Device fault early warning method, device, medium and program product

Abstract

The application provides a device fault early warning method, device, equipment, medium and program product. The method comprises: acquiring first running data of a servo device of a centering mechanism of a battery swap station within a first time window; detecting the first running data by at least two different unsupervised detection algorithms respectively to obtain at least two anomaly detection data of the servo device, the anomaly detection data being used to indicate an anomaly degree of the first running data; determining device health data of the servo device based on the at least two anomaly detection data and the first running data; and generating early warning information of the centering mechanism in a case where the device health data indicates that the servo device has a fault. The method provided by the application can realize early warning of potential faults of the centering mechanism, ensure the reliability of the operation of the centering mechanism, and avoid the case that the servo device cannot supply power to the vehicle due to a fault.

Description

Technical Field

[0001] This application relates to the field of equipment fault detection technology, and in particular to a method, apparatus, equipment, medium and program product for early warning of equipment faults. Background Technology

[0002] With the rapid development of the new energy industry, battery swapping has become a promising energy replenishment method for electric vehicles. By managing the battery pack through battery swapping stations, the vehicle's power can be replenished in a short time, quickly restoring the vehicle's driving range.

[0003] In battery swapping stations, a servo device of a centering mechanism can be used to move the vehicle whose battery needs to be swapped to a preset position for recharging. Therefore, ensuring the reliability of the centering mechanism and avoiding the inability to recharge vehicles due to servo device failure has become a research direction for equipment fault detection technology. Summary of the Invention

[0004] In view of the above problems, this application provides a method, apparatus, equipment, medium and program product for early warning of equipment failure, which can improve the reliability of the operation of the centering mechanism.

[0005] In a first aspect, embodiments of this application provide a method for early warning of equipment failure. The method includes acquiring first operating data of a servo device of a battery swapping station centering mechanism within a first time window; detecting the first operating data using at least two different unsupervised detection algorithms to obtain at least two abnormal detection data of the servo device, wherein the abnormal detection data is used to indicate the degree of abnormality of the first operating data; determining equipment health data of the servo device based on the at least two abnormal detection data and the first operating data; and generating early warning information for the centering mechanism when the equipment health data indicates that the servo device has a failure.

[0006] In the above scheme, by acquiring the first operating data of the servo equipment of the battery swapping station's centering mechanism within the first time window, and using the detection results of the first operating data through at least two unsupervised detection algorithms, the status of the servo equipment is assigned a value, thus obtaining at least two anomaly detection data. In this way, at least two anomaly detection data can be used to indicate the degree of anomaly of the first operating data of the servo equipment, effectively reducing the errors and irrationality that may occur in the process of manually scoring the status of the servo equipment, thereby improving the accuracy of servo equipment detection. Then, based on the anomaly detection data obtained from the detection of the first operating data and the first operating data of the servo equipment, the equipment health data of the servo equipment is determined. The equipment health data is used to determine whether the servo equipment has a fault. If the servo equipment has a fault, an early warning information for the centering mechanism is generated, which can realize early warning of potential faults in the centering mechanism, improve the reliability of the centering mechanism operation, and avoid the inability to replenish power to vehicles due to servo equipment failure.

[0007] In some embodiments, the step of determining the device health data of the servo device based on at least two anomaly detection data and first operating data may specifically include:

[0008] Based on at least two anomaly detection data, determine the correlation data between each pair of at least two different unsupervised detection algorithms;

[0009] The target anomaly detection data detected by the two unsupervised detection algorithms corresponding to the target correlation data and the first running data are reconstructed to obtain the second running data of the servo device within the first time window. The target correlation data includes the correlation data between each two unsupervised detection algorithms whose data value is greater than or equal to a preset threshold.

[0010] Based on the second set of operational data, determine the device health data of the servo device.

[0011] In the above scheme, when detecting the representation learning results, the feature space of the generated running data is optimized by introducing correlation data. When selecting unsupervised detection algorithms, the correlation data between each pair of unsupervised detection algorithms is used to determine the unsupervised detection algorithm to be reconstructed. Introducing the abnormal detection data detected by the unsupervised detection algorithm after data reconstruction into the first running data enhances the feature space, ensures the diversity of the feature space, and alleviates the class imbalance problem in predictive maintenance identification. Thus, the equipment health data of the servo device is determined through the second running data after data reconstruction, which improves the accuracy of determining the equipment health data of the servo device, thereby ensuring the accuracy of the early warning information generated by the centering mechanism. Furthermore, by using the prediction results of the equipment health data of the servo device within the transformation time window, early warning of potential faults of the centering mechanism is achieved.

[0012] In some embodiments of this application, the first running data includes at least two types of statistical data, one type of statistical data corresponding to a key statistical feature; one anomaly detection data corresponds to an unsupervised detection algorithm, and one anomaly detection data includes statistical anomaly detection data for each of the at least two types of statistical data. Based on this, the step of determining the correlation data between each pair of at least two different unsupervised detection algorithms according to at least two anomaly detection data may specifically include:

[0013] Based on the statistical anomaly detection data of each type of statistical data in the anomaly detection data corresponding to an unsupervised detection algorithm, determine the detection accuracy evaluation data corresponding to the unsupervised detection algorithm.

[0014] Based on the detection accuracy evaluation data corresponding to each of the at least two different unsupervised detection algorithms, determine the correlation data between each pair of unsupervised detection algorithms.

[0015] In the above scheme, when determining the correlation data between two unsupervised detection algorithms, different categories of key statistical features and statistical anomaly detection data of each category of key statistical data can be introduced. This can strengthen the feature space, ensure the diversity of the feature space, and alleviate the problem of class imbalance of statistical features in predictive maintenance identification.

[0016] Furthermore, by using unsupervised detection algorithms as anchors and through the detection data of key statistical features of different categories and statistical anomalies of each category of key statistical data, the accuracy of the evaluation data for the detection accuracy of unsupervised detection algorithms can be improved. This, in turn, improves the accuracy of determining the correlation data between each pair of unsupervised detection algorithms based on the evaluation data for the detection accuracy of unsupervised detection algorithms, effectively improving the accuracy of selecting unsupervised detection algorithms.

[0017] In some embodiments, the step of reconstructing the target anomaly detection data and the first running data detected by two unsupervised detection algorithms corresponding to the target relevance data to obtain the second running data of the servo device within the first time window may specifically include:

[0018] The target anomaly detection data detected by each of the two unsupervised detection algorithms corresponding to the target relevance data and the first running data are reconstructed to obtain the second running data.

[0019] or,

[0020] The target anomaly detection data detected by either of the two unsupervised detection algorithms corresponding to the target relevance data and the first running data are reconstructed to obtain the second running data.

[0021] In the above scheme, the target anomaly detection data detected by each of the two unsupervised detection algorithms corresponding to the target relevance data can be used to participate in data reconstruction, thereby improving the accuracy of the second running data. Then, the equipment health data of the servo device can be determined through the second running data after data reconstruction, which can improve the accuracy of determining the equipment health data of the servo device and thus ensure the accuracy of the early warning information generated by the centering mechanism.

[0022] Alternatively, target anomaly detection data detected by either of two unsupervised detection algorithms corresponding to the degree of relevance to the target can be selected for data reconstruction. This reduces data reconstruction resources, computational load, and efficiency, thereby improving the efficiency of generating early warning information for the centering mechanism.

[0023] In some embodiments, the equipment fault early warning method provided in this application may further include:

[0024] Based on the equipment identification information of the centering mechanism of the swapping station, obtain the feature list corresponding to the servo equipment of the centering mechanism of the swapping station. The feature list includes at least two statistical features and the importance value corresponding to each of the at least two statistical features.

[0025] Based on the importance value corresponding to each statistical feature, key statistical features are determined from at least two statistical features, and the importance value of the key statistical features is greater than or equal to the preset importance value.

[0026] In the above scheme, a feature list that contributes most to the fault samples of servo equipment can be obtained. To address the problems of scarce fault samples and difficulty in quantifying the status of servo equipment, and by using a pre-constructed feature list corresponding to the centering equipment of the swapping station, the most suitable key statistical features for the servo equipment can be selected. This adapts to the differences between swapping stations and the differences in the centering equipment. Furthermore, by selecting the unsupervised detection algorithm that best matches the key statistical feature distribution of the centering mechanism, the status of the servo equipment is assigned a value. This enhances the feature space, ensures the diversity of the feature space, effectively reduces the errors and irrationality that may occur in the process of manually scoring the status of equipment, and alleviates the problem of class imbalance of statistical features in predictive maintenance identification.

[0027] In some embodiments, the step of determining the device health data of the servo device based on the second operating data may specifically include:

[0028] The second operating data is input into the classifier. The classifier determines whether the servo device is faulty in the second time window. The time of the first time window is earlier than the time of the second time window. The classifier is trained by the operating data of the servo device of the mechanism in the battery swapping station in the third time window. The time of the third time window is earlier than the time of the first time window.

[0029] Based on the classification results of whether the servo device is faulty within the second time window, determine the device health data of the servo device.

[0030] In the above scheme, the second running data after data reconstruction can be detected by a classifier to determine the equipment health data of the servo device, which can improve the accuracy of determining the equipment health data of the servo device, thereby ensuring the accuracy of the early warning information generated by the centering mechanism. Furthermore, by using the prediction results of the equipment health data of the servo device within the conversion time window, early warning of potential faults of the centering mechanism can be achieved.

[0031] In some embodiments of this application, the first time window includes at least two sub-time windows, and the second running data includes sub-running data within each sub-time window. Based on this, the step of inputting the second running data into a classifier and determining whether the servo device is faulty within the second time window through the classifier may specifically include:

[0032] The sub-run data within each sub-time window is input into the classifier, and the sub-classification result corresponding to the sub-run data within each sub-time window is obtained through the classifier;

[0033] Based on the sub-classification results, determine whether the servo device is faulty within the second time window.

[0034] In the above scheme, a first time window can be divided. Based on the sub-classification results corresponding to the sub-operation data within at least two sub-time windows after the first time window is divided, the classification results of whether the servo device is faulty in the second time window can be statistically analyzed. This avoids the situation where the servo device switches between faulty and non-faulty states due to the unstable health status of its components, and improves the accuracy of determining whether the servo device is faulty in the second time window.

[0035] In some embodiments, the subclassification results in this application include fault results or non-fault results. Based on this, the above-mentioned steps involving determining whether the servo device is faulty within a second time window based on the subclassification results may specifically include:

[0036] Based on the first number of fault results in the subclassification results and the total number of subclassification results, determine the proportion of fault results in the total number of subclassification results;

[0037] The proportion of the fault results in the total number of sub-classification results is determined as the device health data of the servo device.

[0038] In the above scheme, the sub-operational data within at least two sub-time windows after the first time window is divided are detected, and the proportion of multiple sub-classification results within the first time window is used to calculate the proportion of fault results in the total number of sub-classification results. This alleviates the impact of unstable operation data quality of the servo device on the fault results and improves the accuracy of the classification results for determining whether the servo device is faulty in the second time window.

[0039] In some embodiments, the equipment fault early warning method provided in this application may further include:

[0040] If the proportion is greater than or equal to the reference proportion, the classification result of the servo device in the second time window is determined as the classification result of the fault.

[0041] If the percentage is less than the reference percentage, the classification result of the servo device in the second time window is determined to be a non-faulty classification result.

[0042] In the above scheme, the proportion of fault results in the total number of sub-classification results can be calculated by using multiple sub-classification results within the first time window. Combined with the reference proportion, the classification result of the servo device in the second time window can be determined. This can alleviate the impact of unstable operating data quality of the servo device on the fault results and improve the accuracy of determining whether the servo device is faulty in the second time window.

[0043] In some embodiments, the equipment fault early warning method provided in this application may further include:

[0044] Acquire the operating signals of the servo device within the first time window. The operating signals include step sequence signals and at least one of the following target signals: torque signal, displacement signal, and speed signal.

[0045] Based on the step sequence signals and the step sequence annotations corresponding to the step sequence signals during the execution of the work steps by the servo device, the key steps of the servo device during the execution of the work steps are determined.

