Automatic fault detection method for power device in electric tricycle controller

Abstract

The invention provides an automatic fault detection method for a power device in an electro-tricycle controller, and relates to the technical field of fault detection. Multiple pieces of real-time operation data of the power device are acquired through a sensing device group; deep learning is carried out based on the multiple pieces of real-time operation data, a fault detection channel is constructed to carry out fault dynamic detection, and a fault detection parameter set is obtained; performing coverage traversal and classification identification according to the fault detection parameter set to obtain a plurality of fault types; performing diagnosis traceability and fault positioning on the power device based on the fault type and the fault detection parameter set, and determining a fault position information set; and performing operation feedback on a power device in the controller according to the fault position information set, and generating a maintenance suggestion. According to the method and the device, the technical problems of insufficient fault detection precision and inaccurate fault positioning caused by lack of deep analysis capability on detection data in a traditional detection method are solved, accurate positioning of the fault type and the fault position is realized, and the fault detection precision and reliability are improved.

Description

Technical Field

[0001] The present application relates to the technical field of fault detection, and in particular to an automatic fault detection method for power devices in an electric tricycle controller. Background Art

[0002] The electric tricycle controller is responsible for regulating the drive of the motor, controlling the charge and discharge status of the battery, and ensuring the stable operation of the vehicle under different working conditions. Power devices (such as power modules, inverters, etc.) are key components in the controller, and their failure will directly affect the performance and safety of the vehicle.

[0003] At present, the fault detection method of power devices in electric tricycle controllers mainly adopts simple circuit testing and threshold judgment, and monitors the operating status of power devices in real time by installing sensors (such as current, voltage, temperature, etc. sensors, or temperature, vibration sensors). When the detected data exceeds the predetermined normal range (i.e., the threshold), a fault alarm is triggered or the protection state is entered. These detection methods can only provide limited information. When a small, complex or initial fault occurs in the power device, it is difficult to detect and distinguish the fault type in time, and early warning of the fault cannot be achieved. At the same time, there is a lack of in-depth analysis of the cause of the fault, and it is impossible to infer the root cause and specific location of the fault, resulting in a longer maintenance response time, and more serious equipment damage or safety accidents may occur due to the failure to be discovered in time, affecting the normal use of the electric tricycle. Summary of the invention

[0004] The present application provides an automatic fault detection method for power devices in an electric tricycle controller, which solves the technical problem that traditional detection methods lack the ability to deeply analyze detection data, resulting in insufficient fault detection accuracy and inaccurate fault location, and achieves the technical effect of improving the accuracy and reliability of fault detection and accurately locating the fault type and fault location.

[0005] In view of the above problems, the present application provides an automatic fault detection method for power devices in an electric tricycle controller, the method comprising: deploying multiple sensors on the controller to construct a sensor device group, and performing real-time sensing of the power devices in the controller through the sensor device group to obtain multiple real-time operation data; performing deep learning based on the multiple real-time operation data to construct a fault detection channel, synchronizing the multiple real-time operation data to the fault detection channel to perform dynamic fault detection, and obtain a fault detection parameter set; performing coverage traversal according to the fault detection parameter set, classifying and identifying the fault detection parameter set according to the coverage traversal results, and obtaining multiple fault types; diagnosing and tracing the power device based on the multiple fault types and the fault detection parameter set to generate multiple fault sources, locating the fault based on the multiple fault sources, and determining a fault location information set; performing operation feedback on the power device in the electric tricycle controller according to the fault location information set to generate maintenance suggestions.

[0006] One or more technical solutions provided in this application have at least the following technical effects or advantages: By deploying a sensing device group constructed by multiple sensors, the power devices in the controller are sensed in real time, and multiple real-time operation data are obtained, which can provide continuous data support for fault detection. The fault detection channel is constructed using deep learning technology. Deep learning can handle complex, multi-variable relationships, identify nonlinear fault modes that are difficult to capture with traditional methods, and improve the accuracy and sensitivity of fault detection. The multiple real-time operation data are synchronized to the fault detection channel for dynamic fault detection, and a fault detection parameter set is obtained. The real-time operation data can be dynamically analyzed to achieve early identification and early warning of faults. By performing coverage traversal and classification identification on the fault detection parameter set, multiple fault types can be identified. Coverage traversal ensures that all possible fault modes can be considered, while classification identification can accurately classify faults according to different characteristics, improving the accuracy and comprehensiveness of fault identification. Based on the fault type and fault detection parameter set, the power device is diagnosed and traced, and the fault is located based on the multiple fault sources to determine the fault location information set. The root cause of the fault can be deeply excavated from multiple dimensions and multiple angles, improving the accuracy of fault diagnosis and ensuring that maintenance personnel can accurately identify the fault location. Based on the fault location results, specific maintenance suggestions are generated, and relevant personnel are notified through a feedback mechanism to carry out repairs, which helps save maintenance time and improve maintenance efficiency.

