Digitization-based product safety detection method and system
By combining sensor networks and intelligent inspection equipment with anomaly identification models and optimization schemes, the robustness and adaptability issues of multi-source data fusion methods in complex industrial scenarios have been solved, achieving efficient and accurate product safety inspection and optimization.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-20
- Publication Date
- 2026-04-14
AI Technical Summary
In existing technologies, multi-source data fusion methods are not robust enough when facing complex industrial scenarios. Simple weighted averaging is prone to amplifying errors. Rule engines are prone to misjudgment or omission when facing unpreset defect combinations. They are difficult to cope with dynamic changes and complexity of data, resulting in unstable detection results.
The system uses sensor networks to collect environmental and product status data, combines an abnormal environment identification model and intelligent inspection equipment, uses the Mask R-CNN algorithm to identify defects, matches and optimizes schemes through anomaly path knowledge graphs, implements simulation testing and iterative optimization, generates maintenance decisions, and forms a closed-loop inspection process.
It improves the robustness and accuracy of detection, reduces false positives and false negatives, achieves scientific and timely anomaly warnings, reduces maintenance costs, and improves the efficiency and reliability of product safety testing.
Smart Images
Figure CN121860465A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of product safety testing technology, specifically to a digital product safety testing method and system. Background Technology
[0002] In the field of digital product safety inspection, multi-source data fusion technology is a key element in improving the accuracy and stability of inspections, and is widely used in complex tasks in industrial scenarios such as non-destructive testing of automotive parts and defect analysis of electronic components. However, current mainstream data fusion methods still mainly rely on simple weighted averaging or rule engines, which have significant shortcomings in practical applications. Simple weighted averaging assigns fixed weights to data from different sources and then superimposes them, but it is difficult to adapt to dynamic fluctuations in data quality, such as outliers caused by temporary sensor interference or deviations caused by environmental factors. This can amplify errors in actual inspections, thus affecting the reliability of the inspection results. Rule engines rely on manually preset rules for data fusion and decision-making. While they are effective in specific scenarios, their generalization ability is significantly insufficient when facing complex scenarios (such as the inspection of electronic components with multiple types of defects). They are prone to failure due to data distribution shifts or the emergence of new interference factors.
[0003] Traditional data fusion methods suffer from weak robustness, particularly in practical industrial applications. For example, in non-destructive testing of automotive parts, simple weighted averaging can further amplify errors when ultrasonic data is interfered with by material inhomogeneity or infrared data is deviated by ambient temperature fluctuations. Furthermore, rule-based engines often misjudge or miss defects when encountering unpredictable combinations. These issues make existing systems ill-equipped to handle the dynamic changes and complexity of data in industrial scenarios, resulting in inadequate stability and accuracy of test results. Therefore, a more reliable multi-source data fusion mechanism is urgently needed to improve the robustness and adaptability of digital product safety testing systems, thereby better addressing the diverse challenges in industrial settings. Summary of the Invention
[0004] To address the shortcomings of existing technologies, this invention provides a digital-based product safety testing method and system, which solves the problem of weak robustness in existing technologies that rely on simple weighted averages or rule engines and traditional data fusion methods.
[0005] To achieve the above objectives, the present invention provides the following technical solution: a digital-based product safety testing method, comprising: Collect product operating environment data within the target area. If there is an abnormality in the operating environment, generate an environmental anomaly value from the abnormal state data of the abnormal environment. If the environmental anomaly value exceeds the anomaly threshold, send a collection test command to the outside. Collect product status data within the target area. If there is an abnormality in the product status within the target area, the intelligent inspection equipment will collect image and signal data within the target area, identify and classify the defect type, and summarize the acquired defect data into status monitoring data to generate a comprehensive inspection data set. The system uses the comprehensive detection data to filter out abnormal points within the target area, generates abnormal values from the abnormal state data of the abnormal points, and sends an alarm command to the outside if the abnormal value exceeds the preset abnormal threshold. After matching the corresponding optimization scheme from the abnormal paths of product safety, the optimization scheme is simulated and tested to obtain test data. Performance values are generated from the simulation test data. If the performance values do not exceed expectations, the obtained optimization scheme is iteratively optimized. The optimized product is applied in the target area, and comprehensive test data is obtained again. If the improvement of the optimized product is lower than expected, the product is tested after a test interval that meets the constraints. Based on the obtained test data, the maintenance decision library matches the corresponding maintenance strategy for the product anomaly.