[0046] Using the key step sequence as the first dimension, based on the operation signal, the key step sequence statistical feature data corresponding to the key step sequence is determined; and using the operation cycle of the servo device as the second dimension, based on the operation signal, the operation cycle statistical feature data corresponding to the operation cycle of the servo device is determined.

[0047] Based on the statistical feature data of key steps and the statistical feature data of the running cycle, the first running data is generated.

[0048] In the above scheme, by acquiring the operating signals of the servo device during operation, the richness of the quantity and types of data participating in the equipment fault early warning process is ensured, the diversity of the feature space is guaranteed, the problem of class imbalance of statistical features in predictive maintenance identification is alleviated, and the accuracy of the early warning information generated by the centering mechanism is guaranteed.

[0049] In some embodiments, the steps described above, which involve determining the key step sequence statistical feature data corresponding to the key step sequence based on the running signal and using the key step sequence as the first dimension, may specifically include:

[0050] Using the key step sequence as the first dimension, at least one of the following first target signals corresponding to the first dimension is selected from the operation signals: first torque signal, first displacement signal, and first velocity signal;

[0051] Perform data statistics on at least one first target signal to obtain statistical data corresponding to the first target signal;

[0052] The statistical data corresponding to the first target signal is determined as the key step sequence statistical feature data corresponding to the key step sequence.

[0053] In the above scheme, the key step sequence statistical feature data of the key steps that are prone to failure of servo equipment are obtained through the first dimension, so as to ensure the relevance of the data participating in the equipment failure early warning process and the accuracy of the early warning information generated by the centering mechanism.

[0054] In some embodiments, the step of determining the statistical characteristic data of the operating cycle corresponding to the operating cycle of the servo device, based on the operating signal and with the operating cycle of the servo device as the second dimension, may specifically include:

[0055] Taking the operating cycle of the servo device as the second dimension, at least one of the following second target signals corresponding to the second dimension is selected from the operating signals: second torque signal, second displacement signal, and second speed signal;

[0056] Statistical analysis is performed on at least one first target signal to obtain statistical data corresponding to the second target signal;

[0057] The statistical data corresponding to the second target signal is determined as the operating cycle statistical feature data corresponding to the operating cycle of the servo device.

[0058] In the above scheme, the statistical feature data of all operating cycles in the servo device's operating cycle are obtained through the first dimension to ensure the richness and completeness of the quantity and types of data participating in the equipment fault early warning process, to ensure the diversity of the feature space, to alleviate the problem of class imbalance of statistical features in predictive maintenance identification, and to ensure the accuracy of the early warning information generated by the centering mechanism.

[0059] In some embodiments of this application, the at least two different unsupervised detection algorithms include the following at least two algorithms:

[0060] Single-class support vector machine anomaly detection algorithm, K-nearest neighbor classification detection algorithm, local anomaly factor detection algorithm, and histogram-based anomaly value scoring detection algorithm.

[0061] In the above scheme, by using at least two different unsupervised detection algorithms, the accuracy of the detection results of the first running data can be improved. Also, by using at least two different unsupervised detection algorithms to assign values ​​to the state of the servo device, anomaly detection data is obtained. This anomaly detection data can be used to indicate the degree of anomaly of the first running data of the servo device, effectively reducing the errors and irrationality that may occur in the process of manually scoring the state of the servo device, thereby improving the accuracy of servo device detection.

[0062] In some embodiments, the equipment fault early warning method according to this application may further include:

[0063] The first sample operation data and the sample device health data corresponding to the first sample operation data are obtained from the servo device of the centering mechanism of the battery swapping station within the third time window. The first sample operation data includes at least two types of sample statistical data, one type of sample statistical data corresponds to one sample statistical feature, and the sample device health data includes the sample statistical device health data of each type of sample statistical data in at least two types of sample statistical data.

[0064] The random forest classifier model is trained by using statistical data of each class of samples and health data of the sample statistical data of each class of samples until the second reference condition is met, thus obtaining the feature selection model. The model data of the feature selection model includes the sample importance value corresponding to each sample statistical feature.

[0065] Based on the sample importance value corresponding to each sample statistical feature, at least one key sample statistical feature with a sample importance value greater than or equal to a preset sample importance value is selected from at least two sample statistical features;

[0066] Generate a feature list based on at least one key statistical feature of the sample.

[0067] In the above scheme, a sample for detection can be constructed based on the sample operation data of the servo equipment of the swapping station centering mechanism in the third time window and its individual sample equipment health data, such as the sample operation data of the servo equipment when it fails and the data of the previous day, to achieve early warning of equipment. In addition, this scheme uses a random forest classifier model to obtain the feature list that contributes the most to the servo equipment failure samples, in order to deal with the problem of scarce failure samples and difficulty in quantifying the status of servo equipment. Furthermore, the key statistical features most suitable for the servo equipment can be selected from the feature list corresponding to the swapping station centering mechanism in advance, thereby adapting to the differences between swapping stations and the differences in the centering mechanism. Then, by selecting the unsupervised detection algorithm that best matches the key statistical feature distribution of the centering mechanism, the status of the servo equipment is assigned a value, which enhances the feature space, ensures the diversity of the feature space, effectively reduces the errors and irrationality that may occur in the process of manually scoring the status of equipment, and alleviates the problem of class imbalance of statistical features in predictive maintenance identification.

[0068] In some embodiments, the equipment fault early warning method according to this application may further include:

[0069] Obtain the first sample operation data and the sample device health data corresponding to the first sample operation data of the servo device of the battery swapping station centering mechanism within the third time window;

[0070] The first sample running data is detected by at least two different unsupervised detection algorithms to obtain at least two sample anomaly detection data of the servo device. The sample anomaly detection data is used to indicate the degree of anomaly of the first sample running data.

[0071] Based on at least two sample anomaly detection data, determine the sample correlation data between each pair of at least two different unsupervised detection algorithms;

[0072] The sample target anomaly detection data detected by the two unsupervised detection algorithms corresponding to the sample target correlation data and the first sample running data are reconstructed to obtain the second sample running data of the servo device in the third time window. The sample target correlation data includes the sample correlation data between each two unsupervised detection algorithms whose correlation data values ​​are greater than or equal to a preset threshold.

[0073] Based on the second operating data and the health data of the sample equipment, the sample classifier is trained until the first reference training condition is met, and the classifier is obtained.

[0074] In the above scheme, semi-supervised learning can be achieved through unsupervised detection algorithms and supervised learning of classifiers to train the classifier. Then, the classifier is used to detect the second running data after data reconstruction to determine the equipment health data of the servo device. This can improve the accuracy of determining the equipment health data of the servo device, thereby ensuring the accuracy of the early warning information generated by the centering mechanism. Furthermore, by using the prediction results of the equipment health data of the servo device within the conversion time window, early warning of potential faults in the centering mechanism is achieved.

[0075] In some embodiments, the first sample operation data in this application includes sample statistics data of at least one key statistical feature of the sample in the feature list, and the sample device health data includes sample statistics data of the sample device health data. Based on this, the device fault early warning method involved in this application may further include:

[0076] If the first reference training condition is not met, adjust the number of key statistical features of the samples in the feature list and execute the target step;

[0077] The target steps include:

[0078] The first sample running data is detected by at least two different unsupervised detection algorithms to obtain at least two sample anomaly detection data of the servo device. The sample anomaly detection data is used to indicate the degree of anomaly of the first sample running data.

[0079] Based on at least two sample anomaly detection data, determine the sample correlation data between each pair of at least two different unsupervised detection algorithms;

[0080] The sample target anomaly detection data detected by the two unsupervised detection algorithms corresponding to the sample target correlation data and the first sample running data are reconstructed to obtain the second sample running data of the servo device in the third time window. The sample target correlation data includes the sample correlation data between each two unsupervised detection algorithms whose correlation data values ​​are greater than or equal to a preset threshold.

[0081] The sample classifier is trained based on the second set of operational data and the health data of the sample equipment.

[0082] In the above scheme, if the feature list is completed and the training sample classifier does not meet the first reference training condition, the number of key statistical features of the samples in the feature list can be adjusted to optimize the feature space and address the problem of quantifying the state of the servo device. This reduces the dimensionality of the feature space while ensuring the accuracy of the state representation. Then, by selecting the unsupervised detection algorithm that best matches the distribution of key statistical features of the centering mechanism, the servo device state is assigned a value, which enhances the feature space, ensures the diversity of the feature space, effectively reduces the errors and irrationality that may occur in the process of manually scoring the device state, and alleviates the problem of class imbalance of statistical features in predictive maintenance identification.

[0083] Secondly, embodiments of this application provide a device for early warning of equipment failure, comprising:

[0084] The acquisition module is used to acquire the first operating data of the servo device of the battery swapping station centering mechanism within the first time window;

[0085] The detection module is used to detect the first running data using at least two different unsupervised detection algorithms to obtain at least two abnormal detection data of the servo device. The abnormal detection data is used to indicate the degree of abnormality of the first running data.

[0086] The determination module is used to determine the device health data of the servo device based on at least two anomaly detection data and the first operating data;

[0087] The generation module is used to generate early warning information for the centering mechanism when the equipment health data indicates that the servo device has a fault.

[0088] In the above scheme, by acquiring the first operating data of the servo equipment of the battery swapping station's relocation mechanism within the first time window, and using the detection results of the first operating data through an unsupervised detection algorithm, the status of the servo equipment is assigned a value, thus obtaining anomaly detection data. This anomaly detection data can be used to indicate the degree of anomaly in the first operating data of the servo equipment, effectively reducing the errors and irrationality that may occur during the manual scoring of the servo equipment status, thereby improving the accuracy of servo equipment detection. Then, based on the anomaly detection data obtained from the detection of the first operating data and the first operating data of the servo equipment, the equipment health data of the servo equipment is determined. The equipment health data is used to determine whether the servo equipment has a fault. If the servo equipment has a fault, an early warning information for the relocation mechanism is generated, which can realize early warning of potential faults of the servo equipment, improve the reliability of the relocation mechanism operation, and avoid the inability to replenish power to vehicles due to servo equipment failure.

[0089] Thirdly, this application provides a computer device, including a processor and a memory storing computer program instructions, wherein the processor, when executing the computer program instructions, implements the device fault early warning method as described in the first aspect.

[0090] Fourthly, this application provides a readable storage medium storing computer program instructions, which, when executed by a processor, implement the device fault warning method as described in the first aspect.

[0091] Fifthly, this application provides a computer program product in which the instructions, when executed by the processor of an electronic device, cause the electronic device to perform the device fault warning method as described in the first aspect.

[0092] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0093] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the embodiments of this application 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.

[0094] Figure 1 This is a flowchart of a device fault early warning method according to an embodiment of this application;

[0095] Figure 2This is a schematic diagram of the training process of a semi-supervised anomaly detection model in an equipment fault early warning method according to an embodiment of this application;

[0096] Figure 3 This is a flowchart illustrating a device fault early warning method according to an embodiment of this application;

[0097] Figure 4 Schematic diagrams of the structure of the equipment fault early warning device provided in some embodiments of this application;

[0098] Figure 5 A schematic diagram of the hardware structure of a computer device provided for some embodiments of this application. Detailed Implementation

[0099] The embodiments of the technical solution of this application will now be described in detail with reference to the accompanying drawings. These embodiments are only used to more clearly illustrate the technical solution of this application and are therefore merely examples, and should not be used to limit the scope of protection of this application.

[0100] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the application; the terms “comprising” and “having”, and any variations thereof, in the specification, claims, and foregoing description of the drawings are intended to cover non-exclusive inclusion.

[0101] In the description of the embodiments of this application, technical terms such as "first" and "second" are used only to distinguish different objects and should not be construed as indicating or implying relative importance or implicitly specifying the number, specific order, or primary and secondary relationship of the indicated technical features. In the description of the embodiments of this application, "multiple" means two or more, unless otherwise explicitly defined.

[0102] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0103] Unless otherwise specified, all embodiments and optional embodiments of this application can be combined to form new technical solutions.

[0104] Unless otherwise specified, all technical features and optional technical features of this application may be combined to form new technical solutions.