[0007] In summary, this application forms a complete electric tricycle controller power device fault detection system through a series of steps such as multi-sensor data collection, deep learning analysis, fault classification, diagnostic tracing and operation feedback. This system can accurately detect the fault type and locate the fault location, improve the accuracy and reliability of fault detection, and provide maintenance personnel with clear maintenance directions, which helps to improve maintenance efficiency and increase the reliability and service life of electric tricycle controllers.

[0008] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Figure 1 A schematic flow chart of a method for automatic fault detection of power devices in an electric tricycle controller provided in an embodiment of the present application.

[0010] Figure 2 A schematic diagram of a flow chart of obtaining a fault detection parameter set in a method for automatic fault detection of power devices in an electric tricycle controller provided in an embodiment of the present application.

[0011] Figure 3 A schematic diagram of a flow chart of obtaining multiple fault types in a method for automatic fault detection of power devices in an electric tricycle controller provided in an embodiment of the present application. DETAILED DESCRIPTION

[0012] The embodiment of the present application solves the technical problem that traditional detection methods lack the ability to deeply analyze detection data, resulting in insufficient fault detection accuracy and inaccurate fault location, by providing an automatic fault detection method for power devices in an electric tricycle controller. This achieves the technical effect of improving the accuracy and reliability of fault detection and accurately locating the fault type and fault location.

[0013] like Figure 1 As shown, the embodiment of the present application provides an automatic fault detection method for power devices in an electric tricycle controller, the method comprising: Step S1: a plurality of sensors are arranged on the controller to construct a sensor device group, and the power devices in the controller are sensed in real time by the sensor device group to obtain a plurality of real-time operation data.

[0014] Specifically, multiple different types of sensors are arranged around the electric tricycle controller to fully understand the operating status of the power devices in the electric tricycle controller. These sensors include temperature sensors, current sensors, voltage sensors, etc. These sensors are installed at key positions of the controller to form an integrated sensor device group to continuously sense and collect power devices. For example, the current sensor is used to monitor the current size, the voltage sensor is used to monitor the voltage level, and the temperature sensor is used to monitor the temperature of the device. The data collected by the sensor device group is collected to form multiple real-time operation data. These real-time operation data can reflect the immediate working status of the power device and provide basic data for subsequent judgment of whether there is a fault, the type of fault, and the location of the fault.

[0015] Step S2: Perform deep learning based on the multiple real-time operation data, build a fault detection channel, synchronize the multiple real-time operation data to the fault detection channel to perform dynamic fault detection, and obtain a fault detection parameter set.

[0016] Specifically, deep learning is performed based on the multiple real-time operation data obtained in step S1, and a specific deep learning framework, such as TensorFlow or PyTorch, is used to build a model for analyzing these real-time operation data, namely, a fault detection channel. In the specific construction process, it is first necessary to collect a certain amount of historical fault data and normal operation data for training the deep learning model. These data include but are not limited to parameters such as current, voltage, and temperature. Then, these data are used to train a deep learning model, such as a convolutional neural network or a recurrent neural network. The model will automatically adjust internal parameters such as weights according to the characteristics of the training data to identify normal and abnormal operating modes. After the training is completed, the model can automatically learn from the real-time operation data and identify fault characteristics.

[0017] Synchronize multiple real-time operation data to the fault detection channel for dynamic fault detection. The fault detection channel processes the real-time operation data according to the internal algorithm, such as extracting features and classifying the data, mining the feature patterns related to the fault, and outputting the fault detection parameter set. The fault detection parameter set contains all abnormal operation parameters in the real-time operation data, such as abnormal current, voltage surge or abnormal temperature. For example, if the current fluctuation amplitude in the real-time operation data of a power device exceeds the normal range, the fault detection channel will identify the corresponding current fluctuation parameters and add them to the fault detection parameter set.

[0018] Through deep learning and the construction of fault detection channels, the original real-time operation data can be converted into a more targeted fault detection parameter set, which reduces the scope of data analysis for subsequent fault classification, identification and location, allowing subsequent operations to focus on abnormal operating parameters, improving the accuracy and efficiency of fault detection and saving data processing resources.

[0019] Step S3: Perform coverage traversal according to the fault detection parameter set, classify and identify the fault detection parameter set according to the coverage traversal result, and obtain multiple fault types.

[0020] Specifically, coverage traversal refers to a comprehensive scan and analysis of the fault detection parameter set to ensure that every possible fault mode is taken into account. Use data processing and analysis tools to perform coverage traversal on the fault detection parameter set, such as the Pandas library or NumPy library in Python. Taking the Pandas library as an example, the traversal function can be used to comprehensively scan and analyze the fault detection parameter set, record the characteristics and values ​​of each parameter, and the relationship between the parameters, and generate coverage traversal results. Then, match the traversal results with known fault types, map the detected real-time data to specific fault types, and obtain multiple fault types.