[0006] Preferably, several monitoring nodes are selected within the target area of the product, and a sensor network is deployed at the monitoring nodes; The sensor network collects operating environment data at monitoring nodes, including temperature fluctuations, humidity changes, and electromagnetic interference intensity data, and summarizes them to generate a corresponding environmental monitoring data set; Using environmental monitoring data within the environmental monitoring dataset as input, the trained abnormal environment identification model is used to identify abnormal environments, outputting the corresponding environmental anomaly degree, obtaining the environmental anomaly degree of the abnormal environment and the time node that generated the abnormal environment, and summarizing to generate the corresponding environmental anomaly dataset. Environmental anomaly values are generated from the anomaly degree of the abnormal environment and the abnormal time points, as follows: , in, For the first The time interval of each abnormal node. The mean of the time intervals. For the first Anomaly level of the environment, For the corresponding mean, The number of abnormal nodes; weighting coefficient: ,and .
[0007] Preferably, after receiving the data acquisition and testing command, the sensor network in the target area acquires data, including vibration data, current fluctuation data and signal strength data, and the acquired detection data is summarized to obtain a product status monitoring data set; Using product status monitoring data as input, the trained status detection model is used to detect abnormal states. If there is an abnormality in the product status within the target area, an image and signal acquisition command is sent to the outside.
[0008] Preferably, each abnormal point is marked on a digital map, and a pre-trained path planning algorithm plans the corresponding inspection path for the intelligent inspection equipment. After receiving the image and signal acquisition instructions, the intelligent inspection equipment collects data in the target area according to the inspection path to obtain the corresponding product defect data. Using the collected product defect data as input, the pre-trained MaskR-CNN algorithm is used to identify and classify defects, and the defect data is labeled with time and location information.
[0009] Preferably, product defect data is used as input, and the trained defect evaluation model is used to evaluate the degree of defect at each defect location in the target area to obtain the corresponding defect score. If the defect type at the defect location matches the expected defect, and the defect score exceeds the expectation, the corresponding location will be designated as an anomaly point, and an alarm command will be issued to the outside.
[0010] Preferably, anomalies are marked on a digital map, the detection and identification data are verified, the identification results are checked and further verified and labeled, and the locations, defects, and times of defects are summarized to generate an anomaly status data set. The following method is used to generate outlier values from outlier state data within the outlier state dataset: , In the formula: For the first Next and first The distance to the location of the secondary anomaly point The mean distance for The degree of defect at the anomaly point in time. For the corresponding mean, The threshold for the number of outliers. Number of outliers; Weighting coefficient: ,and .
[0011] Preferably, principal component analysis is performed on the product operating environment data to determine the key factors affecting outliers, and an anomaly path identification knowledge graph is pre-constructed using product defect identification as the target term. Upon receiving an alarm command, based on the correspondence between key factors and product design parameters and abnormal paths, the abnormal path identification knowledge graph outputs the abnormal path for product safety.
[0012] Preferably, based on the performance anomaly path matching optimization scheme, the optimization scheme is simulated and tested in various test scenarios, and the obtained test data is summarized to generate a product simulation test data set; Using simulation test data from the product simulation test dataset as input, the trained performance evaluation model is used to evaluate the product performance and generate performance values, which are then used to evaluate the current product performance.
[0013] Preferably, the optimized solution is executed to adjust the product design parameters to obtain the optimized product; after reconstructing the performance values from the product defect data, the performance values before and after optimization at each anomaly point are obtained. Improvement degree is generated from the performance value at the anomaly point. If the improvement degree of the performance value is lower than the preset improvement threshold, several maintenance strategies are pre-defined and aggregated to generate the corresponding maintenance decision library.