[0105] Unless otherwise specified, all steps of this application may be performed sequentially or randomly, preferably sequentially. For example, the method includes steps (a) and (b), indicating that the method may include steps (a) and (b) performed sequentially, or it may include steps (b) and (a) performed sequentially. For example, the mention that the method may also include step (c) indicates that step (c) may be added to the method in any order; for example, the method may include steps (a), (b), and (c), or it may include steps (a), (c), and (b), or it may include steps (c), (a), and (b), etc.

[0106] In the description of the embodiments in this application, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.

[0107] In the description of the embodiments of this application, the term "multiple" refers to two or more (including two), similarly, "multiple sets" refers to two or more (including two sets), and "multiple pieces" refers to two or more (including two pieces).

[0108] In the description of the embodiments of this application, the technical terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," "counterclockwise," "axial," "radial," and "circumferential" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing the embodiments of this application and simplifying the description, and are not intended to indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the embodiments of this application.

[0109] In the description of the embodiments of this application, unless otherwise expressly specified and limited, technical terms such as "installation," "connection," "joining," and "fixing" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. For those skilled in the art, the specific meaning of the above terms in the embodiments of this application can be understood according to the specific circumstances.

[0110] In a battery swapping station, when a vehicle needs a battery swap, the station's centering mechanism pushes the vehicle towards the center to position it in a reference direction, such as the Y-axis, in order to recharge the vehicle. Statistics show that 30% of the failures to recharge vehicles occur at this centering mechanism, but maintenance is only performed after a work order is reported, making preventative measures impossible and impacting the station's operation.

[0111] To achieve predictive maintenance, related technologies typically employ two approaches. First, supervised learning is used, which involves labeling data from both failed and normally functioning institutions to train a predictive model. Second, unsupervised learning is used, which eliminates the need for data labeling and identifies potential faults by analyzing the inherent structure and patterns of the data.

[0112] However, both supervised and unsupervised learning have limitations. Supervised learning heavily relies on labeled datasets. Insufficient data or inaccurate labeling can negatively impact the model's predictive performance. Furthermore, due to the diversity and complexity of equipment failures, the model may struggle to generalize to unseen failure types. Unsupervised learning, on the other hand, does not depend on labeled datasets. Therefore, the model's accuracy can be affected by data quality and the choice of clustering algorithm, potentially misclassifying normal but rare operating patterns as abnormal, thus impacting prediction accuracy. Consequently, ensuring the reliability of equipment failure detection mechanisms has become a key research direction in equipment failure detection technology.

[0113] Based on this, this application proposes a fault early warning method for the servo equipment of the relocation mechanism in a battery swapping station, addressing the fault detection problem of the servo equipment. This method acquires the first operating data of the servo equipment within a first time window. Using an unsupervised detection algorithm, the detection results of the first operating data are used to assign values ​​to the servo equipment status, thus obtaining anomaly detection data. This anomaly detection data indicates the degree of anomaly in the first operating data of the servo equipment, effectively reducing the errors and irrationality that may occur during manual scoring of the servo equipment status, thereby improving the accuracy of servo equipment detection. Then, based on the anomaly detection data obtained from the first operating data and the first operating data of the servo equipment, the equipment health data of the servo equipment is determined. The equipment health data is used to determine whether the servo equipment is faulty. If a fault exists in the servo equipment, an early warning message for the relocation mechanism is generated. This method enables early warning of potential servo equipment faults, improving the reliability of the relocation mechanism and preventing the inability to replenish power to vehicles due to servo equipment failure.

[0114] Based on this, in order to better describe the equipment fault early warning method, apparatus, computer equipment, storage medium and computer program product provided in the embodiments of this application, the following will be combined with the appendix. Figures 1 to 5The following detailed description is provided, and it should be noted that these embodiments are not intended to limit the scope of this application.

[0115] First, combined Figure 1 The device fault early warning method provided in the embodiments of this application will be described in detail.

[0116] Figure 1 This is a flowchart illustrating a device fault early warning method provided in an embodiment of this application.

[0117] like Figure 1 As shown, this equipment fault early warning method can be applied to computer equipment. Specifically, the equipment fault early warning method may include the following steps:

[0118] Step 110: Obtain the first operating data of the servo equipment of the battery swapping station centering mechanism within the first time window; Step 120: Detect the first operating data using at least two different unsupervised detection algorithms to obtain at least two anomaly detection data of the servo equipment, which are used to indicate the degree of anomaly of the first operating data; Step 130: Determine the equipment health data of the servo equipment based on the at least two anomaly detection data and the first operating data; Step 140: If the equipment health data indicates that the servo equipment has a fault, generate early warning information for the centering mechanism.

[0119] In this application embodiment, the first time window can be a running cycle of the servo device. In some embodiments of this application, within the first time window, the first running data can include the running data of the servo device in that running cycle. This running data can include all running data of the servo device executing work steps within that running cycle and running data of key steps within those work steps; it can be understood as global running data and local running data. The key steps can be steps where the servo device is prone to failure. In other embodiments, the first running data can be data divided according to at least one key statistical feature within the running cycle. This data can include statistical data of all running data of the servo device executing work steps within that running cycle corresponding to each key statistical feature, and statistical data of running data of key steps within those work steps; it can be understood as data calculated from global running data according to each key statistical feature and data calculated from local running data according to each key statistical feature. The at least one key statistical feature includes at least one of the following: maximum value feature, minimum value feature, average value feature, standard deviation feature, and root mean square deviation feature. The at least two different unsupervised detection algorithms in the embodiments of this application include the following at least two algorithms: single-class support vector machine anomaly detection algorithm, K-nearest neighbor classification detection algorithm, local anomaly factor detection algorithm, and histogram-based anomaly value scoring detection algorithm.

[0120] Here, we will take the first running data from some of the other embodiments mentioned above, and two different unsupervised detection algorithms, namely, unsupervised detection algorithm 1 is a single-class support vector machine anomaly detection algorithm and unsupervised detection algorithm 2 is a K-nearest neighbor classification detection algorithm, as an example for explanation. The number of key statistical features is d, and the amount of running data in this running cycle is n.

[0121] For example, unsupervised detection algorithm 1 and unsupervised detection algorithm 2 are used to run the first data X∈R respectively. n×d The detection process yields two anomaly detection data points for the servo device: TNS1, output by unsupervised detection algorithm 1, and TNS2, output by unsupervised detection algorithm 2. Next, based on TNS1, TNS2, and X, the device health data of the servo device is determined. This health data can be represented numerically, for example, an F1-score of 87%, which serves as a label characterizing the health of the servo device, enabling the generation of early warning information for the centralization mechanism.

[0122] Therefore, by acquiring the first operating data of the servo equipment of the battery swapping station's centering mechanism within the first time window, and using the detection results of the first operating data through at least two unsupervised detection algorithms, the status of the servo equipment is assigned a value, thus obtaining at least two anomaly detection data. In this way, at least two anomaly detection data can be used to indicate the degree of anomaly in the first operating data of the servo equipment, effectively reducing the errors and irrationality that may occur in the process of manually scoring the status of the servo equipment, thereby improving the accuracy of servo equipment detection. Then, based on the anomaly detection data obtained from the detection of the first operating data and the first operating data of the servo equipment, the equipment health data of the servo equipment is determined. The equipment health data is used to determine whether the servo equipment has a fault. If the servo equipment has a fault, an early warning information for the centering mechanism is generated, which can realize early warning of potential faults of the servo equipment, improve the reliability of the centering mechanism operation, and avoid the inability to replenish power to vehicles due to servo equipment failure.

[0123] The steps described above are explained in detail below.

[0124] Regarding step 110, in this embodiment of the application, the equipment fault early warning method may further include the process of determining the first operating data. Based on this, before step 110, the equipment fault early warning method provided in this embodiment of the application may further include steps 2101 to 2101, as shown below.

[0125] Step 2101: Obtain the running signal of the servo device within the first time window. The running signal includes a step sequence signal and at least one of the following target signals: torque signal, displacement signal, and speed signal.

[0126] In this embodiment, the step sequence signal can be obtained according to the operating cycle of the servo device. For example, taking one operating cycle as an example, the step sequence signal can go from the start state step sequence S to the end state step sequence e. In this embodiment, the step sequence signal can start from 50 and end from 67. For example, step sequence signal 50 corresponds to the step sequence annotation "vehicle centering", step sequence signal 51 corresponds to the step sequence annotation "vehicle centering failure, device returns to standby position", and step sequence signal 67 corresponds to the step sequence annotation "lifting servo goes to working position" and "tire positioning returns to standby position". Based on this, the target signal corresponding to the step sequence signal can be obtained according to the step sequence signal, such as obtaining the torque signal, displacement signal, and speed signal corresponding to at least one step sequence signal from step sequence signal 50 to 67 respectively.

[0127] Step 2102: Based on the step sequence signal and the step sequence annotation corresponding to the step sequence signal during the execution of the work steps by the servo device, determine the key steps of the servo device during the execution of the work steps.

[0128] For example, taking the example from step 2101 above, since the step sequence signal obtained based on step 2101 can be of various kinds—that is, it can be a partial step sequence signal within one running cycle, a partial step sequence signal within two running cycles, or all partial step sequence signals within at least one running cycle—key steps can be selected from the currently obtained step sequence signals according to the reference key steps and reference step sequence annotations. For example, step sequence signal 50, step sequence signal 55, and step sequence signal 67 can all be key steps. The key steps can be steps where the servo device is prone to failure.

[0129] Step 2103: Using the key step sequence as the first dimension, based on the running signal, determine the key step sequence statistical feature data corresponding to the key step sequence; and using the running cycle of the servo device as the second dimension, based on the running signal, determine the running cycle statistical feature data corresponding to the running cycle of the servo device.

[0130] For example, using the key step sequence as the first dimension, at least one target signal corresponding to the key step sequence, such as step sequence signal 50, can be obtained. Statistical data such as the maximum value feature, minimum value feature, average value feature, standard deviation feature, and root mean square feature of the target signal can be calculated according to key statistical features, and these statistical data can be used as the key step sequence statistical feature data of step sequence signal 50. Furthermore, using the servo device's operating cycle as the dimension, the operating data of each step sequence signal within each operating cycle of the servo device can be obtained.

[0131] Here, it can be understood that the data in the first dimension can be local statistical feature data, while the data in the second dimension can be global statistical feature data. Additionally, in this embodiment, the statistical feature data can also carry the following information: equipment identification information of the battery swapping station centering mechanism and the location identifier of the servo device. This allows for reasonable analysis and effective tracking of the statistical feature data of the servo device, providing identification markers for subsequent data processing and analysis of the servo device.

[0132] The following uses torque signals as an example to explain the process of determining the statistical characteristic data of the key steps corresponding to the key steps and the process of determining the statistical characteristic data of the operating cycle corresponding to the operating cycle of the servo device.

[0133] Based on the above division of each operating cycle i and the two dimensions of key steps, statistical features are extracted. The extracted statistical features include at least one of the following: the maximum value statistics MAX, minimum value statistics MIN, average value statistics MEAN, standard deviation statistics STD, and root mean square statistics RMS of displacement signal, velocity signal, and torque signal.

[0134] If x mns This is the displacement signal of the nth step in the mth running cycle. A total of T cycles of displacement signals have been collected. The step sequence set is P′, and the key step sequence set is P, S. ij Let be the number of signals in the j-th step of the i-th cycle.

[0135] Taking the determination of the statistical characteristic data of the operating cycle of the torque signal as an example, it can be achieved through the following formulas (1) to (5):

[0136]

[0137] Taking the determination of key step sequence statistical feature data of torque signal as an example, it can be specifically achieved through the following formulas (6) to (10):

[0138]

[0139] It should be noted that the calculation process for displacement, velocity, and torque signals is the same, and the determination process for the key step sequence statistical characteristic data and the running cycle statistical characteristic data of displacement and velocity signals will not be repeated here.

[0140] Step 2104: Generate the first running data based on the key step sequence statistical feature data and the running cycle statistical feature data.

