A fault criterion model construction method and system based on an AI target detection model

Abstract

The application relates to the technical field of AI target detection model fault criterion model construction, and discloses a fault criterion model construction method and system based on an AI target detection model, which comprises the following steps: obtaining AI target detection model historical operation data, including historical fault data, to provide basic information for the establishment of a fault criterion model; secondly, data cleaning is performed on the AI target detection model historical fault data, and the data is integrated into a data fault set, so that the quality and availability of the data are ensured; then, operation parameters are extracted from the data fault set, and key function parameters are further screened out; subsequently, a functional model is established and a fitting experiment is performed to improve the model performance; in this process, data processing is performed on the functional data set to determine optimal functional data, so that the accuracy of the model is improved; finally, the functional model is used to determine the fault criterion model of the AI target detection model, thereby providing support for the reliability of the establishment of the fault criterion model.

Description

Technical Field

[0001] This invention relates to the field of AI target detection model fault criterion model construction technology, and in particular to a method and system for constructing a fault criterion model based on an AI target detection model. Background Technology

[0002] With the rapid development of computer vision, especially the widespread application of deep learning technology, AI object detection models have achieved significant breakthroughs in image and video processing. These models can detect and locate various target objects with high accuracy, playing a crucial role in both road recognition for autonomous vehicles and product quality inspection in smart manufacturing.

[0003] Currently, data collection and recording in modern industry and technology have become more extensive and sophisticated. Large amounts of operating parameters and performance data are collected in real time and used to monitor the status of equipment or AI systems. However, this data inevitably contains errors, interference, or misjudgments, leading to inaccurate assessments of the status of monitored equipment or AI systems.

[0004] Therefore, there is an urgent need to invent a fault criterion model technology for constructing the status of monitoring equipment or AI systems, in order to solve the problem of inaccurate status judgment of monitoring equipment or AI systems caused by errors, interference or erroneous judgments in data judgment. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for constructing a fault criterion model based on an AI target detection model, which aims to solve the problem of inaccurate state judgment of monitoring equipment or AI systems.

[0006] On one hand, this invention provides a method for constructing a fault criterion model based on an AI target detection model, including:

[0007] Obtain historical runtime data of the AI ​​object detection model to be built;

[0008] Obtain historical fault data from the historical running data of the AI ​​target detection model to be built, and perform data cleaning on the historical fault data;

[0009] Obtain the historical fault data after data cleaning and establish a data fault set;

[0010] Obtain the operating parameters from the data fault set, extract the key functional parameters of the operating parameters, and establish a functional model for fitting experiments;

[0011] Obtain the functional dataset of the functional model in the fitting experiment, and establish a fault criterion model for the AI ​​object detection model, wherein,

[0012] The functional dataset of the functional model is processed, and the optimal functional data is obtained by considering the relationships between the functional data in the functional dataset of the functional model after data processing.

[0013] Obtain the functional model of the optimal functional data, and determine the fault criterion model of the AI ​​target detection model based on the functional model of the optimal functional data.

[0014] Furthermore, when acquiring historical fault data from the historical operational data of the AI ​​target detection model to be constructed, and cleaning the historical fault data, the process includes:

[0015] Obtain historical fault data from the historical operation data of the AI ​​target detection model;

[0016] Duplicate data in the historical fault data of the AI ​​target detection model are removed and cleaned up;

[0017] Obtain historical fault data of the AI ​​target detection model after removing duplicate data, and remove erroneous judgment data from the historical fault data of the AI ​​target detection model after removing duplicate data based on the Z-Score method.

[0018] Furthermore, when removing erroneous judgment data from the historical fault data of the AI ​​target detection model based on the Z-Score method to eliminate duplicate data, the process includes:

[0019] Obtain each feature data point E from the historical fault data of the AI ​​target detection model after removing duplicate data, and set it as E=E1,E2,E3…En, and then apply the formula... Obtain the Z-Score value ΔE for each of the feature data points E, and set ΔE as (E, Zi), i=1,2,3,...i;

[0020] In the formula, Z is the Z-Score of each feature data point E, X is the actual value of each feature data point E in the historical fault data of the AI ​​target detection model, mu is the average value of each feature data point E in the historical fault data of the AI ​​target detection model, and sigma is the standard deviation of each feature data point E in the historical fault data of the AI ​​target detection model.

[0021] Pre-set a first preset Z-Score value △E1 and a second preset Z-Score value △E2.

[0022] Based on the relationship between the Z-Score value △E of the feature data point E and the first preset Z-Score value △E1 and the second preset Z-Score value △E2, it is determined whether the Z-Score value △E of the feature data point E is an outlier.

