Method for detecting components of a server and electronic device

Abstract

The application discloses a kind of server component detection method and electronic equipment, at least one dimension of mechanical data on target component is predicted based on mechanical data prediction model and the historical mechanical data of target component, obtain the mechanical prediction data of target component;According to mechanical prediction data and mechanical anomaly evaluation model, the abnormal prediction score corresponding to target component is obtained;When abnormal prediction score is greater than or equal to the preset failure score corresponding to target component, output component failure warning;When abnormal prediction score is less than preset failure score, and according to abnormal prediction score and preset score interval determine that target component exists abnormal risk, at least one dimension of real-time mechanical data of target component is collected by sensor configured on target component, and at least one dimension of real-time mechanical data corresponding actual anomaly score is obtained by mechanical anomaly evaluation model;According to actual anomaly score, the abnormal detection result of target component is output.

Description

Technical Field

[0001] This application relates to the field of server technology, and in particular to methods for testing server components and electronic devices. Background Technology

[0002] Servers integrate numerous core components such as processors, network cards, hard drives, and memory. These components are highly sensitive to mechanical vibrations, structural displacements, deformation stresses, and mechanical noise during transportation and operation, and are prone to malfunctions that can affect the overall operation of the server. Currently, monitoring of server components primarily involves real-time monitoring of component operating parameters and checking logs for errors to determine faults. This requires continuous collection of operating parameters, is labor-intensive, and can only detect anomalies when they occur. Summary of the Invention

[0003] This application provides a method and electronic device for detecting server components, in order to at least solve the problem that the current related technologies mainly rely on real-time monitoring of component operating parameters and checking for errors in logs to determine faults. This requires continuous collection of operating parameters, which is labor-intensive, and can only be detected when an anomaly occurs.

[0004] This application provides a method for detecting components of a server, including: Based on the mechanical data prediction model and the historical mechanical data of the target component, the mechanical data of at least one dimension of the target component is predicted to obtain the mechanical prediction data of the target component. The mechanical data prediction model is obtained by iterative optimization of the historical measured data and historical prediction data of the first test component. Based on the mechanical prediction data and the mechanical anomaly assessment model, the anomaly prediction score corresponding to the target component is obtained; When the abnormal prediction score is greater than or equal to the preset failure score corresponding to the target component, a component failure warning is output. When the anomaly prediction score is less than the preset failure score, and it is determined that the target component has an anomaly risk based on the anomaly prediction score and the preset score range, the target component is equipped with sensors to collect real-time mechanical data of at least one dimension of the target component, and the actual anomaly score corresponding to the real-time mechanical data of at least one dimension is obtained through the mechanical anomaly assessment model. Based on the actual anomaly score, the anomaly detection results of the target component are output.

[0005] This application also provides a component testing device for a server, comprising: The processing module is used to predict the mechanical data of at least one dimension of the target component based on the mechanical data prediction model and the historical mechanical data of the target component, so as to obtain the mechanical prediction data of the target component. The mechanical data prediction model is obtained by iterative optimization of the historical measured data and historical prediction data of the first test component. The processing module is also used to obtain the anomaly prediction score corresponding to the target component based on the mechanical prediction data and the mechanical anomaly assessment model; The processing module is also used to output a component failure warning when the abnormal prediction score is greater than or equal to the preset failure score corresponding to the target component. The acquisition module is used to collect real-time mechanical data of at least one dimension of the target component through sensors configured on the target component when the anomaly prediction score is less than the preset failure score and the target component is determined to have an anomaly risk based on the anomaly prediction score and the preset score range. The processing module is also used to obtain the actual anomaly score corresponding to real-time mechanical data in at least one dimension through the mechanical anomaly assessment model. The processing module is also used to output the anomaly detection results of the target component based on the actual anomaly score.

[0006] This application also provides an electronic device, including: a memory for storing a computer program; and a processor for executing the computer program to implement the component detection method of any of the above-described servers.

[0007] This application also provides a computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, it implements the steps of the component detection method of any of the above-described servers.

[0008] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of any of the above-described server component detection methods.

[0009] This application utilizes a mechanical data prediction model and historical mechanical data of a target component to predict mechanical data for at least one dimension of the target component, thereby obtaining mechanical prediction data for the target component. The mechanical data prediction model is obtained through iterative optimization using historical measured data and historical prediction data from a first test component. Based on the mechanical prediction data and a mechanical anomaly assessment model, an anomaly prediction score corresponding to the target component is obtained. When the anomaly prediction score is greater than or equal to a preset failure score corresponding to the target component, a component failure warning is output. When the anomaly prediction score is less than the preset failure score, and the target component is determined to have an anomaly risk based on the interval between the anomaly prediction score and the preset score, real-time mechanical data for at least one dimension of the target component is collected using sensors configured on the target component, and the actual anomaly score corresponding to the real-time mechanical data for at least one dimension is obtained through the mechanical anomaly assessment model. Based on the actual anomaly score, the anomaly detection result of the target component is output. In this scheme, a pre-trained mechanical data prediction model is used to predict the mechanical data of the component. The model then calculates an anomaly prediction score. If the anomaly prediction score exceeds the failure threshold, an early warning is issued directly. If the anomaly prediction score does not exceed the failure threshold, but a potential anomaly risk is still detected, then the actual mechanical data of the component can be collected by sensors and anomaly score monitoring can be performed. This eliminates the need for real-time collection of component operating data. Fault determination is only performed based on actual data when an anomaly is suspected through prediction data. This reduces the amount of data collected and allows for early prediction of component failures, improving the efficiency and accuracy of component failure detection. Attached Figure Description

[0010] To more clearly illustrate the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 The flowchart of a server component detection method provided in this application embodiment Figure 1 ; Figure 2 The flowchart of a server component detection method provided in this application embodiment Figure 2 ; Figure 3 A structural diagram of a server component detection device provided in an embodiment of this application; Figure 4 This is a structural diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0012] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the protection scope of this application.

[0013] It should be noted that, in the description of this application, 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 a process, method, article, or apparatus. The terms "first," "second," etc., in this application are used to distinguish similar objects and are not used to describe a specific order or sequence.

[0014] With the development of artificial intelligence (AI) technology, the demand for computing power is increasing, and AI servers have also experienced rapid development. AI servers integrate a large number of core components such as graphics processing unit (GPU) cards, high-speed serial computer expansion bus standard (Peripheral Component Interconnect Express, PCIe) cards (network cards, Redundant Array of Independent Disks (RAID) cards, hard drives, and memory). These components are extremely sensitive to operating parameters such as mechanical vibration, structural displacement, deformation stress, and mechanical noise during transportation and operation. Abnormal mechanical conditions can easily lead to poor contact, component loss, link loss, and abnormal operation, directly affecting the transportation and operational reliability of the entire server.

