A multi-dimensional intelligent monitoring system and method for a relay flexible production line

Abstract

The application discloses a relay flexible production line multidimensional intelligent monitoring system and method, relates to the relay production detection technical field, and includes: a multi-source data acquisition module, which is used for synchronously collecting appearance image data, electrical parameter data, mechanical characteristic data and production line environment data in a relay production process; a multi-task detection model module, which is in communication connection with the multi-source data acquisition module, adopts a shared backbone network and an independent output head architecture, is based on the data collected by the multi-source data acquisition module, and concurrently executes appearance defect detection, electrical parameter regression prediction and mechanical characteristic evaluation prediction tasks. The relay flexible production line multidimensional intelligent monitoring system and method, through a hybrid mechanism of "zero sample parameter mapping and few sample fine tuning", combines a multi-task shared model architecture, greatly shortens the changeover time from more than 30 minutes of the tradition to within 5 minutes, and greatly improves the response speed of the flexible production line.

Description

Technical Field

[0001] This invention relates to the field of relay production and testing technology, specifically to a multi-dimensional intelligent monitoring system and method for flexible relay production lines. Background Technology

[0002] As a core control component in electronic devices, the production quality of relays directly determines the reliability of end-user equipment. With the diversification of market demands, relay production is gradually moving towards multi-variety, small-batch, and customized production. However, traditional relay production line testing systems suffer from the following technical challenges:

[0003] Low changeover efficiency: For different relay models, manual adjustment of testing parameters and replacement of tooling fixtures are required, resulting in long changeover times and failing to meet the rapid changeover requirements of flexible production.

[0004] Single testing dimension: Existing testing methods often focus on a single dimension (such as only testing appearance or only testing electrical parameters), which cannot comprehensively cover multiple quality indicators such as appearance, electrical, and mechanical aspects, and are prone to missed detections and false detections.

[0005] Poor environmental adaptability: The impact of fluctuations in the production environment (such as temperature and humidity) on the test results was not fully considered, resulting in unstable test accuracy.

[0006] Low level of intelligence: It relies on manual judgment, which is costly and inconsistent in standards; the detection data is not used in a closed loop, and the model cannot self-optimize. Summary of the Invention

[0007] The purpose of this invention is to provide a multi-dimensional intelligent monitoring system and method for relay flexible production lines to solve the problems mentioned in the background art.

[0008] To achieve the above objectives, the present invention provides the following technical solution: a multi-dimensional intelligent monitoring system and method for a relay flexible production line, comprising: The multi-source data acquisition module is used to simultaneously acquire appearance image data, electrical parameter data, mechanical characteristic data, and production line environment data during the relay production process; The multi-task detection model module communicates with the multi-source data acquisition module and adopts an architecture with a shared backbone network and independent output heads. Based on the data acquired by the multi-source data acquisition module, it concurrently executes tasks such as appearance defect detection, electrical parameter regression prediction, and mechanical characteristic evaluation prediction. The model rapid adaptation module is coupled with the multi-task detection model module and has a built-in model library. It is used to adjust the model parameters of the multi-task detection model module based on the mapping relationship between model parameters and model weights, so as to realize rapid model change adaptation for different models of relays. The defect classification and judgment module is connected to the multi-source data acquisition module and the multi-task detection model module respectively. It is used to receive environmental data and multi-task inference results, calculate a comprehensive score using a preset weight allocation strategy and environmental correction factor, and output quality level and handling instructions based on the comprehensive score. The data closure and incremental learning module is connected to the defect classification and judgment module and the multi-task detection model module. It is used to collect samples containing classification results and review data, and to screen difficult samples to trigger incremental training of the multi-task detection model module in order to achieve continuous optimization of the model. The central control and scheduling module communicates with each module and is used to coordinate the start-up, operation, stop and data interaction of each module, and control the execution unit to sort the relays according to the processing instructions.