[0141] For example, key step sequence statistical feature data and runtime cycle statistical feature data can be used as the first runtime data.

[0142] Therefore, by acquiring the operating signals of the servo device during operation, the quantity and variety of data involved in the equipment fault early warning process can be guaranteed, the diversity of the feature space can be ensured, the imbalance of statistical features in predictive maintenance identification can be alleviated, and the accuracy of the early warning information generated by the centering mechanism can be guaranteed.

[0143] In some embodiments, the step of determining the key step sequence statistical feature data corresponding to the key step sequence based on the running signal, with the key step sequence as the first dimension, in step 2103 above may specifically include steps 21031 to 21033.

[0144] Step 21031: Using the key step sequence as the first dimension, select at least one of the following first target signals corresponding to the first dimension from the running signals: first torque signal, first displacement signal, and first speed signal.

[0145] Step 21032: Perform data statistics on at least one first target signal to obtain statistical data corresponding to the first target signal.

[0146] For example, formulas (6) to (10) involved in step 2103 above can be used. It should be noted that the calculation process of displacement signal, velocity signal and torque signal is the same, and the determination process of key step sequence statistical characteristic data and running cycle statistical characteristic data of displacement signal and velocity signal will not be repeated here.

[0147] Step 21033: The statistical data corresponding to the first target signal is determined as the key step sequence statistical feature data corresponding to the key step sequence.

[0148] For example, the key step sequence statistical feature data of key step sequence 50 may include the MAX value of the torque signal. mn MIN mn MEAN mn STD mn and RMS mn ; and, the MAX of the speed signal mn MIN mn MEAN mn STD mn and RMS mn ; and, the MAX of the displacement signal mn MIN mn MEAN mn STD mn and RMS mn .

[0149] Therefore, by obtaining the key step sequence statistical feature data of the critical steps that are prone to failure in servo devices through the first dimension, the data participating in the device failure early warning process can be targeted, and the early warning information generated by the centering mechanism can be guaranteed to be accurate.

[0150] In some embodiments, the step of determining the statistical characteristic data of the operating cycle corresponding to the operating cycle of the servo device based on the operating signal, with the operating cycle of the servo device as the second dimension, in step 2103 above may specifically include steps 21034 to 21036.

[0151] Step 21034: Taking the operating cycle of the servo device as the second dimension, select at least one of the following second target signals corresponding to the second dimension from the operating signals: second torque signal, second displacement signal, and second speed signal.

[0152] Step 21035: Perform data statistics on at least one first target signal to obtain statistical data corresponding to the second target signal.

[0153] For example, formulas (1) to (5) involved in step 2103 above can be used. It should be noted that the calculation process of displacement signal, velocity signal and torque signal is the same, and the determination process of key step sequence statistical characteristic data and running cycle statistical characteristic data of displacement signal and velocity signal will not be repeated here.

[0154] Step 21036: Determine the statistical data corresponding to the second target signal as the operating cycle statistical feature data corresponding to the operating cycle of the servo device.

[0155] For example, the operational cycle statistical characteristic data of the key step 50 may include the MAX value of the torque signal. mn MIN mn MEAN mn STD mn and RMS mn ; and, the MAX of the speed signal mn MIN mn MEAN mn STD mn and RMS mn ; and, the MAX of the displacement signal mn MIN mn MEAN mn STD mn and RMS mn .

[0156] Therefore, by obtaining all the statistical feature data of the servo device's operating cycle in the first dimension, we can ensure the richness and completeness of the quantity and types of data involved in the equipment fault early warning process, ensure the diversity of the feature space, alleviate the problem of class imbalance of statistical features in predictive maintenance identification, and ensure the accuracy of the early warning information generated by the centering mechanism.

[0157] Regarding step 120, the at least two different unsupervised detection algorithms in this application embodiment include the following at least two algorithms:

[0158] One-Class Support Vector Machine (OCSVM) anomaly detection algorithm, K-Nearest Neighbors (KNN) classification detection algorithm, Local Outlier Factor (LOF) algorithm, and Histogram-Based Outlier Score (HBOS) algorithm.

[0159] Therefore, by using at least two different unsupervised detection algorithms, the accuracy of the detection results of the first running data can be improved. Furthermore, by assigning values ​​to the state of the servo device using at least two different unsupervised detection algorithms, abnormal detection data can be obtained. This abnormal detection data can be used to indicate the degree of abnormality of the first running data of the servo device, effectively reducing the errors and irrationality that may occur during the process of manually scoring the state of the servo device, thereby improving the accuracy of servo device detection.

[0160] In some embodiments, step 130 may specifically include steps 1301 to 1303.

[0161] Step 1301: Based on at least two anomaly detection data, determine the correlation data between each pair of unsupervised detection algorithms in at least two different unsupervised detection algorithms.

[0162] Step 1302: Perform data reconstruction processing on the target anomaly detection data and the first running data detected by the two unsupervised detection algorithms corresponding to the target relevance data to obtain the second running data of the servo device within the first time window. The target relevance data includes the relevance data in the relevance data between each two unsupervised detection algorithms whose data value is greater than or equal to a preset threshold.

[0163] Step 1303: Determine the device health data of the servo device based on the second running data.

[0164] For example, steps 1301 to 1303 above are explained using two anomaly detection data, TNS1 and TNS2, as an example. Specifically, the correlation between unsupervised detection algorithm 1 and unsupervised detection algorithm 2 is determined based on TNS1 and TNS2. TNS1, TNS2, and the first running data X can be reconstructed to obtain second running data, such as (X + TNS1 + TNS2). Based on this, the device health data of the servo device can be determined according to the second running data, such as (X + TNS1 + TNS2).

[0165] Therefore, when detecting the representation learning results, the feature space of the generated running data is optimized by introducing correlation data. When selecting unsupervised detection algorithms, the correlation data between each pair of unsupervised detection algorithms is used to determine the unsupervised detection algorithm to be reconstructed. Introducing the abnormal detection data detected by the unsupervised detection algorithm after data reconstruction into the first running data enhances the feature space, ensures its diversity, and alleviates the class imbalance problem in predictive maintenance identification. Thus, the accuracy of determining the equipment health data of the servo device can be improved by using the second running data after data reconstruction, thereby ensuring the accuracy of the early warning information generated by the centering mechanism. Furthermore, by using the prediction results of the equipment health data of the servo device within the transformation time window, early warning of potential faults in the centering mechanism is achieved.

[0166] In some embodiments of this application, the first running data in the embodiments of this application includes at least two types of statistical data, one type of statistical data corresponds to a key statistical feature; one anomaly detection data corresponds to an unsupervised detection algorithm, and one anomaly detection data includes statistical anomaly detection data of each of the at least two types of statistical data.

[0167] For example, if at least two types of statistical data may include maximum value statistical data and average value statistical data, where the maximum value statistical data corresponds to the maximum value feature and the average value statistical data corresponds to the average value feature, the anomaly detection data TNS1 of unsupervised detection algorithm 1 may include statistical anomaly detection data TNS11 of maximum value statistical data and statistical anomaly detection data TNS12 of average value statistical data. The anomaly detection data TNS2 of unsupervised detection algorithm 2 may include statistical anomaly detection data TNS21 of maximum value statistical data and statistical anomaly detection data TNS22 of average value statistical data.

[0168] Based on this, step 1301 mentioned above may specifically include steps 13011 and 13012.

[0169] Step 13011: Based on the statistical anomaly detection data of each type of statistical data in the anomaly detection data corresponding to an unsupervised detection algorithm, determine the detection accuracy evaluation data corresponding to the unsupervised detection algorithm.

[0170] For example, taking the above example again, the statistical anomaly detection data of two types of statistical data in an unsupervised detection algorithm, namely unsupervised detection algorithm 1, namely TNS11 and TNS12, can be used to determine the detection accuracy evaluation data corresponding to unsupervised detection algorithm 1. ACC can be used. ic The expression ACC represents the unsupervised detection algorithm, where i is the identifier (1) and c is the number of key statistical features. The accuracy evaluation data corresponding to unsupervised detection algorithm 1 can be represented as ACC. 12 Similarly, the statistical anomaly detection data of the two types of statistical data in unsupervised detection algorithm 2, namely TNS21 and TNS22, can be used to determine the detection accuracy evaluation data corresponding to unsupervised detection algorithm 2. This can be represented as ACC. 22 , among which, ACC i ≥0.

[0171] Based on this, the Receiver Operating Characteristic (ROC) algorithm can be introduced, that is, the detection accuracy evaluation data (ACC) corresponding to an unsupervised detection algorithm can be determined through the following formulas (11) to (12). i :

[0172] ACC ic = ROC(Φ i (X) T ,y c ),c=0,1 (11)

[0173]

[0174] Where, Φ i (X)R n×1 For statistical anomaly detection data for each type of statistical data, y c The device health data corresponding to the first set of running data is identified by 0 for non-faulty data and 1 for faulty data. N represents the total amount of running data within the first time window. c This represents the amount of data in the c-th category of statistical data.

[0175] Step 13012: Based on the detection accuracy evaluation data corresponding to each of the at least two different unsupervised detection algorithms, determine the correlation data between each pair of unsupervised detection algorithms.

[0176] For example, the detection accuracy evaluation data ACC1 corresponding to unsupervised detection algorithm 1 and the detection accuracy evaluation data ACC2 corresponding to unsupervised detection algorithm 2 are used as examples to illustrate the determination of the correlation data between unsupervised detection algorithm 1 and unsupervised detection algorithm 2. Here, the correlation data can be represented by the Pearson coefficient.

[0177] Based on this, the correlation data between each pair of unsupervised detection algorithms in at least two different unsupervised detection algorithms can be determined by the following formula (13):

[0178] Define the selection function Ψ as:

[0179]

[0180] Where, stΦ j ∈{S},ρ(Φ i ,Φ j ) is the Pearson coefficient, used to measure the correlation between any two unsupervised detection algorithms, and S is the original feature space.

[0181] Therefore, when determining the correlation data between any two unsupervised detection algorithms, different categories of key statistical features and statistical anomaly detection data for each category of key statistical data can be introduced. This enhances the feature space, ensures its diversity, and alleviates the class imbalance problem of statistical features in predictive maintenance identification. Furthermore, using unsupervised detection algorithms as anchors, the accuracy of determining the detection accuracy evaluation data for unsupervised detection algorithms can be improved through different categories of key statistical features and statistical anomaly detection data for each category. This, in turn, improves the accuracy of determining the correlation data between any two unsupervised detection algorithms based on the detection accuracy evaluation data, effectively enhancing the accuracy of selecting unsupervised detection algorithms.

[0182] In some embodiments of this application, the data reconstruction process in step 1302 mentioned above is provided in at least two ways. Based on this, step 1302 may specifically include step 13021 or step 13022, as shown below.

[0183] Step 13021: Perform data reconstruction processing on the target anomaly detection data detected by each of the two unsupervised detection algorithms corresponding to the target relevance data and the first running data to obtain the second running data.

[0184] For example, we will still use two unsupervised detection algorithms, namely Unsupervised Detection Algorithm 1 and Unsupervised Detection Algorithm 2, with the first running data being X. That is, the target anomaly detection data of Unsupervised Detection Algorithm 1 is TNS1, and the target anomaly detection data of Unsupervised Detection Algorithm 2 is TNS2. Based on this, we can concatenate the two target anomaly detection data and the first running data X, that is, concatenate TNS1, TNS2, and X to obtain the second running data (X+TNS1+TNS2). It should be noted that the order of data concatenation is not limited; that is, the second running data can also be (TNS1+TNS2+X), (TNS1+X+TNS2), etc.

[0185] Therefore, the target anomaly detection data detected by each of the two unsupervised detection algorithms corresponding to the target relevance data can be used to participate in data reconstruction, thereby improving the accuracy of the second running data. Then, the equipment health data of the servo device can be determined through the second running data after data reconstruction, which can improve the accuracy of determining the equipment health data of the servo device and thus ensure the accuracy of the early warning information generated by the centering mechanism.

[0186] Alternatively, in step 13022, the target anomaly detection data detected by either of the two unsupervised detection algorithms corresponding to the target relevance data and the first running data are reconstructed to obtain the second running data.