[0015] Currently, server mechanical reliability testing in the industry generally adopts a traditional single vibration testing device combined with manual judgment. This approach cannot simultaneously monitor related multi-dimensional mechanical parameters such as displacement, strain, and mechanical noise, making it difficult to fully reproduce the actual mechanical stress and operating conditions of the server's core components. Furthermore, it cannot accurately identify potential hidden risks near the vibration failure threshold, resulting in poor fault warning timeliness. Test results can only be judged by whether there is damage to the mechanical structure after testing and whether there are errors in the test logs. Problems can only be discovered after mechanical failure triggers hardware anomalies, making it impossible to achieve early prediction and prevention.

[0016] Meanwhile, current testing technologies lack the ability to analyze historical test data and have self-optimization mechanisms for models, making them unsuitable for diverse testing scenarios involving different server models, testing conditions, and hardware assembly processes. Furthermore, existing technologies cannot establish standardized judgment criteria based on historical test data; when vibration exceeds limits, test failures can only be manually recorded, failing to utilize historical test data effectively. They also cannot automatically match similar failure cases to generate targeted structural optimization solutions. In summary, the current level of intelligence, precision, and adaptability in detecting mechanical faults in server components is severely inadequate.

[0017] To enable those skilled in the art to better understand the present application, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0018] like Figure 1 As shown, Figure 1 A flowchart of a server component detection method provided for embodiments of this application is shown. The server component detection method may include the following steps: 101. Based on the mechanical data prediction model and the historical mechanical data of the target component, predict the mechanical data of at least one dimension of the target component to obtain the mechanical prediction data of the target component.

[0019] It should be noted that the mechanical data prediction model can be obtained by iterative optimization using historical measured data and historical prediction data of the first test component. The first test component can be any component in the server with a known operating status, used as sample data for model training. The historical measured data can be data actually collected by various sensors on the first test component, while the historical prediction data can be data predicted by the mechanical data prediction model for the first test component. Here, the mechanical data prediction model can be understood as a model that is constantly learning, optimizing and updating. Initially, a currently common open-source data prediction model can be used as the initial model, and then the initial model can be iteratively optimized using a large amount of historical measured data and historical prediction data of the first test component to obtain the mechanical data prediction model.

[0020] In this embodiment of the application, the mechanical data of a target component can be predicted using a mechanical data prediction model. The target component can be any component in the server that needs to be fault detected. During the prediction, the historical mechanical data of the target component can be referenced to predict the mechanical data of the target component. The mechanical data can include at least one dimension. It is understood that during the operation and transportation of the target component, various mechanical movements will affect the health status of the target component, such as component vibration, displacement, external environmental noise, strain, etc., thereby affecting the target component from at least one dimension. Therefore, the mechanical data of the target component in at least one dimension can be predicted to obtain the mechanical prediction data of the target component. The mechanical prediction data can naturally also include prediction data in at least one dimension.

[0021] 102. Based on the mechanical prediction data and the mechanical anomaly assessment model, obtain the anomaly prediction score corresponding to the target component.

[0022] In this embodiment of the application, after obtaining the mechanical prediction data of the target component, it is necessary to determine whether the target component currently has an anomaly and what the specific degree of the anomaly is based on the mechanical prediction data. In order to adapt to various server core components and establish a unified evaluation scale, anomaly evaluation can be performed through a scoring mode. Therefore, the mechanical prediction data can be analyzed based on the mechanical anomaly evaluation model to obtain the anomaly prediction score corresponding to the target component.

[0023] It should be noted that this mechanical anomaly assessment model can also be a model pre-trained using test components. The mechanical data of the component is input into this model, which analyzes and evaluates the data to obtain the corresponding anomaly prediction score for the component. This anomaly prediction score can be a specific numerical value, or a percentage score, ranging from 0 to 100. A higher anomaly prediction score indicates a higher degree of anomaly, suggesting that the component's health condition is very poor, possibly indicating failure or impending failure. Conversely, a lower anomaly prediction score indicates a lower degree of anomaly, suggesting that the component's health condition is relatively high, in a normal state, or with only minor anomalies.

[0024] 103. When the abnormal prediction score is greater than or equal to the preset failure score corresponding to the target component, output a component failure warning.

[0025] In this embodiment, after obtaining the anomaly prediction score, it can be compared with the corresponding score thresholds at each level for the target component. Since different components may have varying degrees of tolerance for anomalies, some components will experience significant disruptions to normal operation even with minor anomalies, resulting in relatively lower score thresholds for each level. Conversely, some components may have minimal impact on operation due to minor anomalies, leading to relatively higher score thresholds for each level. Therefore, the score threshold for the target component can be determined. Generally, component failure is a gradual process. Initially, there may be minor faults that have some impact on operation, but the effect is minimal. If left unaddressed, the failure will worsen until the component fails and cannot continue operating. Therefore, for the target component, the preset failure score is the highest threshold, representing the most severe failure. If the anomaly prediction score is greater than or equal to the preset failure score for the target component, it may be considered that the target component has failed and is completely inoperable. In this case, a component failure warning can be directly output to remind staff to repair or replace the target component.

[0026] 104. When the anomaly prediction score is less than the preset failure score, and the target component is determined to have an anomaly risk based on the anomaly prediction score and the preset score range, the target component is equipped with sensors to collect real-time mechanical data of at least one dimension of the target component, and the actual anomaly score corresponding to the real-time mechanical data of at least one dimension is obtained through the mechanical anomaly assessment model.

[0027] In this embodiment, if the anomaly prediction score is less than the preset failure score, it indicates that the anomaly of the target component has not yet reached the level of failure. In this case, it is possible to further determine whether there is an anomaly risk in the target component based on the anomaly prediction score and the preset score range. Here, the anomaly risk can be an anomaly level that is lower than failure. If there is an anomaly risk, it is also necessary to handle it. However, since the anomaly prediction score is obtained through a mechanical data prediction model, it may not be completely accurate. When it is determined that there is an anomaly risk, the anomaly can be further judged based on the actual situation of the target component. Therefore, at least one dimension of real-time mechanical data of the target component can be collected through the sensors configured on the target component. This real-time mechanical data is the actual mechanical data of the current target component. Then, the real-time mechanical data can also be analyzed and scored by the mechanical anomaly assessment model to obtain the actual anomaly score.