[0021] Through coverage traversal and classification identification, fault detection can be transformed from single abnormality detection to specific fault type identification, and ensure that no potential fault type is missed, which not only improves the accuracy and comprehensiveness of fault detection, but also provides more specific guidance for subsequent fault handling. By identifying the specific fault type, maintenance personnel can locate the problem more quickly and take targeted maintenance measures, thereby improving maintenance efficiency and the reliability of electric tricycles.

[0022] Step S4: diagnose and trace the power device based on the multiple fault types and the fault detection parameter set to generate multiple fault sources, locate the fault based on the multiple fault sources, and determine the fault location information set.

[0023] Specifically, the fault source refers to the root cause of the power device failure, such as damage to a specific component. The fault location information set is a collection of all fault location information, which lists in detail the specific location of each fault in the power device, the fault type, related operating parameters and other data.

[0024] First, the power device is diagnosed and traced based on multiple fault types and fault detection parameter sets. This process can use a rule-based system, which predefines a series of rules based on the working principle of the power device, previous fault cases, etc. When the fault type and fault detection parameter set are input, the system performs reasoning and analysis based on these rules. If the fault type is abnormal power output, and the fault detection parameter set shows that the voltage of several power transistors is abnormal, then through the system's rule reasoning, it may be determined that there is a problem with the drive circuit related to these transistors, thereby generating the fault source. After identifying multiple fault sources, further analyze and determine the specific location of the fault in the power device to form fault location information.

[0025] Through diagnostic tracing and fault location, detailed information about the cause and location of the fault can be provided, providing maintenance personnel with clear maintenance goals and helping to quickly restore the normal operation of the electric tricycle.

[0026] Step S5: providing operation feedback to the power devices in the electric tricycle controller according to the fault location information set, and generating maintenance suggestions.

[0027] Specifically, according to the fault location information set, the power devices in the electric tricycle controller are fed back for operation, the real-time operating status at the fault location is monitored, and guiding maintenance suggestions are generated based on the monitoring results and sent to maintenance personnel. For example, if the pin solder joint is cold-soldered, the maintenance suggestion may be to re-solder the pin and check whether the welding process meets the requirements; if the cold-soldering problem occurs multiple times, it may also be necessary to check whether there are defects in the design of the circuit board, such as whether there is excessive thermal stress.

[0028] Operation feedback and maintenance suggestions can provide direct and targeted guidance for the maintenance of power devices in the electric tricycle controller, making repair or maintenance work more efficient and accurate, helping to reduce the repair time of power devices and increase their service life, thereby improving the reliability and stability of the entire electric tricycle controller.

[0029] Furthermore, in step S2, deep learning is performed based on the multiple real-time operation data to build a fault detection channel, including: Step S21: performing characteristic analysis of the power device based on the plurality of real-time operating data to generate a plurality of operating characteristics.

[0030] Step S22: Retrieve a historical fault data set of the power device, integrate the multiple real-time operation data with the historical fault data set based on the multiple operation characteristics, and obtain a data training set and a data test set according to the integration result.

[0031] Step S23: input the data training set into the autoencoder for unsupervised learning training to generate data training results, set a test accuracy threshold according to the data test set, and test and verify the data training results with the data test set until the data training results reach the test accuracy threshold, thereby obtaining the fault detection channel.

[0032] Specifically, feature analysis refers to extracting key information or patterns that can represent the state of power devices from real-time operation data. Multiple operating features can be extracted through feature analysis. These operating features can describe the characteristic parameters of power devices under different working conditions, including but not limited to the current peak value, voltage fluctuation range, temperature rise rate, etc. of power devices. Feature analysis can use relevant feature selection methods in data mining and analysis tools (such as the Scikit-learn library in Python). For example, for current data in real-time operation data, the current operation characteristics of power devices can be characterized by calculating statistical features such as the mean, variance, and rate of change of the current; for temperature data, the maximum, minimum, and heating rate of temperature can be analyzed. By accurately identifying the key operating characteristics of power devices, high-quality input data is provided for subsequent deep learning model training.

[0033] The historical fault data set refers to the power device fault events and related data recorded in the past, including the operating data, fault type, fault location and other information when the fault occurred. These historical fault data provide reference and comparison samples for current fault detection and can be used to train and verify the fault detection model. The historical fault data set of the power device is retrieved from the controller's historical database. According to multiple operating characteristics, the real-time operating data is integrated with the historical fault data set. Through operations such as data merging and data alignment, the real-time operating data and historical fault data with the same operating characteristics are aggregated to form a comprehensive data set. Using data partitioning techniques such as cross-validation or random partitioning, the integrated data set is divided into two parts, one for training the fault detection model (i.e., data training set) and the other for testing the performance of the model (i.e., data test set). For example, random sampling is used to divide according to a certain ratio, with 80% of the data as the data training set and 20% of the data as the data test set.