[0009] Preferably, the backbone network of the multi-task detection model module includes a C2f module and an SPPF module for extracting shared features of appearance, electrical, and mechanical aspects; the neck feature fusion network of the multi-task detection model module adopts a PAN-FPN structure for realizing multi-scale feature fusion; the output head of the multi-task detection model module includes an appearance defect detection head, an electrical parameter regression head, and a mechanical characteristic evaluation head, and the three output heads share the backbone network and the neck feature fusion network.

[0010] Preferably, the model quick adaptation module includes: Model library, used to store the standard parameters of various relay models; A zero-sample adaptation unit, connected to the model library, is used to calculate and adjust the weight parameters of the corresponding output head in the multi-task detection model module according to the standard parameters and through a preset mathematical mapping model, so as to achieve rapid adaptation without new model samples. A few-sample fine-tuning unit, connected to the multi-task detection model module, is used to load a small number of new model samples when the adaptation accuracy of the zero-sample adaptation unit does not reach a preset threshold, and to fine-tune and train at least some network layers of the output head and neck network of the multi-task detection model module.

[0011] Preferably, the defect classification and judgment module determines the weights of appearance, electrical, mechanical, and environmental dimensions based on the analytic hierarchy process, and the environmental correction factor is calculated based on the degree to which the ambient temperature and / or humidity deviate from a preset threshold.

[0012] Preferably, the data closure and incremental learning module includes: The sample collection unit is used to collect model inference data, grading results, manual review labels, and revision records; A difficult sample screening unit, connected to the sample collection unit, is used to execute at least one screening strategy to label difficult samples. The screening strategy includes: samples whose labels are inconsistent with the model classification results after manual review; samples whose defects are detected by the model but are classified as qualified; and samples whose defects are not detected by the model but are classified as requiring repair or scrap. An incremental training unit, connected to the difficult sample screening unit, is used to perform incremental training on the multi-task detection model module using the difficult samples when the number of difficult samples reaches a preset threshold. The model update unit is used to synchronize the incrementally trained model parameters to the multi-task detection model module and the model fast adaptation module.

[0013] Preferably, the multi-source data acquisition module includes: The appearance image acquisition unit includes multiple industrial cameras at different angles and a diffuse light source; The electrical parameter acquisition unit includes a resistance tester, a voltage tester, and an oscilloscope; The mechanical characteristic acquisition unit includes a displacement sensor, a pressure sensor, and a timing acquisition card; The environmental data acquisition unit includes temperature and humidity sensors and dust sensors.

[0014] A multi-dimensional intelligent monitoring method for a flexible relay production line, applied to the aforementioned multi-dimensional intelligent monitoring system for a flexible relay production line, the method comprising: In response to production instructions, the multi-source data acquisition module is controlled to synchronously acquire data from the relay under test in different dimensions and production line environment data; The collected data is input into the multi-task detection model module, features are extracted through a shared backbone network, and the appearance defect detection results, electrical parameter prediction values ​​and mechanical characteristic prediction values ​​are output concurrently through independent output heads; The defect classification and judgment module calculates a comprehensive score based on the environmental data, the appearance defect detection results, the predicted values ​​of electrical parameters and the predicted values ​​of mechanical characteristics, and generates a quality level and handling instructions based on the comprehensive score. The control execution unit sorts and processes the relays to be tested according to the processing instructions; The data closure and incremental learning module uses historical detection data, grading results, and verification data to screen difficult samples and incrementally optimize the multi-task detection model module.

[0015] Preferably, prior to the step of responding to the production instruction, a changeover adaptation step is also included: Receive new model switching command; Obtain the standard parameters of the new model from the model library; The zero-sample adaptation unit calculates and adjusts the weight parameters of the multi-task detection model module based on the standard parameters. Verify the detection accuracy of the adapted model; If the detection accuracy is lower than a preset threshold, the multi-task detection model module is fine-tuned using a small number of samples of the new model through the few-sample fine-tuning unit.