[0014] This invention also provides a digital-based product safety testing system for implementing the aforementioned digital-based product safety testing method, comprising: The environmental anomaly analysis unit collects the product's operating environment data within the target area. If there is an anomaly in the operating environment, it generates an environmental anomaly value from the anomaly status data of the anomaly environment. If the environmental anomaly value exceeds the anomaly threshold, it sends a collection test command to the outside. The status detection unit collects product status data within the target area. If there is an abnormality in the product status within the target area, the intelligent inspection equipment collects image and signal data within the target area, identifies and classifies the defect type, and summarizes the acquired defect data into status monitoring data to generate a comprehensive detection data set. The abnormal alarm unit uses the comprehensive detection data to filter out abnormal points in the target area, generates abnormal values from the abnormal state data of the abnormal points, and sends an alarm command to the outside if the abnormal value exceeds the preset abnormal threshold. The simulation unit is optimized. After matching the corresponding optimization schemes from the abnormal paths of product safety, the optimization schemes are simulated and tested to obtain test data. Performance values are generated from the simulation test data. If the performance values do not exceed expectations, the obtained optimization schemes are iteratively optimized. The maintenance decision unit applies the optimized product to the target area, re-tests and obtains comprehensive test data. If the improvement of the optimized product is lower than expected, the product is tested after a test interval that meets the constraints. Based on the obtained test data, the maintenance decision library matches the corresponding maintenance strategy for the product anomaly.
[0015] Beneficial effects
[0016] This invention enhances detection robustness by dynamically fusing multi-source data. It utilizes sensor networks to collect environmental data and combines this data with an anomaly identification model to generate environmental anomaly values, accurately quantifying the degree of environmental anomalies. This overcomes the limitations of traditional detection methods that rely on simple weighted averages or rule engines, avoiding misjudgments caused by dynamic data fluctuations and adapting to the dynamic changes of complex environments. Simultaneously, it improves the accuracy of defect identification and anomaly judgment. Data is collected through intelligent inspection equipment, and the Mask R-CNN algorithm is used to accurately identify and classify defects. After secondary verification, annotation, and a defect evaluation model to generate scores, anomaly point location becomes more precise, reducing missed and false positives. Furthermore, it achieves… The scientific and timely nature of anomaly warnings is ensured by comprehensively considering multiple factors through anomaly value calculation formulas to quantify the severity of anomalies, ensuring accurate triggering of alarm commands and reducing invalid alarms. The optimization and maintenance process is more efficient, using anomaly path knowledge graph matching optimization solutions. Through simulation testing and iterative optimization, product performance is rapidly improved. Maintenance strategies are dynamically formulated based on the degree of improvement, and combined with constraint-based detection intervals, maintenance costs are reduced and maintenance targeting is improved. Furthermore, the entire process is digitized and intelligent, reducing manual intervention and integrating multi-dimensional data to form a closed loop from detection to optimized maintenance, significantly improving the efficiency and reliability of product safety testing, making it suitable for complex industrial scenarios. Attached Figure Description
[0017] Figure 1 This is a system structure block diagram of the present invention. Detailed Implementation
[0018] 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.
[0019] This invention provides a digital-based product safety testing method and system, the specific implementation of which is described in conjunction with the appendix. Figure 1 Please provide a detailed explanation.
[0020] Multiple monitoring nodes are deployed within the target area, and a sensor network is installed at each node to collect environmental data. This sensor network can monitor temperature fluctuations, humidity changes, and electromagnetic interference intensity in real time. The data acquired by the sensor network is aggregated to generate an environmental monitoring dataset. This environmental monitoring dataset is fed as input data into a trained anomaly identification model, which analyzes the data and outputs a corresponding environmental anomaly score. The environmental anomaly score, combined with the time point at which the anomaly occurred, generates an environmental anomaly dataset. Further, an environmental anomaly value is calculated based on the data in the environmental anomaly dataset, using the following formula: , in, For the first The time interval of each abnormal node. The mean of the time intervals. For the first Anomaly level of the environment, For the corresponding mean, The number of abnormal nodes; weighting coefficient: ,and The environmental anomaly value E calculated using the above formula can quantify the degree of anomaly in the operating environment. If the environmental anomaly value exceeds the preset anomaly threshold, a data acquisition and testing command is sent externally, triggering the subsequent status detection process.