[0023] When △E1≤△E<△E2, the Z-Score value △E of the feature data point E is determined to be a normal value.

[0024] When △E < △E1, the Z-Score value △E of the feature data point E is determined to be an outlier, and the feature data point E is removed.

[0025] When △E > △E2, the Z-Score value △E of the feature data point E is determined to be an outlier, and the feature data point E is removed.

[0026] Furthermore, when acquiring the cleaned historical fault data and establishing the data fault set, the process includes:

[0027] Obtain the historical fault data of the AI ​​target detection model after removing erroneous judgment data, including each fault data and the data features of each fault data.

[0028] Clustering is performed based on the data characteristics of each fault data, and a fault dataset is established based on each fault data under each cluster.

[0029] Furthermore, when acquiring the operating parameters from the data fault set, extracting the key functional parameters of the operating parameters, and establishing a functional model for fitting experiments, the process includes:

[0030] Obtain non-fault data from the historical operating data of the AI ​​target detection model to be built, and obtain each operating parameter from the non-fault data;

[0031] Obtain the minimum and maximum values ​​of each operating parameter in the non-fault data, and establish the operating parameter thresholds of the AI ​​target detection model based on the minimum and maximum values ​​of each operating parameter in the non-fault data;

[0032] Obtain each operating parameter in the data fault set, compare each operating parameter in the data fault set with the operating parameter threshold of the AI ​​target detection model, and obtain the operating parameters in the data fault set that are not in the operating parameter threshold of the AI ​​target detection model;

[0033] Each operating parameter in the data fault set that is not within the operating parameter threshold of the AI ​​target detection model is identified as the operating highly relevant parameter in the data fault set.

[0034] Furthermore, when determining each operating parameter in the data fault set that is not within the operating parameter threshold of the AI ​​target detection model as the operating highly relevant parameter in the data fault set, this includes:

[0035] The weight coefficients of each highly correlated operational parameter in the fault data set are obtained based on the linear regression calculation method.

[0036] Based on the relationship between the weight coefficients of each highly correlated operational parameter, the parameter with the largest absolute value of its weight coefficient is determined as the key functional parameter of the data fault set.

[0037] Furthermore, when obtaining the weight coefficients of each highly correlated operational parameter in the data fault set based on the linear regression calculation method, it includes:

[0038] Obtain the data variable Y for each fault data in the fault data set;

[0039] And calculate the equation based on linear regression. Obtain the weight coefficients of each highly correlated operational parameter in the data fault set;

[0040] In the formula, Set the parameters with high correlation for each operation in the data fault set. = 1, 2,… p and β are the weighting coefficients of each highly correlated parameter in the data fault set, and β is set to β0, β1, β2…βp;

[0041] Where p = 1, 2, 3, ..., p.

[0042] Furthermore, when processing the functional dataset of the functional model and obtaining the optimal functional data based on the relationships between the functional data in the processed functional dataset, the process includes:

[0043] Obtain the functional data of each function in the functional dataset of the functional model in the fitting experiment, and remove duplicate data in each function data in the functional dataset;

[0044] Clustering was used to exclude outlier data in each function of the functional dataset after removing duplicate data, and the value of k in k-means clustering was solved by using the idea of ​​quadratic clustering.

[0045] Furthermore, the secondary clustering includes:

[0046] Obtain the functional data and the variation range between each functional data in the functional dataset, and use the direct clustering method to obtain the clustering result as {C1,C2,...,Ck}, and denote the cluster center of each cluster as {W1,W2,...,Wk};

[0047] Obtain the cluster center Wmax with the largest value in {W1, W2, ..., Wk}, and the cluster Cmax corresponding to the largest cluster center Wmax;

[0048] Get the distance di between Wmax and each cluster center in {W1, W2, ..., Wk} except Wmax, and set di = Wmax - Wi;

[0049] Based on the relationship between the distance di of each cluster center and a preset threshold T, the optimal functional data in the functional dataset is obtained; wherein,

[0050] When di≥T, it is determined that the distance di between the cluster center Wi and the maximum cluster center Wmax is greater than or equal to the preset threshold T. The cluster center Wi and the maximum cluster center Wmax have a large difference in feature space, and the cluster Ci corresponding to the cluster center Wi is excluded.

[0051] When di < T, it is determined that the distance di between the cluster center Wi and the maximum cluster center Wmax is less than the preset threshold T, the cluster center Wi and the maximum cluster center Wmax have little difference in feature space, and the cluster Ci corresponding to the cluster center Wi is merged with the maximum cluster Cmax into a new cluster;

[0052] The final number of clusters retained is the value of k in the k-means algorithm. The retained clusters are the optimal functional data in the functional dataset.

