Instrument data analysis device based on edge calculation

Abstract

The invention relates to the technical field of industrial automation and intelligent sensing, and discloses an instrument data analysis device based on edge computing, which comprises a data acquisition and preprocessing module, a co-processing module, a cloud optimization module and an output display and interaction module. Through a lightweight-enhanced collaborative deep learning model and a self-adaptive data processing technology, each link of industrial instrument data processing can be comprehensively covered, common conductivity and temperature time sequence data are concerned, factors such as a sensor drift state and process fluctuation characteristics are also included, the model reasoning and data preprocessing process is deeply optimized, and the data processing efficiency is improved. The concentration soft measurement result can more accurately reflect the real concentration of the industrial solution, and a more scientific basis is provided for process control and quality monitoring of industrial production.

Description

Technical Field

[0001] This invention relates to the fields of industrial automation and intelligent sensing technology, and more specifically discloses an instrument data analysis device based on edge computing. Background Technology

[0002] While existing soft measurement technologies for solution concentrations represent an improvement over traditional methods, they still fall short in distributed industrial scenarios with multiple instruments, making it difficult to meet the demands for high precision, high reliability, and adaptability in the industrial context.

[0003] The prior art patent document with authorization announcement number CN120510959A discloses a "method for constructing a concentration prediction model for a mixed solution and a soft measurement method for concentration". The construction method includes: obtaining multiple sample data of the mixed solution through experiments, the sample data including temperature data, density data and concentration data; fitting based on all temperature data, density data and corresponding concentration data to obtain a target correspondence, and determining an initial prediction model based on the target correspondence; training the initial prediction model based on all temperature data and density data to obtain temporary concentration data; optimizing the initial prediction model based on the temporary concentration data and corresponding concentration data, and using the optimized initial prediction model as the final concentration prediction model.

[0004] While existing technologies obtain temperature, density, and concentration sample data of mixed solutions through experiments, construct prediction models through fitting, and optimize using loss functions, reducing manual table lookups to minimize human error and improve data reliability, and achieving rapid conversion from temperature and density data to concentration prediction, thus improving real-time performance and accuracy to some extent, and requiring only the storage of model parameters with low hardware requirements, they also have certain applicability. However, existing technologies rely on laboratory experimental data to build initial models, making them unsuitable for distributed multi-instrument layouts in industrial settings. They cannot integrate multi-point data to mine spatial correlations, limiting the accuracy of global concentration prediction. Furthermore, isolated processing of single-point data makes measurement continuity susceptible to disruptions due to localized failures. Additionally, model updates rely on offline experimental data optimization, lacking a dynamic collaborative mechanism between the edge and cloud, which is insufficient for industrial applications. When process parameters change (e.g., solution type switching), the system cannot quickly adapt to the new scenario, requiring downtime for updates that disrupt production continuity. Furthermore, it cannot utilize cloud-based global datasets to optimize the model and improve generalization capabilities. Additionally, it lacks an adaptive compensation mechanism to address the drift problem inherent in long-term industrial sensor operation. When sensor drift occurs, input data distortion leads to a significant increase in concentration prediction errors, requiring manual recalibration and incurring high maintenance costs. Data processing lacks a dynamic window to adapt to the dynamic fluctuations of industrial processes. When the process is stable, data redundancy easily occurs, reducing inference efficiency; when the process fluctuates rapidly, prediction lag occurs, failing to reflect real-time concentration. Finally, the system lacks sensor health diagnostics, making it impossible to monitor sensor operating status in real time, detect sensor anomalies promptly, and issue warnings, further impacting the accuracy of measurement data and the controllability of the production process. Summary of the Invention

[0005] The present invention mainly provides an instrument data analysis device based on edge computing, which can solve the problems mentioned in the background art.

