Sorting assembly line monitoring system for returned intelligent electric meters

Abstract

The invention discloses a returned intelligent electric meter sorting assembly line monitoring system. The system comprises a feature fusion sorting module, a digital mapping control module, a chain type tracing management module, a state diagnosis module and an environment management and control module. The feature fusion sorting module collects ammeter data through an industrial camera array, a microphone array, a structured light scanner and an infrared thermal imager, and performs feature fusion classification; the digital mapping control module calculates sorting process control parameters by using a distributed control model and an optimization algorithm; the chain type traceability management module adopts a block chain technology to record data and realize information retrieval; the state diagnosis module performs equipment fault prediction in combination with a data prediction technology; the environment management and control module monitors key parameters of the sorting environment. According to the system, the automation and intelligence level of an electricity meter sorting assembly line is improved, and the problems that an existing system is low in sorting efficiency, lagged in equipment monitoring and the like are solved.

Description

Technical Field

[0001] The invention belongs to the technical field of smart meter processing, and in particular relates to a monitoring system for a sorting line of returned smart meters. Background Art

[0002] With the continuous advancement of smart grid construction, smart meters have been widely used in power systems. In actual operation, a large number of smart meters need to be returned due to equipment failure, replacement and other reasons. These returned meters need to go through multiple links such as sorting, testing and maintenance. However, the existing returned smart meter processing system has the following technical problems:

[0003] 1. The traditional manual sorting mode is inefficient:

[0004] Due to the wide variety of smart meters, models and specifications, manual sorting is not only time-consuming and laborious, but also prone to classification errors, which affects the accuracy of subsequent processing. Manual recording leads to data distortion, which cannot meet the needs of refined management and affects processing efficiency and quality.

[0005] 2. The automated sorting line system is complex and collaboration is difficult:

[0006] The processing flow of returned smart meters involves multiple links, such as transportation, testing, disassembly, and labeling, which require coordinated control of various types of equipment, including conveyor belts, robots, pressure test equipment, appearance inspection equipment, laser engraving machines, etc. In the existing automated assembly line system, there is a lack of effective integration and coordination between various types of equipment, forming a complex and error-prone automated system, which makes it difficult to ensure the efficiency and accuracy of the sorting process.

[0007] 3. Lack of a unified monitoring platform makes management difficult:

[0008] Existing systems often fail to provide a centralized and unified monitoring platform, making it difficult to monitor and manage equipment and processes in real time. Since the sorting line involves many types of equipment and is widely distributed, with many data collection points, it is difficult for existing systems to detect and handle various abnormal situations in a timely manner, resulting in delayed fault response, low production efficiency and equipment utilization.

[0009] 4. Difficulties in tracing the entire process:

[0010] The various data generated during the processing of returned smart meters have not been effectively linked and integrated, making it difficult to trace the entire process from entering the processing link to the final completion of the processing. This not only limits the ability of quality management and continuous improvement, but also increases the difficulty of tracing the potential source of failure.

[0011] 5. Equipment status monitoring is delayed and maintenance response is slow:

[0012] The existing system lacks real-time monitoring of equipment operating status, resulting in delayed response when equipment failure occurs. The lack of effective evaluation of equipment performance makes it difficult to implement preventive maintenance measures, thereby increasing the frequency of equipment failures and repair costs.

[0013] 6. Lack of security monitoring:

[0014] In the process of handling returned smart meters, there are potential sources of danger such as mechanical movement and high-voltage testing. Existing systems usually lack effective safety monitoring of the operating area, personnel entry and exit, and abnormal working conditions, and cannot detect and handle safety hazards in a timely manner, affecting operational safety.

[0015] 7. Inadequate environmental monitoring affects the quality of treatment:

[0016] During the processing of returned smart meters, environmental factors such as temperature, humidity, air pressure, noise, etc. have a significant impact on the processing quality. However, the existing system fails to effectively monitor these environmental factors, resulting in unstable processing results.

[0017] 8. Insufficient remote management capabilities:

[0018] Most of the current systems can only be managed on-site and lack remote monitoring, diagnosis and technical support capabilities, which limits management efficiency and technical service levels. In particular, when operating across regions or diagnosing equipment remotely, timely response and effective intervention cannot be achieved.

[0019] Therefore, there is an urgent need to develop a monitoring system for the sorting line of returned smart meters, which can improve the level of automation and intelligence in the processing of returned meters and solve the above-mentioned technical problems. Summary of the invention

[0020] The present invention provides a monitoring system for a returned smart meter sorting line, which solves technical problems such as low manual sorting efficiency, complex line control, untraceable processing process, lagging equipment status monitoring, and lack of environmental safety monitoring in the returned smart meter sorting line.