[0053] On the other hand, this embodiment of the invention also provides a fault criterion model construction system based on an AI target detection model, applicable to the fault criterion model construction methods based on AI target detection models in the above embodiments, including:

[0054] The data acquisition module is electrically connected to the running database of the AI ​​target detection model. The data acquisition module is used to acquire the historical running data of the AI ​​target detection model to be built.

[0055] The first data processing module is electrically connected to the data acquisition module. The first data processing module is used to acquire historical fault data from the historical running data of the AI ​​target detection model to be constructed, and to clean the historical fault data. The first data processing module is also used to acquire the cleaned historical fault data and establish the data fault set.

[0056] The second data processing module, electrically connected to the first data processing module, is used to acquire operating parameters from the data fault set, extract key functional parameters from the operating parameters, and establish a functional model for fitting experiments. The second data processing module is also used to acquire the functional dataset of the functional model in the fitting experiments and establish a fault criterion model for the AI ​​object detection model.

[0057] The second data processing module is further configured to process the functional dataset of the functional model and obtain the optimal functional data by determining the relationship between the functional data in the functional dataset of the functional model after data processing. The second data processing module is further configured to obtain the functional model of the optimal functional data and determine the fault criterion model of the AI ​​target detection model based on the functional model of the optimal functional data.

[0058] Compared with existing technologies, the present invention provides a method and system for constructing a fault criterion model based on an AI object detection model. The advantages are as follows: By collecting and analyzing historical operational data of the model, a comprehensive understanding of the model's performance and behavior can be obtained. This helps identify the model's performance under different conditions, providing benchmark data for further improvement. Secondly, extracting historical fault data from the historical operational data and cleaning it is a crucial step. Data cleaning eliminates potential errors or anomalies, ensuring that the model is built based on accurate and reliable information. Subsequently, a fault data set is established, which helps to centralize fault data in a manageable dataset, facilitating subsequent analysis and modeling. After obtaining the fault data set, the operational parameters are further analyzed to extract key functional parameters, and a functional model is built for fitting experiments. This process helps to understand the relationship between parameters and model performance, providing guidance for subsequent optimization. Through fitting experiments, a functional dataset of the functional model is obtained, laying the foundation for constructing the fault criterion model. This model will help to better understand model behavior, especially its performance when facing faults. Finally, by processing and analyzing the functional dataset of the functional model, the optimal functional data can be determined, which will be used as the fault criterion model for the AI ​​object detection model. This will help improve the stability and robustness of the model, reduce the risk of failure, and thus enhance the reliability and practicality of the model. Attached Figure Description

[0059] Figure 1 This is a flowchart of a fault criterion model construction method based on an AI target detection model according to an embodiment of the present invention.

[0060] Figure 2 This is a structural block diagram of a fault criterion model construction system based on an AI target detection model according to an embodiment of the present invention. Detailed Implementation

[0061] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the disclosure to those skilled in the art. It should be noted that, unless otherwise specified, embodiments and features in the embodiments of the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0062] like Figure 1 As shown, the present invention provides a method for constructing a fault criterion model based on an AI target detection model, comprising:

[0063] Step S100: Obtain the historical running data of the AI ​​object detection model to be built.

[0064] Step S200: Obtain historical fault data from the historical running data of the AI ​​target detection model to be built, and clean the historical fault data.

[0065] Step S300: Obtain historical fault data after data cleaning and establish a data fault set.

[0066] Step S400: Obtain the operating parameters from the data fault set, extract the key functional parameters of the operating parameters, and establish a functional model for fitting experiments.

[0067] Step S500: Obtain the functional dataset of the functional model in the fitting experiment, and establish the fault criterion model of the AI ​​target detection model. Specifically, the functional dataset of the functional model is processed, and the relationship between the functional data in the functional dataset of the functional model after data processing is used to obtain the optimal functional data. The functional model of the optimal functional data is obtained, and the functional model of the optimal functional data is determined as the fault criterion model of the AI ​​target detection model.

[0068] Understandably, in step S100, by acquiring historical operational data, a benchmark for the model can be established to understand its performance under various scenarios, providing a basis for subsequent improvements. Step S200 extracts and cleans historical fault data from historical operational data, which is crucial for accurately capturing model problems and anomalies, contributing to the model's stability and reliability. Step S300 establishes a dedicated fault data set, making fault data easier to manage and analyze, providing a clear data source for fault analysis and modeling. Through step S400, the operational parameters in the fault data set are analyzed, key functional parameters are extracted, and a functional model is built. This helps to understand the relationship between parameters and model performance, as well as factors that may lead to failures. Step S500 further constructs a fault criterion model for the AI ​​object detection model. By processing the functional dataset of the functional model, the optimal functional data is determined, thereby improving the model's performance and robustness. This process enables better prediction and diagnosis of potential faults, reduces model instability in practical applications, and improves its usability and maintainability.