[0006] To address the aforementioned technical problems, this invention provides the following technical solution, more specifically, an edge computing-based instrument data analysis device, comprising: a data acquisition and preprocessing module, a collaborative processing module, a cloud optimization module, and an output display and interaction module. The data acquisition and preprocessing module acquires relevant data during instrument operation and preprocesses the acquired raw data to complete preliminary calculations. The collaborative processing module receives information output from the data acquisition and preprocessing module, integrates and fuses the received multi-source information, performs deep computation based on the fused information, monitors the instrument's operating status, and enables data transmission between different modules. The cloud optimization module receives data transmitted from the collaborative processing module and manages the data uniformly, optimizes and adjusts calculation-related models, and simultaneously sends the optimized models or related parameters to the corresponding modules. The output display and interaction module enables human-machine interactive control operations and displays relevant information during the instrument data analysis process.

[0007] Furthermore, the data acquisition and preprocessing module includes: a distributed sensing acquisition module, a preprocessing module, and a lightweight inference module; Distributed sensing and acquisition module: Employs conductivity sensors, temperature sensors, and pressure sensors to acquire instrument operating data in real time; Preprocessing module: The raw data is processed using Hampel filter, moving average filter, and Z-Score normalization; Lightweight Inference Module: Employs a lightweight causal convolutional neural network and simplified gated loop units to perform preliminary inference on the preprocessed data and output preliminary calculation results.

[0008] Furthermore, the collaborative processing module includes: a data fusion module, a multi-branch inference module, a sensor health diagnosis module, and a communication module; Data fusion module: Employs timestamp alignment and multi-source data matrix construction techniques to integrate and fuse data from multiple nodes; Multi-branch inference module: Deep computation and state monitoring are performed through CNN-GRU-Attention network and feature-wise linear modulation; Sensor health diagnosis module: It uses mean deviation analysis and standard deviation threshold determination technology to perform health diagnosis on the sensor status; Communication module: Enables data transmission between modules via Ethernet and 4G / 5G wireless communication.

[0009] Furthermore, the cloud optimization module includes: a data management module, a model optimization module, and a model distribution and update module; Data management module: Employs global dataset storage and data classification indexing technology to uniformly manage transmitted data; Model optimization module: Employs global amplitude pruning, INT8 integer quantization, and batch normalization layer fusion techniques, while combining optimizers and loss functions to optimize and adjust the model; Model delivery and update module: Delivers optimized models or parameters using the STM32Cube.AI tool and TensorFlow Lite Micro conversion technology.

[0010] Furthermore, the output display and interaction module includes: an interaction control module and a visualization module; Interactive control module: It allows interaction via a touchscreen, is also equipped with physical buttons, and supports interface connection with field PLCs; Visualization module: includes local display screen and cloud-based web visualization platform.

[0011] According to another aspect of the present invention, an edge computing-based instrument data analysis method is provided. This method is implemented based on the above-described edge computing-based instrument data analysis device and specifically includes: S1. Acquire data during instrument operation, perform preprocessing operations on the raw data, and output preliminary calculation results; S2. Receive preliminary calculation results and related data, integrate and fuse multi-source data, perform in-depth calculations based on the fused data to monitor the instrument's operating status, and realize data transmission between different functional modules; S3. Receive the transmitted data, manage the data in a unified manner, optimize and adjust the model used for calculation, and send the optimized model or related parameters to the corresponding functional modules. S4. Enable human-machine interaction control operations, display relevant information during instrument data analysis, and complete information interaction and result presentation between humans and machines.

[0012] Furthermore, in S1, the acquired instrument operation data includes instrument conductivity data, temperature data, and pressure data. The preprocessing operations performed on the raw data include using a Hampel filter to remove outliers, a moving average filter to reduce noise interference, and Z-Score normalization to unify the data volume. The preliminary calculation is implemented through a lightweight inference module, and a lightweight causal convolutional neural network is used to extract local features of the data and a simplified gated loop unit to mine the temporal correlation of the data. Finally, the preliminary calculation results are output.

[0013] Furthermore, in S2, the integration and fusion of multi-source data adopts timestamp alignment technology to achieve time synchronization of data from different nodes. Combined with multi-source data matrix construction technology, the preliminary calculation results and original preprocessed data are integrated into a structured data matrix. Deep computation based on the fused data is carried out through a CNN-GRU-Attention network. At the same time, feature-wise linear modulation technology is used to optimize the feature interaction effect, thereby achieving accurate monitoring of the instrument's operating status.