[0021] Upon receiving the data acquisition and testing command, the sensor network within the target area begins collecting product status data, including vibration data, current fluctuation data, and signal strength data. This data is aggregated to generate a product status monitoring dataset, which is then fed into the trained status detection model. The status detection model analyzes the product status monitoring data to determine if any abnormal conditions exist within the target area. If an abnormal condition is detected, it sends out image and signal acquisition commands. Upon receiving these commands, the intelligent inspection equipment plans an inspection path based on a pre-trained path planning algorithm and acquires image and signal data within the target area. The acquired data is then processed to generate a product defect dataset.
[0022] To identify and classify product defect data, a pre-trained Mask R-CNN algorithm is used. This algorithm accurately identifies defect types and labels defect data with time and location information. Subsequently, a trained defect evaluation model is used to assess the defect severity at each defect location within the target area, generating a defect score. If the defect type at a certain location matches the expected defect, and the defect score exceeds the expected value, the location is marked as an anomaly, and an alarm is issued. The location, defect severity, and time of occurrence of the anomaly are aggregated to generate an anomaly state dataset. Based on the data in the anomaly state dataset, outliers are calculated. The formula is as follows: , in, For the first Next and first The distance to the location of the secondary anomaly point The mean distance for The degree of defect at the anomaly point in time. For the corresponding mean, The threshold for the number of outliers. Number of outliers; Weighting coefficient: ,and Outliers calculated using the above formula It can quantify the severity of anomalies. If an anomaly value exceeds a preset anomaly threshold, an alarm command is triggered by the anomaly alarm unit.
[0023] Upon receiving an alarm command, the system enters the abnormal path matching phase. First, principal component analysis is performed on the product operating environment data to identify key factors influencing outliers. Based on the correspondence between key factors and product design parameters, a pre-constructed abnormal path identification knowledge graph outputs abnormal paths for product safety. Corresponding optimization schemes are matched according to the abnormal paths, and simulation tests are conducted on these schemes. During simulation testing, the optimization schemes are verified under multiple test scenarios, acquiring test data and generating a product simulation test dataset. The data in the product simulation test dataset is used as input to the trained performance evaluation model to generate performance values. These performance values are used to evaluate the effectiveness of the optimization schemes; if the performance values do not meet expectations, the optimization schemes are iteratively optimized.
[0024] After multiple iterations and optimizations, the final optimization plan was determined and applied to products within the target area. The optimized products were then re-run within the target area, and the system again collected comprehensive testing data to evaluate the optimization effect. If the improvement of the optimized products was less than expected, it was necessary to set testing intervals based on constraints and perform testing at testing nodes that met the constraints. The constraint method for the testing intervals is as follows: , in, The number of product inspections within the target area. For the first The environmental outlier. This represents the mean of environmental outliers. This represents the difference between the time point and the outlier. This is the expected value of the outlier difference. The difference threshold; This is an indicator function that takes a value of 1 when the condition is true and 0 otherwise; weighting coefficients. And satisfy The detection interval C calculated using the above formula ensures the scientific validity and rationality of the detection.
[0025] After the inspection interval ends, the system re-inspects the product and, based on the acquired inspection data, matches appropriate maintenance strategies to product anomalies using the maintenance decision library. The maintenance decision library contains various maintenance strategies, which are pre-defined based on feature extraction results from the product defect data. To evaluate the effectiveness of the maintenance strategies, the system generates an improvement score by calculating performance values at anomaly points. The formula for calculating the improvement score is as follows: , in, and These represent the numbers before and after the intervention. Performance values at each outlier point and For the corresponding mean, This is the anomaly point number. The number of outliers. For the first The median improvement rate of each outlier This represents the mean of the median improvement rates. The improvement rate is calculated using the above formula. It can quantify the effectiveness of maintenance strategies.