[0016] Preferably, the step of the defect classification and determination module in calculating the comprehensive score includes: The severity of appearance defects, the degree of electrical parameter exceedance, and the degree of abnormal mechanical characteristics are quantified to obtain the corresponding severity values; Calculate environmental correction factors based on environmental data; Calculate the overall score S: S = Severity of appearance defect × 0.35 + Degree of electrical parameter exceedance × 0.4 + Degree of abnormal mechanical characteristics × 0.25 + C_env × 0.05; S∈[0,0.5]→Excellent, S∈(0.5,1.0]→Good, S∈(1.0,2.0]→Qualified, S∈(2.0,3.0]→Rework, S∈(3.0,4.0]→Scrapped.

[0017] Preferably, the step of the data closure and incremental learning module in screening difficult samples includes at least one of the following strategies: Level 1 screening: Screening samples whose labels, as manually verified, do not match the model's grading results; Secondary screening: Screening samples that are detected as defective by the multi-task detection model module but are deemed qualified by the defect classification judgment module; Three-level screening: Screening samples for which the multi-task detection model module did not detect defects but the defect classification judgment module determined that they should be reworked or scrapped.

[0018] Compared with the prior art, the beneficial effects of the present invention are: This multi-dimensional intelligent monitoring system and method for relay flexible production lines, through a hybrid mechanism of "zero-sample parameter mapping and few-sample fine-tuning" combined with a multi-task sharing model architecture, significantly reduces changeover time from more than 30 minutes to less than 5 minutes, greatly improving the response speed of flexible production lines.

[0019] It achieves simultaneous intelligent detection of multiple dimensions of indicators, including appearance, electrical, and mechanical aspects. By introducing an environmental correction factor, it effectively suppresses fluctuations in detection accuracy under complex industrial environments and significantly improves the rates of missed detection and false detection.

[0020] From data collection, intelligent analysis, and automatic grading to sorting execution and model self-optimization, a complete closed loop is formed, which greatly reduces the reliance on manual labor and ensures the consistency of judgment standards.

[0021] Through an incremental learning loop based on difficult sample mining, the system can continuously optimize itself using production data, thereby continuously improving detection accuracy as the production process progresses, and possessing the long-term capability to adapt to process changes and new product development. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of the architecture of the multi-dimensional intelligent monitoring system for the relay flexible production line of the present invention; Figure 2 This is a schematic diagram of the improved YOLOv8 architecture of the slow multi-task detection model module of the present invention. Figure 3 This is a flowchart illustrating the multi-dimensional intelligent monitoring method for the relay flexible production line of the present invention. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] Please see Figures 1-3 This invention provides a technical solution: a multi-dimensional intelligent monitoring system and method for a relay flexible production line, comprising: The multi-source data acquisition module is used to simultaneously acquire appearance image data, electrical parameter data, mechanical characteristic data, and production line environment data during the relay production process; The multi-task detection model module communicates with the multi-source data acquisition module and adopts an architecture with a shared backbone network and independent output heads. Based on the data acquired by the multi-source data acquisition module, it concurrently executes tasks such as appearance defect detection, electrical parameter regression prediction, and mechanical characteristic evaluation prediction. The model rapid adaptation module is coupled with the multi-task detection model module and has a built-in model library. It is used to adjust the model parameters of the multi-task detection model module based on the mapping relationship between model parameters and model weights, so as to realize rapid model change adaptation for different models of relays. The defect classification and judgment module is connected to the multi-source data acquisition module and the multi-task detection model module respectively. It is used to receive environmental data and multi-task inference results, calculate a comprehensive score using a preset weight allocation strategy and environmental correction factor, and output quality level and handling instructions based on the comprehensive score. The data closure and incremental learning module is connected to the defect classification and judgment module and the multi-task detection model module. It is used to collect samples containing classification results and review data, and to screen difficult samples to trigger incremental training of the multi-task detection model module in order to achieve continuous optimization of the model. The central control and scheduling module communicates with each module and is used to coordinate the start-up, operation, stop and data interaction of each module, and control the execution unit to sort the relays according to the processing instructions.

[0025] The backbone network of the multi-task detection model module includes a C2f module and an SPPF module, which are used to extract shared features of appearance, electrical, and mechanical aspects. The neck feature fusion network of the multi-task detection model module adopts a PAN-FPN structure to achieve multi-scale feature fusion. The output head of the multi-task detection model module includes an appearance defect detection head, an electrical parameter regression head, and a mechanical characteristic evaluation head. The three output heads share the backbone network and the neck feature fusion network.