[0014] Furthermore, in S3, the unified management of transmitted data adopts global dataset storage technology to achieve centralized data retention, and combines data classification indexing technology to improve data retrieval efficiency. The optimization and adjustment of the model used for calculation adopts global amplitude pruning, INT8 integer quantization, and batch normalization layer fusion technology, and at the same time, it is combined with Adam optimizer and MSE loss function to achieve model accuracy iteration.

[0015] Furthermore, in S4, the display of instrument data analysis-related information is presented in real time on the local display screen, showing the preliminary calculation results and real-time instrument operating parameters. The cloud-based Web visualization platform displays the global instrument operating status distribution, model optimization logs, and historical data analysis curves, realizing full-dimensional information interaction and result presentation between humans and machines.

[0016] The beneficial effects of this invention's edge computing-based instrument data analysis device are as follows: Through a lightweight-enhanced collaborative deep learning model and adaptive data processing technology, it comprehensively covers all aspects of industrial instrument data processing. It not only focuses on common conductivity and temperature time-series data but also incorporates factors such as sensor drift states and process fluctuation characteristics. Furthermore, it deeply optimizes model inference and data preprocessing, enabling concentration soft measurement results to more accurately reflect the true concentration of industrial solutions, providing a more scientific basis for process control and quality monitoring in industrial production. In addition, through adaptive data fusion and sensor health diagnosis technology, it achieves effective integration of multi-distributed instrument data, breaking down the barriers of isolated data processing. This allows for collaborative analysis and complementary verification between data from various measurement points, improving the accuracy and reliability of concentration measurement and enhancing the full lifecycle management level of industrial sensors. Simultaneously, through dynamic window partitioning and model hierarchical optimization technology, it provides more flexible and efficient data processing methods, accurately adapting to the dynamic fluctuations and real-time requirements of industrial processes. Moreover, the continuous optimization mechanism based on hierarchical models and global datasets enables rapid model iteration and updates when process changes or model performance degrades, greatly improving the adaptability of industrial soft measurement and its ability to cope with complex operating conditions. Attached Figure Description

[0017] The present invention will now be described in further detail with reference to the accompanying drawings and specific implementation methods.

[0018] Figure 1 This is a schematic diagram of the system framework; Figure 2 This is a flowchart illustrating the method. Detailed Implementation

[0019] The present invention will be described in detail below with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in the present application can be combined with each other.

[0020] According to one aspect of the invention, such as Figures 1-2 As shown, an edge computing-based instrument data analysis device is provided, comprising: Step 1: Data Acquisition, Preprocessing, and Output of Preliminary Calculation Results Acquire data during instrument operation, perform preprocessing on the raw data, and output preliminary calculation results; Specifically, the acquired instrument operation data includes instrument conductivity data, temperature data, and pressure data. The preprocessing operations performed on the raw data include using a Hampel filter to remove outliers, a moving average filter to reduce noise interference, and Z-Score normalization to unify the data volume. The preliminary calculation is implemented through a lightweight inference module, and a lightweight causal convolutional neural network is used to extract local features of the data and simplify the gated loop unit to mine the temporal correlation of the data. Finally, the preliminary calculation results are output.