[0187] For example, we will still use two unsupervised detection algorithms, namely unsupervised detection algorithm 1 and unsupervised detection algorithm 2, with the first running data being X. That is, the target anomaly detection data of unsupervised detection algorithm 1 is TNS1, and the target anomaly detection data of unsupervised detection algorithm 2 is TNS2. Based on this, we can arbitrarily choose one of TNS1 and TNS2 and concatenate it with X to obtain the second running data (X+TNS1) or (X+TNS2).

[0188] Therefore, target anomaly detection data detected by either of the two unsupervised detection algorithms corresponding to the degree of relevance to the target can be selected to participate in data reconstruction. This reduces data reconstruction resources, reduces the amount of data reconstruction computation, improves the efficiency of data reconstruction processing, and thus improves the efficiency of generating early warning information for the centering mechanism.

[0189] In addition, this application embodiment also provides a method for determining the above-mentioned key statistical features. Based on this, the equipment fault early warning method provided in this application embodiment may further include steps 1501 and 1502.

[0190] Step 1501: Based on the equipment identification information of the battery swapping station centering mechanism, obtain a feature list corresponding to the servo equipment of the battery swapping station centering mechanism. The feature list includes at least two statistical features and the importance value corresponding to each of the at least two statistical features.

[0191] For example, the equipment identification information of the centering mechanism of the battery swapping station can be represented as number 001. Number 001 can correspond to the codes of 4 servo devices, namely 0-left front position, 1-right front position, 2-left rear position, and 3-right front position. Based on these equipment identification information, the feature list corresponding to the equipment identification information is obtained from the feature list database. The feature list can include 3 statistical features, such as maximum value feature, minimum value feature, and average value feature, as well as the importance value of 0.7 corresponding to the maximum value feature, the application degree value of 0.8 corresponding to the minimum value feature, and the importance value of 0.9 corresponding to the average value feature.

[0192] Step 1502: Based on the importance value corresponding to each statistical feature, determine the key statistical feature from at least two statistical features, wherein the importance value of the key statistical feature is greater than or equal to the preset importance value.

[0193] For example, if the preset importance value is 0.8, the key statistical features can be the minimum value feature and the average value feature.

[0194] Therefore, through steps 1501 and 1502 above, a feature list that contributes most to the servo device fault samples can be obtained. To address the problems of scarce fault samples and difficulty in quantifying the servo device status, and by using a feature list corresponding to the pre-constructed battery swapping station centering mechanism, the most suitable key statistical features for the servo device can be selected. This adapts to the differences between battery swapping stations and the differences in the centering mechanism. Furthermore, by selecting the unsupervised detection algorithm that best matches the key statistical feature distribution of the centering mechanism, the servo device status is assigned a value. This enhances the feature space, ensures the diversity of the feature space, effectively reduces the errors and irrationality that may occur during the manual scoring of the device status, and alleviates the problem of class imbalance of statistical features in predictive maintenance identification.

[0195] It should be noted that the equipment fault early warning method provided in this application embodiment also provides a process for determining a feature list. Based on this, the equipment fault early warning method involved in this application embodiment may also include steps 1601 to 1604, as shown below.

[0196] Step 1601: Obtain the first sample operation data and the sample device health data corresponding to the first sample operation data of the servo device of the battery swapping station centering mechanism within the third time window; wherein, the first sample operation data includes at least two types of sample statistical data, one type of sample statistical data corresponds to one sample statistical feature, and the sample device health data includes the sample statistical device health data of each type of sample statistical data in at least two types of sample statistical data.

[0197] For example, the first sample operation data may include classifying the equipment status based on the data of the servo equipment 24 hours before the failure, dividing the sample operation cycle into different time periods, and according to the sample operation cycle, obtaining the sample operation cycle statistical feature data of the sample torque signal, sample displacement signal or sample speed signal within each sample operation cycle, as well as the sample key step sequence statistical feature data of the key steps of the sample torque signal, sample displacement signal or sample speed signal.

[0198] The sample statistical features corresponding to the sample torque signal may include maximum value features, minimum value features, average value features, standard deviation features, and root mean square (RMS) features. The corresponding sample statistical data can be the sample maximum value statistical data (MAX) corresponding to the maximum value feature, the sample minimum value statistical data (MIN) corresponding to the minimum value feature, the sample average value statistical data (MEAN) corresponding to the average value feature, the sample standard deviation statistical data (STD) corresponding to the standard deviation feature, and the sample root mean square (RMS) statistical data corresponding to the RMS feature. It also involves obtaining the sample equipment health data corresponding to each type of sample statistical data, i.e., whether it is faulty or not. In some embodiments, sample statistical features can be further extracted according to two dimensions: the sample running cycle i and the sample key step sequence. The extracted sample statistical features may include the sample running cycle statistical feature data of the sample torque signal, sample displacement signal, or sample velocity signal, and the sample key step sequence statistical feature data of the sample torque signal, sample displacement signal, or sample velocity signal. Assume x... mns For the displacement signal of the nth step in the mth period, a total of T sample periods were collected. The step set is P′, and the key step set is P, S. ij Let be the number of signals in the j-th step of the i-th cycle. The statistical characteristic data of the sample running cycle for determining the running cycle of the sample torque signal, sample displacement signal or sample velocity signal mentioned above can be referred to the above formulas (1) to (5), and the statistical characteristic data of the key step sequence for determining the key step sequence of the sample torque signal, sample displacement signal or sample velocity signal mentioned above can be referred to the above formulas (6) to (10).

[0199] It should be noted that the sample step sequence signals of the servo device for obtaining the centering mechanism mentioned above can be determined according to the flow of the servo device's execution steps, and a corresponding step sequence signal is set for each step, for example, using 50 as the start state step sequence s and 67 as the end state step sequence e, thereby dividing the servo device into independent sample operation cycles. Based on the action principle and the comparison of the signal curve trends of each step before and after the fault, its key steps are determined, as shown in Table 1:

[0200] Table 1 Servo Device Step Sequence Analysis

[0201]

[0202] Here, we analyze the attribute characteristics of the centering mechanism besides signal characteristics. Since each battery swapping station's ID is unique, it is used as the characteristic representing the battery swapping station. Because the centering mechanism has four corresponding servos for each car tire, the servos at different positions are encoded as follows: 0 - front left position, 1 - front right position, 2 - rear left position, 3 - front right position, which serve as the servo position characteristics. This completes the construction of the first sample of operational data.

[0203] In some embodiments, the first sample running data can be divided into a training sample running dataset and a test sample running dataset, and the training sample running dataset and the test sample running dataset can be further divided into statistical data for each type of sample and sample statistical device health data for each type of sample statistical data.

[0204] Step 1602: Train the random forest classifier model using statistical data of each class of samples and health data of the sample statistical equipment corresponding to the statistical data of each class of samples until the second reference condition is met, and obtain the feature selection model. The model data of the feature selection model includes the sample importance value corresponding to each sample statistical feature.

[0205] For example, the extracted statistical data of each class of samples is used as the input variable X. Simultaneously, the sample statistical data of the equipment health data Y, along with the statistical data of each class of samples, is used to train a random forest classifier model. A random seed is set to ensure the repeatability of the results until a second reference condition is met, thus obtaining the feature selection model. The second reference condition can be that the difference between the output variable Y and the reference output variable is less than or equal to a preset difference level.

[0206] Step 1603: Based on the sample importance value corresponding to each sample statistical feature, select at least one key sample statistical feature from at least two sample statistical features whose sample importance value is greater than or equal to the preset sample importance value.

[0207] For example, if the preset sample importance value is 0.7, then the key statistical features of the samples corresponding to the sample importance values ​​greater than or equal to 0.7 can be selected from the sample importance values ​​corresponding to at least two sample statistical features.

[0208] Step 1604: Generate a feature list based on at least one key statistical feature of the samples. For example, select the top d key statistical features with a sample importance value greater than or equal to 0.7 from the sample importance values ​​corresponding to at least two sample statistical features to generate the feature list. This feature list corresponds to the battery swapping station and the centering mechanism and servo equipment within the battery swapping station.

[0209] Therefore, based on the sample operation data of the servo equipment of the swapping station centering mechanism in the third time window and its individual sample equipment health data, such as the sample operation data of the servo equipment when it fails and the data of the previous day, a sample for detection can be constructed to realize early warning of equipment. In addition, this scheme uses a random forest classifier model to obtain the feature list that contributes the most to the servo equipment failure samples. In order to deal with the problem of scarce failure samples and difficulty in quantifying the status of servo equipment, the key statistical features most suitable for the servo equipment can be selected through the feature list corresponding to the swapping station centering mechanism in advance. This adapts to the differences between swapping stations and the differences in the centering mechanism. Then, by selecting the unsupervised detection algorithm that best matches the key statistical feature distribution of the centering mechanism, the status of the servo equipment is assigned a value, which enhances the feature space, ensures the diversity of the feature space, effectively reduces the errors and irrationality that may be generated in the process of manually scoring the status of equipment, and alleviates the problem of class imbalance of statistical features in predictive maintenance identification.

[0210] In this embodiment of the application, a classifier can be used to determine whether the servo device is faulty in the second time window. Based on this, the above step 1303 may specifically include step 13031 and step 13032.

[0211] Step 13031: Input the second operating data into the classifier. Through the classifier, determine the classification result of whether the servo device is faulty in the second time window. The time of the first time window is earlier than the time of the second time window. The classifier is trained by the operating data of the servo device of the mechanism in the battery swapping station in the third time window. The time of the third time window is earlier than the time of the first time window.

[0212] The classifier in this embodiment may include an ensemble learning algorithm based on gradient boosting decision tree (GBDT), namely the XGBoost classifier.

[0213] For example, taking the second running data as (X+TNS1+TNS2) as an example, the second running data is input into the XGBoost classifier. The XGBoost classifier determines whether the servo device is faulty in the second time window. If the classification result output by the XGBoost classifier is 0, it means that the servo device is expected to be non-faulty in the second time window. If the classification result output by the XGBoost classifier is 1, it means that the servo device is expected to be faulty in the second time window.

[0214] Step 13032: Determine the device health data of the servo device based on the classification results of whether the servo device is faulty within the second time window.

[0215] Therefore, the second running data after data reconstruction can be detected by a classifier to determine the equipment health data of the servo device, which can improve the accuracy of determining the equipment health data of the servo device, thereby ensuring the accuracy of the early warning information generated by the centering mechanism. Furthermore, by using the prediction results of the equipment health data of the servo device within the conversion time window, early warning of potential faults of the centering mechanism can be achieved.

[0216] In some embodiments, the first time window in this application includes at least two sub-time windows, and the second running data includes sub-running data within each sub-time window. Based on this, step 13031 may specifically include steps 130311 and 130312, as shown below.

[0217] Step 130311: Input the sub-run data within each sub-time window into the classifier. Through the classifier, obtain the sub-classification result corresponding to the sub-run data within each sub-time window.

[0218] For example, if the number of sub-time windows is 4, then the sub-running data within 4 sub-time windows will be input into the XGBoost classifier. The XGBoost classifier will obtain the sub-classification result corresponding to the sub-running data within each of the 4 sub-time windows, that is, the sub-classification result corresponding to the sub-running data within the first sub-time window is 0, the sub-classification result corresponding to the sub-running data within the third sub-time window is 0, the sub-classification result corresponding to the sub-running data within the third sub-time window is 0, and the sub-classification result corresponding to the sub-running data within the first sub-time window is 1.

[0219] Step 130312: Based on the sub-classification results, determine the classification result of whether the servo device is faulty within the second time window.

[0220] For example, the classification result of whether the servo device is faulty in the second time window can be determined based on the sub-classification result corresponding to the sub-operation data in the first sub-time window being 0, the sub-classification result corresponding to the sub-operation data in the third sub-time window being 0, and the sub-classification result corresponding to the sub-operation data in the first sub-time window being 1.