[0021] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0022] A monitoring system for a returned smart meter sorting line includes: a feature fusion sorting module, a digital mapping control module, a chain traceability management module, a status diagnosis module and an environmental control module; wherein:

[0023] The feature fusion sorting module is located at the input end of the system. It collects the appearance image of the meter through the industrial camera array, the mechanical operation acoustic signal through the microphone array, the three-dimensional point cloud data through the structured light scanner, and the temperature field data through the infrared thermal imager. It uses a hierarchical attention network (Swin Transformer) to process image data, Mel spectrum analysis to process acoustic signals, a point cloud neural network (PointNet++) to process three-dimensional point clouds, and a region growing algorithm to process temperature fields. It uses a weighted attention mechanism to fuse the above features and transmit the classification results to the digital mapping control module.

[0024] The digital mapping control module receives the classification results and equipment status data, builds the equipment status migration model using the time Petri net, realizes cross-equipment status consensus based on the Byzantine Fault Tolerance (PBFT) mechanism, calculates the sorting process control parameters in combination with the distributed optimization control algorithm based on the alternating direction multiplier method (ADMM), renders the equipment operation status in real time through WebAssembly technology, and sends the control parameters to the execution device;

[0025] The chain traceability management module collects the meter serial number through a barcode scanner, obtains the classification identification and process time data from the digital mapping control module, uses sharded blockchain technology to establish a data storage structure, uses a directed acyclic graph to record the meter transfer process, and establishes a time series index based on the logarithmic structure merge (LSM) tree. The mapping relationship between meter processing and equipment status is constructed through the graph database Neo4j to provide operation data for the status diagnosis module;

[0026] The status diagnosis module collects equipment operating parameters through speed sensors and vibration sensors, builds a parameter time series prediction model using a bidirectional gated recurrent network (BiGRU), uses a domain adversarial network to eliminate parameter distribution differences between devices, extracts fault feature frequency bands through wavelet packet decomposition, generates equipment maintenance decisions in combination with a proximal strategy optimization algorithm, and transmits fault warning information to the digital mapping control module;

[0027] The environmental control module collects the temperature and humidity parameters of the sorting environment through the temperature and humidity sensor network, obtains personnel location data using a lidar scanner, identifies work behavior using a spatiotemporal convolutional network, predicts motion trajectories using a Kalman filter algorithm, and monitors the sorting environmental parameters based on a computational fluid dynamics model based on hierarchical decomposition. The monitoring information of the sorting operation site is transmitted to the digital mapping control module for linkage processing.

[0028] Beneficial effects of the present invention:

[0029] 1. Improve sorting efficiency and accuracy

[0030] By integrating multiple sensors (industrial camera array, microphone array, structured light scanner and infrared thermal imager) into the feature fusion sorting module for data collection and processing, the present invention realizes the rapid classification of returned smart meters, improves the sorting efficiency and reduces the occurrence of misclassification. This automated method significantly improves the overall sorting speed and processing accuracy.

[0031] 2. Simplify system control and optimization

[0032] The digital mapping control module optimizes the coordination of device states through the time Petri net and Byzantine fault tolerance mechanism (PBFT), and simplifies the control process across devices. Combined with the distributed optimization algorithm of the alternating direction multiplier method (ADMM), the control parameters in the sorting process are optimized, thereby improving the system response speed and reducing the coordination delay between devices.

[0033] 3. Realize full data traceability

[0034] The chain traceability management module uses sharded blockchain technology and directed acyclic graph (DAG) to record the meter transfer process, and combines the graph database Neo4j to build a mapping relationship between the meter and the device status, ensuring the full traceability of the returned meter processing process. This innovation improves data transparency and provides data support for subsequent quality control and management.

[0035] 4. Improve equipment operation monitoring and fault diagnosis

[0036] The status diagnosis module can identify potential equipment failures in advance by collecting equipment operating parameters in real time and using a bidirectional gated recurrent network (BiGRU) to predict equipment status. It also extracts fault feature frequency bands using wavelet packet decomposition technology, optimizes equipment maintenance decisions, and helps reduce equipment failure rates and maintenance costs.

[0037] 5. Strengthen environmental monitoring and safety management

[0038] The environmental control module collects working environment data through a variety of sensors, and combines the spatiotemporal convolutional network to identify working behaviors, and monitors the temperature and humidity of the working environment, personnel location and other factors in real time. This technology improves environmental safety, ensures the safety protection of the working site, and reduces potential safety risks caused by environmental changes or improper personnel behavior. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 A schematic diagram of the system architecture of the present invention is shown;

[0040] Figure 2 A system workflow diagram of the present invention is shown. DETAILED DESCRIPTION

[0041] Exemplary embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0042] Combination Figure 1 The present invention provides a monitoring system for sorting production lines of returned smart meters, and the system composition is as follows:

[0043] 1. Feature fusion sorting module

[0044] The feature fusion sorting module is the input of the system. It uses a multimodal sensor array to collect feature information of returned smart meters, extracts and fuses features through a deep learning model, and finally achieves accurate classification of meters. The implementation process of this module includes the following steps:

[0045] A1). Multimodal data acquisition

[0046] This module collects multimodal data simultaneously through four sensor arrays. The parameters of the specific sensors and their collected data are set as follows:

[0047] a). Image data acquisition:

[0048] An array of four high-resolution industrial cameras is used, with a resolution of no less than 2048×1536 pixels, arranged in a regular tetrahedron to obtain a full-range image of the electric meter.

