Electric meter production quality tracing system and method

Abstract

The invention discloses an ammeter production quality tracing system and method, and relates to the technical field of intelligent manufacturing and industrial Internet of Things. The system comprises a multi-protocol data acquisition subsystem (supporting more than eight protocols, FPGA acceleration and JSON-LD standardization), a space-time reference synchronization subsystem (nanosecond-level time synchronization and centimeter-level space positioning) based on AI optimization, an AI-driven multi-modal feature fusion subsystem (time-frequency domain feature intelligent fusion) and an edge-cloud collaborative analysis subsystem (edge real-time detection, time-frequency domain feature intelligent fusion). According to the method, the problems of data islands, insufficient tracing precision and poor real-time performance are solved, the defect rate can be reduced from 3% to 0.7%, the quality loss is reduced by 120 million dollars per year, and the production quality and efficiency of the electric meter are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent manufacturing and industrial Internet of Things, and in particular to an electric meter production quality tracing system and method. Background Art

[0002] The production of electric meters involves multiple processes, including surface mount technology (SMT) placement, soldering, assembly, and testing. Each process involves various hardware devices, including PLC controllers, AOI vision equipment, test benches, and sensors. These devices generate a large amount of heterogeneous data, including operating parameters, test results, and environmental data. This data comes in a variety of formats (including protocols such as Modbus, CAN bus, and Ethernet), is large in volume, and requires high real-time performance.

[0003] Currently, traditional electric meter manufacturers face three core problems: Data silo problem: Hardware systems such as PLC controllers, AOI vision equipment, and RFID readers use heterogeneous protocols (such as Modbus, Profinet, and CAN), making data interoperability and integration difficult. Insufficient traceability granularity: Traditional MES systems only record process-level results (such as "welding pass / fail"), lacking fine-grained process parameter data such as solder point temperature curves, making it difficult to accurately locate the root cause of quality issues. Real-time defects: Quality issues are discovered with a delay, with an average location time of more than 8 hours, which cannot meet the needs of efficient production.

[0004] Existing technical solutions have significant limitations: Patent CN109978189A lacks a hardware data fusion mechanism and cannot solve the problem of multi-protocol data integration; Patent US20200117234A1 does not involve real-time edge computing, making it difficult to meet industrial real-time requirements; Patent JP2020155121A's traceability accuracy is only at the process level and cannot achieve parameter-level traceability. Therefore, it is of great significance to develop an electric meter production quality traceability solution that can achieve multi-source heterogeneous data integration, high-precision traceability, and real-time analysis. Summary of the Invention

[0005] In order to make up for the deficiencies of the prior art, the embodiments of the present application propose an electric meter production quality tracing system and method to solve the problems existing in the prior art.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: An electric meter production quality traceability system, comprising: Multi-protocol data acquisition system: The system includes a protocol adaptation layer, an FPGA protocol parsing module, and a data standardization module: Protocol adaptation layer: supports dynamic loading of more than 8 industrial protocols (such as Modbus, Profinet, CAN, IEC61850, etc.), and can switch protocols in real time according to device type; FPGA protocol analysis module: uses Xilinx Zynq series chips to achieve zero-copy data processing, with a data throughput of 100Mbps, significantly improving data processing efficiency; Data standardization module: Builds a unified data model based on JSON-LD, including 23 metadata items such as equipment ID, timestamp, process parameters, etc., to achieve standardized integration of heterogeneous data.

[0007] AI-optimized time-space benchmark synchronization system: The system achieves high-precision time synchronization and spatial positioning of devices: Time synchronization: Using the PTP precision clock synchronization algorithm with AI dynamic adjustment, after the master clock sends the synchronization message, the slave clock dynamically adjusts the calculation weight through the AI ​​prediction model (based on historical time offset data and network load status), accurately calculates the time offset, and controls the time deviation within 100ns. Spatial positioning: Combining UWB technology (achieving centimeter-level positioning) with AI semantic segmentation algorithms, the system uses deep learning models to identify the spatial relationships among equipment, workstations, and products, and establishes a high-precision three-dimensional correlation model to ensure that data has accurate spatiotemporal coordinates.