[0221] Therefore, a first time window can be defined, and the sub-classification results corresponding to the sub-operation data within at least two sub-time windows after the first time window is defined can be used to statistically classify whether the servo device is faulty in the second time window. This avoids the situation where the servo device switches between faulty and non-faulty states due to the unstable health status of its components, and improves the accuracy of determining whether the servo device is faulty in the second time window.

[0222] In some embodiments, the subclassification results in this application include fault results or non-fault results. Based on this, step 13032 may specifically include steps 130321 and 130322, as shown below.

[0223] Step 130321: Based on the first number of fault results in the subclassification results and the total number of subclassification results, determine the proportion of fault results in the total number of subclassification results.

[0224] For example, the proportion of fault results in the total number of subcategories can be determined using the following formula (14): HealthValue

[0225]

[0226] Here, count_0 and count_1 represent the counts of the sub-classification prediction results corresponding to the sub-run data within the sub-time window that are 0 and 1, respectively.

[0227] Step 130322: Determine the proportion of fault results in the total number of sub-classification results as the device health data of the servo device.

[0228] Therefore, by detecting the sub-operational data within at least two sub-time windows after the first time window is divided, and taking the sub-classification results within the first time window to participate in the calculation of the proportion of fault results in the total number of sub-classification results, the impact of unstable operation data quality of servo devices on fault results is mitigated, and the accuracy of the classification results for determining whether the servo device is faulty in the second time window is improved.

[0229] It should be noted that the embodiments of this application also provide a process for training a classifier in the equipment fault early warning method provided in the embodiments of this application. Based on this, the equipment fault early warning method involved in the embodiments of this application may also include steps 1701 to 1705, as shown below.

[0230] Step 1701: Obtain the first sample running data and the sample device health data corresponding to the first sample running data of the servo device of the battery swapping station centering mechanism within the third time window.

[0231] For example, such as Figure 2 As shown, based on the key statistical features of the samples obtained in steps 1601 to 1604 above, the first sample running data extracted according to the key statistical features of the samples is used as the input variable X∈R. n×d Unsupervised representation learning is performed, where X represents the set of n data points and d features, and y is the health data of the sample device, with 1 representing a fault point and 0 representing a non-fault point.

[0232] Step 1702: The first sample running data is detected by at least two different unsupervised detection algorithms to obtain at least two sample anomaly detection data of the servo device. The sample anomaly detection data is used to indicate the degree of anomaly of the first sample running data.

[0233] For example, during the learning process, KNN, LOF, HBOS, and OCSVM can be used as unsupervised anomaly detection methods to generate representation results. The unsupervised representation learning process is defined as a score mapping function Φ, and each method outputs a real-valued vector. As the corresponding outlier score TNS i This is a new data representation describing the degree of anomaly of the data points, which can be shown in the following formula (15). The anomaly detection outputs of different methods are used to construct a transformation function matrix, which is the sample anomaly detection data:

[0234]

[0235] Step 1703: Based on at least two sample anomaly detection data, determine the sample correlation data between each pair of unsupervised detection algorithms in at least two different unsupervised detection algorithms.

[0236] For example, for Φ i Given n data points of (X), let N be the number of samples. c The number of samples in class c is represented by ROC. The correlation between the samples is calculated using formulas (11) to (13) above, which will not be elaborated here.

[0237] Step 1704: Perform data reconstruction processing on the sample target anomaly detection data detected by the two unsupervised detection algorithms corresponding to the sample target correlation data and the first sample running data to obtain the second sample running data of the servo device in the third time window. The sample target correlation data includes the sample correlation data in the sample correlation data between each two unsupervised detection algorithms where the data value of the correlation data is greater than or equal to a preset threshold.

[0238] For example, based on accuracy and a selection function, this embodiment of the application selects p (p≤k) TNSs with the highest scores from the generated anomaly score matrix Φ(X), and merges them with the first sample running data X to form new second sample running data. This allows the feature space to maintain a balance between accuracy and diversity.

[0239] Step 1705: Based on the second running data and the health data of the sample equipment, train the sample classifier until the first reference training condition is met, and obtain the classifier.

[0240] For example, an XGBoost classifier is trained on a new feature space using second running data, and its output is used as a prediction of the device's operating status. The sample classifier is evaluated by maximizing the F1-score and ROC-AUC score, and it is determined whether the difference between the output of the sample classifier and the sample device health data is less than or equal to a reference difference. If the difference is less than or equal to the reference difference, the first reference training condition is met, and the classifier is obtained.

[0241] It should be noted that since the centering mechanism has four corresponding servos for each vehicle tire, the analysis of the design of the four servos reveals that the motion parameters are logically consistent. Considering that training the model separately for the four servos would be redundant, but the states of different servos cannot be completely identical, this embodiment encodes the servo devices to enable unified training and reduce the waste of training resources.

[0242] Therefore, semi-supervised learning can be achieved through supervised learning of unsupervised detection algorithms and classifiers to train the classifier. The classifier is then used to detect the second running data after data reconstruction to determine the equipment health data of the servo device. This improves the accuracy of determining the equipment health data of the servo device, thereby ensuring the accuracy of the early warning information generated by the centering mechanism. Furthermore, by using the prediction results of the equipment health data of the servo device within the conversion time window, early warning of potential faults in the centering mechanism is achieved.

[0243] It should be noted that the first sample running data in this application embodiment includes sample statistical data of at least one key statistical feature of the sample in the feature list, and the sample device health data includes sample statistical device health data of the sample statistical data. Based on this, during the training process, there may be situations where the first reference training conditions are not met. Based on this, the device fault early warning method may also include step 1706, adjusting the number of key statistical features of the sample in the feature list and executing the target step when the first reference training conditions are not met.

[0244] The target steps include:

[0245] The first sample running data is detected by at least two different unsupervised detection algorithms to obtain at least two sample anomaly detection data of the servo device. The sample anomaly detection data is used to indicate the degree of anomaly of the first sample running data.

[0246] Based on at least two sample anomaly detection data, determine the sample correlation data between each pair of at least two different unsupervised detection algorithms;

[0247] The sample target anomaly detection data detected by the two unsupervised detection algorithms corresponding to the sample target correlation data and the first sample running data are reconstructed to obtain the second sample running data of the servo device in the third time window. The sample target correlation data includes the sample correlation data between each two unsupervised detection algorithms whose correlation data values ​​are greater than or equal to a preset threshold.

[0248] The sample classifier is trained based on the second set of operational data and the health data of the sample equipment.

[0249] For example, if the condition is not met, the d value of the feature list is adjusted, and the above target steps are repeated until the first reference training condition is met.

[0250] Therefore, if the feature list is completed and the training sample classifier does not meet the first reference training condition, the number of key statistical features of the samples in the feature list can be adjusted to address the problem of scarce fault samples and difficulty in quantifying the status of servo equipment. This allows for the selection of the most suitable key statistical features for the servo equipment. Furthermore, by selecting the unsupervised detection algorithm that best matches the distribution of key statistical features of the centering mechanism, the status of the servo equipment can be assigned a value. This improves the accuracy of feature space construction for centering servo samples, effectively reduces the errors and irrationality that may occur during manual scoring of equipment status, and alleviates the problem of class imbalance of statistical features in predictive maintenance identification.

[0251] Regarding step 140, in this embodiment of the application, before step 140, the device fault early warning method may also include step 1801 or step 1802, as shown below.

[0252] Step 1801: If the percentage is greater than or equal to the reference percentage, determine that the servo device characterizes a fault in the servo device.

[0253] Step 1802: If the percentage is less than the reference percentage, determine that the servo device characterization servo device has no fault.

[0254] For example, with a reference percentage of 50%, if the percentage is greater than or equal to 50%, the servo device is determined to be faulty. Conversely, if the percentage is less than 50%, the servo device is determined to be without fault.

[0255] Therefore, by using multiple sub-classification results within the first time window to calculate the proportion of fault results in the total number of sub-classification results, and combining this with the reference proportion, the classification result of the servo device within the second time window can be determined. This can mitigate the impact of unstable operating data quality of the servo device on the fault results and improve the accuracy of determining whether the servo device is faulty within the second time window.

[0256] To better illustrate the device fault early warning method provided in the embodiments of this application, the following is combined with... Figure 3 Please provide a detailed explanation.

[0257] Step 301: Obtain the first sample operating data and the corresponding sample device health data of the servo equipment of the battery swapping station centering mechanism within the third time window. For details, please refer to step 1601 above, which will not be repeated here.

[0258] Step 302: Construct a feature selection model, which can be referred to in step 1602 above.

[0259] Step 303: Obtain the feature list. For details, please refer to steps 1603 to 1604 above. They will not be repeated here.

[0260] Step 304: Complete the training of the semi-supervised anomaly detection model. For details, please refer to steps 1701 to 1705 above, which will not be repeated here.

[0261] Step 305: Evaluate the semi-supervised anomaly detection model. Specifically, refer to the description of satisfying the first reference training condition in step 1705 above, and step 1706, which will not be repeated here.

[0262] Step 306: If the first reference training condition is met, proceed to step 307. If the first reference training condition is not met, return to step 302.

[0263] Step 307, complete the construction of the semi-supervised anomaly detection model, i.e. Figure 2 The content shown.

[0264] Step 308: Obtain the first operating data of the servo device of the battery swapping station centering mechanism within the first time window.

[0265] Step 309, use the first running data as... Figure 2 X in the middle, through, as Figure 2 The semi-supervised anomaly detection model shown determines the device health data of the servo device. Figure 2 The equipment status prediction results are as follows. The specific process can be found in steps 110 to 130 above.

[0266] Step 310: If the device health data indicates that the servo device is faulty, generate a warning message for the centering mechanism.

[0267] Therefore, in this embodiment, the action sequence and corresponding signals of the equipment can be analyzed, key steps can be selected to form an operating cycle, and then the statistical features of each operating cycle and the statistical features of a single step signal can be extracted, thereby effectively capturing the regularity and anomalies in the equipment operation process. Secondly, to address the problem of scarce fault samples and difficulty in quantifying equipment status, this solution uses random forest to obtain a feature list that contributes the most to equipment fault samples. Then, a semi-supervised anomaly detection model is constructed. After learning and selecting the representation of the equipment status, the representation learning results are used to enhance the feature space and train a classifier to judge the equipment status. Finally, considering the shortcomings of existing label generation methods, the equipment health conversion score is calculated based on the status prediction results within the time window, thereby achieving early warning of potential faults in the centering mechanism.

[0268] This application also provides a device for early warning of equipment failure, specifically combined with... Figure 4 Please provide a detailed explanation.

[0269] Figure 4 This is a schematic diagram of the structure of a device fault early warning device provided in one embodiment of this application.

[0270] In some embodiments of this application, Figure 4 The device fault warning device shown can be installed in the computer equipment provided in the embodiments of this application.

[0271] Based on this, such as Figure 4 As shown, the equipment fault early warning device 40 may specifically include:

[0272] The acquisition module 401 is used to acquire the first running data of the servo device of the battery swapping station centering mechanism within the first time window;

[0273] The detection module 402 is used to detect the first running data using at least two different unsupervised detection algorithms to obtain at least two abnormal detection data of the servo device. The abnormal detection data is used to indicate the degree of abnormality of the first running data.

[0274] The determination module 403 is used to determine the device health data of the servo device based on at least two anomaly detection data and the first operating data;

[0275] The generation module 404 is used to generate early warning information for the centering mechanism when the equipment health data indicates that the servo device has a fault.

[0276] In this way, by acquiring the first operating data of the servo equipment of the battery swapping station's centering mechanism within the first time window, and using the detection results of the first operating data through an unsupervised detection algorithm, the status of the servo equipment is assigned a value, thus obtaining anomaly detection data. This anomaly detection data can be used to indicate the degree of anomaly in the first operating data of the servo equipment, effectively reducing the errors and irrationality that may occur during the manual scoring of the servo equipment status, thereby improving the accuracy of servo equipment detection. Then, based on the anomaly detection data obtained from the detection of the first operating data and the first operating data of the servo equipment, the equipment health data of the servo equipment is determined. The equipment health data is used to determine whether the servo equipment has a fault. If the servo equipment has a fault, an early warning information for the centering mechanism is generated, which can realize early warning of potential faults of the servo equipment, improve the reliability of the centering mechanism operation, and avoid the inability to replenish power to vehicles due to servo equipment failure.