[0049] The image collected by each camera at time t is:

[0050]

[0051] in, represents the image captured by the i-th camera at time t, represents the set of real numbers, Indicates the image height, Indicates the image width, and 3 indicates the three RGB channels. Image data is collected through hardware synchronization triggering to ensure the timing consistency of image collection.

[0052] b). Acoustic signal collection:

[0053] A circular array consisting of 8 omnidirectional microphones is used with a sampling frequency of 48kHz. The collected acoustic signal data is expressed as:

[0054]

[0055] in, represents the acoustic signal collected by the jth microphone at time t, Indicates the number of sampling points, with a typical value of 2048. The microphone array and image sensor work synchronously to ensure that the timing of the acoustic signal matches the image data.

[0056] c). Point cloud data collection:

[0057] The three-dimensional point cloud data of the electric meter is collected by combining a structured light scanner with a binocular camera, which is expressed as:

[0058]

[0059] in, represents the three-dimensional point cloud data of the electric meter collected at time t, is the number of point clouds, each point contains three-dimensional coordinates in space , k is an index variable, taking the value [1, N]. During acquisition, laser scanning and image synchronization are used to ensure the integrity and accuracy of the 3D point cloud data.

[0060] d). Temperature field data collection:

[0061] The temperature field data is obtained by using an infrared thermal imager with a resolution of , expressed as:

[0062]

[0063] in, represents the temperature field data collected at time t, Indicates the resolution of the temperature field. The acquisition process of this sensor is synchronized with other sensors to ensure that the ambient temperature information is obtained at the same time.

[0064] A2). Feature extraction and processing

[0065] Different methods are used to extract and process the collected multimodal data.

[0066] a). Image data processing:

[0067] For image data, Swin Transformer is used for processing. The input image patch sequence is used Represented, the attention calculation formula is:

[0068]

[0069] in, , , are query, key, and value matrices respectively, is the scaling factor, is the relative position bias matrix. The specific parameter settings of Swin Transformer are: patch size (16×16), number of Transformer layers (4 layers), and number of hidden units in each layer (256) to ensure the effective extraction of image features.

[0070] b).Acoustic signal processing:

[0071] For acoustic signals, feature extraction is performed through Mel spectrum analysis. The Mel spectrum calculation formula is:

[0072]

[0073] in, is the linear frequency, The parameter settings of Mel spectrum analysis include the number of Mel filters (40) and the frequency band range (0 to 24kHz) to adapt to the operating environment of the returned electricity meter.

[0074] c). Point cloud data processing:

[0075] For 3D point cloud data, the PointNet++ network is used for processing, using a hierarchical sampling and grouping strategy. Layer features are used It means that the calculation formula is:

[0076]

[0077] in, For the Layer features, is a learnable feature extraction function, where a multi-layer perceptron (MLP) is used. Indicate point The neighborhood set of and x represent the coordinates of the i-th point and the current point respectively. The parameter settings of PointNet++ include the neighborhood size (16 points) and the number of layers (3 layers) to ensure effective feature extraction of point cloud data.

[0078] A3). Feature fusion decision

[0079] In order to fuse the feature data of different modalities, a weighted attention mechanism is adopted. The characteristics of the mode are , then the attention weight calculation formula is:

[0080]

[0081] in, For the The attention weight of the modality, , is the learnable weight matrix, , Respectively mode, the eigenvector of the kth mode, and exp represents the natural exponential function. Through this mechanism, the model is able to automatically adjust the weight of each mode to ensure that the most informative mode dominates the final decision.

[0082] The fused feature vector is:

[0083]

[0084] in, is the fused feature vector.

[0085] A4). Classification decision

[0086] Finally, the Softmax function is used for classification decision. The fused features are set as , the number of categories is , then the classification probability is:

[0087]

[0088] in, For the The probability of the class, , is the weight matrix.

[0089] 2. Digital mapping control module

[0090] The digital mapping control module is the core control unit of the system. It is responsible for receiving the classification results of the feature fusion sorting module and interacting with each execution device in real time to ensure the efficient and coordinated operation of the sorting line. This module builds a virtual mapping through digital twin technology to achieve real-time monitoring and optimized control of physical equipment. The specific implementation process is as follows:

[0091] B1). Equipment status modeling

[0092] The time Petri net is used to construct the equipment state transition model. It is defined as:

[0093]

[0094] in:

[0095] is a set of positions, indicating the possible states of the device (such as standby, running, fault, etc.);

[0096] is a transition set, representing the events that trigger state changes;

[0097] It is a flow relationship, describing the state transition path;

[0098] is a weight function, which indicates the priority of the conversion condition;

[0099] is the initial identifier, indicating the initial state of the system;

[0100] For the time interval function, define the time constraints of state transition.