[0034] An autoencoder is a special neural network architecture that is mainly used for unsupervised learning tasks. The autoencoder is used to learn the feature representation in the operating data of the power device, so as to detect faults. The test accuracy threshold is a pre-set standard value to measure the accuracy of the data training results. Only when the data training results reach this threshold under the test of the data test set, the constructed fault detection channel is considered reliable. Use a deep learning framework such as TensorFlow or PyTorch to build an autoencoder model, and input the data training set into the autoencoder for unsupervised learning training. During the training process, the autoencoder will continuously adjust the internal weight parameters to minimize the difference between the input data and the output data, thereby generating data training results. Then, the data test set is used to test and verify the data training results until the data training results reach the test accuracy threshold. During the test process, if the data training results do not reach the threshold, it is necessary to adjust the parameters of the autoencoder (such as increasing the training rounds, adjusting the network structure, etc.) and retrain until a fault detection channel that meets the test accuracy threshold is obtained.

[0035] The above steps utilize the unsupervised learning ability of the autoencoder, and through continuous training and test verification, ultimately obtain an accurate fault detection channel, which can effectively perform dynamic fault detection on power devices and improve the accuracy and efficiency of fault detection.

[0036] Further, such as Figure 2 As shown, in step S2 of the embodiment of the present application, the multiple real-time operation data are synchronized to the fault detection channel for dynamic fault detection to obtain a fault detection parameter set, including: Step S24: segmenting the plurality of real-time operation data according to the historical fault data set to generate a plurality of batches of real-time operation data.

[0037] Step S25: Synchronize the multiple batches of real-time operation data to the fault detection channel and perform dynamic analysis in combination with the historical fault data set to obtain fault-generated dynamic data.

[0038] Step S26: using the fault-generated dynamic data as an index, searching and identifying the plurality of real-time operation data to obtain data retrieval results.

[0039] Step S27: extracting a plurality of operation fault data according to the data retrieval result, performing data confidence analysis, and generating a data confidence value.

[0040] Step S28: screening and integrating the plurality of operating fault data according to the data confidence value to obtain the fault detection parameter set.

[0041] Specifically, according to the characteristics or patterns of the historical fault data set, multiple real-time operation data are segmented according to data segmentation, and the multiple real-time operation data are split into multiple batches according to the similarities of different working conditions or fault types when the fault occurs, to generate multiple batches of real-time operation data, each batch of data corresponds to a specific historical fault pattern, so as to facilitate more targeted analysis later. Exemplarily, the historical fault data set has been classified according to the fault type, and the real-time operation data can be segmented according to the characteristics of the fault type. For example, if there is data classification about overheating faults and short-circuit faults in the historical fault data set, the temperature-related data in the real-time operation data can be segmented into a batch of data related to overheating fault analysis, and the data related to current and voltage anomalies can be segmented into another batch of data related to short-circuit fault analysis, etc.

[0042] Multiple batches of real-time operation data are synchronized to the fault detection channel and combined with the historical fault data set for dynamic analysis, thereby generating fault generation dynamic data. These fault generation dynamic data reflect the dynamic change characteristics of the fault generation process and contain the law or trend information of fault development. The built-in data analysis algorithm is used in the fault detection channel to analyze the data. For example, for each batch of real-time operation data, combined with the development trajectory of similar faults in the historical fault data set, analyze whether there is a similar trend or abnormal change in the current data, so as to obtain the fault generation dynamic data.

[0043] Using the fault-generated dynamic data as an index, the key information in the fault-generated dynamic data (such as specific fault feature values ​​or change trends) is used to search and match in multiple real-time operation data to find the real-time operation data records related to the fault-generated dynamic data, thereby obtaining data retrieval results. This data retrieval result is a subset of data related to the fault that is filtered out from the original real-time operation data, including multiple operation fault data.

[0044] Extract multiple operation fault data from the data retrieval results for data confidence analysis to determine the source stability, data integrity, and consistency with other related data of multiple operation fault data. Based on the judgment results, generate a corresponding data confidence value for each operation fault data to quantify the credibility of the operation fault data. The higher the data confidence value, the more reliable the data. The data confidence value can be used to screen out more credible operation fault data to construct a fault detection parameter set. Among them, data confidence analysis can use statistical methods or machine learning algorithms (such as Bayesian networks) to evaluate the confidence of the data. If a certain operation fault data can be obtained from multiple data sources and corroborate each other, then its confidence is relatively high; if the data is missing or contradictory to other data, then its confidence is low.