[0069] It can be seen that by collecting historical data, cleaning fault data, analyzing parameters, building functional models, and establishing fault criterion models, a more reliable and high-performance fault criterion model for AI object detection can be created. This process not only reduces the occurrence of faults and problems but also improves the stability and reliability of the model in practical applications, thus providing a solid foundation for judging the state of the AI ​​object detection model.

[0070] Specifically, in some embodiments of the present invention, when obtaining historical fault data from the historical operating data of the AI ​​target detection model to be constructed and cleaning the historical fault data, the process includes: obtaining historical fault data from the historical operating data of the AI ​​target detection model; removing and cleaning duplicate data from the historical fault data of the AI ​​target detection model; obtaining the historical fault data of the AI ​​target detection model after removing and cleaning duplicate data, and removing erroneous judgment data from the historical fault data of the AI ​​target detection model after removing and cleaning duplicate data based on the Z-Score method.

[0071] Specifically, in some embodiments of the present invention, when removing erroneous judgment data from the historical fault data of an AI target detection model that removes duplicate data based on the Z-Score method, the process includes: obtaining each feature data point E in the historical fault data of the AI ​​target detection model that removes duplicate data, setting E=E1,E2,E3…En, and applying the formula... Obtain the Z-Score value ΔE for each feature data point E, and set ΔE as (E, Zi), i = 1, 2, 3, ... i; where Z is the Z-Score of each feature data point E, X is the actual value of each feature data point E in the historical fault data of the AI ​​target detection model, mu is the average value of each feature data point E in the historical fault data of the AI ​​target detection model, and sigma is the standard deviation of each feature data point E in the historical fault data of the AI ​​target detection model; a first preset Z-Score value ΔE1 and a second preset Z-Score value ΔE2 are preset, and the Z-Score value is determined based on the Z-Score of each feature data point E. The relationship between the Z-Score value △E and the first preset Z-Score value △E1 and the second preset Z-Score value △E2 is used to determine whether the Z-Score value △E of feature data point E is an outlier. When △E1≤△E<△E2, the Z-Score value △E of feature data point E is considered normal. When △E<△E1, the Z-Score value △E of feature data point E is considered an outlier, and feature data point E is removed. When △E>△E2, the Z-Score value △E of feature data point E is considered an outlier, and feature data point E is removed.

[0072] Understandably, removing duplicate data from historical failure data during the data cleaning phase helps reduce redundant information in the dataset, ensuring data quality and efficiency in subsequent analysis and modeling processes. Secondly, further filtering of historical failure data using the Z-Score method effectively identifies and removes outlier data points. Z-Score is a statistical method commonly used to detect outliers; it determines whether data is an outlier by calculating the deviation between a data point and the mean and standard deviation of the dataset. This helps eliminate invalid data that may be caused by errors or other anomalies, thereby improving data accuracy and reliability. Finally, integrating these steps into the data cleaning process ensures that the historical failure dataset used is more reliable, providing more dependable input data for subsequent AI object detection model building, thus improving model performance and reliability. This data cleaning method helps reduce erroneous judgments caused by noisy data, thereby improving the accuracy and stability of the model in practical applications.

[0073] Specifically, in some embodiments of the present invention, when acquiring historical fault data after data cleaning and establishing a data fault set, the following steps are included: acquiring each fault data and data features of the historical fault data of the AI ​​target detection model after removing erroneous judgment data; performing clustering based on the data features of each fault data; and establishing a fault dataset based on each fault data under each cluster.

[0074] Understandably, acquiring cleaned historical failure data and its features provides a more practically meaningful dataset for analysis, including detailed information on different failure scenarios. This helps to gain a more comprehensive understanding of model performance issues and accurately identify different types of failure modes. Secondly, clustering similar failure data points using data features helps to organize the failure data into a more structured form. This clustering helps to identify failure patterns and find commonalities, thereby better understanding the model's weaknesses and problems. Finally, building a failure dataset will facilitate further analysis and modeling efforts. It makes data on different failure modes more easily used for model improvement, optimization, and prediction, thus improving model performance and reliability. This approach helps extract valuable information from historical failure data to improve and maintain AI object detection models, reduce the occurrence of potential failures, and enhance the model's stability and usability in practical applications.