[0026] Throughout the implementation process, the environmental anomaly analysis unit is responsible for collecting and analyzing operating environment data, the status detection unit is responsible for collecting and analyzing product status data, the anomaly alarm unit is responsible for screening anomalies and issuing alarm commands, the optimization simulation unit is responsible for matching optimization schemes and conducting simulation tests, and the maintenance decision-making unit is responsible for formulating maintenance strategies and evaluating their effectiveness. These units collaborate efficiently through data flow, ensuring the consistency and accuracy of the entire detection process.
[0027] Furthermore, to improve detection efficiency and accuracy, this invention also incorporates digital mapping technology. The locations of anomalies are marked on the digital map, and the detected and identified data undergoes verification and secondary validation. This method allows for rapid location of anomalies and targeted handling. Simultaneously, the system boasts a high degree of intelligence, automating the entire process from data acquisition to anomaly handling, significantly reducing the need for manual intervention.
[0028] In summary, this invention generates a comprehensive detection dataset through dynamic weight allocation of multi-source heterogeneous data. Combined with steps such as outlier calculation, outlier path matching, optimization scheme simulation testing, and maintenance strategy matching, it achieves comprehensive detection and optimization of product safety. This method not only improves detection efficiency and accuracy but also effectively solves the technical problems existing in the background art, possessing significant practical application value.
[0029] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, 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.
[0030] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.
Claims
1. A digital-based product safety testing method, characterized in that, include: Collect product operating environment data within the target area. If there is an abnormality in the operating environment, generate an environmental anomaly value from the abnormal state data of the abnormal environment. If the environmental anomaly value exceeds the anomaly threshold, send a collection test command to the outside. Collect product status data within the target area. If there is an abnormality in the product status within the target area, the intelligent inspection equipment will collect image and signal data within the target area, identify and classify the defect type, and summarize the acquired defect data into status monitoring data to generate a comprehensive inspection data set. The system uses the comprehensive detection data to filter out abnormal points within the target area, generates abnormal values from the abnormal state data of the abnormal points, and sends an alarm command to the outside if the abnormal value exceeds the preset abnormal threshold. After matching the corresponding optimization scheme from the abnormal paths of product safety, the optimization scheme is simulated and tested to obtain test data. Performance values are generated from the simulation test data. If the performance values do not exceed expectations, the obtained optimization scheme is iteratively optimized. The optimized product is applied in the target area, and comprehensive test data is obtained again. If the improvement of the optimized product is lower than expected, the product is tested after a test interval that meets the constraints. Based on the obtained test data, the maintenance decision library matches the corresponding maintenance strategy for the product anomaly.
2. The digital-based product safety testing method according to claim 1, characterized in that, Select several monitoring nodes within the target area of the product and deploy a sensor network at the monitoring nodes; The sensor network collects operating environment data at monitoring nodes, including temperature fluctuations, humidity changes, and electromagnetic interference intensity data, and summarizes them to generate a corresponding environmental monitoring data set; Using environmental monitoring data within the environmental monitoring dataset as input, the trained abnormal environment identification model is used to identify abnormal environments, outputting the corresponding environmental anomaly degree, obtaining the environmental anomaly degree of the abnormal environment and the time node that generated the abnormal environment, and summarizing to generate the corresponding environmental anomaly dataset. Environmental anomaly values are generated from the anomaly degree of the abnormal environment and the abnormal time points, as follows: ， in, For the first The time interval of each abnormal node. The mean of the time intervals. For the first Anomaly level of the environment, For the corresponding mean, The number of abnormal nodes; weighting coefficient: ,and .
3. The digital-based product safety testing method according to claim 1, characterized in that, Upon receiving the data acquisition and testing command, the sensor network within the target area collects data, including vibration data, current fluctuation data, and signal strength data. The acquired detection data is then aggregated to obtain a product status monitoring data set. Using product status monitoring data as input, the trained status detection model is used to detect abnormal states. If there is an abnormality in the product status within the target area, an image and signal acquisition command is sent to the outside.