[0028] It should be noted that the sensors configured on the target component can be sensors of at least one dimension, such as vibration sensors for collecting vibration data, displacement sensors for collecting displacement data, noise sensors for collecting noise data, strain sensors for collecting strain force data, etc. Each component can be configured with at least one sensor to collect data according to its own mechanical requirements. For the target component, each sensor can collect real-time mechanical data of one corresponding dimension, thereby obtaining real-time mechanical data of at least one dimension. The mechanical anomaly assessment model can fuse the real-time mechanical data of at least one dimension and analyze them together to obtain the actual anomaly score. The actual anomaly score corresponding to the target component has only one score, which is obtained after fusing the data of at least one dimension.

[0029] 105. Based on the actual anomaly score, output the anomaly detection results of the target component.

[0030] In this embodiment of the application, the actual anomaly score can be considered as the accurate score of the current target component, which can accurately reflect the degree of anomaly of the target component. Therefore, the anomaly detection result of the target component can be determined based on the actual anomaly score, and the anomaly detection result can be output.

[0031] In this embodiment, a trained mechanical data prediction model is used to predict the mechanical data of the component. An anomaly prediction score is calculated by the model. If the anomaly prediction score exceeds the failure threshold, an early warning is directly output. If the anomaly prediction score does not exceed the failure threshold, but a potential anomaly risk is still detected, the actual mechanical data of the component can be collected by sensors and anomaly score monitoring can be performed. This eliminates the need to collect component operating data in real time. Fault determination is only performed based on actual data when an anomaly is suspected based on the predicted data. This reduces the amount of data collected and allows for early prediction of component faults, improving the efficiency and accuracy of component fault detection.

[0032] like Figure 2 As shown, Figure 2 Another flowchart of a server component detection method provided for embodiments of this application, the server component detection method may further include the following steps: 201. Obtain the measured data of the first test component through the sensor configured on the first test component.

[0033] In this embodiment of the application, the first test component is also a component in the server. Therefore, the first test component is also equipped with a sensor, which can collect the measured data of the first test component. The measured data may include mechanical data in at least one dimension.

[0034] 202. Using the initial mechanical data prediction model, the first test component is predicted to obtain the predicted data of the first test component.

[0035] In this embodiment of the application, the initial mechanical data prediction model can be an untrained general prediction model or a mechanical data prediction model that has been trained and needs to be iteratively optimized. In this case, the initial mechanical data prediction model can be used to predict the first test component. Specifically, the prediction can be achieved by monitoring the operation of the first test component, thereby obtaining the prediction data of the first test component. The prediction data can also include mechanical data of at least one dimension.

[0036] 203. Based on the difference between the measured data and the predicted data, the initial mechanical data prediction model is iteratively optimized to obtain the mechanical data prediction model.

[0037] In this embodiment, after obtaining the measured data and the predicted data, since the measured data is actually collected by the sensor, it can be considered as the currently accurate data. The process of iteratively optimizing the initial mechanical data prediction model is to make the predicted data continuously approach the accurate measured data. Therefore, the initial mechanical data prediction model can be iteratively optimized based on the difference between the measured data and the predicted data. Specifically, the model parameters of the initial mechanical data prediction model can be updated based on the difference between the measured data and the predicted data. Then, data prediction and data collection can continue. If the difference between the measured data and the predicted data decreases at this time, it means that the model parameter update just performed is in the right direction. Then, the model parameters can be updated again, and the above steps can be iterated until the final predicted data is very close to the measured data, or the predicted data is directly equal to the measured data. At this point, the initial mechanical data prediction model can be considered to have been trained successfully, and the mechanical data prediction model is obtained.

[0038] In some embodiments, the initial mechanical data prediction model is iteratively optimized based on the difference between the measured data and the predicted data to obtain the mechanical data prediction model. Specifically, this may include: when the difference between the measured data and the predicted data is greater than or equal to the predicted data difference threshold, adjusting the model parameters of the initial mechanical data prediction model based on the difference between the measured data and the predicted data, and continuing to calculate the difference between the measured data and the predicted data until it is detected that the difference between the measured data and the predicted data is less than the predicted data difference threshold, thereby obtaining the mechanical data prediction model.

[0039] It should be noted that the mechanical data prediction model is used to predict the mechanical data of components. Therefore, the evaluation criterion for the mechanical data prediction model is that the predicted data and the actual measured data are the same, or the difference is within a very small range. Thus, during the iterative optimization of the initial mechanical data prediction model, the difference between the measured data and the predicted data can be calculated. This difference is then compared with a predicted data difference threshold, which can be a pre-set threshold. If the difference between the measured data and the predicted data is less than this threshold, the predicted data can be considered quite close to the measured data, indicating that the model optimization is complete. Conversely, if the difference between the measured data and the predicted data is greater than or equal to the predicted data difference threshold, it indicates that the initial mechanical data... The prediction model still needs adjustment. Therefore, the model parameters of the initial mechanical data prediction model can be adjusted based on the difference. Then, the initial mechanical data prediction model with adjusted parameters continues to predict the mechanical data of the first test component, obtaining the predicted data. Simultaneously, the measured data of the first test component is collected through sensors, and the difference is calculated again. This process is repeated. If the difference between the measured data and the predicted data is greater than or equal to the predicted data difference threshold, further optimization is needed, and the model parameters of the initial mechanical data prediction model are adjusted. Finally, when the difference between the predicted data and the measured data obtained from a certain adjusted initial mechanical data prediction model is less than the predicted data difference threshold, the optimization is considered complete, and the mechanical data prediction model is obtained.

[0040] In this embodiment, since the mechanical data prediction model is used to predict the mechanical data of a component, the predicted data obtained by the mechanical data prediction model needs to be as close as possible to the measured data, or even equal to the measured data. Therefore, the initial mechanical data prediction model can be optimized in advance by repeatedly using the difference between the predicted data and the measured data to ensure that the difference between the predicted data and the measured data of the final mechanical data prediction model is less than the prediction data difference threshold, thereby improving the accuracy of mechanical data prediction and also improving the accuracy of component fault detection.

[0041] In some embodiments, after the mechanical data prediction model is trained and put into practical use, the actual state of each component will change at any time due to various factors affecting the server during production, transportation, and operation. Therefore, the prediction accuracy of the mechanical data prediction model will also be affected. In actual use, it is still necessary to update the mechanical data prediction model at regular intervals. Therefore, according to a preset cycle, mechanical verification data of at least one dimension of the target component at the first moment can be collected by sensors configured on the target component; mechanical prediction data of at least one dimension of the target component at the first moment can be determined by the mechanical data prediction model; and the mechanical data prediction model can be updated according to the difference between the mechanical verification data and the mechanical prediction data.