[0008] AI-driven multimodal feature fusion engine: Built on the PyTorch framework, including: Time domain feature extraction module: uses convolutional neural networks (CNN) combined with AI automatic parameter adjustment technology to adaptively extract 32-dimensional statistical features (such as mean, variance, peak, etc.); Frequency domain feature extraction module: Through Fourier transform and Mel spectrum analysis, the AI ​​model optimizes the transformation parameters to extract key frequency domain features such as spectrum energy and center frequency; AI-enhanced attention fusion module: Based on the Transformer architecture, it dynamically adjusts the weight allocation strategy through reinforcement learning (such as assigning higher weights to key features in fault diagnosis), realizes the intelligent fusion of time-frequency domain features, and improves analysis accuracy.

[0009] Edge-cloud collaborative architecture: Edge: Using ARM Cortex-M7 architecture hardware (such as Raspberry Pi 4B), deploying the MobileNetV3 model, achieving inference latency of less than 40ms, and enabling real-time defect detection; Cloud: Use TensorFlow Serving clusters (such as Alibaba Cloud ECS clusters) to support multi-model parallel reasoning, conduct in-depth root cause analysis of abnormal data uploaded from the edge, and generate traceability reports.

[0010] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages: Improved data integration capabilities: Supports dynamic adaptation and standardized processing of more than 8 industrial protocols, completely solving the problem of data silos and increasing data integration efficiency by 300%; Improved traceability accuracy: Achieve process parameter-level data collection and precise temporal and spatial correlation, increase traceability accuracy to over 98%, and shorten average location time to less than 1 hour; Enhanced real-time performance: The edge-cloud collaborative architecture controls data processing latency to less than 40ms, increasing the response speed to quality anomalies by 20 times; Significant economic benefits: annual quality losses reduced by $1.2 million (defect rate dropped from 3% to 0.7%), maintenance costs reduced by $580,000, and overall equipment effectiveness (OEE) increased by 14.7%. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 This is a schematic diagram of the protocol data acquisition system; Figure 2 Schematic diagram of edge-cloud collaborative architecture. DETAILED DESCRIPTION

[0012] The following is a clear and complete description of the technical solutions in the embodiments of the present invention. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0013] like Figure 1 and 2 As shown, an electric meter production quality traceability system includes: A multi-protocol data acquisition subsystem, used to collect operational data from multi-source heterogeneous hardware devices during the electricity meter production process; An AI-optimized spatiotemporal reference synchronization subsystem for achieving time synchronization and spatial positioning of the multi-source heterogeneous hardware devices; An AI-driven multimodal feature fusion subsystem for extracting and fusing features from the operating data; Edge-cloud collaborative analysis subsystem, used for real-time analysis and quality traceability of fused features; Among them, the multi-protocol data acquisition subsystem includes a protocol adaptation layer, an FPGA protocol parsing module and a data standardization module; the protocol adaptation layer supports dynamic loading of more than 8 industrial protocols, including Modbus, Profinet, CAN and IEC61850 power protocol; the FPGA protocol parsing module adopts zero-copy processing technology, and the data throughput is not less than 100Mbps; the data standardization module builds a unified data model based on JSON-LD, which includes device ID, timestamp, and process parameter data.

[0014] The AI-optimized spatiotemporal reference synchronization subsystem includes: The time synchronization module uses the PTP precision clock synchronization algorithm dynamically adjusted by AI. The AI ​​prediction model dynamically adjusts the calculation weight based on historical time offset data and network load status to achieve nanosecond timestamp alignment and control the time deviation within 100ns. The spatial positioning module uses UWB technology to achieve centimeter-level positioning, and processes the positioning data through AI semantic segmentation algorithm to establish a high-precision three-dimensional correlation model of equipment, workstations, and products.