4. The digital-based product safety testing method according to claim 3, characterized in that, Each anomaly point is marked on a digital map, and a pre-trained path planning algorithm plans the corresponding inspection path for the intelligent inspection equipment. After receiving the image and signal acquisition instructions, the intelligent inspection equipment collects data in the target area according to the inspection path to obtain the corresponding product defect data. Using the collected product defect data as input, the pre-trained MaskR-CNN algorithm is used to identify and classify defects, and the defect data is labeled with time and location information.
5. The digital-based product safety testing method according to claim 4, characterized in that, Using product defect data as input, the trained defect evaluation model is used to evaluate the degree of defect at each defect location within the target area and obtain the corresponding defect score. If the defect type at the defect location matches the expected defect, and the defect score exceeds the expectation, the corresponding location will be designated as an anomaly point, and an alarm command will be issued to the outside.
6. The digital-based product safety testing method according to claim 5, characterized in that, Mark the anomalies on the digital map, verify the detection and identification data, check the identification results and perform secondary verification and annotation; summarize the anomaly location, defect severity and defect occurrence time to generate an anomaly status data set. The following method is used to generate outlier values from outlier state data within the outlier state dataset: ， In the formula: For the first Next and first The distance to the location of the secondary anomaly point The mean distance for The degree of defect at the anomaly point in time. For the corresponding mean, The threshold for the number of outliers. Number of outliers; Weighting coefficient: ,and .
7. The digital-based product safety testing method according to claim 6, characterized in that, Principal component analysis was performed on product operating environment data to identify key factors affecting outliers. Using product defect identification as the target term, an anomaly path identification knowledge graph was pre-constructed. Upon receiving an alarm command, based on the correspondence between key factors and product design parameters and abnormal paths, the abnormal path identification knowledge graph outputs the abnormal path for product safety.
8. The digital-based product safety testing method according to claim 1, characterized in that, Based on the performance anomaly path matching optimization scheme, the optimization scheme is simulated and tested in various test scenarios. The obtained test data is then summarized to generate a product simulation test data set. Using simulation test data from the product simulation test dataset as input, the trained performance evaluation model is used to evaluate the product performance and generate performance values, which are then used to evaluate the current product performance.
9. The digital-based product safety testing method according to claim 1, characterized in that, The optimized solution is implemented to adjust the product design parameters and obtain the optimized product. After reconstructing performance values from product defect data, obtain the performance values before and after optimization at each anomaly point; Improvement degree is generated from the performance value at the anomaly point. If the improvement degree of the performance value is lower than the preset improvement threshold, several maintenance strategies are pre-defined and aggregated to generate the corresponding maintenance decision library.
10. A digital-based product safety testing system, used to implement the digital-based product safety testing method according to any one of claims 1-9, characterized in that, include: The environmental anomaly analysis unit collects the product's operating environment data within the target area. If there is an anomaly in the operating environment, it generates an environmental anomaly value from the anomaly status data of the anomaly environment. If the environmental anomaly value exceeds the anomaly threshold, it sends a collection test command to the outside. The status detection unit collects product status data within the target area. If there is an abnormality in the product status within the target area, the intelligent inspection equipment collects image and signal data within the target area, identifies and classifies the defect type, and summarizes the acquired defect data into status monitoring data to generate a comprehensive detection data set. The abnormal alarm unit uses the comprehensive detection data to filter out abnormal points in the target area, generates abnormal values from the abnormal state data of the abnormal points, and sends an alarm command to the outside if the abnormal value exceeds the preset abnormal threshold. The simulation unit is optimized. After matching the corresponding optimization schemes from the abnormal paths of product safety, the optimization schemes are simulated and tested to obtain test data. Performance values are generated from the simulation test data. If the performance values do not exceed expectations, the obtained optimization schemes are iteratively optimized. The maintenance decision unit applies the optimized product to the target area, re-tests and obtains comprehensive test data. If the improvement of the optimized product is lower than expected, the product is tested after a test interval that meets the constraints. Based on the obtained test data, the maintenance decision library matches the corresponding maintenance strategy for the product anomaly.