[0042] It should be noted that the preset period can be a user-defined duration, such as 5 days, 10 days, or one month. The process of updating the mechanical data prediction model is similar to the training process described above. It is also updated based on the difference between the actual collected data and the predicted data. Therefore, at least one dimension of the mechanical verification data of the target component can be collected using sensors configured on the target component at the first moment. Then, the mechanical data prediction model is used to determine at least one dimension of the mechanical prediction data of the target component at the first moment. After obtaining the mechanical verification data and the mechanical prediction data, if the difference between the mechanical verification data and the mechanical prediction data is still less than the prediction data difference threshold, it indicates that the current data prediction of the mechanical data prediction model is relatively accurate, and an update is not necessary. However, if the difference between the mechanical verification data and the mechanical prediction data is greater than or equal to the prediction data difference threshold, it indicates that the current data prediction of the mechanical data prediction model is starting to deviate, and an update of the mechanical data prediction model is required. This update process also involves adjusting the model parameters of the mechanical data prediction model and then updating iteratively based on the difference between the collected mechanical verification data and the mechanical prediction data.

[0043] In this embodiment of the application, since the state of the components in the server changes at any time during operation, in order to prevent the mechanical data prediction model from having large deviations in data prediction, the mechanical data prediction model can be periodically verified. If it is found that the difference between the actual collected mechanical verification data and the mechanical prediction data is greater than or equal to the prediction data difference threshold, the mechanical data prediction model is updated to ensure the accuracy of the mechanical data prediction model as much as possible.

[0044] 204. Based on the mechanical data prediction model and the historical mechanical data of the target component, predict the mechanical data of at least one dimension of the target component to obtain the mechanical prediction data of the target component.

[0045] 205. Based on the mechanical prediction data of at least one dimension, extract the prediction feature vector corresponding to at least one dimension respectively.

[0046] In this embodiment, during the process of determining the anomaly prediction score of a target component based on mechanical prediction data of at least one dimension, a prediction feature vector corresponding to at least one dimension can be extracted from the mechanical prediction data of at least one dimension. Here, specific core features can be extracted for different dimensions. For example, for vibration dimension data, time-domain peak values ​​and frequency-domain resonant frequency band energy features can be extracted; for displacement dimension data, dynamic offset and deformation fluctuation period features can be extracted; for noise dimension data, noise values ​​and abnormal spikes / whistles can be extracted; and for strain dimension data, maximum stress values ​​and fatigue load coefficient features can be extracted. In other words, for the target component, the mechanical data of the dimension configured is collected based on the sensors used, thereby extracting the corresponding prediction features and forming the prediction feature vector.

[0047] 206. Based on the predicted feature vector corresponding to at least one dimension and the mechanical anomaly assessment model, obtain the anomaly prediction score of the target component.

[0048] In this embodiment of the application, after obtaining the predicted feature vector corresponding to at least one dimension, the predicted feature vector corresponding to at least one dimension can be input into the mechanical anomaly assessment model so that the mechanical anomaly assessment model can analyze and evaluate the predicted feature vector corresponding to at least one dimension to obtain the anomaly prediction score of the target component.

[0049] In some embodiments, an anomaly prediction score for a target component is obtained based on the predicted feature vector corresponding to at least one dimension and the mechanical anomaly assessment model. Specifically, this may include: determining the dimension weights corresponding to at least one dimension based on the risk detection requirements of the target component; fusing the mechanical prediction data of at least one dimension based on the dimension weights to obtain a predicted feature vector; and obtaining the anomaly prediction score for the target component based on the predicted feature vector and the mechanical anomaly assessment model.

[0050] It should be noted that since mechanical prediction data can include at least one dimension, if multiple dimensions exist, the mechanical prediction data for these dimensions needs to be fused. During the fusion process, the weights corresponding to each dimension need to be considered. Since some dimensions have a greater impact on the component and are more important during anomaly assessment, their corresponding weights are higher. Conversely, some dimensions have a smaller impact and are less important during anomaly assessment, so their corresponding weights are lower. Therefore, based on the risk detection requirements of the target component, the weights corresponding to at least one dimension can be determined. These risk detection requirements primarily indicate which dimension of the target component's risk is of greater concern. After determining the weights corresponding to each dimension, the mechanical prediction data for at least one dimension can be fused based on these weights to obtain a prediction feature vector. Then, the prediction feature vector can be input into the mechanical anomaly assessment model to obtain the anomaly prediction score for the target component.

[0051] In some embodiments, before fusing mechanical prediction data of at least one dimension, the mechanical data of each dimension have different dimensions, that is, the value range of mechanical data of each dimension is different, which will affect the fusion of data of each dimension. Therefore, the mechanical prediction data of each dimension can be normalized first to map the mechanical prediction data of each dimension to the same data range, and then weighted fusion can be performed to eliminate the difference in data dimensions.

[0052] In this embodiment, different data features corresponding to each dimension can be extracted separately, and then weighted and fused to reflect the data of the dimensions of interest for fault detection as much as possible. Then, the fused predicted feature vector is input into the mechanical anomaly assessment model for anomaly assessment, which effectively improves the accuracy and reliability of anomaly assessment.

[0053] In some embodiments, the mechanical anomaly assessment model is similar to the mechanical data prediction model, both of which are models trained using sample data from test components. Specifically, mechanical sample data and fault detection data of multiple second test components during operation can be obtained; based on the mechanical sample data and fault detection data, the initial mechanical anomaly assessment model is trained to obtain the mechanical anomaly assessment model.

[0054] It should be noted that during the training of the mechanical anomaly assessment model, real mechanical data samples of test components in the server can be used for learning. This mechanical anomaly assessment model can be based on a general industrial multimodal pre-trained large model, and then fine-tuned for the lightweight domain of server mechanical reliability testing. It adopts a dual mode of LoRA low-rank matrix fine-tuning + reliability mechanical testing-specific instruction fine-tuning, without modifying the underlying general basic parameters of the large model, only optimizing multimodal feature fusion and fault risk inference, avoiding model overfitting and reducing training costs.