[0021] In the distributed sensing acquisition process, conductivity data is acquired through conductivity sensors adapted to industrial scenarios, temperature data is acquired through industrial-grade temperature sensors, and pressure data is acquired through high-precision pressure sensors. The sampling frequency is dynamically adjusted according to process requirements to obtain raw conductivity, temperature, and pressure data in real time. After the acquired data is processed by the signal conditioning circuit, a digital signal conforming to industrial standards is output to ensure the integrity and real-time nature of the raw data and provide a high-quality data source for subsequent preprocessing. Furthermore, the Hampel filtering in the preprocessing stage strictly follows the median absolute deviation (MAD) calculation logic, first calculating the degree of data fluctuation using the following formula: In the formula, For the first Median absolute deviation of data points For the first in the sliding window One set of raw data, The median of the data within the sliding window. The sliding window width is adaptively adjusted based on data fluctuations to enhance resistance to sudden noise. Data deviating from the normal range is identified as outliers and replaced. Moving average filtering reduces random noise interference through smoothing. Z-score normalization unifies the data magnitude using the following formula: In the formula, The normalized version Time data, For the first The raw data at that moment This represents the mean of the corresponding data sequence. To ensure the comparability of different types of sensor data, the standard deviation of the corresponding data sequences is used to clear data format obstacles for subsequent inference calculations; The lightweight inference module's lightweight causal convolutional neural network extracts local features from the data through depthwise convolution and pointwise convolution. The depthwise convolution formula is: In the formula, This represents the nth output feature of the l-th depthwise convolution layer. Let i be the weight parameter of the l-th depthwise convolution. For the input normalized conductivity data, the point convolution formula is: In the formula, For the nth output feature of the l-th point convolution, Let be the weight matrix of the l-th layer point convolution. The bias term of the l-th layer point convolution can effectively capture the local correlation features of conductivity and pressure data, simplifying the gating loop unit by updating the gate: Reset Door: Candidate state: Final state: By mining the temporal correlation features of temperature and pressure data, the two work together to achieve efficient preliminary inference on the preprocessed data, and the time consumption of a single inference meets the real-time requirements of industry. After the initial calculations are completed, the lightweight inference module outputs the preliminary concentration values. At the same time, it organizes the preprocessed normalized data and the preliminary calculation results into a structured data package according to the timestamp order and uploads it synchronously to the subsequent collaborative processing stage. This ensures the continuity and integrity of data transmission, provides standardized input for the integration and fusion of multi-source data, and further guarantees the accuracy and reliability of subsequent deep calculations.

[0022] Step 2: Multi-source data integration and fusion, deep computing, and cross-module data transmission Receive preliminary calculation results and related data, integrate and fuse multi-source data, perform in-depth calculations based on the fused data to monitor the instrument's operating status, and realize data transmission between different functional modules; Specifically, the integration of multi-source data employs timestamp alignment technology to achieve time synchronization of data from different nodes. Combined with multi-source data matrix construction technology, the preliminary calculation results and raw preprocessed data are integrated into a structured data matrix. Deep computation based on the fused data is carried out through a CNN-GRU-Attention network. At the same time, feature-wise linear modulation technology is used to optimize the feature interaction effect, thereby achieving accurate monitoring of the instrument's operating status.

[0023] The received preprocessed data (such as normalized conductivity, temperature, and pressure data) and preliminary calculation results are first used as a benchmark, and linear interpolation is used to synchronize and calibrate the data of other nodes that have time deviations, ensuring that the data of all nodes are consistent in the time dimension. Then, according to the multi-source data matrix construction technology, the synchronized data of each node is organized into a structured data matrix according to the "node-parameter" dimension, which includes the normalized conductivity, temperature, and pressure data of each node and the preliminary calculation results. This matrix realizes the structured integration of multi-node data and provides a unified data input for deep computing. Before conducting deep computation, the drift of sensor data at each node is determined based on mean deviation analysis technology: the deviation of the mean of conductivity, temperature and pressure data in the current period of each node from the mean of the initial calibration stage is calculated. If the deviation exceeds the standard deviation threshold of the corresponding data, the sensor is determined to be drifting. At this time, the drifted data is compensated, and the first and second order differences are calculated on the compensated temperature data to form a feature vector containing the original value and the difference, which provides richer temporal information for deep feature extraction of the temperature branch. Furthermore, deep computation based on fused data is carried out through a CNN-GRU-Attention network with branching: the conductivity branch inputs the compensated conductivity data and extracts local correlation features through causal convolution, with the convolution process following depthwise separable convolution logic; the temperature branch inputs the above feature vectors into the GRU network to mine the long-term temporal dependency features of the data; in the feature fusion stage, feature-wise linear modulation (FiLM) technology is used to optimize the interaction effect, and the temperature branch outputs the modulation factor to dynamically modulate the features extracted by the conductivity branch; the modulated conductivity features and the temperature branch features are input together into the fully connected layer, and the instrument operating status parameters (such as precise concentration values) are calculated and output to achieve accurate monitoring of the instrument operating status; Finally, after completing deep computation and status monitoring, the communication module enables data transmission between different functional modules: on the one hand, multi-source data matrices, precise concentration, and sensor health status (e.g., normal / drift) are uploaded to the cloud optimization module via Ethernet or 4G / 5G wireless communication to provide data support for model optimization; on the other hand, precise concentration values ​​and sensor health status are synchronously transmitted to the output display and interaction module, and key operating parameters are fed back to the industrial control loop through the interface, ensuring the real-time performance and availability of data transmission, forming a complete closed loop of "data fusion - deep computation - status monitoring - data transmission".