[0277] The equipment fault early warning device 40 in the embodiments of this application will be described in detail below.

[0278] In some embodiments of this application, the determining module 403 may specifically be used to determine the correlation data between each pair of unsupervised detection algorithms in at least two different unsupervised detection algorithms based on at least two anomaly detection data.

[0279] The target anomaly detection data detected by the two unsupervised detection algorithms corresponding to the target correlation data and the first running data are reconstructed to obtain the second running data of the servo device within the first time window. The target correlation data includes the correlation data between each two unsupervised detection algorithms whose data value is greater than or equal to a preset threshold.

[0280] Based on the second set of operational data, determine the device health data of the servo device.

[0281] In some embodiments of this application, the determining module 403 may be specifically used to determine the detection accuracy evaluation data corresponding to an unsupervised detection algorithm based on the statistical anomaly detection data of each type of statistical data in the anomaly detection data corresponding to an unsupervised detection algorithm when the first running data includes at least two types of statistical data, one type of statistical data corresponds to a key statistical feature, one anomaly detection data corresponds to an unsupervised detection algorithm, and the anomaly detection data includes statistical anomaly detection data of each type of statistical data in the at least two types of statistical data.

[0282] Based on the detection accuracy evaluation data corresponding to each of the at least two different unsupervised detection algorithms, determine the correlation data between each pair of unsupervised detection algorithms.

[0283] In some embodiments of this application, the determining module 403 may be specifically used to perform data reconstruction processing on the target anomaly detection data detected by each of the two unsupervised detection algorithms corresponding to the target relevance data and the first running data to obtain the second running data;

[0284] or,

[0285] The target anomaly detection data detected by either of the two unsupervised detection algorithms corresponding to the target relevance data and the first running data are reconstructed to obtain the second running data.

[0286] In some embodiments of this application, the acquisition module 401 can also be used to acquire a feature list corresponding to the servo device of the swap station centering mechanism based on the device identification information of the swap station centering mechanism. The feature list includes at least two statistical features and an importance value corresponding to each of the at least two statistical features.

[0287] The determination module 403 can also be used to determine key statistical features from at least two statistical features based on the importance value corresponding to each statistical feature, wherein the importance value of the key statistical feature is greater than or equal to a preset importance value.

[0288] In some embodiments of this application, the determining module 403 can be specifically used to input the second operating data into the classifier, and through the classifier, determine the classification result of whether the servo device is faulty in the second time window. The time of the first time window is earlier than the time of the second time window. The classifier is trained by the operating data of the servo device of the mechanism in the battery swapping station in the third time window. The time of the third time window is earlier than the time of the first time window.

[0289] Based on the classification results of whether the servo device is faulty within the second time window, determine the device health data of the servo device.

[0290] In some embodiments of this application, the determining module 403 can be specifically used to input the sub-running data in each sub-time window into a classifier when the first time window includes at least two sub-time windows and the second running data includes sub-running data in each sub-time window, and obtain the sub-classification result corresponding to the sub-running data in each sub-time window through the classifier;

[0291] Based on the sub-classification results, determine whether the servo device is faulty within the second time window.

[0292] In some embodiments of this application, the determining module 403 may be specifically used to determine the proportion of fault results in the total number of subclassification results based on a first number of fault results and the total number of subclassification results when the subclassification results include fault results or non-fault results.

[0293] The proportion of the fault results in the total number of sub-classification results is determined as the device health data of the servo device.

[0294] In some embodiments of this application, the determining module 403 can also be used to determine the classification result of the servo device in the second time window as the classification result of the fault when the proportion is greater than or equal to the reference proportion.

[0295] If the percentage is less than the reference percentage, the classification result of the servo device in the second time window is determined to be a non-faulty classification result.

[0296] In some embodiments of this application, the acquisition module 401 can also be used to acquire the running signal of the servo device within a first time window, the running signal including a step sequence signal and at least one of the following target signals: torque signal, displacement signal, and speed signal;

[0297] The determination module 403 can also be used to determine the key steps of the servo device in the process of executing the work steps based on the step sequence signal and the step sequence annotation corresponding to the step sequence signal during the execution of the work steps by the servo device.

[0298] The determination module 403 can also be used to determine the key step sequence statistical feature data corresponding to the key step sequence based on the running signal, with the key step sequence as the first dimension; and to determine the running cycle statistical feature data corresponding to the running cycle of the servo device based on the running signal, with the running cycle of the servo device as the second dimension.

[0299] The generation module 404 can also be used to generate first running data based on key step sequence statistical feature data and running cycle statistical feature data.

[0300] In some embodiments of this application, the determining module 403 may be specifically used to filter at least one of the following first target signals corresponding to the first dimension from the running signals, with the key step sequence as the first dimension: a first torque signal, a first displacement signal, and a first speed signal;

[0301] Perform data statistics on at least one first target signal to obtain statistical data corresponding to the first target signal;

[0302] The statistical data corresponding to the first target signal is determined as the key step sequence statistical feature data corresponding to the key step sequence.

[0303] In some embodiments of this application, the determining module 403 may be specifically used to filter at least one of the following second target signals corresponding to the second dimension from the running signals, taking the running cycle of the servo device as the second dimension: a second torque signal, a second displacement signal, and a second speed signal.

[0304] Statistical analysis is performed on at least one first target signal to obtain statistical data corresponding to the second target signal;

[0305] The statistical data corresponding to the second target signal is determined as the operating cycle statistical feature data corresponding to the operating cycle of the servo device.

[0306] In some embodiments of this application, at least two different unsupervised detection algorithms in these embodiments include the following at least two algorithms:

[0307] Single-class support vector machine anomaly detection algorithm, K-nearest neighbor classification detection algorithm, local anomaly factor detection algorithm, and histogram-based anomaly value scoring detection algorithm.

[0308] In some embodiments of this application, the acquisition module 401 can also be used to acquire the first sample running data and the sample device health data corresponding to the first sample running data of the servo device of the battery swapping station centering mechanism within a third time window; wherein, the first sample running data includes at least two types of sample statistical data, one type of sample statistical data corresponds to one sample statistical feature, and the sample device health data includes the sample statistical device health data of each type of sample statistical data in at least two types of sample statistical data.

[0309] The equipment fault early warning device 40 also includes a training module, which is used to train the random forest classifier model by using the statistical data of each type of sample and the sample statistical equipment health data of each type of sample statistical data until the second reference condition is met, and obtain the feature selection model. The model data of the feature selection model includes the sample importance value corresponding to each sample statistical feature.

[0310] The equipment fault early warning device 40 also includes a screening module, which is used to screen at least one key statistical feature of a sample from at least two sample statistical features based on the sample importance value corresponding to each sample statistical feature. The key statistical feature of a sample is greater than or equal to a preset sample importance value.

[0311] The generation module 404 can also be used to generate a feature list based on at least one key statistical feature of a sample.

[0312] In some embodiments of this application, the acquisition module 401 can also be used to acquire the first sample running data of the servo device of the battery swapping station centering mechanism within a third time window and the sample device health data corresponding to the first sample running data.

[0313] The detection module 402 can also be used to detect the first sample running data by using at least two different unsupervised detection algorithms to obtain at least two sample anomaly detection data of the servo device. The sample anomaly detection data is used to indicate the degree of anomaly of the first sample running data.

[0314] The determination module 403 can also be used to determine, based on at least two sample anomaly detection data, the sample correlation data between each pair of unsupervised detection algorithms in at least two different unsupervised detection algorithms;

[0315] The equipment fault early warning device 40 also includes a reconstruction module, which is used to perform data reconstruction processing on the sample target anomaly detection data and the first sample running data detected by the two unsupervised detection algorithms corresponding to the sample target correlation data, to obtain the second sample running data of the servo device in the third time window. The sample target correlation data includes the sample correlation data in the sample correlation data between each two unsupervised detection algorithms where the data value of the correlation data is greater than or equal to a preset threshold.

[0316] The equipment fault early warning device 40 also includes a training module, which is used to train a sample classifier based on the second operating data and sample equipment health data until the first reference training condition is met, and thus obtain the classifier.

[0317] In some embodiments of this application, the equipment fault early warning device 40 further includes an adjustment module, which is used to adjust the number of sample key statistical features in the feature list and execute the target step when the first sample running data includes sample statistical data of at least one sample key statistical feature in the feature list, the sample equipment health data includes sample statistical equipment health data of the sample statistical data, and the first reference training condition is not met.

[0318] The target steps include:

[0319] The first sample running data is detected by at least two different unsupervised detection algorithms to obtain at least two sample anomaly detection data of the servo device. The sample anomaly detection data is used to indicate the degree of anomaly of the first sample running data.

[0320] Based on at least two sample anomaly detection data, determine the sample correlation data between each pair of at least two different unsupervised detection algorithms;

[0321] The sample target anomaly detection data detected by the two unsupervised detection algorithms corresponding to the sample target correlation data and the first sample running data are reconstructed to obtain the second sample running data of the servo device in the third time window. The sample target correlation data includes the sample correlation data between each two unsupervised detection algorithms whose correlation data values ​​are greater than or equal to a preset threshold.

[0322] The sample classifier is trained based on the second set of operational data and the health data of the sample equipment.

[0323] This application also provides a computer device. (Specifically combined with...) Figure 5 Please provide a detailed explanation.

[0324] Figure 5 This is a schematic diagram of the structure of a computer device provided in one embodiment of this application.

[0325] like Figure 5 As shown, the computer device may include the computer device for executing the device fault early warning method involved in the embodiments of this application. The computer device may include a processor 501 and a memory 502 storing computer program instructions.

[0326] Specifically, the processor 501 may include a central processing unit (CPU), an application-specific integrated circuit (ASTC), or one or more integrated circuits that can be configured to implement the embodiments of this application.

[0327] Memory 502 may include a large-capacity memory for data or instructions. For example, and not limitingly, memory 502 may include a hard disk drive (HDD), a floppy disk drive, flash memory, optical disk, magneto-optical disk, magnetic tape, or a Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, memory 502 may include removable or non-removable (or fixed) media. Where appropriate, memory 502 may be internal or external to the integrated gateway disaster recovery device. In a particular embodiment, memory 502 is non-volatile solid-state memory. In a particular embodiment, memory 502 includes solid-state storage (ROM). Where appropriate, the ROM may be a mask-programmed ROM, a programmable ROM (PROM), an erasable PROM (EPROM), an electrically erasable PROM (EEPROM), an electrically rewritable ROM (EAROM), or flash memory, or a combination of two or more of these.

[0328] The processor 501 reads and executes computer program instructions stored in the memory 502 to implement any of the device fault early warning methods in the above embodiments.

[0329] In one example, the computer device may also include a communication interface 503 and a bus 510. Wherein, as... Figure 5 As shown, the processor 501, memory 502, and communication interface 503 are connected through bus 510 and complete communication with each other.

[0330] The communication interface 503 is mainly used to realize communication between various modules, devices, units and / or equipment in the embodiments of this application.

[0331] Bus 510 includes hardware, software, or both, that couples components of a flow control device together. For example, and not limitingly, the bus may include an Accelerated Graphics Port (AGP) or other graphics bus, an Enhanced Industry Standard System (ETSA) bus, a Front Side Bus (FSB), an HyperTransport (HT) interconnect, an Industry Standard System (TSA) bus, an Infinite Bandwidth Interconnect, a Low Pin Count (LPC) bus, a memory bus, a Microchannel System (MCA) bus, a Peripheral Component Interconnect (PCT) bus, a PCT-Express (PCT-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local (VLB) bus, or other suitable buses, or combinations of two or more of these. Where appropriate, bus 510 may include one or more buses. Although specific buses are described and illustrated in embodiments of this application, this application contemplates any suitable bus or interconnect.

[0332] The data processing device can execute the device fault early warning method in the embodiments of this application, thereby achieving a combination of Figures 1 to 5 The equipment failure early warning method and apparatus are described.