[0045] Filter and integrate multiple operation fault data according to the data confidence value. Set a confidence threshold, such as 0.8. If the confidence value of the operation fault data is greater than this threshold, it will be filtered out. Then integrate the filtered data, remove duplicate or redundant data, and finally obtain the fault detection parameter set, which provides more accurate and representative data for subsequent fault type identification, diagnosis and tracing.

[0046] Further, such as Figure 3 As shown, step S3 includes: Step S31: introducing operation scenario parameters of the power device, performing multi-dimensional coverage analysis on the fault detection parameter set according to the operation scenario parameters, and determining a multi-dimensional coverage mode.

[0047] Step S32: Map the fault detection parameter set to the multi-dimensional coverage mode for traversal analysis to generate a coverage traversal result.

[0048] Step S33: performing data association analysis on the fault detection parameter set according to the coverage traversal result to generate a plurality of association factors.

[0049] Step S34: performing cluster analysis on the fault detection parameter set according to the multiple correlation factors to obtain the multiple fault types.

[0050] Specifically, the operating scenario parameters are scenario-related parameters that describe the operating environment and working conditions of the power device. For example, for the power device in the electric tricycle controller, the operating scenario parameters may include ambient temperature, load size, working hours, etc. These parameters will affect the operating state of the power device and the possibility of failure. The multi-dimensional coverage mode is a mode or framework for coverage analysis of the fault detection parameter set determined after considering multiple dimensions of the operating scenario parameters. It describes how to comprehensively analyze the fault detection parameter set in multiple dimensions.

[0051] The operating scenario parameters of the power device are obtained from the data collected by the sensor or the pre-set working conditions. For example, the ambient temperature parameters are obtained through the temperature sensor, and the load size parameters are obtained through the load sensor. Then, a multi-dimensional coverage analysis of the fault detection parameter set is performed based on the operating scenario parameters. The matrix analysis method can be used to take each operating scenario parameter as a dimension of the matrix and the fault detection parameter set as an element of the matrix. The distribution of the fault detection parameter set under each dimension is analyzed to determine the multi-dimensional coverage pattern and provide a more comprehensive analysis perspective for fault detection. By considering different operating scenarios, the accuracy and adaptability of fault detection can be improved.

[0052] The fault detection parameter set is mapped to the multidimensional coverage pattern, and each element of the fault detection parameter set is checked and analyzed in turn according to the order set by the multidimensional coverage pattern. During the traversal process, the state and characteristics of each element in the multidimensional coverage pattern are recorded to generate the coverage traversal results. Through traversal analysis, it can be ensured that no possible fault type is missed, thereby improving the comprehensiveness of fault detection.

[0053] The correlation factor is a factor that represents the correlation relationship between data obtained through data association analysis. These factors reflect the internal connection of the data in the fault detection parameter set, which can be a mathematical relationship, a logical relationship, or a relational expression with actual physical meaning. According to the coverage traversal results, the fault detection parameter set is subjected to data association analysis. The association rule algorithm in data mining, such as the Apriori algorithm, can be used to analyze the frequent item sets between the elements of the fault detection parameter set in the coverage traversal results, find out the correlation relationship between the data, and thus generate multiple correlation factors. For example, if it is found in the coverage traversal results that a certain fault detection parameter often appears at the same time or has a certain quantitative relationship with several other parameters, the correlation factor between these parameters can be determined.

[0054] Clustering algorithms are used to perform cluster analysis on the fault detection parameter set according to multiple correlation factors. First, the basis for clustering is determined based on the correlation factors, and then the elements of the fault detection parameter set are clustered according to similarity. For example, if the correlation factors indicate that some fault detection parameters are related to temperature and others are related to current, then the fault detection parameter set can be divided into a category related to temperature and a category related to current, etc., based on these correlation factors, to obtain multiple fault types. Through cluster analysis, different fault modes can be clearly distinguished and identified, providing clear guidance for fault diagnosis and repair.

[0055] Further, step S34 includes: Step S341: Perform density analysis based on the multiple correlation factors and the coverage traversal result to obtain a correlation density coefficient.

[0056] Step S342: performing cluster analysis on the fault detection parameter set according to the association density coefficient in combination with the multiple association factors to generate multiple data clusters.

[0057] Step S343: performing fault identification according to the multiple data clusters in combination with the historical fault data set to generate multiple fault labels.

[0058] Step S344: Match the multiple fault labels with the fault detection parameter set to determine the multiple fault types.