[0024] Step 3: Data Management, Model Optimization, and Parameter Distribution Receive transmitted data, manage the data in a unified manner, optimize and adjust the model used for calculation, and send the optimized model or related parameters to the corresponding functional modules; Specifically, the unified management of transmitted data adopts global dataset storage technology to achieve centralized data retention, and data classification indexing technology is combined to improve data retrieval efficiency. The optimization and adjustment of the model used for calculation adopts global amplitude pruning, INT8 integer quantization, and batch normalization layer fusion technology, while the Adam optimizer and MSE loss function are used to achieve model accuracy iteration.

[0025] The system receives multi-source data matrices, precise concentration values, and sensor health status data transmitted from the collaborative processing module. It also combines real concentration values ​​obtained from laboratory tests and uses global dataset storage technology to centralize and retain these data to avoid data loss due to dispersion. At the same time, based on data classification and indexing technology, the data is classified, labeled, and indexed according to data source (e.g., different edge gateways), data type (e.g., conductivity, temperature, and pressure data), and time period. When data needs to be retrieved later, it can be quickly located through the index, which greatly improves data retrieval efficiency and provides efficient data support for model optimization. The optimization and adjustment of the model used for calculation first involves dividing the global dataset into training and validation samples. The MSE loss function is then used to measure the deviation between the model's predicted values ​​and the actual concentration values. The formula for the MSE loss function is as follows: In the formula, This is the loss value. To predict concentration values ​​for the model, Using the true concentration value, the model parameters are iteratively updated with the Adam optimizer to gradually reduce the loss value, thereby improving the model accuracy and ensuring that the model can more accurately adapt to the concentration prediction needs of industrial scenarios. Meanwhile, after the model optimization is completed, global amplitude pruning technology is used to remove redundant connections with small weight amplitudes in the model, reduce the number of model parameters to reduce storage consumption, and convert model parameters from floating point to integer through INT8 integer quantization technology to reduce the model's demand on edge device computing power. Batch normalization layer parameters are fused into the front convolutional layer or fully connected layer to simplify the model calculation process, avoid inference delay caused by multi-layer data transmission, and ensure that the optimized model can run efficiently on edge hardware. In addition, the compressed model is converted into a parameter file format compatible with the coprocessing module using the STM32Cube.AI tool or TensorFlow Lite Micro conversion technology. During the conversion process, the compatibility between the model structure and the hardware interface is automatically checked, and an adaptation report is generated. If there are incompatible items, the conversion parameters are adjusted in time to ensure that the model can be loaded and executed inference tasks normally on the coprocessing module, avoiding the model being unusable due to format issues. Finally, the model delivery and update module delivers the converted model parameter file via Ethernet or 4G / 5G wireless communication. After receiving the file, it first verifies its integrity. Once confirmed to be correct, it synchronizes it to all subordinate edge nodes. The entire update process does not require downtime. Edge modules can seamlessly load the new model after completing the current inference task. At the same time, the cloud records the time, version, and optimization metrics (such as the accuracy improvement ratio) of each model update, forming a model update log. This facilitates the tracking of the iteration process and provides a reference for the direction of future model optimization.

[0026] Step 4: Human-computer interaction control, data analysis, information display, and result presentation. Enables human-machine interactive control operations, displays relevant information during instrument data analysis, and completes information interaction and result presentation between humans and machines; Specifically, the display of instrument data analysis-related information is presented in real time on the local display screen, showing preliminary calculation results and real-time instrument operating parameters. The cloud-based web visualization platform displays the global instrument operating status distribution, model optimization logs, and historical data analysis curves, realizing full-dimensional information interaction and result presentation between humans and machines.