[0101] For the first The state transition equation of a device is:

[0102]

[0103] in: express Time equipment The state vector of Indicates control actions; Indicates the influence of environmental factors; is the state transition function.

[0104] B2). Cross-device state consensus

[0105] The state consistency between devices is achieved based on the PBFT (Practical Byzantine Fault Tolerance) mechanism, ensuring that the system can still operate normally when some devices fail. The consensus process includes:

[0106] a). Request phase:

[0107]

[0108] in: For equipment control instructions; is the timestamp; Identify the device; For digital signature.

[0109] b). Response in the pre-preparation phase:

[0110]

[0111] in: Number the views; is the serial number; To request a summary; Sign for the master node.

[0112] B3). Distributed Optimization Control

[0113] The ADMM (Alternating Direction Method of Multipliers) algorithm is used to implement distributed optimization control, which decomposes the global control problem into multiple local sub-problems:

[0114] a). Objective function:

[0115]

[0116]

[0117] in:

[0118] For the The local objective function of a device represents the device Optimization objectives under certain constraints;

[0119] is the global coordination goal, which means that the global control system ensures overall coordination while satisfying each local goal;

[0120] For the A local variable of a device, indicating the device state or behavior variables;

[0121] is a global variable, which represents the unified control variable of all devices in the distributed system;

[0122] Indicates finding the minimum value of an expression.

[0123] b). Iterative solution process:

[0124] renew:

[0125]

[0126] renew:

[0127]

[0128] renew:

[0129]

[0130] in:

[0131] represents the value of the i-th variable in the k+1-th iteration;

[0132] , Express 、zFind the solution that minimizes the subsequent expression , the value of z;

[0133] , represents the k value in the kth and k+1th iterations;

[0134] represents the dual variable of the i-th device in the k-th and k+1-th iterations;

[0135] is the step size parameter, which controls the penalty factor in the update process;

[0136] is the number of iterations, indicating the progress of optimization;

[0137] For the A dual variable for each device that represents the difference between a local variable and a global variable.

[0138] c). Control parameter output:

[0139]

[0140] in:

[0141] For equipment The position parameter indicates the spatial position of the device;

[0142] is the movement speed parameter, indicating the movement speed of the device;

[0143] is the force parameter, which indicates the force or torque applied by the device;

[0144] It is the working mode parameter, indicating the operating mode of the device.

[0145] B4). Real-time state rendering

[0146] WebAssembly technology is used to achieve real-time rendering of the device's operating status, with a data update frequency of no less than 60Hz. Rendering frame rate adaptive control equation:

[0147]

[0148] in:

[0149] is the target frame rate, which indicates the frame rate expected by the rendering system;

[0150] For the maximum frame rate limit, set it to 60fps;

[0151] For the minimum frame rate requirement, set it to 24fps;

[0152] is the current CPU processing capacity, indicating the adequacy of computing resources;

[0153] It is a rendering load indicator that reflects the complexity of calculations during the rendering process.

[0154] This module generates the optimal control strategy in real time by coordinating with the input data of the feature fusion sorting module, and passes the control parameters to the execution device. At the same time, the module provides the equipment operation status information to the chain traceability management module to support full-process data tracking.

[0155] 3. Chain tracking management module

[0156] The chain tracking management module uses multi-source data collection, distributed storage and graph database technology to achieve full-process data tracking of returned smart meters in the sorting line. This module ensures that the processing of each returned meter is traceable, queryable and evaluable. The specific implementation process is as follows:

[0157] C1. Electricity meter identification information collection

[0158] At the entrance of the sorting line, an industrial-grade QR code scanner (resolution ≥ 3072 × 2048 pixels) is used to collect meter identification information. The identification information structure of each returned smart meter is:

[0159]

[0160] in:

[0161] MID is a 16-digit meter serial number in the segmented coding format "AABB-CCDD-EEFF-GGHH": AA: manufacturer code (such as 01 for Zhejiang manufacturers, 02 for Jiangsu manufacturers), BB: production year (such as 20 for 2020), CC-DD: model code (such as 1001 for a level 1 single-phase meter, 2001 for a level 1 three-phase meter), EE-FF: batch number (such as 01-01 for the first batch of the year, the first group), GG-HH: serial number (such as 00-01 for the first in the batch);

[0162] Type is a 4-digit meter type code: 1001: single-phase local fee control smart meter, 1002: single-phase remote fee control smart meter, 2001: three-phase local fee control smart meter, 2002: three-phase remote fee control smart meter;

[0163] Batch is an 8-digit batch code, and the production date is recorded in the "YYYYMMDD" format;

[0164] It is the millisecond timestamp of entering the system;

[0165] Status is the current status of the meter: 0: waiting for detection, 1: detecting, 2: detection completed, 3: classified, 4: processing completed.