[0055] The general-purpose industrial multimodal pre-trained large model can be an open-source multimodal base model adapted to industrial scenarios. The core modules of this model can include: a visual encoder, which uses a ViT structure to complete feature extraction of multimodal visual data such as server vibration spectrum diagrams, stress cloud diagrams, and assembly real-world images; a text encoder, which completes semantic feature encoding of reliability test standards, fault description texts, and test process instructions; and a post-interaction feature fusion module, which is used to achieve fine-grained alignment and matching of visual features and text features. The three are connected in series through a cross-modal attention layer, which has the basic understanding and retrieval capabilities of multimodal information in general industrial scenarios.

[0056] Based on this, lightweight fine-tuning can be carried out specifically for the server mechanical reliability testing scenario required in this application. The underlying general basic parameters of the above-mentioned large model are not changed throughout the process. Only the two local modules in the later interactive feature fusion module, the multimodal feature fusion layer and the fault risk inference layer, are optimized using a dual mode of LoRA low-rank matrix fine-tuning and reliability mechanical testing-specific instruction fine-tuning: by inserting a low-rank adaptation matrix into the feature fusion layer, only the low-rank sub-matrix, which accounts for 0.8% of the total parameters, is trained. At the same time, a test sample instruction set of tens of thousands of samples containing specific fault labels such as server shaft wear, sheet metal deformation, and abnormal fan vibration is input to complete the targeted fine-tuning. This avoids the model overfitting problem caused by full parameter training and reduces the overall training cost by more than 70%, finally obtaining a mechanical anomaly assessment model adapted to the server mechanical reliability testing scenario.

[0057] In some embodiments, during model training using multiple second test components as samples, mechanical sample data during operation can be acquired. This mechanical sample data can be accurate mechanical data collected by sensors. Specifically, the mechanical sample data may include: specification parameters of the second test components, normal mechanical data, faulty mechanical data, critical anomaly data, and corresponding fault labels and anomaly scores. Furthermore, fault detection data can be acquired, which can be understood as indicating faults present during the operation of the second test components. Using this mechanical sample data and fault detection data, an initial mechanical anomaly assessment model is trained. The mechanical sample data is determined as the input to the initial mechanical anomaly assessment model, and the fault detection data is determined as the output of the initial mechanical anomaly assessment model, thereby training the initial mechanical anomaly assessment model to obtain the mechanical anomaly assessment model.

[0058] Furthermore, based on mechanical sample data and fault detection data, the initial mechanical anomaly assessment model is trained to obtain a mechanical anomaly assessment model, which may specifically include: Mechanical sample data and fault detection data are divided into training sample set and validation sample set. The initial mechanical anomaly assessment model is trained using the training sample set and validated using the validation sample set to obtain the mechanical anomaly assessment model.

[0059] It should be noted that during the training of the initial mechanical anomaly assessment model, training alone is insufficient; the trained model also needs to be validated. Therefore, the mechanical sample data and fault detection data can be divided into a training sample set and a validation sample set, with the training sample set containing more data than the validation sample set. Typically, the training and validation sample sets can be divided in an 8:2 or 7:3 ratio. First, the initial mechanical anomaly assessment model is trained using the training sample set. Then, the trained initial mechanical anomaly assessment model is validated using the validation sample set. If the validation passes, the mechanical anomaly assessment model is obtained. If the validation fails, the initial mechanical anomaly assessment model can be retrained until it passes validation, thus obtaining the final mechanical anomaly assessment model.

[0060] In this embodiment, the mechanical anomaly assessment model is trained in advance using actual mechanical data and actual fault conditions of the test components as samples, and a large amount of sample data is divided into training set and validation set, thereby obtaining an accurate mechanical anomaly assessment model and effectively improving the accuracy of component fault detection.

[0061] 207. When the abnormal prediction score is greater than or equal to the preset failure score corresponding to the target component, output a component failure warning.

[0062] 208. When the anomaly prediction score is less than the preset failure score, determine the target score range in which the anomaly prediction score is located.

[0063] In this embodiment of the application, if the anomaly prediction score is less than the preset failure score, it means that the target component has not yet failed, but it is still impossible to determine whether the target component is normal or has a certain risk of anomaly. Therefore, the target score range in which the anomaly prediction score is located can be further determined. This score range can be determined in advance based on multiple thresholds.

[0064] In some embodiments, when performing fault detection on a component, the detection results typically include: component normal, component failure, and component with abnormal risk. Therefore, in addition to comparing the abnormal prediction score with the preset failure score to determine whether it has failed, it is also necessary to compare it with a preset safety threshold to determine whether it is operating normally. Therefore, when the abnormal prediction score is less than the preset failure score, the target score range in which the abnormal prediction score is located is determined. Specifically, this may include: when the abnormal prediction score is less than the preset failure score, determining whether the abnormal prediction score is greater than the preset safety threshold; and when the abnormal prediction score is greater than the preset safety threshold, determining the target score range in which the abnormal prediction score is located.

[0065] It should be noted that a preset safety threshold can be set for the target component. If the anomaly prediction score is less than or equal to this preset safety threshold, the target component is considered to be operating normally. If the anomaly prediction score is greater than the preset safety threshold, it indicates that the target component is malfunctioning. The preset failure score can be considered the most severe failure threshold. If the anomaly prediction score is very high, even exceeding or equal to the preset safety threshold, the target component can be directly considered to have failed. If the anomaly prediction score is between the preset safety threshold and the preset failure score, the target component is considered to have an anomaly, but not yet to the point of failure. To accurately measure the degree of anomaly, the interval between the preset safety threshold and the preset failure score can be further divided. For example, a first-level alarm threshold, a second-level alarm threshold, etc., can be set to divide the interval between the preset safety threshold and the preset failure score into more intervals, thus classifying the anomalies of the target component into multiple levels. Therefore, the target score interval in which the anomaly prediction score falls can be determined.

[0066] For example, assuming the anomaly prediction score is in the range of 0 to 100, the preset safety threshold can be 60 points, the preset failure score can be 90 points, and the first-level alarm threshold can be set to 70 points and the second-level alarm threshold to 80 points between 60 and 90 points. Therefore, the 0 to 100 can be divided into 5 intervals through the above thresholds. 0 to 60 can be considered as the target component is operating normally, 61 to 70 can be considered as the target component is in first-level anomaly, 71 to 80 can be considered as the target component is in second-level anomaly, 81 to 89 can be considered as the target component is at high risk of anomaly, and 90 to 100 can be considered as the target component has failed.

[0067] 209. When the detection result corresponding to the target score interval is a component abnormality, it is determined that the target component has an abnormal risk.