[0027] It receives precise concentration values, sensor health status, and pre-processed conductivity, temperature, and pressure data from the depth calculation module. Basic control is achieved through an interactive interface consisting of a touch screen and physical function keys: users can directly trigger sensor calibration commands on the interface, and the system calls the benchmark data from the initial calibration stage to complete the comparison calibration. Users can also manually set the start and stop times and display cycles for data acquisition. The interface responds to operations in real time and provides feedback on the execution status. If the sensor drifts or the data is abnormal, a pop-up window will be used to prioritize the prompt, ensuring the accuracy and timeliness of the operation. In addition, the data analysis information display adopts a layered presentation logic: real-time data is displayed intuitively through the numerical panel, showing the current conductivity, temperature, pressure values ​​and precise concentration values, and is accompanied by a dynamic dashboard indicating the process range of the concentration value. Historical data is presented as a line graph showing the trend of change over time, and the trend curves of different parameters can be switched to clearly show the data fluctuation pattern. The sensor health status is displayed in a graphical form, with different colors used to distinguish between normal and drift states. The equipment operating status can be quickly identified without complicated operations, which meets the needs of efficient information acquisition in industrial scenarios. It also supports data export and remote interaction functions: users can export the analysis data for a specified time period into commonly used industrial format files through the interface, which is convenient for offline archiving and secondary analysis. By connecting to the cloud data channel, the interface of the remote terminal can be displayed synchronously. Authorized users can view real-time running data and historical records on the remote terminal. Multi-level permission management is set, with different operation permissions corresponding to different control and viewing ranges, taking into account both the convenience of interaction and data security. The final result presents a complete "monitoring-evaluation-feedback" output: the core display shows the comparison results between the accurate concentration value and the process standard threshold, and simultaneously marks the sensor health status assessment conclusions, providing on-site operators with an intuitive basis for judging the operating status. For cases exceeding the threshold, the handling suggestions related to similar historical operating conditions are also presented. At the same time, the key results are synchronously fed back to the connected PLC control system, and the execution feedback status of control commands is displayed in real time on the interactive interface. The operation process and result data are automatically logged, which facilitates subsequent traceability and process optimization, giving full play to the practical value of data analysis.

[0028] Of course, the above description is not a limitation of the present invention, and the present invention is not limited to the examples given above. Any changes, modifications, additions or substitutions made by those skilled in the art within the scope of the present invention are also within the protection scope of the present invention.

Claims

1. An instrument data analysis device based on edge computing, characterized in that, The device includes: a data acquisition and preprocessing module, a collaborative processing module, a cloud optimization module, and an output display and interaction module. The data acquisition and preprocessing module acquires relevant data during instrument operation and preprocesses the raw data to perform preliminary calculations. The collaborative processing module receives information output from the data acquisition and preprocessing module, integrates and fuses the received multi-source information, performs deep calculations based on the fused information, monitors the instrument's operating status, and facilitates data transmission between different modules. The cloud optimization module receives data transmitted from the collaborative processing module, manages the data uniformly, optimizes and adjusts calculation-related models, and simultaneously sends the optimized models or relevant parameters to the corresponding modules. The output display and interaction module enables human-machine interactive control operations and displays relevant information during the instrument data analysis process.

2. The instrument data analysis device based on edge computing according to claim 1, characterized in that: The data acquisition and preprocessing module includes: a distributed sensor acquisition module, a preprocessing module, and a lightweight inference module; Distributed sensing and acquisition module: Employs conductivity sensors, temperature sensors, and pressure sensors to acquire instrument operating data in real time; Preprocessing module: The raw data is processed using Hampel filter, moving average filter, and Z-Score normalization; Lightweight Inference Module: Employs a lightweight causal convolutional neural network and simplified gated loop units to perform preliminary inference on the preprocessed data and output preliminary calculation results.

