%. The adjacent normal data is input into the model to obtain the accurate predicted value, which replaces the temporary supplementary data. After verifying the compensation effect, it is updated to the preprocessing cache. S205: Read the complete dataset after compensation by S204, classify and count the differences by data type, unify the units, unify the data format as "parameter name-value-unit-acquisition time-device number-register address", reallocate the register address, unify the alarm threshold and level, and build a unified semantic model; after verifying the standardization effect, store the data in the preprocessing buffer. S206: Read the time-series dataset standardized by S205, and use Delta coding for lossless compression. The core formula is: ,in The difference between adjacent standardized data. , To standardize data for adjacent time points, only the first raw data and subsequent differences are stored. The data packets are encapsulated and compressed, and the data source is labeled. After verifying the compression effect, the data packets are sent to the dual-link transmission module and backed up to the local solid-state drive of the edge gateway.

5. The method for acquiring multi-protocol silicon carbide wafer production data based on edge computing according to claim 4, characterized in that, The specific logical steps of S3 are as follows: S301: Start the dual-link transmission module and initialize industrial Ethernet and 5G / industrial WiFi; Configure the MQTT protocol and calculate the link switching threshold. ,in To switch thresholds, =50ms, verify link and protocol configuration to ensure normal standby; S302: Reads the compressed data packet from S2 and adds a transmission identifier. It prioritizes transmission via Industrial Ethernet using the MQTT protocol. The transmission delay formula is... ,in For the total delay, For transmission delay, In the event of a link interruption, the system will switch to a wireless link within 1 second. The system detects the signal, receives confirmation, and then completes the transmission. S303: Construct a three-level redundancy backup system. The first level is a local backup for the edge gateway, and the formula used is: ,in Available for at least 7 days, the secondary backup is a RAID5 edge server cluster, and the recovery formula is: Level 3 is a cloud-based off-site backup, and its encryption verification formula is: This ensures that the data is not tampered with and that execution is synchronized at all three levels. S304: Verify the transmission and backup results. The transmission success rate formula is: Verify the consistency of the hash values ​​of the three-level backup data and record the logs in the operation and maintenance module.

6. The method for acquiring multi-protocol silicon carbide wafer production data based on edge computing according to claim 5, characterized in that, The execution logic steps of S4 are as follows: S401: Perform local verification on the edge side. The edge server reads the compressed data transmitted by S3 and the local backup data, verifies the data integrity and format standardization, compares the hash values ​​of the data packets before and after transmission to confirm that the data has not been tampered with, and then performs remote verification on the cloud. The cloud platform reads the core data uploaded by the edge server and performs a three-way comparison with the edge server backup data and the edge gateway local backup data to verify the data consistency. If data inconsistency, missing data or tampering is found, the data retransmission mechanism is immediately triggered, instructing the edge gateway to retransmit the corresponding data. If the retransmission fails, an operation and maintenance warning is triggered, and inconsistent data is corrected to ensure that the data is accurate. S402: Establish an evaluation system covering four dimensions: data accuracy, integrity, time series consistency, and anomaly rate, and clarify the evaluation standards for each dimension: data accuracy error ≤ ±0.5%, integrity ≥ 99.5%, time series consistency error ≤ 1ms, and anomaly rate ≤ 0.5%. Collect data for each dimension in real time, dynamically evaluate the pre-processed and backed-up data, generate real-time data quality reports, and clarify the compliance status and existing problems for each dimension. S403: Based on the data quality assessment report, optimize system parameters accordingly: If accuracy is not up to standard, adjust the Q and R parameters of the dual filtering and the moving average window size in S201; if the anomaly rate is too high, optimize the parameters of the random forest anomaly identification model and the process threshold in S202; if timing consistency is not up to standard, adjust the calibration cycle of clock synchronization in S103; if integrity is not up to standard, optimize the dual-link transmission switching threshold in S3 and the missing data compensation model parameters in S204. The optimization cycle is 2-3 days to ensure that data quality continues to improve after parameter adjustments. S404: Real-time monitoring of the operational status of the acquisition layer, edge gateway layer, edge server layer, cloud platform layer, and transmission links, including device operating parameters, link transmission status, and component load; if a component failure is detected, the location and type of the failure are immediately marked, triggering the corresponding warning, and the failure information is pushed to the operation and maintenance management module. Operation and maintenance personnel can troubleshoot the problem according to the failure prompts, repair the failure in a timely manner, ensure the normal operation of all system components, and ensure that the entire process of data acquisition, transmission, and processing is uninterrupted.

7. A multi-protocol silicon carbide wafer production data acquisition system based on edge computing, used to implement the method described in any one of claims 1-6, characterized in that, It includes a data acquisition layer, an edge gateway layer, an edge server layer, a cloud platform layer, and an operation and maintenance management module; The acquisition layer includes a multi-protocol acquisition node, an old equipment renovation module, a high-precision sensor group, and a clock synchronization module. The multi-protocol acquisition node is used to connect various protocol devices on the production line and collect data. The old equipment renovation module is used to renovate old "dumb" devices and realize the conversion between analog and digital signals. The high-precision sensor group is used to collect environmental and production status parameters and adopts a high-temperature and electromagnetic interference resistant design. The clock synchronization module is used to realize the unified clock synchronization of all components. The edge gateway layer includes a multi-protocol adaptive parsing module, a data preprocessing module, a cache queue module, a dual-link transmission module, and a local backup module. The multi-protocol adaptive parsing module is used for parsing and converting multi-protocol data, the data preprocessing module is used for data noise reduction, anomaly handling, missing data compensation, and standardization conversion, the cache queue module is used for caching collected data, the dual-link transmission module is used for implementing dual-link redundant transmission, and the local backup module is used for local data storage. The edge server layer includes a data central management module, a redundancy backup module, a data verification module, and a dynamic optimization module. The data central management module is used for centralized data management and classified storage, the redundancy backup module is used for data cluster backup, the data verification module is used for dual data verification, and the dynamic optimization module is used to optimize the acquisition parameters and model parameters and monitor the system operating status. The cloud platform layer includes a data storage module, a data quality assessment module, and a data application interface module. The data storage module is used for off-site encrypted backup of core data, the data quality assessment module is used for real-time data quality assessment and report generation, and the data application interface module is used to interface with MES and APC systems to realize closed-loop data application. The operation and maintenance management module is used to remotely monitor the status of various system components, locate faults, perform batch operation and maintenance, and upgrade configurations.

8. The multi-protocol silicon carbide wafer production data acquisition system based on edge computing according to claim 7, characterized in that, The data preprocessing module includes a dual filtering unit, an anomaly identification and correction unit, a missing data compensation unit, and a data standardization unit. The anomaly identification and correction unit has a built-in process threshold library and a lightweight machine learning anomaly identification model, while the missing data compensation unit has a built-in LSTM neural network prediction model.

Images