[0166] c2). Sharded blockchain data storage

[0167] The distributed storage system is constructed using sharding blockchain technology to achieve data immutability and traceability. Block structure definition:

[0168]

[0169] The block header contains key information:

[0170]

[0171] in:

[0172] : 32-byte preceding block hash value generated using the SHA256 algorithm;

[0173] : UTC timestamp of block generation (accurate to milliseconds);

[0174] : The Merkle tree root hash value constructed by all transaction data in the block;

[0175] : 4-byte slice identifier, used to identify the business slice to which the data belongs;

[0176] Difficulty: The proof-of-work difficulty value is dynamically adjusted based on the system's processing power.

[0177] The block data body contains:

[0178]

[0179] in:

[0180] Records is a collection of process processing records;

[0181] States is the state record of related devices;

[0182] Params is the environment parameter record.

[0183] c3). Electricity meter flow track record

[0184] Use a directed acyclic graph (DAG) to record the flow of electricity meters between various processes in the sorting line:

[0185]

[0186] in:

[0187] V is a set of process nodes, including 5 key processes: S001 (appearance inspection): inspect the integrity of the meter shell, the status of the display screen, and the status of the terminal block; S002 (voltage withstand test): perform a 4kV voltage withstand test to ensure insulation performance; S003 (function test): verify the metering function, communication function, and fee control function; S004 (accuracy verification): perform error testing at the three current points of 0.05Ib, Ib, and Imax; S005 (classification and marking): classify and laser mark according to the test results.

[0188] The flow edge set E records the transfer relationship between processes:

[0189]

[0190] in, Represents the transition relationship from process i to process j.

[0191] The weight matrix W records the time information of the flow between processes:

[0192]

[0193] in Indicates the standard flow time from process i to process j (unit: seconds).

[0194] c4). Time series data index

[0195] Building a multi-layer time series index structure based on the LSM (Log-Structured Merge-tree) tree:

[0196]

[0197] Configuration parameters of each layer:

[0198] L0 layer: 128MB capacity, resident in memory, used as a real-time write buffer;

[0199] L1 layer: 1 GB capacity, composed of 10 SSTable files, compression ratio 1:10;

[0200] L2 layer: 10 GB capacity, consisting of 100 SSTable files, with a compression ratio of 1:10;

[0201] L3 layer: 100 GB capacity, consisting of 1,000 SSTable files, with a compression ratio of 1:10.

[0202] c5). Electricity meter processing data association

[0203] The mapping relationship between meter processing and equipment operation status is established through the Neo4j graph database:

[0204]

[0205] in: v is a set of electric meter nodes, including the identification information of electric meters; It is a collection of device nodes, including device operation parameters; To process the relationship edge set, the processing time and parameters are recorded; φ is the attribute mapping function, which defines the attribute set of nodes and edges.

[0206] The data of each processing link is ensured to be tamper-proof through blockchain technology, making the entire process traceable; at the same time, based on the correlation analysis capabilities of the graph database, abnormal processes and problematic equipment can be quickly located, thereby improving the operating efficiency and quality control level of the sorting line.

[0207] 4. Status diagnosis module

[0208] The status diagnosis module collects and analyzes equipment operating parameters in real time to monitor the status of key equipment in the sorting line and provide fault warnings. The module integrates multiple types of sensor data, builds a prediction model, and provides data support for equipment maintenance decisions.

[0209] D1). Equipment parameter collection

[0210] The data collected by the speed sensor is defined as a time series:

[0211]

[0212] in:

[0213] It represents the angular velocity value collected by the i-th speed sensor at time t, in rad / s;

[0214] M is the total number of speed sensors;

[0215] t is the sampling time point, and the sampling frequency is 1kHz;

[0216] Three-axis acceleration data collected by the vibration sensor:

[0217]

[0218]

[0219] in:

[0220] represents the three-axis acceleration vector collected by the j-th vibration sensor at time t;

[0221] N is the total number of vibration sensors;

[0222] , , Respectively represent the acceleration values ​​of the X, Y, and Z axes, in m / s²

[0223] The sampling frequency is 10kHz.

[0224] D2). Time series prediction model

[0225] The Bidirectional Gated Recurrent Unit (BiGRU) network is used to build a parameter prediction model, and its forward propagation process is:

[0226] Reset gate calculation:

[0227]

[0228] Update gate calculation:

[0229]

[0230] Candidate hidden state calculation:

[0231]

[0232] Hidden state update:

[0233]

[0234] in:

[0235] is the input feature vector at time t, including speed and vibration data;

[0236] , is the hidden state vector at time t and time t-1;

[0237] , , is the weight matrix;

[0238] , , is the bias vector;

[0239] σ(·) is the sigmoid activation function;

[0240] ⊙ represents the Hadamard product (element-wise multiplication).

[0241] D3). Domain adversarial adaptation

[0242] In order to eliminate the differences in parameter distribution between different devices, a domain adversarial training mechanism is introduced. Define feature extractor and domain classifier , the objective function for:

[0243]

[0244] in: is the source domain data (distribution of known device parameters)

[0245] is the target domain data (device parameter distribution to be adapted);

[0246] , They are the expectations on the source domain data and the target domain data respectively;

[0247] represents the feature extraction map;

[0248] Represents the domain discriminator output.