[0059] Specifically, density analysis refers to evaluating the distribution density of data points in multidimensional space. Through density analysis, the correlation density coefficients of each correlation factor can be obtained. These correlation density coefficients are quantitative indicators used to measure the similarity or aggregation trend between different correlation factors, reflecting the distribution density of the correlation factors in the coverage traversal results. Density analysis is performed on the correlation factors and coverage traversal results, using a distance-based density analysis method. For example, for each correlation factor, the kernel density estimation formula is used to calculate its point density in the coverage traversal results. If the points of the correlation factor in a certain area are relatively concentrated, then the density of the area is relatively large. The density of each correlation factor is obtained through density calculation, and then the correlation density coefficient is comprehensively obtained.

[0060] Use hierarchical clustering algorithm or density-based spatial clustering algorithm (DBSCAN) etc. to cluster the fault detection parameter set according to the association density coefficient combined with multiple association factors. Taking DBSCAN algorithm as an example, the neighborhood radius is determined according to the association density coefficient. If the association density coefficient is large, it means that the data is closely connected, so the neighborhood radius can be appropriately reduced; conversely, if the association density coefficient is small, the neighborhood radius can be appropriately increased. Determine the minimum number of points based on experience or the scale of the data, for example, it can be set to a certain percentage of the total number of points in the data set. Determine the distance measurement method between data points based on multiple association factors, for example, assign different weights to each association factor based on the importance of the association factor, and then calculate the distance measurement between the eigenvalues ​​of the association factor, and adjust it in combination with the association factor weight. Perform hierarchical clustering based on the neighborhood radius, the minimum number of points, and the distance between the data, and divide the data points in the fault detection parameter set into different data clusters. The fault detection parameter set in each data cluster has a high similarity.

[0061] Fault labels are labels used to mark the fault types corresponding to data clusters, and are a form of representation of data cluster fault classification results. Fault identification is performed based on multiple data clusters combined with historical fault data sets to generate multiple fault labels. First, the characteristics of each data cluster are analyzed, such as the value range of the data in the data cluster, the relationship between the data, etc. Then, it is compared with the historical fault data set to find the historical fault types similar to the data cluster characteristics, so as to generate a fault label for the data cluster. For example, if the data characteristics within a data cluster are similar to the overheating fault data characteristics in the historical fault data set, then this data cluster is labeled with the "overheating fault" label.

[0062] Match multiple fault labels with the fault detection parameter set. For each parameter in the fault detection parameter set, determine the corresponding fault type according to the fault label corresponding to the data cluster to which it belongs. For example, if a fault detection parameter belongs to a data cluster marked as "short circuit fault", then the fault type corresponding to this parameter is short circuit fault.

[0063] Through cluster analysis and fault identification, a clear fault classification can be provided for the data points of each fault detection parameter set, thus achieving accurate fault diagnosis.

[0064] Further, step S4 includes: Step S41: Analyze the data correlation between the multiple fault types and the fault detection parameter set to obtain multiple parameter correlation coefficients.

[0065] Step S42: setting data tracing rules according to the multiple parameter correlation coefficients in combination with the multiple fault types, wherein the data tracing rules include multiple tracing paths, and the multiple tracing paths correspond to the multiple fault types.

[0066] Step S43: using the data tracing rule, tracing the multiple fault types to the power devices according to the multiple tracing paths to generate the multiple fault sources.

[0067] Step S44: infer the position of the power device according to the multiple fault sources in combination with the multiple parameter correlation coefficients to determine the fault area information, and analyze the fault area information in combination with the spatial position information of the power device to determine the fault location information set.

[0068] Specifically, statistical methods (such as Pearson correlation coefficient, Spearman rank correlation coefficient, etc.) are used to analyze the correlation between multiple fault types and fault detection parameter sets (such as current, temperature, voltage, etc.), and the correlation coefficients between the fault detection parameter sets and each fault type are calculated, thereby obtaining multiple parameter correlation coefficients. Each parameter correlation coefficient reflects the degree of association between each fault type and the fault detection parameter. Through correlation analysis, the relationship between the fault type and the fault detection parameter can be accurately grasped, providing an important basis for subsequent data tracing and fault location.

[0069] Data traceability rules are used to guide the operation of tracing the faults of power devices. They contain multiple traceability paths. Each traceability path corresponds to a specific fault type, and clarifies the tracing direction and method from the fault type to the fault source. Data traceability rules are set according to the calculated correlation coefficients of multiple parameters and multiple fault types. For example, if the correlation coefficient between a certain fault type and a certain fault detection parameter is very high, then when setting the traceability path, the traceability can be started from this fault detection parameter. For different fault types, different traceability paths are constructed according to their correlation with the fault detection parameters. For example, for fault type A, if it has a large correlation coefficient with fault detection parameters x1 and x2, then the traceability path can start from the abnormal situation of x1 and x2, and gradually trace back to the internal structure or external factors of the power device that may cause this abnormality, thereby constructing a data traceability rule containing multiple traceability paths and corresponding to multiple fault types. By setting data traceability rules, a clear direction and method are provided for fault tracing, making the traceability process more organized and efficient.