[0075] Specifically, in some embodiments of the present invention, when taking the operating parameters from the data fault set, extracting the key functional parameters of the operating parameters, and establishing a functional model for fitting experiments, the following steps are included: obtaining non-fault data from the historical operating data of the AI ​​target detection model to be constructed, and obtaining each operating parameter in the non-fault data; obtaining the minimum and maximum values ​​of each operating parameter in the non-fault data, and establishing the operating parameter threshold of the AI ​​target detection model based on the minimum and maximum values ​​of each operating parameter in the non-fault data; obtaining each operating parameter from the data fault set, comparing each operating parameter from the data fault set with the operating parameter threshold of the AI ​​target detection model, and obtaining the operating parameters from the data fault set that are not within the operating parameter threshold of the AI ​​target detection model; and determining each operating parameter from the data fault set that is not within the operating parameter threshold of the AI ​​target detection model as the operating highly correlated parameters in the data fault set.

[0076] Understandably, by acquiring non-fault data and various operating parameters from historical operational data, a baseline can be created representing the model's performance and parameter range under normal operating conditions. This provides crucial insight into the model's expected behavior. Secondly, by obtaining the minimum and maximum values ​​of each operating parameter in the non-fault data and establishing operating parameter thresholds based on these values, clear boundaries can be set for the model's normal operating range. This helps identify any anomalies deviating from the normal range. Finally, by comparing the operating parameters in the fault data set with the operating parameter thresholds, parameters that do not conform to the normal operating range can be identified, thus determining the operating parameters in the fault data set that are highly correlated with model performance. This approach enables more accurate problem diagnosis, identifies fault-related parameters, and facilitates model improvement and optimization to enhance its performance and stability.

[0077] Specifically, in some embodiments of the present invention, when determining each operating parameter in the data fault set that is not in the operating parameter threshold of the AI ​​target detection model as the operating highly correlated parameter in the data fault set, the method includes: obtaining the weight coefficient of each operating highly correlated parameter in the data fault set based on the linear regression calculation method; and determining the highly correlated parameter with the largest absolute value of the weight coefficient of each operating highly correlated parameter as the key functional parameter of the data fault set according to the size relationship between the weight coefficients of each operating highly correlated parameter.

[0078] Specifically, in some embodiments of the present invention, obtaining the weight coefficients of each highly correlated operational parameter in the data fault set based on the linear regression calculation method includes: obtaining the data variable Y of each fault data in the data fault set; and calculating the weight coefficients according to the linear regression equation. Obtain the weight coefficients of each highly correlated operational parameter in the data fault set; where, Set the parameters with high correlation for each operation in the data fault set. = 1, 2,… p and β are the weighting coefficients of each highly correlated parameter in the data fault set, and β is set to β0, β1, β2…βp; where p = 1, 2, 3, …, p.

[0079] Understandably, by obtaining the weight coefficients of highly correlated parameters based on linear regression calculations, the impact of these parameters on the performance of the target detection model can be quantitatively measured. This helps identify which parameters play a key role in model performance, thus providing a better understanding of how the model works. Secondly, based on the relationship between the weight coefficients, it's possible to determine which parameter has the greatest influence on model performance, considering it a critical functional parameter. This allows for more targeted improvement and optimization of the model, focusing on addressing the most critical performance issues and improving the model's effectiveness and maintainability. Finally, by identifying critical functional parameters, changes in these parameters can be more easily monitored and tracked during model monitoring and maintenance, enabling timely identification and intervention of potential problems. This helps reduce the risk of failures and improves the model's reliability and usability.

[0080] It can be seen that data analysis and weight coefficient calculation helped determine the most critical functional parameters in the data fault set, thus providing targeted guidance for optimizing and maintaining the criterion model of the AI ​​object detection model, and improving the performance and reliability of the model.

[0081] Specifically, in some embodiments of the present invention, when processing the functional dataset of a functional model and obtaining the optimal functional data by analyzing the relationships between functional data in the functional dataset of the functional model after data processing, the following steps are included: obtaining each functional data in the functional dataset of the functional model in the fitting experiment, and removing duplicate data from each functional data in the functional dataset; using clustering to exclude abnormal data in each functional data in the functional dataset after removing duplicate data, and using the idea of ​​quadratic clustering to solve the value of k in k-means clustering.