[0026] The model rapid adaptation module includes: Model library, used to store the standard parameters of various relay models; A zero-sample adaptation unit, connected to the model library, is used to calculate and adjust the weight parameters of the corresponding output head in the multi-task detection model module according to the standard parameters and through a preset mathematical mapping model, so as to achieve rapid adaptation without new model samples. A few-sample fine-tuning unit, connected to the multi-task detection model module, is used to load a small number of new model samples when the adaptation accuracy of the zero-sample adaptation unit does not reach a preset threshold, and to fine-tune and train at least some network layers of the output head and neck network of the multi-task detection model module.

[0027] The defect classification and judgment module determines the weights of appearance, electrical, mechanical and environmental dimensions based on the analytic hierarchy process. The environmental correction factor is calculated based on the degree to which the ambient temperature and / or humidity deviate from the preset threshold.

[0028] The data closure and incremental learning module includes: The sample collection unit is used to collect model inference data, grading results, manual review labels, and revision records; A difficult sample screening unit, connected to the sample collection unit, is used to execute at least one screening strategy to label difficult samples. The screening strategy includes: samples whose labels are inconsistent with the model classification results after manual review; samples whose defects are detected by the model but are classified as qualified; and samples whose defects are not detected by the model but are classified as requiring repair or scrap. An incremental training unit, connected to the difficult sample screening unit, is used to perform incremental training on the multi-task detection model module using the difficult samples when the number of difficult samples reaches a preset threshold. The model update unit is used to synchronize the incrementally trained model parameters to the multi-task detection model module and the model fast adaptation module.

[0029] 6. The multi-dimensional intelligent monitoring system for a flexible relay production line according to claim 5, characterized in that: the multi-source data acquisition module includes: The appearance image acquisition unit includes multiple industrial cameras at different angles and a diffuse light source; The electrical parameter acquisition unit includes a resistance tester, a voltage tester, and an oscilloscope; The mechanical characteristic acquisition unit includes a displacement sensor, a pressure sensor, and a timing acquisition card; The environmental data acquisition unit includes temperature and humidity sensors and dust sensors.

[0030] The entire system consists of six core modules: Data acquisition module: Simultaneously acquires data such as the relay's appearance image, electrical performance parameters, mechanical action parameters, and the temperature and humidity of the production environment to ensure information synchronization.

[0031] Intelligent Detection Model Module: This is the "brain" of the system, employing a shared deep learning network to simultaneously process all types of data. This network can output judgments on multiple aspects at once, such as whether the relay has visual defects, whether electrical parameters are qualified, and whether mechanical actions meet standards. This "one network handling all tasks" design is key to enabling rapid switching between product models.

[0032] Quick Changeover Module: This module comes into play when the production line needs to produce a different model of relay. It internally stores standard parameters for various models. During a changeover, the system first automatically adjusts the settings of the detection model based on the parameters of the new model ("zero-sample adaptation"). If the accuracy is insufficient after adjustment, the model is then quickly fine-tuned using a very small number of samples of the new model ("few-sample fine-tuning"). This method can reduce changeover preparation time from the traditional tens of minutes to less than five minutes.

[0033] Quality Grading Module: This module receives test results and environmental data, and performs a comprehensive quality score and grading for the relays. It comprehensively considers the impact of different aspects such as appearance, electrical, and mechanical properties on quality (assigning different weights), and fine-tunes the score based on current environmental conditions (such as excessively high temperatures) to make the grading results more reliable. Ultimately, the product is classified into levels such as "Excellent," "Good," "Qualified," "Reworkable," and "Scrapped," and corresponding processing instructions are provided.

[0034] Self-learning optimization module: The system possesses self-improvement capabilities. It automatically collects "difficult" cases during detection that are ambiguous in judgment, inconsistent with manual review, or ultimately require rework. When such cases accumulate to a certain number, the system will use them to automatically retrain the detection model, thereby making the model increasingly accurate and achieving continuous optimization.