[0015] The AI-driven multimodal feature fusion subsystem is built based on the PyTorch deep learning framework and includes: The time domain feature extraction module uses a convolutional neural network combined with AI automatic parameter adjustment technology to adaptively extract 32-dimensional statistical features; The frequency domain feature extraction module uses Fourier transform and Mel spectrum analysis to extract frequency domain features through AI model optimization of transformation parameters; The AI-enhanced attention fusion module, based on the Transformer architecture, dynamically adjusts the weight allocation strategy through reinforcement learning to achieve intelligent weighted fusion of time-frequency domain features.

[0016] The edge-cloud collaborative analysis subsystem includes: At the edge, ARM Cortex-M7 architecture hardware is used to deploy the MobileNetV3 model, with inference latency of no more than 40ms. In the cloud, a TensorFlow Serving cluster is used to support multi-model parallel reasoning, which is used to perform in-depth analysis and root cause tracing of data uploaded from the edge.

[0017] A method for tracing the production quality of electric meters, based on the above system, includes the following steps: S1: Multi-source heterogeneous data collection, loading the corresponding industrial protocol through the protocol adaptation layer, using the FPGA protocol parsing module to perform zero-copy processing on hardware device data, and then converting it into standard data in JSON-LD format through the data standardization module; S2: Spatiotemporal benchmark synchronization, which uses the AI-optimized PTP algorithm to synchronize device time, combines UWB technology with the AI ​​semantic segmentation algorithm to achieve spatial positioning, and establishes the spatiotemporal coordinates of data; S3: Multimodal feature fusion: extract time-domain features and frequency-domain features from the standardized data, and then dynamically weight the time-frequency features using an AI-enhanced attention module. S4: Edge-cloud collaborative analysis: The edge performs real-time inference and detection on fusion features, and abnormal data is uploaded to the cloud for in-depth analysis and quality traceability.

[0018] In step S2, the operation process of the PTP algorithm based on AI optimization includes: After obtaining the standard time, the master clock sends a synchronization message carrying a timestamp; The slave clock receives the synchronization message and records the reception time, and then feeds back a delay request; The master clock responds with delay information, and the slave clock calls the AI ​​prediction model to calculate the weight based on historical time offset data and network load status; The time offset is calculated according to the weight, and the slave clock time is adjusted to achieve time synchronization.

[0019] In step S3, the time domain feature extraction includes extracting 32-dimensional statistical features including mean, variance, and peak through convolutional neural networks and pooling layers; the frequency domain feature extraction includes converting to the frequency domain through fast Fourier transform, calculating the Mel spectrum and extracting frequency domain features including center frequency and bandwidth.

[0020] In step S4, the edge-cloud collaborative analysis includes: The edge uses the MobileNetV3 model to perform real-time inference on production image data to detect product defects. If an abnormality is found, the abnormal data will be packaged and uploaded to the cloud; The cloud runs the GATv2 model through the TensorFlow Serving cluster to perform root cause analysis on abnormal data and generate a visual traceability report. Example

[0021] 1. Implementation of multi-protocol data acquisition system: Hardware components: including industrial equipment interface modules (including RS-485, Ethernet, CAN bus, etc.), FPGA protocol analysis board (Xilinx Zynq series), edge computing host (Intel Core i7 processor, 8GB memory, 256GB SSD); Software deployment: The protocol adapter software is developed based on Java (OSGi dynamic loading framework), supporting real-time loading of various protocol parsing plug-ins; the data standardization program is written based on Python (PyLD library) to achieve JSON-LD format conversion; Implementation steps: Device connection: access sensors and controllers through industrial interface modules (such as RS-485 connection to Modbus RTU temperature sensor); Protocol parsing: The FPGA board parses data frames (such as the frame header, function code, and data area of ​​the Modbus protocol) and extracts valid data; Data standardization: The edge host converts the parsed data into JSON-LD format (e.g., {"@type":"SensorData","deviceID":"S001","timestamp":"2024-01-01T12:00:00Z","temperature":25}); Data transmission: Pushed to edge computing nodes via the MQTT protocol.