[0082] Specifically, in some embodiments of the present invention, secondary clustering includes: obtaining each functional data and the variation range between each functional data in the functional dataset, and using the direct clustering method to obtain the clustering result as {C1,C2,...,Ck}, and denoteing the cluster center of each cluster as {W1,W2,...,Wk}; obtaining the cluster center Wmax with the largest value in {W1,W2,...,Wk}, and the cluster Cmax corresponding to the largest cluster center Wmax; obtaining the distance di between Wmax and each cluster center in {W1,W2,...,Wk} other than Wmax, and setting di = Wmax - Wi; and obtaining the optimal cluster center in the functional dataset based on the relationship between the obtained distance di of each cluster center and a preset threshold T. Functional data; where, when di≥T, it is determined that the distance di between cluster center Wi and the maximum cluster center Wmax is greater than or equal to a preset threshold T, cluster center Wi and the maximum cluster center Wmax have a large difference in feature space, and the cluster Ci corresponding to cluster center Wi is excluded; when di<T, it is determined that the distance di between cluster center Wi and the maximum cluster center Wmax is less than a preset threshold T, cluster center Wi and the maximum cluster center Wmax have a small difference in feature space, and the cluster Ci corresponding to cluster center Wi is merged with the maximum cluster Cmax into a new cluster; the final number of clusters retained is obtained, which is the value of k in the k-means algorithm, and the retained clusters are the optimal functional data in the functional dataset.

[0083] Understandably, removing duplicate data from the functional dataset ensures its cleanliness and high quality. This helps reduce the interference of redundant information on analysis and modeling, improving data usability. Secondly, clustering is used to eliminate outliers. By dividing the data into different clusters, we can better understand the distribution and characteristics of the data. The use of quadratic clustering solves the problem of selecting the k-value in k-means clustering, which helps to more accurately divide the data clusters and eliminate outliers that do not meet the model's requirements. Most importantly, by identifying the optimal functional data, we can accurately select the data points that have the greatest impact on model performance. This helps optimize the functional model, improve its predictive performance, reduce model errors, and enhance the model's usability and reliability in practical applications.

[0084] In summary, this invention provides a method for constructing a fault criterion model based on an AI object detection model. By collecting and analyzing the model's historical operating data, it can gain a comprehensive understanding of the model's performance and behavior. This helps identify the model's performance under different conditions and provides benchmark data for further improvement. Secondly, extracting historical fault data from the historical operating data and cleaning it is a crucial step. By cleaning the data, potential errors or anomalies can be eliminated, ensuring that the model is built based on accurate and reliable information. Subsequently, a data fault set is established, which helps to centralize fault data in a manageable dataset, facilitating subsequent analysis and modeling. After obtaining the data fault set, the operating parameters are further analyzed to extract key functional parameters, and a functional model is built for fitting experiments. This process helps to understand the relationship between parameters and model performance, providing guidance for subsequent optimization. Through fitting experiments, a functional dataset of the functional model is obtained, laying the foundation for constructing the fault criterion model. This model will help to better understand the model's behavior, especially its performance when facing faults. Finally, by processing and analyzing the functional dataset of the functional model, the optimal functional data can be determined, which will be used as the fault criterion model for the AI ​​object detection model. This will help improve the stability and robustness of the model, reduce the risk of failure, and thus enhance the reliability and practicality of the model.

[0085] like Figure 2As shown, some embodiments of the present invention also provide a fault criterion model construction system based on an AI target detection model, which is applicable to the fault criterion model construction method based on an AI target detection model in the above embodiments of the invention, including: a data acquisition module, a first data processing module, and a second data processing module. The data acquisition module is electrically connected to the running database of the AI ​​target detection model. The data acquisition module is used to acquire historical running data of the AI ​​target detection model to be built. The first data processing module is electrically connected to the data acquisition module. The first data processing module is used to acquire historical fault data from the historical running data of the AI ​​target detection model to be built, and to clean the historical fault data. The first data processing module is also used to acquire the cleaned historical fault data and establish a data fault set. The second data processing module is electrically connected to the first data processing module. The second data processing module is used to acquire running parameters from the data fault set, extract key functional parameters of the running parameters, and establish a functional model for fitting experiments. The second data processing module is also used to acquire the functional dataset of the functional model in the fitting experiments and establish a fault criterion model for the AI ​​target detection model. The second data processing module is also used to process the functional dataset of the functional model and obtain the optimal functional data based on the relationships between the functional data in the functional dataset of the functional model after data processing. The second data processing module is also used to obtain the functional model of the optimal functional data and determine the fault criterion model of the AI ​​target detection model based on the functional model of the optimal functional data.