[0249] D4). Fault feature extraction

[0250] The vibration signal is decomposed by wavelet packet, the decomposition level is 3, and the db4 wavelet basis is used. Energy feature calculation:

[0251]

[0252] in: is the energy value of the i-th frequency band; is the wavelet packet coefficient of the i-th frequency band; is the number of coefficients in the ith frequency band.

[0253] Define the fault feature vector:

[0254]

[0255] in:

[0256] Indicates the energy values ​​of different frequency bands;

[0257] is the total energy; n is the number of frequency bands (n=8 in this embodiment)

[0258] D5). Maintenance decision making

[0259] The Proximal Policy Optimization (PPO) algorithm is used to generate equipment maintenance decisions. Value function definition:

[0260]

[0261] in:

[0262] is the value function, which means from the current state Expected return at the start;

[0263] is the discount factor, which takes a value of 0.95 and determines the impact of future returns on current decisions;

[0264] represents the strategy ratio at time step t+k.

[0265] Strategy function:

[0266]

[0267] in:

[0268] Indicates the current policy (given by the parameter Control) in state Take action The probability of

[0269] is the device state vector at time t;

[0270] For maintenance action, γ is the discount factor (value is 0.95);

[0271] θ is the policy network parameter;

[0272] Output of the policy network.

[0273] Policy update goals:

[0274]

[0275] in:

[0276] is the clipping objective function of PPO, which represents the objective function optimized during policy update. By maximizing this function, the policy parameter θ is adjusted to make the policy update more stable. is the strategy ratio; t is the estimated value of the advantage function, , where is the state-action value function, which means that in the state Next select action The cumulative return after is the state value function, which means that in the state The expected return of executing the current strategy. Represents the clipping operation in PPO, which is used to limit the change range of the strategy ratio and prevent the strategy from being updated too much; ε is the clipping parameter (value 0.2); Indicates expected value.

[0277] D6). Fault warning information

[0278] The generated fault warning information includes the unique device identifier, warning timestamp, fault type code, fault severity (level 1-5), and recommended maintenance measures.

[0279] Adaptive update of warning thresholds:

[0280]

[0281] in: Represent the new and old thresholds respectively; α is the smoothing factor (value is 0.8); is the average of the most recent N measurements; N is the sliding window size (value is 100).

[0282] 5.Environmental control module

[0283] The environmental control module is responsible for all-round monitoring and management of the working environment of the sorting line to ensure the stability of the equipment operating environment and the safety of operators. The module collects environmental parameters in real time through a multi-level sensor network and realizes intelligent control in combination with a deep learning algorithm.

[0284] E1). Environmental parameter collection network

[0285] The temperature and humidity sensor network adopts a hierarchical mesh topology structure, and the collected data is defined as:

[0286]

[0287]

[0288] in:

[0289] is the total number of sensor nodes (in this embodiment , ensuring that there is at least one monitoring point in a 3m×3m area);

[0290] For the The temperature value measured by each node has a measurement range of [−10°C, 50°C] and an accuracy of ±0.3°C;

[0291] is the relative humidity value, the measurement range is [0%, 100%], and the accuracy is ±2% RH;

[0292] It is the atmospheric pressure value, the measuring range is [85 kPa, 106 kPa], and the accuracy is ±0.1 kPa;

[0293] is the sampling time point, and the sampling frequency is 0.2 Hz.

[0294] The sensor spatial distribution weight matrix is ​​defined as:

[0295]

[0296]

[0297] in:

[0298] For Node and The Euclidean distance between

[0299] is the spatial correlation attenuation coefficient, which takes a value of 1.5;

[0300] Represents the spatial weight coefficient between nodes.

[0301] E2). Personnel positioning and trajectory tracking

[0302] The parameters of the lidar scanner are set as follows: horizontal scanning angle: 360°; vertical scanning angle: 30°; angular resolution: 0.1°; ranging accuracy: ±30mm; scanning frequency: 20Hz.

[0303] Point cloud data preprocessing:

[0304]

[0305] in:

[0306] is a single data point in the point cloud, including coordinate;

[0307] is the centroid of the local area point cloud;

[0308] is the outlier filtering threshold, which is set to 0.15 m;

[0309] is the total amount of point cloud data.

[0310] E3). Operational behavior identification

[0311] The spatiotemporal convolutional network (ST-GCN) is used to process personnel motion data, and the graph convolution operation is defined as:

[0312]

[0313] in:

[0314] For Node The neighborhood set of ; is the normalization coefficient; is the input feature; is a learnable weight matrix; is the spatial relationship label.

[0315] The time dimension is modeled as:

[0316]

[0317] in:

[0318] For the The feature matrix of the layer; is the adjacency matrix; is the degree matrix; For the The parameter matrix of the layer; is the ReLU activation function.