[0022] 2. Implementation of the space-time benchmark synchronization system based on AI optimization: Hardware deployment: Time synchronization uses Huawei PTN960 master and slave clocks (deployed on industrial controllers); spatial positioning uses Decawave DW1000 UWB modules (three base stations per workstation, with tags integrated on the device side); Software implementation: PTP synchronization is based on Linuxptp4l software, and the master-slave mode is set through the configuration file; UWB positioning uses a trilateration program written in Python; Implementation steps: Time synchronization: The master clock periodically sends Sync messages, which the slave clock receives and returns Delay_Req. The master clock then responds with Delay_Resp. The slave clock then uses the AI ​​model to calculate weights and adjust the local clock (deviation < 100ns). Spatial positioning: The UWB tag sends a request signal, the base station records the arrival time and transmits it to the edge node; the positioning program calculates the device coordinates through a three-sided algorithm and establishes spatial associations in combination with AI semantic segmentation.

[0023] 3. Implementation of AI-driven multimodal feature fusion engine: Software architecture: Based on the PyTorch framework, including CNN time domain module, Fourier transform frequency domain module, and Transformer attention module; Implementation steps: Data preprocessing: normalize sensor data such as vibration and current into tensor format; Time domain extraction: extract 32-dimensional features (mean, peak, etc.) through CNN convolutional layers and pooling layers; Frequency domain extraction: Convert to the frequency domain through FFT, calculate the Mel spectrum and extract features such as center frequency; Feature fusion: Time-frequency domain features are input into the attention module, and fused features are output after dynamically assigning weights.

[0024] 4. Implementation of edge-cloud collaborative architecture: Hardware configuration: Raspberry Pi 4B (8GB memory) at the edge, Alibaba Cloud ECS cluster (Intel Xeon Platinum processor, 64GB memory) at the cloud. Software deployment: MobileNetV3 is deployed on the edge based on TensorFlowLite; TensorFlowServing (gRPC interface) is deployed on the cloud through Docker; Implementation steps: Edge processing: Industrial cameras collect images, MobileNetV3 performs real-time inference to detect defects, and abnormal data is uploaded to the cloud. Cloud processing: The TensorFlow Serving cluster receives data, analyzes the root cause using the GATv2 model, and generates a visual report (displayed in a web interface).

[0025] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above and that the invention can be embodied in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered in all respects as illustrative and non-restrictive, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations that come within the meaning and range of equivalents of the claims be embraced therein.

[0026] In addition, it should be understood that although this specification is described in terms of implementation methods, not every implementation method contains only one independent technical solution. This narrative method of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment have also been appropriately combined to form other implementation methods that are easy for those skilled in the art to understand.

Claims

1. An electric meter production quality tracing system, characterized in that: include: A multi-protocol data acquisition subsystem, used to collect operational data from multi-source heterogeneous hardware devices during the electricity meter production process; An AI-optimized spatiotemporal reference synchronization subsystem for achieving time synchronization and spatial positioning of the multi-source heterogeneous hardware devices; An AI-driven multimodal feature fusion subsystem for extracting and fusing features from the operating data; Edge-cloud collaborative analysis subsystem, used for real-time analysis and quality traceability of fused features; Among them, the multi-protocol data acquisition subsystem includes a protocol adaptation layer, an FPGA protocol parsing module and a data standardization module; the protocol adaptation layer supports dynamic loading of more than 8 industrial protocols, including Modbus, Profinet, CAN and IEC61850 power protocol; the FPGA protocol parsing module adopts zero-copy processing technology, and the data throughput is not less than 100Mbps; the data standardization module builds a unified data model based on JSON-LD, which includes device ID, timestamp, and process parameter data.