[0068] In this embodiment of the application, if the detection result corresponding to the target score interval is a component abnormality, and the abnormality prediction score of the target component is exactly in the target score interval, then the target component can be considered to have an abnormal risk.

[0069] In some embodiments, since the target component is considered to have failed when the anomaly prediction score is greater than or equal to the preset failure score, if the anomaly prediction score is less than the preset failure score but very close to it, the target component can be considered to be on the verge of failure, and an early warning can be given. Therefore, when the detection result corresponding to the target score interval is high anomaly risk, the difference between the anomaly prediction score and the preset failure score is determined, and the preset failure score is the upper limit of the score interval corresponding to high anomaly risk. When the difference between the anomaly prediction score and the preset failure score is detected to be less than or equal to the failure difference threshold, it is determined that the target component has a failure risk.

[0070] It should be noted that if the detection result corresponding to the target score interval is high anomaly risk, and the upper limit of the score interval corresponding to high anomaly risk is the preset failure score, that is, high anomaly risk is an anomaly with an anomaly degree lower than failure but closest to failure. Therefore, it can be considered that the target component may be on the verge of failure. Thus, the difference between the anomaly prediction score and the preset failure score can be calculated. If the difference is less than the failure difference threshold, it means that the anomaly prediction score is already very close to the preset failure score. Then, it is considered that the target component has a failure risk.

[0071] Furthermore, after determining that the target component has a failure risk when the difference between the detected abnormal prediction score and the preset failure score is less than or equal to the failure difference threshold, the following may also be included: Based on the anomaly prediction score, identify the anomaly dimensions in the target component that pose a failure risk; based on the anomaly dimensions, determine the corresponding improvement strategies; and output the improvement strategies.

[0072] It should be noted that when a target component is identified as having a failure risk, it means that the target component has not yet failed, but is on the verge of failure. Therefore, the specific abnormal dimension with failure risk can be identified from at least one dimension corresponding to the target component, that is, which aspect of the data is significantly abnormal. This requires judging through the mechanical data corresponding to each specific dimension. After identifying the specific abnormal dimension, an improvement strategy corresponding to the abnormal dimension can be determined and output, so that the staff can maintain the target component according to the improvement strategy.

[0073] In this embodiment of the application, multiple thresholds of different degrees can be set in advance for the abnormality score of the component, thereby dividing multiple score intervals. In this way, when the target component has not yet failed, the degree of abnormality can be determined according to the score interval in which the abnormality prediction score of the target component is located; and when the target component is on the verge of failure, early warning maintenance can be performed in advance, thereby accurately determining the degree of abnormality of the target component, and further monitoring the failure threshold situation, making improvements in advance, and avoiding component failure.

[0074] 210. Collect real-time mechanical data of at least one dimension of the target component using sensors configured on the target component, and extract the real-time feature vector corresponding to at least one dimension from each dimension.

[0075] 211. Input the real-time feature vector corresponding to at least one dimension into the mechanical anomaly assessment model to obtain the actual anomaly score.

[0076] In this embodiment, the specific process of collecting real-time mechanical data of at least one dimension of the target component and then obtaining the actual anomaly score based on the mechanical anomaly assessment model is the same as the specific process of "obtaining the anomaly prediction score corresponding to the target component based on mechanical prediction data and the mechanical anomaly assessment model" mentioned earlier. Both involve collecting real-time mechanical data of at least one dimension, extracting real-time feature vectors from at least one dimension, and then performing feature extraction for each dimension and combining the dimensional weights of each dimension to obtain a real-time feature vector. This real-time feature vector can then be input into the mechanical anomaly assessment model to obtain the actual anomaly score output by the model. Further details are omitted here.

[0077] 212. Based on the actual anomaly score, output the anomaly detection results of the target component.

[0078] 213. When the abnormality detection result indicates that there is an abnormality in the target component, determine the abnormality handling plan for the target component.

[0079] 214. Output exception handling solution.

[0080] In this embodiment of the application, when an abnormality is detected in the target component, some abnormality handling schemes can also be provided. The abnormality handling schemes can indicate how to maintain or improve the target component in the face of the abnormality at this time, and then output the abnormality handling schemes so that the staff can handle the target component according to the abnormality handling schemes.

[0081] It should be noted that steps 214 and 212 can be output synchronously. That is, after obtaining the actual anomaly score, the current anomaly detection result of the target component can be determined based on the score range in which the actual anomaly score is located. Here, you can refer to the process of determining the target score range in which the anomaly prediction score is located as described above. Then, based on the anomaly detection result, the corresponding anomaly handling solution can be obtained. Finally, the anomaly detection result and the anomaly handling solution can be output together.

[0082] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods according to the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method.

[0083] like Figure 3 As shown, embodiments of this application also provide a server component testing apparatus, which may include: The processing module 301 is used to predict the mechanical data of at least one dimension of the target component based on the mechanical data prediction model and the historical mechanical data of the target component, so as to obtain the mechanical prediction data of the target component. The mechanical data prediction model is obtained by iterative optimization of the historical measured data and historical prediction data of the first test component. The processing module 301 is also used to obtain the anomaly prediction score corresponding to the target component based on the mechanical prediction data and the mechanical anomaly assessment model; The processing module 301 is also used to output a component failure warning when the abnormal prediction score is greater than or equal to the preset failure score corresponding to the target component; The acquisition module 302 is used to collect real-time mechanical data of at least one dimension of the target component through sensors configured on the target component when the anomaly prediction score is less than the preset failure score and the target component is determined to have an anomaly risk based on the anomaly prediction score and the preset score range. The processing module 301 is also used to obtain the actual anomaly score corresponding to real-time mechanical data in at least one dimension through the mechanical anomaly assessment model. The processing module 301 is also used to output the anomaly detection results of the target component based on the actual anomaly score.

[0084] In some embodiments, the acquisition module 302 is further configured to acquire measured data of the first test component through a sensor configured on the first test component; The processing module 301 is also used to predict the first test component through the initial mechanical data prediction model to obtain the prediction data of the first test component; The processing module 301 is also used to iteratively optimize the initial mechanical data prediction model based on the difference between the measured data and the predicted data to obtain the mechanical data prediction model.

[0085] In some embodiments, the processing module 301 is specifically used to adjust the model parameters of the initial mechanical data prediction model according to the difference between the measured data and the predicted data when the difference between the measured data and the predicted data is greater than or equal to the prediction data difference threshold, and continue to calculate the difference between the measured data and the predicted data until it is detected that the difference between the measured data and the predicted data is less than the prediction data difference threshold, thereby obtaining the mechanical data prediction model.