[0035] The central control and scheduling module serves as the system's command center, responsible for coordinating the orderly operation of all modules, controlling the start and stop of the production line, and directing equipment such as robotic arms to automatically sort relays according to the classification results (e.g., sending them to the qualified area, rework area, or scrap area).

[0036] A multi-dimensional intelligent monitoring method for a flexible relay production line, applied to the aforementioned multi-dimensional intelligent monitoring system for a flexible relay production line, the method comprising: In response to production instructions, the multi-source data acquisition module is controlled to synchronously acquire data from the relay under test in different dimensions and production line environment data; The collected data is input into the multi-task detection model module, features are extracted through a shared backbone network, and the appearance defect detection results, electrical parameter prediction values ​​and mechanical characteristic prediction values ​​are output concurrently through independent output heads; The defect classification and judgment module calculates a comprehensive score based on the environmental data, the appearance defect detection results, the predicted values ​​of electrical parameters and the predicted values ​​of mechanical characteristics, and generates a quality level and handling instructions based on the comprehensive score. The control execution unit sorts and processes the relays to be tested according to the processing instructions; The data closure and incremental learning module uses historical detection data, grading results, and verification data to screen difficult samples and incrementally optimize the multi-task detection model module.

[0037] Prior to the step of responding to production instructions, a changeover adaptation step is also included: Receive new model switching command; Obtain the standard parameters of the new model from the model library; The zero-sample adaptation unit calculates and adjusts the weight parameters of the multi-task detection model module based on the standard parameters. Verify the detection accuracy of the adapted model; If the detection accuracy is lower than a preset threshold, the multi-task detection model module is fine-tuned using a small number of samples of the new model through the few-sample fine-tuning unit.

[0038] The steps for calculating the comprehensive score by the defect classification and determination module include: The severity of appearance defects, the degree of electrical parameter exceedance, and the degree of abnormal mechanical characteristics are quantified to obtain the corresponding severity values; Calculate environmental correction factors based on environmental data; Calculate the overall score S: S = Severity of appearance defect × 0.35 + Degree of electrical parameter exceedance × 0.4 + Degree of abnormal mechanical characteristics × 0.25 + C_env × 0.05; S∈[0,0.5]→Excellent, S∈(0.5,1.0]→Good, S∈(1.0,2.0]→Qualified, S∈(2.0,3.0]→Rework, S∈(3.0,4.0]→Scrapped.

[0039] The steps of the data loop closure and incremental learning module for screening difficult samples include at least one of the following strategies: Level 1 screening: Screening samples whose labels, as manually verified, do not match the model's grading results; Secondary screening: Screening samples that are detected as defective by the multi-task detection model module but are deemed qualified by the defect classification judgment module; Three-level screening: Screening samples for which the multi-task detection model module did not detect defects but the defect classification judgment module determined that they should be reworked or scrapped.

[0040] The main process of the monitoring method based on the above system is as follows: Preparation and model changeover: If a production model needs to be switched, the system calls the quick model changeover module, which uses model parameters and a small number of samples to complete the adaptation of the detection model within minutes.

[0041] Synchronous data acquisition: When the relay arrives at the detection station, the system synchronously triggers all sensors and cameras to acquire their appearance, electrical, mechanical and environmental data at once.

[0042] Joint intelligent inspection: The collected multi-dimensional data is input into the intelligent inspection model. After one calculation, the model provides comprehensive inspection results for appearance defects, electrical parameters, and mechanical characteristics in parallel.

[0043] Comprehensive grading decision: The grading module combines environmental data to perform weighted fusion calculations on the test results, obtains a comprehensive score, and maps it to a specific quality level (such as "rework").

[0044] Automatic sorting execution: Based on the determined level, the system automatically controls the sorting equipment to send the relay to the corresponding subsequent processing area.

[0045] Closed-loop learning optimization: The system runs continuously in the background, collects production data, and filters "difficult samples" for regular training and updating of the detection model, so that the system's detection capabilities continue to evolve.