[0319] E4). Motion trajectory prediction

[0320] The improved Kalman filter algorithm is used for trajectory prediction. The state equation is:

[0321]

[0322] The observation equation is:

[0323]

[0324] in:

[0325] for The state vector at time ;

[0326] is the state transfer matrix;

[0327] is the control input matrix;

[0328] is the control vector;

[0329] is process noise, which conforms to normal distribution ;

[0330] is the observation vector;

[0331] is the observation matrix;

[0332] is the observation noise, which conforms to the normal distribution .

[0333] The forecast covariance is updated as:

[0334]

[0335] The Kalman gain is calculated as:

[0336]

[0337] E5). Environmental parameter flow field analysis

[0338] The computational fluid dynamics model based on hierarchical decomposition mainly includes the following equations:

[0339] a). Continuity equation:

[0340]

[0341] b). Momentum equation:

[0342]

[0343] c). Energy equation:

[0344]

[0345] in:

[0346] is the density of air, 1.225 kg / m³;

[0347] is the velocity vector field;

[0348] is the pressure field;

[0349] is the dynamic viscosity, 1.789×1 Pa·s;

[0350] is the gravitational acceleration vector;

[0351] is the specific heat capacity at constant pressure, 1005 J / (kg·K);

[0352] is the thermal conductivity, 0.0257 W / (m·K);

[0353] is the viscous dissipation function.

[0354] E6). Environmental abnormality warning

[0355] The anomaly detection index based on multi-parameter fusion is defined as:

[0356]

[0357] in:

[0358] For the The measured values ​​of environmental parameters; is the historical mean of the parameter; is the standard deviation of the parameter; is the weight coefficient; is the total number of environmental parameters.

[0359] The warning levels are as follows:

[0360] Level 1 warning:

[0361] Level 2 warning:

[0362] Level 3 warning:

[0363] Level 4 warning: .

[0364] This module transmits environmental monitoring information to the digital mapping control module in real time, supports environmental adaptive control of the sorting line, and provides environmental parameter support for the status diagnosis module, thereby realizing the correlation analysis between equipment status and environmental conditions.

[0365] Combination Figure 2 , the workflow of the present invention is as follows:

[0366] S1. Electricity meter feature collection and classification

[0367] The system first collects the feature information of the returned smart meter through a multimodal sensor array. Four industrial cameras with a resolution of 2048×1536 pixels are arranged in a regular tetrahedron to achieve all-round image acquisition of the meter appearance; a ring-arranged array of eight microphones with a sampling frequency of 48kHz collects 2048 acoustic signals; a structured light scanner generates three-dimensional point cloud data, each point contains accurate spatial coordinate information; an infrared thermal imager collects temperature field data with a resolution of M×M; the collected multi-source data are input into a special processing model respectively, the image data is processed by the Swin Transformer network, and the hierarchical features are extracted using the attention mechanism; the acoustic signal is converted to the Mel frequency domain through Mel spectrum analysis; the point cloud data is processed by the PointNet++ network, and the geometric features are extracted using a hierarchical sampling strategy; the temperature field is analyzed by the regional growing algorithm; finally, the weighted attention mechanism is used to fuse all features into a unified feature vector, and the classification decision is output through the softmax function.

[0368] S2. Equipment control and coordinated scheduling

[0369] The digital mapping control module receives the classification results and builds a device state migration model based on the time Petri net; the model includes a position set P (indicating device state), a transition set T (indicating trigger events), a flow relationship F, and a time interval function I, to achieve accurate modeling of device state; the PBFT mechanism is used to achieve state consensus between devices, and state confirmation is completed through four stages: request, pre-preparation, preparation, and submission. The ADMM algorithm implements distributed optimization control, decomposes the global control problem into multiple local sub-problems, and obtains the optimal control parameters through iterative solution; WebAssembly technology is used to achieve a state rendering frequency of 60Hz to ensure the real-time performance of the monitoring interface.

[0370] S3. Full process data tracking

[0371] The 16-digit serial number of the meter is collected through an industrial-grade scanner (resolution 3072×2048 pixels), including manufacturer code, production year, model code, batch number and serial number information; sharding blockchain technology is used to store and process data, and the block contains a 32-byte prefix hash value, timestamp, Merkle tree root value and shard identifier; a directed acyclic graph is constructed to record the transfer relationship of five processes (appearance inspection, voltage resistance test, function inspection, accuracy verification, classification marking); a four-layer index structure is established based on the LSM tree to realize data organization from L0 (128MB) to L3 (100GB); a mapping relationship between meter processing and equipment status is established through the Neo4j graph database to support full process traceability.

[0372] S4. Equipment status monitoring

[0373] Speed ​​sensors (1kHz sampling, ±1rpm accuracy) and triaxial vibration sensors (10kHz sampling, ±50g range) are installed on the core equipment of the assembly line to collect speed and vibration data respectively. The BiGRU neural network builds a 24-hour prediction model, using 128-dimensional hidden layer states for bidirectional feature extraction; three-layer wavelet packet decomposition is performed through the db4 wavelet basis function to extract eight frequency band energy features; the domain adversarial network eliminates individual differences in equipment and achieves unified modeling; the proximal strategy optimization algorithm generates the optimal maintenance strategy based on equipment health assessment and maintenance cost within a 4-hour decision cycle.