[0086] In some embodiments, the acquisition module 302 is further configured to acquire mechanical verification data of at least one dimension of the target component at a first moment by means of a sensor configured on the target component at a preset period. The processing module 301 is also used to determine mechanical prediction data of at least one dimension of the target component at the first moment through the mechanical data prediction model. The processing module 301 is also used to update the mechanical data prediction model based on the difference between the mechanical verification data and the mechanical prediction data.

[0087] In some embodiments, the processing module 301 is specifically used to extract the prediction feature vector corresponding to at least one dimension based on the mechanical prediction data of at least one dimension. The processing module 301 is specifically used to obtain the anomaly prediction score of the target component based on the predicted feature vector corresponding to at least one dimension and the mechanical anomaly assessment model.

[0088] In some embodiments, the processing module 301 is specifically used to determine the dimension weights corresponding to at least one dimension based on the risk detection requirements of the target component. The processing module 301 is specifically used to fuse mechanical prediction data of at least one dimension based on dimension weights to obtain a prediction feature vector; The processing module 301 is specifically used to obtain the anomaly prediction score of the target component based on the predicted feature vector and the mechanical anomaly evaluation model.

[0089] In some embodiments, the acquisition module 302 is further configured to acquire mechanical sample data of multiple second test components during operation, as well as fault detection data; The processing module 301 is also used to train the initial mechanical anomaly assessment model based on mechanical sample data and fault detection data to obtain the mechanical anomaly assessment model.

[0090] In some embodiments, the processing module 301 is specifically used to divide the mechanical sample data and fault detection data into a training sample set and a validation sample set. The processing module 301 is specifically used to train the initial mechanical anomaly assessment model using a training sample set and to verify the initial mechanical anomaly assessment model using a verification sample set, thereby obtaining the mechanical anomaly assessment model.

[0091] In some embodiments, the processing module 301 is further configured to determine the target score range in which the anomaly prediction score is located when the anomaly prediction score is less than the preset failure score. The processing module 301 is also used to determine that the target component has an abnormal risk when the detection result corresponding to the target score interval is a component abnormality.

[0092] In some embodiments, the processing module 301 is specifically used to determine whether the abnormal prediction score is greater than a preset safety threshold when the abnormal prediction score is less than a preset failure score. The processing module 301 is specifically used to determine the target score range in which the anomaly prediction score is located when the anomaly prediction score is greater than a preset safety threshold.

[0093] In some embodiments, the processing module 301 is specifically used to determine the difference between the anomaly prediction score and the preset failure score when the detection result corresponding to the target score interval is of high anomaly risk, wherein the preset failure score is the upper limit of the score interval corresponding to the high anomaly risk. The processing module 301 is specifically used to determine that the target component has a failure risk when the difference between the detected abnormal prediction score and the preset failure score is less than or equal to the failure difference threshold.

[0094] In some embodiments, the processing module 301 is further configured to determine the anomaly dimension in the target component that has a failure risk based on the anomaly prediction score; The processing module 301 is also used to determine the improvement strategy corresponding to the anomaly dimension based on the anomaly dimension; Processing module 301 is also used to output improvement strategies.

[0095] In some embodiments, the acquisition module 302 is specifically used to acquire real-time mechanical data of at least one dimension of the target component through sensors configured on the target component, and extract real-time feature vectors corresponding to at least one dimension from the at least one dimension respectively. The processing module 301 is specifically used to input the real-time feature vector corresponding to at least one dimension into the mechanical anomaly assessment model to obtain the actual anomaly score.

[0096] In some embodiments, the processing module 301 is further configured to determine an anomaly handling scheme for the target component when the anomaly detection result indicates that an anomaly exists in the target component; The processing module 301 is also used to output exception handling solutions.

[0097] For a description of the features in the embodiment corresponding to the component detection device for the server, please refer to the relevant description in the embodiment corresponding to the component detection method for the server, which will not be repeated here.

[0098] like Figure 4 As shown, embodiments of this application also provide an electronic device, including a memory 401 and a processor 402. The memory 401 stores a computer program, and the processor 402 is configured to run the computer program to perform the steps in any of the above-described server component detection method embodiments.

[0099] Embodiments of this application also provide a computer-readable storage medium storing a computer program, wherein the computer program is configured to execute the steps in any of the above-described embodiments of the server component detection method when running.

[0100] In one exemplary embodiment, the aforementioned computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard disk, magnetic disk, or optical disk.

[0101] The embodiments of this application also provide a computer program product, which includes a computer program that, when executed by a processor, implements the steps in any of the above-described server component detection method embodiments.

[0102] Embodiments of this application also provide another computer program product, including a non-volatile computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps in any of the above-described server component detection method embodiments.

[0103] Any of the components, modules, units, parts, methods, and operations described herein can be implemented using software, firmware, hardware (e.g., fixed logic circuitry), manual processing, or any combination thereof. Alternatively or additionally, any functionality described herein can be performed at least in part by one or more hardware logic components, such as, but not limited to, processors (CPUs), field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), system-on-a-chip (SoCs), complex programmable logic devices (CPLDs), microprocessors (MCUs), etc. The terms "system," "computing device," or "apparatus" as used herein encompass various means, devices, and machines for processing data, including, for example, one or more programmable processors, computers, SoCs, or combinations thereof. The apparatus may also include code that creates an execution environment for the computer program in question, such as code constituting processor firmware, protocol stacks, database management systems, operating systems, cross-platform runtime environments, virtual machines, or combinations thereof. The aforementioned computer program (also known as a program, software, software application, app, script, or code) can be written in any form of programming language, including compiled or interpreted languages, declarative or procedural languages, and can be deployed in any form, including as a standalone program or as a module, component, subroutine, object, or other unit suitable for a computing environment.

[0104] Those skilled in the art will further recognize that the units and algorithm 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. To clearly illustrate the interchangeability of 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 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 this application.

[0105] The foregoing has provided a detailed description of a server component detection method and electronic device provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the embodiments above are merely for the purpose of helping to understand the method and its core ideas. It should be noted that those skilled in the art can make various improvements and modifications to this application without departing from its principles, and these improvements and modifications also fall within the protection scope of the claims of this application.