[0086] It is understood that the fault criterion model construction system based on AI target detection model in the embodiments of the present invention is applicable to the fault criterion model construction method based on AI target detection model in the above embodiments of the invention. Therefore, the fault criterion model construction system based on AI target detection model and the fault criterion model construction method based on AI target detection model in the above embodiments of the invention have the same beneficial effects, and therefore will not be described again.

[0087] The above is only one embodiment of the present invention, but it cannot be used to limit the scope of the present invention. Any structural changes made based on the present invention, as long as they do not lose the essence of the present invention, should be considered to fall within the protection scope of the present invention and be subject to its restrictions.

[0088] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process and related descriptions of the system described above can be found in the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0089] It should be noted that the system provided in the above embodiments is only illustrated by the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the modules or steps in the embodiments of the present invention can be further decomposed or combined. For example, the modules in the above embodiments can be merged into one module, or further divided into multiple sub-modules to complete all or part of the functions described above. The names of the modules and steps involved in the embodiments of the present invention are only for distinguishing the various modules or steps and are not considered as an improper limitation of the present invention.

[0090] Those skilled in the art will recognize that the modules and method steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. The programs corresponding to the software modules and method steps can be placed in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disks, removable disks, CD-ROMs, or any other form of storage medium known in the art. To clearly illustrate the interchangeability of electronic hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in electronic hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the invention.

[0091] The term "comprising" or any other similar term is intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus / device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent in such process, method, article, or apparatus / device.

[0092] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after such changes or substitutions will all fall within the scope of protection of the present invention.

[0093] The above are merely preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention.

Claims

1. A method for constructing a fault criterion model based on an AI target detection model, characterized in that, include: Obtain historical running data of the AI ​​object detection model to be built, wherein the AI ​​object detection model is used for image and video processing; Obtain historical fault data from the historical running data of the AI ​​target detection model to be built, and perform data cleaning on the historical fault data; Obtain the historical fault data after data cleaning and establish a data fault set; Obtain the operating parameters from the data fault set, extract the key functional parameters of the operating parameters, and establish a functional model for fitting experiments; Obtain the functional dataset of the functional model in the fitting experiment, and establish a fault criterion model for the AI ​​object detection model, wherein, The functional dataset of the functional model is processed, and the optimal functional data is obtained based on the relationship between the functional data in the functional dataset of the functional model after data processing. Obtain the functional model of the optimal functional data, and determine the functional model of the optimal functional data as the fault criterion model of the AI ​​target detection model; When processing the functional dataset of the functional model and obtaining the optimal functional data based on the relationships between the functional data in the processed functional dataset, the process includes: Obtain the functional data of each function in the functional dataset of the functional model in the fitting experiment, and remove duplicate data in each function data in the functional dataset; Clustering was used to exclude outlier data in each function of the functional dataset after removing duplicate data, and the value of k in k-means clustering was solved by using the idea of ​​quadratic clustering. The secondary clustering includes: Obtain the functional data and the variation range between each functional data in the functional dataset, and use the direct clustering method to obtain the clustering result as {C1, C2, ..., Ck}, and denote the cluster center of each cluster as {W1, W2, ..., Wk}; obtain the cluster center Wmax with the largest value in {W1, W2, ..., Wk}, and the cluster Cmax corresponding to the largest cluster center Wmax; obtain the distance di between Wmax and each cluster center in {W1, W2, ..., Wk} other than Wmax, and set di = Wmax - Wi; according to the relationship between the obtained distance di of each cluster center and the preset threshold T, obtain the optimal functional data in the functional dataset; wherein, When di≥T, it is determined that the distance di between the cluster center Wi and the maximum cluster center Wmax is greater than or equal to the preset threshold T. The cluster center Wi and the maximum cluster center Wmax have a large difference in feature space, and the cluster Ci corresponding to the cluster center Wi is excluded. When di < T, it is determined that the distance di between the cluster center Wi and the maximum cluster center Wmax is less than the preset threshold T, the cluster center Wi and the maximum cluster center Wmax have little difference in feature space, and the cluster Ci corresponding to the cluster center Wi is merged with the maximum cluster Cmax into a new cluster; Obtain the final number of clusters retained, i.e., the value of k in the k-means algorithm. The retained clusters are the optimal functional data in the functional dataset. When acquiring the operating parameters from the data fault set, extracting the key functional parameters of the operating parameters, and establishing a functional model for fitting experiments, the process includes: Obtain non-fault data from the historical operating data of the AI ​​object detection model to be built, and obtain each operating parameter from the non-fault data; Obtain the minimum and maximum values ​​of each operating parameter in the non-fault data, and establish the operating parameter thresholds of the AI ​​target detection model based on the minimum and maximum values ​​of each operating parameter in the non-fault data; Obtain each operating parameter in the data fault set, compare each operating parameter in the data fault set with the operating parameter threshold of the AI ​​target detection model, and obtain the operating parameters in the data fault set that are not within the operating parameter threshold of the AI ​​target detection model; Each operating parameter in the data fault set that is not within the operating parameter threshold of the AI ​​target detection model is identified as the operating highly relevant parameter in the data fault set.