[0046] In summary, this invention effectively solves the industry problems of slow changeover, incomplete testing, poor adaptability, and inability to self-improve in flexible relay production through multi-module collaboration and closed-loop design.

[0047] It should be noted that, in this document, the terms “comprising,” “including,” or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0048] This article uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only for the purpose of helping to understand the method and core ideas of the present invention. The above descriptions are only preferred embodiments of the present invention. It should be noted that due to the limitations of textual expression, while there are objectively infinite specific structures, those skilled in the art can make several improvements, modifications, or changes without departing from the principles of the present invention, and can also combine the above technical features in an appropriate manner. These improvements, modifications, changes, or combinations, or the direct application of the inventive concept and technical solution to other situations without modification, should all be considered within the scope of protection of the present invention.

Claims

1. A multi-dimensional intelligent monitoring system for a flexible relay production line, characterized in that: include: The multi-source data acquisition module is used to simultaneously acquire appearance image data, electrical parameter data, mechanical characteristic data, and production line environment data during the relay production process; The multi-task detection model module communicates with the multi-source data acquisition module and adopts an architecture with a shared backbone network and independent output heads. Based on the data acquired by the multi-source data acquisition module, it concurrently executes tasks such as appearance defect detection, electrical parameter regression prediction, and mechanical characteristic evaluation prediction. The model rapid adaptation module is coupled with the multi-task detection model module and has a built-in model library. It is used to adjust the model parameters of the multi-task detection model module based on the mapping relationship between model parameters and model weights, so as to realize rapid model change adaptation of different relay models. The defect classification and judgment module is connected to the multi-source data acquisition module and the multi-task detection model module respectively. It is used to receive environmental data and multi-task inference results, calculate a comprehensive score using a preset weight allocation strategy and environmental correction factor, and output quality level and handling instructions based on the comprehensive score. The data closure and incremental learning module is connected to the defect classification and judgment module and the multi-task detection model module. It is used to collect samples containing classification results and review data, and to screen difficult samples to trigger incremental training of the multi-task detection model module in order to achieve continuous optimization of the model. The central control and scheduling module communicates with each module and is used to coordinate the start-up, operation, stop and data interaction of each module, and control the execution unit to sort the relays according to the processing instructions.

2. The multi-dimensional intelligent monitoring system for a relay flexible production line according to claim 1, characterized in that: The backbone network of the multi-task detection model module includes a C2f module and an SPPF module, which are used to extract shared features of appearance, electrical, and mechanical aspects. The neck feature fusion network of the multi-task detection model module adopts a PAN-FPN structure to achieve multi-scale feature fusion. The output head of the multi-task detection model module includes an appearance defect detection head, an electrical parameter regression head, and a mechanical characteristic evaluation head. The three output heads share the backbone network and the neck feature fusion network.

3. The multi-dimensional intelligent monitoring system for a flexible relay production line according to claim 2, characterized in that: The model rapid adaptation module includes: Model library, used to store the standard parameters of various relay models; A zero-sample adaptation unit, connected to the model library, is used to calculate and adjust the weight parameters of the corresponding output head in the multi-task detection model module according to the standard parameters and through a preset mathematical mapping model, so as to achieve rapid adaptation without new model samples. A few-sample fine-tuning unit, connected to the multi-task detection model module, is used to load a small number of new model samples when the adaptation accuracy of the zero-sample adaptation unit does not reach a preset threshold, and to fine-tune and train at least some network layers of the output head and neck network of the multi-task detection model module.

4. The multi-dimensional intelligent monitoring system for a relay flexible production line according to claim 3, characterized in that: The defect classification and judgment module determines the weights of appearance, electrical, mechanical and environmental dimensions based on the analytic hierarchy process. The environmental correction factor is calculated based on the degree to which the ambient temperature and / or humidity deviate from the preset threshold.