Claims

1. A method for testing server components, characterized in that, include: Based on the mechanical data prediction model and the historical mechanical data of the target component, the mechanical data of at least one dimension of the target component is predicted to obtain the mechanical prediction data of the target component. The mechanical data prediction model is obtained by iterative optimization through the historical measured data and historical prediction data of the first test component. Based on the mechanical prediction data and the mechanical anomaly assessment model, the anomaly prediction score corresponding to the target component is obtained; When the anomaly prediction score is greater than or equal to the preset failure score corresponding to the target component, a component failure warning is output. When the anomaly prediction score is less than the preset failure score, and it is determined that the target component has an anomaly risk based on the anomaly prediction score and the preset score range, real-time mechanical data of at least one dimension of the target component is collected by the sensor configured on the target component, and the actual anomaly score corresponding to the real-time mechanical data of the at least one dimension is obtained through the mechanical anomaly assessment model. Based on the actual anomaly score, the anomaly detection result of the target component is output.

2. The server component testing method according to claim 1, characterized in that, Before predicting the mechanical data of the target component based on the mechanical data prediction model and historical mechanical data of the target component, the method further includes: The measured data of the first test component are obtained through the sensors configured on the first test component; The first test component is predicted using an initial mechanical data prediction model to obtain the predicted data for the first test component. Based on the difference between the measured data and the predicted data, the initial mechanical data prediction model is iteratively optimized to obtain the mechanical data prediction model.

3. The server component testing method according to claim 2, characterized in that, The step of iteratively optimizing the initial mechanical data prediction model based on the difference between the measured data and the predicted data to obtain the mechanical data prediction model includes: When the difference between the measured data and the predicted data is greater than or equal to the predicted data difference threshold, the model parameters of the initial mechanical data prediction model are adjusted according to the difference between the measured data and the predicted data, and the difference between the measured data and the predicted data is calculated again until the difference between the measured data and the predicted data is detected to be less than the predicted data difference threshold, thus obtaining the mechanical data prediction model.

4. The server component testing method according to claim 3, characterized in that, The method further includes: According to a preset cycle, mechanical verification data of at least one dimension of the target component is collected at a first moment using sensors configured on the target component. The mechanical data prediction model is used to determine mechanical prediction data for at least one dimension of the target component at the first moment. The mechanical data prediction model is updated based on the difference between the mechanical verification data and the mechanical prediction data.

5. The server component testing method according to claim 1, characterized in that, The step of obtaining the anomaly prediction score corresponding to the target component based on the mechanical prediction data and the mechanical anomaly assessment model includes: Based on the mechanical prediction data of at least one dimension, extract the prediction feature vectors corresponding to the at least one dimension respectively; Based on the predicted feature vector corresponding to the at least one dimension and the mechanical anomaly assessment model, the anomaly prediction score of the target component is obtained.

6. The server component testing method according to claim 5, characterized in that, The step of obtaining the anomaly prediction score of the target component based on the feature prediction data corresponding to the at least one dimension and the mechanical anomaly assessment model includes: Based on the risk detection requirements of the target component, determine the dimension weights corresponding to each of the at least one dimension; Based on the dimensional weights, the mechanical prediction data of at least one dimension are fused to obtain a prediction feature vector; Based on the predicted feature vector and the mechanical anomaly assessment model, the anomaly prediction score of the target component is obtained.

7. The server component testing method according to claim 1, characterized in that, Before obtaining the anomaly prediction score corresponding to the target component based on the mechanical prediction data and the mechanical anomaly assessment model, the method further includes: Acquire mechanical sample data and fault detection data of multiple second test components during operation; Based on the mechanical sample data and the fault detection data, the initial mechanical anomaly assessment model is trained to obtain the mechanical anomaly assessment model.

8. The server component testing method according to claim 7, characterized in that, The step of training the initial mechanical anomaly assessment model based on the mechanical sample data and the fault detection data to obtain the mechanical anomaly assessment model includes: The mechanical sample data and the fault detection data are divided into a training sample set and a validation sample set; The initial mechanical anomaly assessment model is trained using the training sample set and validated using the validation sample set to obtain the mechanical anomaly assessment model.

9. The server component testing method according to claim 1, characterized in that, Before collecting real-time mechanical data of at least one dimension of the target component using sensors configured on the target component when the anomaly prediction score is less than the preset failure score, and the target component is determined to have an anomaly risk based on the anomaly prediction score and the preset score range, the method further includes: When the anomaly prediction score is less than the preset failure score, the target score range in which the anomaly prediction score is located is determined; When the detection result corresponding to the target score interval is a component abnormality, it is determined that the target component has an abnormal risk.

10. The server component testing method according to claim 9, characterized in that, When the anomaly prediction score is less than the preset failure score, determining the target score range in which the anomaly prediction score falls includes: When the anomaly prediction score is less than the preset failure score, it is determined whether the anomaly prediction score is greater than the preset safety threshold. If the anomaly prediction score is greater than the preset safety threshold, the target score range in which the anomaly prediction score is located is determined.

11. The server component testing method according to claim 9, characterized in that, The step of determining that the target component has an abnormal risk when the detection result corresponding to the target scoring interval is a component abnormality includes: When the detection result corresponding to the target score interval is high anomaly risk, the difference between the anomaly prediction score and the preset failure score is determined, and the preset failure score is the upper limit of the score interval corresponding to the high anomaly risk. When the difference between the abnormal prediction score and the preset failure score is less than or equal to the failure difference threshold, it is determined that the target component has a failure risk.

12. The server component testing method according to claim 11, characterized in that, After determining that the target component has a failure risk when the difference between the detected abnormal prediction score and the preset failure score is less than or equal to a failure difference threshold, the method further includes: Based on the anomaly prediction score, the anomaly dimension with failure risk in the target component is determined; Based on the anomaly dimension, determine the improvement strategy corresponding to the anomaly dimension; Output the improved strategy.

13. The server component testing method according to claim 1, characterized in that, The step of collecting real-time mechanical data of at least one dimension of the target component using sensors configured on the target component, and obtaining the actual anomaly score corresponding to the real-time mechanical data of the at least one dimension using the mechanical anomaly assessment model, includes: The system collects real-time mechanical data of the target component in at least one dimension using sensors configured on the target component, and extracts real-time feature vectors corresponding to the at least one dimension from each dimension. The real-time feature vector corresponding to the at least one dimension is input into the mechanical anomaly assessment model to obtain the actual anomaly score.

14. The server component testing method according to claim 1, characterized in that, After outputting the anomaly detection result of the target component based on the actual anomaly score, the method further includes: When the anomaly detection result indicates that the target component has an anomaly, an anomaly handling plan for the target component is determined; Output the exception handling solution.

15. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor, configured to implement the steps of the component detection method for a server as described in any one of claims 1 to 14 when executing the computer program.

Images