[0333] Furthermore, in conjunction with the equipment fault early warning methods in the above embodiments, this application embodiment can provide a computer-readable storage medium for implementation. The computer-readable storage medium stores computer program instructions; when these computer program instructions are executed by a processor, they implement any of the equipment fault early warning methods in the above embodiments.

[0334] It should be clarified that this application is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of this application is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of this application.

[0335] The functional blocks shown in the above block diagram can be implemented as hardware, software, firmware, or a combination thereof. When implemented in hardware, they can be, for example, electronic circuits, application-specific integrated circuits (ASTCs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of this application are programs or code segments used to perform the required tasks. Programs or code segments can be stored on a machine-readable medium or transmitted over a transmission medium or communication link via data signals carried on a carrier wave. "Machine-readable medium" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, fiber optic media, radio frequency (RF) links, etc. Code segments can be downloaded via computer networks such as the Internet, intranets, etc.

[0336] It should also be noted that the exemplary embodiments mentioned in this application describe methods or systems based on a series of steps or apparatus. However, this application is not limited to the order of the above steps; that is, the steps can be performed in the order mentioned in the embodiments, or in a different order, or several steps can be performed simultaneously.

[0337] The above are merely specific embodiments of this application. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, modules, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. It should be understood that the protection scope of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the protection scope of this application.

[0338] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and not to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. These modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application, and they should all be covered within the scope of the claims and specification of this application. In particular, as long as there is no structural conflict, the various technical features mentioned in the embodiments can be combined in any way. This application is not limited to the specific embodiments disclosed herein, but includes all technical solutions falling within the scope of the claims.

Claims

1. A method for early warning of equipment failure, characterized in that, include: Acquire the first operating data of the servo device of the battery swapping station centering mechanism within the first time window; The first running data is detected by at least two different unsupervised detection algorithms to obtain at least two abnormal detection data of the servo device, and the abnormal detection data is used to indicate the degree of abnormality of the first running data. Based on the at least two anomaly detection data and the first operating data, the device health data of the servo device is determined; When the device health data indicates that the servo device is faulty, a warning message is generated for the centering mechanism.

2. The method according to claim 1, characterized in that, The step of determining the device health data of the servo device based on the anomaly detection data and the first operating data includes: Based on the at least two anomaly detection data, determine the correlation data between each pair of unsupervised detection algorithms in the at least two different unsupervised detection algorithms; The target anomaly detection data detected by the two unsupervised detection algorithms corresponding to the target relevance data and the first running data are reconstructed to obtain the second running data of the servo device within the first time window. The target relevance data includes the relevance data in the relevance data between each pair of unsupervised detection algorithms whose data value is greater than or equal to a preset threshold. Based on the second operating data, determine the device health data of the servo device.

3. The method according to claim 2, characterized in that, The first running data includes at least two types of statistical data, one type of statistical data corresponds to a key statistical feature; one anomaly detection data corresponds to one unsupervised detection algorithm, and one anomaly detection data includes statistical anomaly detection data for each of the at least two types of statistical data; The step of determining the correlation data between each pair of unsupervised detection algorithms in the at least two different unsupervised detection algorithms based on the at least two anomaly detection data includes: Based on the statistical anomaly detection data of each type of statistical data in the anomaly detection data corresponding to the unsupervised detection algorithm, the detection accuracy evaluation data corresponding to the unsupervised detection algorithm is determined. Based on the detection accuracy evaluation data corresponding to each of the at least two different unsupervised detection algorithms, determine the correlation data between each pair of unsupervised detection algorithms.

4. The method according to claim 2, characterized in that, The step of reconstructing the target anomaly detection data detected by two unsupervised detection algorithms corresponding to the target relevance data and the first running data to obtain the second running data of the servo device within the first time window includes: The target anomaly detection data detected by each of the two unsupervised detection algorithms corresponding to the target relevance data and the first running data are reconstructed to obtain the second running data. or, The second running data is obtained by reconstructing the target anomaly detection data detected by either of the two unsupervised detection algorithms corresponding to the target relevance data and the first running data.

5. The method according to claim 3, characterized in that, The method further includes: Based on the equipment identification information of the battery swapping station centering mechanism, a feature list corresponding to the servo equipment of the battery swapping station centering mechanism is obtained. The feature list includes at least two statistical features and an importance value corresponding to each of the at least two statistical features. Based on the importance value corresponding to each statistical feature, a key statistical feature is determined from the at least two statistical features, wherein the importance value of the key statistical feature is greater than or equal to a preset importance value.

6. The method according to claim 2 or 3, characterized in that, The step of determining the device health data of the servo device based on the second operating data includes: The second operating data is input into the classifier. The classifier determines whether the servo device is faulty in the second time window. The time of the first time window is earlier than the time of the second time window. The classifier is trained by the operating data of the servo device of the mechanism in the battery swapping station in the third time window. The time of the third time window is earlier than the time of the first time window. Based on the classification results of whether the servo device is faulty within the second time window, the device health data of the servo device is determined.

7. The method according to claim 6, characterized in that, The first time window includes at least two sub-time windows, and the second running data includes sub-running data within each of the sub-time windows; The step of inputting the second operating data into the classifier and determining, through the classifier, whether the servo device is faulty within the second time window includes: The sub-running data within each sub-time window is input into the classifier, and the sub-classification result corresponding to the sub-running data within each sub-time window is obtained through the classifier; Based on the sub-classification results, a classification result is determined as to whether the servo device is faulty within the second time window.

8. The method according to claim 7, characterized in that, The sub-classification results include fault results or non-fault results; the determination of the servo device's health data based on the classification result of whether the servo device is faulty within the second time window includes: Based on the first number of fault results in the sub-classification results and the total number of sub-classification results, determine the proportion of fault results in the total number of sub-classification results; The proportion of the fault results in the total number of sub-classification results is determined as the device health data of the servo device.

9. The method according to claim 8, characterized in that, The method further includes: If the percentage is greater than or equal to the reference percentage, it is determined that the servo device is faulty. If the percentage is less than the reference percentage, it is determined that the servo device is fault-free.

10. The method according to claim 1, characterized in that, Before acquiring the first operating data of the servo device of the battery swapping station centering mechanism within the first time window, the method further includes: The servo device is acquired within the first time window, and the operating signal includes a step sequence signal and at least one of the following target signals: torque signal, displacement signal, and speed signal. Based on the step sequence signal and the step sequence annotation corresponding to the step sequence signal during the execution of the work steps by the servo device, the key steps of the servo device during the execution of the work steps are determined. Using the key step sequence as the first dimension, based on the running signal, determine the key step sequence statistical feature data corresponding to the key step sequence; and using the running cycle of the servo device as the second dimension, based on the running signal, determine the running cycle statistical feature data corresponding to the running cycle of the servo device. The first running data is generated based on the key step sequence statistical feature data and the running cycle statistical feature data.

11. The method according to claim 10, characterized in that, The step of determining the key step sequence statistical feature data corresponding to the key step sequence based on the running signal, using the key step sequence as the first dimension, includes: Using the key step sequence as the first dimension, at least one of the following first target signals corresponding to the first dimension is selected from the running signals: first torque signal, first displacement signal, and first velocity signal; Perform data statistics on the at least one first target signal to obtain statistical data corresponding to the first target signal; The statistical data corresponding to the first target signal is determined as the key step sequence statistical feature data corresponding to the key step sequence.

12. The method according to claim 10, characterized in that, The step of determining the statistical characteristic data of the operating cycle corresponding to the operating cycle of the servo device, based on the operating signal and using the operating cycle of the servo device as the second dimension, includes: Taking the operating cycle of the servo device as the second dimension, at least one of the following second target signals corresponding to the second dimension is selected from the operating signals: second torque signal, second displacement signal, and second speed signal; Perform data statistics on the at least one first target signal to obtain statistical data corresponding to the second target signal; The statistical data corresponding to the second target signal is determined as the operating cycle statistical feature data corresponding to the operating cycle of the servo device.

13. The method according to claim 1, characterized in that, The at least two different unsupervised detection algorithms include at least the following two algorithms: Single-class support vector machine anomaly detection algorithm, K-nearest neighbor classification detection algorithm, local anomaly factor detection algorithm, and histogram-based anomaly value scoring detection algorithm.

14. The method according to claim 5, characterized in that, The method further includes: The first sample operation data and the sample device health data corresponding to the first sample operation data of the servo device of the battery swapping station centering mechanism within the third time window are obtained; wherein, the first sample operation data includes at least two types of sample statistical data, one type of sample statistical data corresponds to one sample statistical feature, and the sample device health data includes the sample statistical device health data of each type of sample statistical data in the at least two types of sample statistical data; The random forest classifier model is trained using the statistical data of each class of samples and the health data of the sample statistical devices corresponding to the statistical data of each class of samples until the second reference condition is met, thereby obtaining a feature selection model. The model data of the feature selection model includes the sample importance value corresponding to each sample statistical feature. Based on the sample importance value corresponding to each of the sample statistical features, at least one key sample statistical feature with a sample importance value greater than or equal to a preset sample importance value is selected from at least two sample statistical features; The feature list is generated based on the at least one key statistical feature of the sample.

15. The method according to claim 6, characterized in that, The method further includes: Acquire the first sample operation data and the sample device health data corresponding to the first sample operation data of the servo device of the battery swapping station centering mechanism within the third time window; The first sample running data is detected by at least two different unsupervised detection algorithms to obtain at least two sample anomaly detection data of the servo device. The sample anomaly detection data is used to indicate the degree of anomaly of the first sample running data. Based on the at least two sample anomaly detection data, determine the sample correlation data between each pair of unsupervised detection algorithms in the at least two different unsupervised detection algorithms; The sample target anomaly detection data detected by the two unsupervised detection algorithms corresponding to the sample target correlation data and the first sample running data are reconstructed to obtain the second sample running data of the servo device in the third time window. The sample target correlation data includes the sample correlation data in the sample correlation data between each two unsupervised detection algorithms where the data value of the correlation data is greater than or equal to a preset threshold. Based on the second operating data and the health data of the sample device, a sample classifier is trained until the first reference training condition is met, thus obtaining the classifier.

16. The method according to claim 15, characterized in that, The first sample running data includes sample statistics data of at least one key statistical feature of the sample in the feature list, and the sample device health data includes sample statistics device health data of the sample statistics data; the method further includes: If the first reference training condition is not met, adjust the number of key statistical features of the samples in the feature list and execute the target step; The target step includes: The first sample running data is detected by at least two different unsupervised detection algorithms to obtain at least two sample anomaly detection data of the servo device. The sample anomaly detection data is used to indicate the degree of anomaly of the first sample running data. Based on the at least two sample anomaly detection data, determine the sample correlation data between each pair of unsupervised detection algorithms in the at least two different unsupervised detection algorithms; The sample target anomaly detection data detected by the two unsupervised detection algorithms corresponding to the sample target correlation data and the first sample running data are reconstructed to obtain the second sample running data of the servo device in the third time window. The sample target correlation data includes the sample correlation data in the sample correlation data between each two unsupervised detection algorithms where the data value of the correlation data is greater than or equal to a preset threshold. The sample classifier is trained based on the second operating data and the sample device health data.

17. A device for early warning of equipment failure, characterized in that, include: The acquisition module is used to acquire the first operating data of the servo device of the battery swapping station centering mechanism within the first time window; The detection module is used to detect the first running data using at least two different unsupervised detection algorithms to obtain at least two abnormal detection data of the servo device, wherein the abnormal detection data is used to indicate the degree of abnormality of the first running data. The determination module is used to determine the device health data of the servo device based on the at least two anomaly detection data and the first operating data; The generation module is used to generate early warning information for the centering mechanism when the device health data indicates that the servo device has a fault.

18. A computer device, characterized in that, include: The processor and the memory storing computer program instructions, wherein the processor, when executing the computer program instructions, implements the steps of the device fault early warning method as described in any one of claims 1-16.

19. A computer-readable storage medium, characterized in that, A computer program is stored on the computer-readable storage medium, which, when executed by a processor, implements the steps of the device fault early warning method as described in any one of claims 1-16.

Images