[0374] S5. Environmental monitoring and safety management

[0375] 24 sensor nodes are deployed in a standard 3m×3m monitoring unit to build an environmental monitoring network. Temperature sensors (-10°C to 50°C, ±0.3°C), humidity sensors (0-100%RH, ±2%RH), and pressure sensors (85-106kPa, ±0.1kPa) are synchronously sampled at a frequency of 0.2Hz; LiDAR (360° horizontal × 30° vertical, 0.1° resolution, 20Hz scanning) is used to track the operation area in real time, and a 12-layer spatiotemporal convolutional network is used to analyze the operation behavior, identifying 8 types of standard actions and 5 types of abnormal behaviors; a computational fluid dynamics model (0.1m spatial accuracy, 1s time step) is used to calculate the distribution of environmental parameter fields to support precise control of environmental equipment.

[0376] In summary, the returned smart meter sorting assembly line monitoring system proposed in the present invention realizes comprehensive monitoring and intelligent management of the returned smart meter processing process through the organic coordination of five functional modules: feature fusion sorting, digital mapping control, chain tracking management, status diagnosis and environmental control. The system adopts advanced technologies such as multimodal sensing, deep learning, distributed storage, and status prediction to solve technical problems such as low efficiency of traditional manual sorting, complex assembly line control, difficult to trace the processing process, lagging equipment status monitoring, and lack of environmental safety monitoring. Practical applications have shown that the system significantly improves the sorting efficiency and accuracy, realizes predictive maintenance of equipment status, ensures the safety and stability of the working environment, and provides reliable technical support for the standardized processing of returned smart meters. The present invention has good practicality and promotion value, and is of great significance to improving the management level of power equipment.

[0377] Although this document shows the specific application of the present invention through specific embodiments, these embodiments are only for illustrative purposes and are not intended to limit the scope of protection of the present invention. The scope of protection of the present invention is defined by the claims, and any appropriate modification, equivalent substitution or improvement based on the principle of the present invention should be deemed to fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be given the broadest interpretation to cover all such modifications, equivalent structures and functions.

Claims

1. A monitoring system for a returned smart meter sorting line, characterized in that: include: A feature fusion sorting module, which includes an industrial camera array, a microphone array, a structured light scanner and an infrared thermal imager, and is used to collect multimodal feature information of the electric meter; The industrial camera array processes image data through a hierarchical attention network, the microphone array processes acoustic signals through Mel spectrum analysis, the structured light scanner processes three-dimensional point clouds through a point cloud neural network, and the infrared thermal imager processes temperature fields through a region growing algorithm, and the above features are fused through a weighted attention mechanism to form a classification result; A digital mapping control module is used to receive the classification results and device status data, construct a device status migration model through a time Petri net, achieve cross-device status consensus based on a Byzantine fault tolerance mechanism, and calculate sorting process control parameters in combination with a distributed optimization control algorithm based on an alternating direction multiplier method; The chain traceability management module collects meter serial numbers through barcode scanners, uses sharded blockchain technology to establish a data storage structure, uses directed acyclic graphs to record meter transfer processes, establishes a time series index based on a logarithmic structure merge tree, and constructs a mapping relationship between meter processing and device status through a graph database; The status diagnosis module collects equipment operating parameters through speed sensors and vibration sensors, builds a parameter time series prediction model using a bidirectional gated recurrent network, uses a domain adversarial network to eliminate parameter distribution differences between devices, extracts fault feature frequency bands through wavelet packet decomposition, and generates equipment maintenance decisions in combination with a proximal strategy optimization algorithm; The environmental control module collects sorting environment parameters through a temperature and humidity sensor network, obtains personnel location data using lidar scanning, uses a spatiotemporal convolutional network to identify work behaviors, uses the Kalman filter algorithm to predict motion trajectories, and monitors sorting environment parameters based on a hierarchical decomposition computational fluid dynamics model.

2. The monitoring system according to claim 1, characterized in that: The weighted attention mechanism in the feature fusion sorting module includes a relative position bias matrix and a learnable weight matrix, and the learnable weight matrix is ​​used to calculate the weights of each modal feature.

3. The monitoring system according to claim 1, characterized in that: The Byzantine fault tolerance mechanism in the digital mapping control module includes four processing stages: request, pre-prepare, prepare and submit.

4. The monitoring system according to claim 1, characterized in that: The directed acyclic graph in the chain traceability management module includes five process nodes: appearance inspection, voltage resistance test, function inspection, accuracy verification and classification marking.

5. The monitoring system according to claim 1, characterized in that: The wavelet packet decomposition in the state diagnosis module adopts db4 wavelet basis to perform three-layer decomposition to extract fault characteristics of eight frequency bands.

Images