[0070] The power device is traced through data traceability rules. For each fault type, the operation is performed according to its corresponding traceability path. For example, if the traceability path starts from the fault detection parameter x1, then combined with the working principle and circuit structure of the power device, check which part of the power device or which external factor caused the abnormality of parameter x1. The analysis is gradually deepened along the traceability path, and finally multiple fault sources are generated.

[0071] The location of the power device is estimated based on multiple fault sources combined with multiple parameter correlation coefficients. If the fault type of a certain fault source identified has a large correlation coefficient with a certain fault detection parameter, and it is known that this fault detection parameter is related to a certain area of ​​the power device, then it can be inferred that the fault may occur in this area. For example, if the fault source has a large correlation coefficient with the fault detection parameter representing the temperature of the power device, and the temperature sensor is located in a specific area of ​​the power device, then it can be inferred that the fault area is this area. Then, based on the fault area information combined with the spatial location information of the power device (such as layout diagram, location mark, etc.), a comprehensive analysis is performed to determine the precise location of the fault, and the precise location information of multiple fault sources, that is, the fault location information set, is obtained to guide maintenance personnel to perform targeted maintenance and troubleshooting operations.

[0072] Further, step S5 includes: Step S51: Based on the fault location information set, real-time operation feedback is performed on the power devices in the electric tricycle controller to generate feedback response data.

[0073] Step S52: performing a fault impact analysis on the operation control strategy according to the feedback response data in combination with the plurality of parameter correlation coefficients to generate an operation impact data set.

[0074] Step S53: Determine whether the operation impact data set is greater than or equal to an expected impact threshold.

[0075] Step S54: When the operation impact data set is greater than or equal to the expected impact threshold, a self-recovery instruction is activated, and data security of the power device is restored according to the fault location information set through the self-recovery instruction, and the maintenance suggestion is generated according to the recovery result.

[0076] Step S55: When the operation impact data set is less than the expected impact threshold, the remote data terminal is started, the fault location information set is transmitted to the remote data terminal for analysis, and the maintenance suggestion is generated.

[0077] Specifically, after obtaining the fault location information set, the operating status information of the power device at the fault location in the electric tricycle controller is monitored, its operating status data is collected in real time, and feedback response data is generated, which can indicate whether the current state of the power device is normal and whether there are any abnormal conditions that require immediate attention. For example, if the fault location information set indicates that there may be a fault at a specific pin of the power device, the operating parameters such as the voltage and current of the pin can be obtained through a sensor or a monitoring circuit, and these parameters can be organized into feedback response data.

[0078] The operation control strategy refers to the strategy used by the controller to manage the operation of the electric tricycle system, including the control methods and algorithms for power devices, loads, temperature, current, etc. The operation control strategy of the electric tricycle is analyzed in combination with the feedback response data and parameter correlation coefficients to evaluate the specific impact of the fault on the operation of the controller power device. Then, an operation impact data set is generated based on the analysis results. This data set contains information such as the degree of impact of the fault on various aspects of the operation control strategy, providing a basis for subsequent decision-making.

[0079] The expected impact threshold is a standard set in advance based on the performance requirements, safety standards and other factors of the electric tricycle to determine whether the impact of the fault has reached a level that requires specific actions. The operational impact data set is compared with the expected impact threshold, and corresponding countermeasures are taken based on the comparison results.

[0080] When the operating impact data set is greater than or equal to the expected impact threshold, the self-recovery instruction is activated to perform a recovery operation on the power device according to the fault location information set. The self-recovery instruction can be a pre-written software program or a set of operation commands stored in the controller. For example, the self-recovery instruction can reset the power device at the fault location, recalibrate parameters, or execute a specific repair algorithm. Maintenance recommendations are generated based on the results of the recovery operation, which will guide subsequent maintenance work. For example, if the recovery is successful, it can be recommended to optimize the relevant software algorithm to prevent similar faults from happening again; if the recovery fails, it is recommended to replace the power device or perform a more in-depth hardware inspection.

[0081] When the operating impact data set is less than the expected impact threshold, the remote data terminal is started. The fault location information set is transmitted to the remote data terminal through network communication technology (such as Bluetooth, Wi-Fi or cellular network). The remote data terminal refers to a remote system or service that can receive and process fault data, which can be a cloud server or a remote monitoring center. At the remote data terminal, more powerful data analysis tools and algorithms can be used to analyze the fault location information set, and maintenance recommendations can be generated by combining a large amount of historical data and expert knowledge. For example, if the fault location information set shows that a small area of ​​the power device has a minor fault, the remote data terminal may recommend regular inspections or local optimization adjustments based on the handling methods of similar faults in historical data.