3. The instrument data analysis device based on edge computing according to claim 1, characterized in that: The collaborative processing module includes: a data fusion module, a multi-branch inference module, a sensor health diagnosis module, and a communication module; Data fusion module: Employs timestamp alignment and multi-source data matrix construction techniques to integrate and fuse data from multiple nodes; Multi-branch inference module: Deep computation and state monitoring are performed through CNN-GRU-Attention network and feature-wise linear modulation; Sensor health diagnosis module: It uses mean deviation analysis and standard deviation threshold determination technology to perform health diagnosis on the sensor status; Communication module: Enables data transmission between modules via Ethernet and 4G / 5G wireless communication.

4. The instrument data analysis device based on edge computing according to claim 1, characterized in that: The cloud-based optimization module includes: a data management module, a model optimization module, and a model distribution and update module; Data management module: Employs global dataset storage and data classification indexing technology to uniformly manage transmitted data; Model optimization module: Employs global amplitude pruning, INT8 integer quantization, and batch normalization layer fusion techniques, while combining optimizers and loss functions to optimize and adjust the model; Model delivery and update module: Delivers optimized models or parameters using the STM32Cube.AI tool and TensorFlow Lite Micro conversion technology.

5. The instrument data analysis device based on edge computing according to claim 1, characterized in that: The output display and interaction module includes: an interaction control module and a visualization module; Interactive control module: It allows interaction via a touchscreen, is also equipped with physical buttons, and supports interface connection with field PLCs; Visualization module: includes local display screen and cloud-based web visualization platform.

6. A method for analyzing instrument data based on edge computing, characterized in that, This method is implemented based on an edge computing-based instrument data analysis device as described in any one of claims 1-5, and specifically includes: S1. Acquire data during instrument operation, perform preprocessing operations on the raw data, and output preliminary calculation results; S2. Receive preliminary calculation results and related data, integrate and fuse multi-source data, perform in-depth calculations based on the fused data to monitor the instrument's operating status, and realize data transmission between different functional modules; S3. Receive the transmitted data, manage the data in a unified manner, optimize and adjust the model used for calculation, and send the optimized model or related parameters to the corresponding functional modules. S4. Enable human-machine interaction control operations, display relevant information during instrument data analysis, and complete information interaction and result presentation between humans and machines.

7. The instrument data analysis method based on edge computing according to claim 6, characterized in that: In step S1, the acquired instrument operation data includes instrument conductivity data, temperature data, and pressure data. The preprocessing operations performed on the raw data include using a Hampel filter to remove outliers, using a moving average filter to reduce noise interference, and using Z-Score normalization to unify the data volume. The preliminary calculation is implemented through a lightweight inference module, using a lightweight causal convolutional neural network to extract local features of the data, and using a simplified gated loop unit to mine the temporal correlation of the data. Finally, the preliminary calculation results are output.

8. The instrument data analysis method based on edge computing according to claim 6, characterized in that: In S2, the integration and fusion of multi-source data adopts timestamp alignment technology to achieve time synchronization of data from different nodes. Combined with multi-source data matrix construction technology, the preliminary calculation results and original preprocessed data are integrated into a structured data matrix. Deep calculation based on the fused data is carried out through CNN-GRU-Attention network. At the same time, feature-wise linear modulation technology is used to optimize the feature interaction effect, so as to achieve accurate monitoring of the instrument's operating status.

9. The instrument data analysis method based on edge computing according to claim 6, characterized in that: In S3, the unified management of transmitted data adopts global dataset storage technology to achieve centralized data retention, and data classification indexing technology is combined to improve data retrieval efficiency. The optimization and adjustment of the model used for calculation adopts global amplitude pruning, INT8 integer quantization, and batch normalization layer fusion technology, and is combined with Adam optimizer and MSE loss function to achieve model accuracy iteration.

10. The instrument data analysis method based on edge computing according to claim 6, characterized in that: In S4, the display of instrument data analysis-related information is presented in real time on the local display screen, showing the preliminary calculation results and real-time instrument operating parameters. The cloud-based Web visualization platform displays the global instrument operating status distribution, model optimization logs, and historical data analysis curves, realizing full-dimensional information interaction and result presentation between humans and machines.

Images