5. The multi-dimensional intelligent monitoring system for a flexible relay production line according to claim 4, characterized in that: The data closure and incremental learning module includes: The sample collection unit is used to collect model inference data, grading results, manual review labels, and revision records; A difficult sample screening unit, connected to the sample collection unit, is used to execute at least one screening strategy to label difficult samples. The screening strategy includes: samples whose labels are inconsistent with the model classification results after manual review; samples whose defects are detected by the model but are classified as qualified; and samples whose defects are not detected by the model but are classified as requiring repair or scrap. An incremental training unit, connected to the difficult sample screening unit, is used to perform incremental training on the multi-task detection model module using the difficult samples when the number of difficult samples reaches a preset threshold. The model update unit is used to synchronize the incrementally trained model parameters to the multi-task detection model module and the model fast adaptation module.

6. The multi-dimensional intelligent monitoring system for a relay flexible production line according to claim 5, characterized in that: The multi-source data acquisition module includes: The appearance image acquisition unit includes multiple industrial cameras at different angles and a diffuse light source; The electrical parameter acquisition unit includes a resistance tester, a voltage tester, and an oscilloscope; The mechanical characteristic acquisition unit includes a displacement sensor, a pressure sensor, and a timing acquisition card; The environmental data acquisition unit includes temperature and humidity sensors and dust sensors.

7. A multi-dimensional intelligent monitoring method for a flexible relay production line, characterized in that, The method, applied to a multi-dimensional intelligent monitoring system for a relay flexible production line as described in any one of claims 1-6, comprises: In response to production instructions, the multi-source data acquisition module is controlled to synchronously acquire data from the relay under test in different dimensions and production line environment data; The collected data is input into the multi-task detection model module, features are extracted through a shared backbone network, and the appearance defect detection results, electrical parameter prediction values ​​and mechanical characteristic prediction values ​​are output concurrently through independent output heads; The defect classification and judgment module calculates a comprehensive score based on the environmental data, the appearance defect detection results, the predicted values ​​of electrical parameters and the predicted values ​​of mechanical characteristics, and generates a quality level and handling instructions based on the comprehensive score. The control execution unit sorts and processes the relays to be tested according to the processing instructions; The data closed-loop and incremental learning module uses historical detection data, grading results, and verification data to screen difficult samples and incrementally optimize the multi-task detection model module.

8. The multi-dimensional intelligent monitoring method for a relay flexible production line according to claim 7, characterized in that: Prior to the step of responding to production instructions, a changeover adaptation step is also included: Receive new model switching command; Obtain the standard parameters of the new model from the model library; The zero-sample adaptation unit calculates and adjusts the weight parameters of the multi-task detection model module based on the standard parameters. Verify the detection accuracy of the adapted model; If the detection accuracy is lower than a preset threshold, the multi-task detection model module is fine-tuned using a small number of samples of the new model through the few-sample fine-tuning unit.

9. The multi-dimensional intelligent monitoring method for a relay flexible production line according to claim 8, characterized in that: The steps for calculating the comprehensive score by the defect classification and determination module include: The severity of appearance defects, the degree of electrical parameter exceedance, and the degree of abnormal mechanical characteristics are quantified to obtain the corresponding severity values; Calculate environmental correction factors based on environmental data; Calculate the overall score S: S = Severity of appearance defect × 0.35 + Degree of electrical parameter exceedance × 0.4 + Degree of abnormal mechanical characteristics × 0.25 + C_env × 0.05; S∈[0,0.5]→Excellent, S∈(0.5,1.0]→Good, S∈(1.0,2.0]→Qualified, S∈(2.0,3.0]→Rework, S∈(3.0,4.0]→Scrapped.

10. The multi-dimensional intelligent monitoring method for a relay flexible production line according to claim 9, characterized in that: The steps of the data closure and incremental learning module for screening difficult samples include at least one of the following strategies: Level 1 screening: Screening samples whose labels, as manually verified, do not match the model's grading results; Secondary screening: Screening samples that are detected as defective by the multi-task detection model module but are deemed qualified by the defect classification judgment module; Three-level screening: Screening samples for which the multi-task detection model module did not detect defects but the defect classification judgment module determined that they should be reworked or scrapped.

Images