2. The method for constructing a fault criterion model based on an AI target detection model as described in claim 1, characterized in that, When acquiring historical fault data from the historical running data of the AI ​​object detection model to be built, and cleaning the historical fault data, the process includes: Obtain historical fault data from the historical operation data of the AI ​​target detection model; Duplicate data in the historical fault data of the AI ​​target detection model are removed and cleaned up; Obtain historical fault data of the AI ​​target detection model after removing and cleaning up duplicate data, and remove erroneous judgment data from the historical fault data of the AI ​​target detection model after removing and cleaning up duplicate data based on the Z-Score method.

3. The method for constructing a fault criterion model based on an AI target detection model as described in claim 2, characterized in that, When removing erroneous judgment data from the historical fault data of the AI ​​object detection model based on the Z-Score method to eliminate duplicate data, the following steps are included: Obtain each feature data point E from the historical fault data of the AI ​​object detection model after removing and cleaning up duplicate data, and set it as... And according to the formula Obtain the Z-Score value ΔE for each feature data point E, and set ΔE as... ; In the formula: Z is the Z-Score of each feature data point E, X is the actual value of each feature data point E in the historical fault data of the AI ​​target detection model, μ is the average value of each feature data point E in the historical fault data of the AI ​​target detection model, and σ is the standard deviation of each feature data point E in the historical fault data of the AI ​​target detection model. A first preset Z-Score value ΔE1 and a second preset Z-Score value ΔE2 are set in advance. Based on the relationship between the Z-Score value ΔE of the feature data point E and the first preset Z-Score value ΔE1 and the second preset Z-Score value ΔE2, it is determined whether the Z-Score value ΔE of the feature data point E is an outlier. When ΔE1≤ΔE<ΔE2, the Z-Score value ΔE of the feature data point E is determined to be a normal value; When ΔE < ΔE1, the Z-Score value ΔE of the feature data point E is determined to be an outlier, and the feature data point E is removed. When ΔE > ΔE2, the Z-Score value ΔE of the feature data point E is determined to be an outlier, and the feature data point E is removed.

4. The method for constructing a fault criterion model based on an AI target detection model as described in claim 3, characterized in that, When acquiring the historical fault data after data cleaning and establishing the data fault set, the process includes: Obtain each fault data and the data features in the historical fault data of the AI ​​target detection model after removing erroneous judgment data; Clustering is performed based on the data characteristics of each fault data, and a fault dataset is established based on each fault data under each cluster.

5. The method for constructing a fault criterion model based on an AI target detection model as described in claim 4, characterized in that, When determining the operating parameters in the data fault set that are not within the operating parameter thresholds of the AI ​​object detection model as the operating parameters with high correlation in the data fault set, this includes: The weight coefficients of each highly correlated operational parameter in the fault data set are obtained based on the linear regression calculation method. Based on the relationship between the weight coefficients of each highly correlated operational parameter, the parameter with the largest absolute value of its weight coefficient is determined as the key functional parameter of the data fault set.

6. A fault criterion model construction system based on an AI target detection model, applicable to the fault criterion model construction method based on an AI target detection model as described in any one of claims 1-5, characterized in that, include: The data acquisition module is electrically connected to the running database of the AI ​​target detection model. The data acquisition module is used to acquire the historical running data of the AI ​​target detection model to be built. The first data processing module is electrically connected to the data acquisition module. The first data processing module is used to acquire historical fault data from the historical running data of the AI ​​target detection model to be built, and to clean the historical fault data. The first data processing module is also used to acquire the cleaned historical fault data and establish the data fault set. The second data processing module, electrically connected to the first data processing module, is used to acquire operating parameters from the data fault set, extract key functional parameters from the operating parameters, and establish a functional model for fitting experiments. The second data processing module is also used to acquire the functional dataset of the functional model in the fitting experiments and establish a fault criterion model for the AI ​​object detection model. The second data processing module is further configured to process the functional dataset of the functional model, and obtain the optimal functional data based on the relationship between the functional data in the functional dataset of the functional model after data processing. The second data processing module is further configured to obtain the functional model of the optimal functional data, and determine the functional model of the optimal functional data as the fault criterion model of the AI ​​target detection model.

Images