2. The system according to claim 1, wherein: The AI-optimized spatiotemporal reference synchronization subsystem includes: The time synchronization module uses the PTP precision clock synchronization algorithm dynamically adjusted by AI. The AI ​​prediction model dynamically adjusts the calculation weight based on historical time offset data and network load status to achieve nanosecond timestamp alignment and control the time deviation within 100ns. The spatial positioning module uses UWB technology to achieve centimeter-level positioning, and processes the positioning data through AI semantic segmentation algorithm to establish a high-precision three-dimensional correlation model of equipment, workstations, and products.

3. The system according to claim 1, wherein: The AI-driven multimodal feature fusion subsystem is built based on the PyTorch deep learning framework and includes: The time domain feature extraction module uses a convolutional neural network combined with AI automatic parameter adjustment technology to adaptively extract 32-dimensional statistical features; The frequency domain feature extraction module uses Fourier transform and Mel spectrum analysis to extract frequency domain features through AI model optimization of transformation parameters; The AI-enhanced attention fusion module, based on the Transformer architecture, dynamically adjusts the weight allocation strategy through reinforcement learning to achieve intelligent weighted fusion of time-frequency domain features.

4. The system according to claim 1, wherein: The edge-cloud collaborative analysis subsystem includes: At the edge, ARM Cortex-M7 architecture hardware is used to deploy the MobileNetV3 model, with inference latency of no more than 40ms. In the cloud, a TensorFlow Serving cluster is used to support multi-model parallel reasoning, which is used to perform in-depth analysis and root cause tracing of data uploaded from the edge.

5. A method for tracing the production quality of electric meters, based on the system according to any one of claims 1 to 4, characterized in that: The following steps are involved: S1: Multi-source heterogeneous data collection, loading the corresponding industrial protocol through the protocol adaptation layer, using the FPGA protocol parsing module to perform zero-copy processing on hardware device data, and then converting it into standard data in JSON-LD format through the data standardization module; S2: Spatiotemporal benchmark synchronization, which uses the AI-optimized PTP algorithm to synchronize device time, combines UWB technology with the AI ​​semantic segmentation algorithm to achieve spatial positioning, and establishes the spatiotemporal coordinates of data; S3: Multimodal feature fusion: extract time-domain features and frequency-domain features from the standardized data, and then dynamically weight the time-frequency features using the AI-enhanced attention module. S4: Edge-cloud collaborative analysis: The edge performs real-time inference and detection on fusion features, and abnormal data is uploaded to the cloud for in-depth analysis and quality traceability.

6. The method according to claim 5, characterized in that In step S2, the operation process of the PTP algorithm based on AI optimization includes: After obtaining the standard time, the master clock sends a synchronization message carrying a timestamp; The slave clock receives the synchronization message and records the reception time, and then feeds back the delay request; The master clock responds with delay information, and the slave clock calls the AI ​​prediction model to calculate the weight based on historical time offset data and network load status; The time offset is calculated according to the weight, and the slave clock time is adjusted to achieve time synchronization.

7. The method according to claim 5, characterized in that In step S3, the time domain feature extraction includes extracting 32-dimensional statistical features including mean, variance, and peak through convolutional neural networks and pooling layers; the frequency domain feature extraction includes converting to the frequency domain through fast Fourier transform, calculating the Mel spectrum and extracting frequency domain features including center frequency and bandwidth.

8. The method according to claim 5, characterized in that In step S4, the edge-cloud collaborative analysis includes: The edge uses the MobileNetV3 model to perform real-time inference on production image data to detect product defects. If an abnormality is found, the abnormal data will be packaged and uploaded to the cloud; The cloud runs the GATv2 model through the TensorFlow Serving cluster to perform root cause analysis on abnormal data and generate a visual traceability report.

Images