[0082] The above steps generate an operation impact data set through operation feedback and fault impact analysis, and take corresponding measures for faults of different severity in combination with the expected impact threshold. It can not only handle serious faults in a timely manner to ensure operation safety, but also reasonably analyze minor faults and generate maintenance suggestions, thereby improving the reliability and service life of power devices in the electric tricycle controller.

[0083] Further, step S52 includes: Step S521: performing weight analysis on the feedback response data according to the multiple parameter correlation coefficients to determine multiple weight coefficients.

[0084] Step S522: performing a fault response analysis on the operation control strategy based on the multiple weight coefficients to determine multiple fault feedback response data.

[0085] Step S523: setting an impact analysis target according to the operation control strategy and the plurality of parameter correlation coefficients.

[0086] Step S524: performing a regression analysis on the impact of the operation control strategy fault in accordance with the impact analysis target and in combination with the plurality of fault reaction response data to generate the operation impact data set.

[0087] Specifically, for each feedback response data, a weight analysis is performed according to its corresponding parameter correlation coefficient, and the parameter correlation coefficients of multiple feedback response data are normalized to obtain the weight coefficient corresponding to each feedback response data. Each weight coefficient reflects the relative importance of the corresponding feedback response data in the whole. Through weight analysis, the importance of each feedback response data in the operation control strategy can be reasonably determined, providing an accurate weight basis for subsequent analysis.

[0088] The operation control strategy is analyzed for fault response using the determined multiple weight coefficients to evaluate the response of the operation control strategy when a fault occurs. The collected feedback response data is analyzed using a weighted method. Different parameters are given different degrees of importance according to the weight coefficients of the parameters to determine multiple fault feedback response data.

[0089] The impact analysis target refers to the target set during fault analysis for quantitatively evaluating the impact of faults during operation, which is used to indicate the direction and focus of the analysis. For electric tricycle controllers, the impact analysis target may include analyzing the impact of faults on the vehicle's maximum speed, acceleration, battery life, and driving safety. The operation control strategy includes multiple components, such as speed control, torque control, current limiting, and other links. Each link corresponds to a specific type of operating parameter. Specific impact analysis targets are set based on the parameter correlation coefficients of these operating parameters. For example, if the parameter correlation coefficient of a certain fault detection parameter is high, then the impact analysis target can be set to focus on analyzing the impact of the fault on the related links of the fault detection parameter in the operation control strategy.

[0090] According to the set impact analysis objectives, the operation control strategy is subjected to fault impact regression analysis in combination with multiple fault reaction response data. First, according to the characteristics of the operation control strategy and the nature of the data, the regression analysis model, such as linear regression or nonlinear regression, is determined in combination with historical operation data. The corresponding fault feedback response data is substituted into the regression model as an independent variable to analyze the degree of impact of the fault on each impact analysis objective. These impact analysis results are summarized to generate an operation impact data set. The fault impact regression analysis comprehensively analyzes the impact of the fault on various aspects of the operation control strategy. The generated operation impact data set provides an important basis for subsequent judgment of the severity of the fault and taking corresponding measures.

[0091] In summary, the automatic fault detection method for power devices in the electric tricycle controller provided in the embodiment of the present application has the following technical effects: The embodiment of the present application realizes intelligent and precise fault detection of power devices in the electric tricycle controller by combining multi-sensor data acquisition with deep learning. By acquiring multi-dimensional sensor data in real time and performing deep learning analysis, various complex fault types can be efficiently identified. Through coverage traversal and classification identification, potential faults are fully captured, and through fault tracing and positioning, the source of the fault is accurately located to generate accurate fault location information. Finally, operation feedback is performed in combination with the fault location information, and maintenance suggestions are generated according to actual conditions to provide guidance for maintenance personnel. Overall, the embodiment of the present application significantly improves the accuracy, reliability, and efficiency of fault detection, while providing accurate guidance for the maintenance process, improving the reliability and stability of the electric tricycle controller, and ensuring the safe and efficient operation of the electric tricycle.

[0092] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An automatic fault detection method for power devices in an electric tricycle controller, characterized in that: The method comprises: Arrange multiple sensors on the controller to form a sensor device group, and use the sensor device group to perform real-time sensing on power devices in the controller to obtain multiple real-time operation data; Perform deep learning based on the multiple real-time operation data, build a fault detection channel, synchronize the multiple real-time operation data to the fault detection channel to perform dynamic fault detection, and obtain a fault detection parameter set; Performing coverage traversal according to the fault detection parameter set, classifying and identifying the fault detection parameter set according to the coverage traversal result, and obtaining multiple fault types; Performing diagnosis and tracing of the power device based on the multiple fault types and the fault detection parameter set to generate multiple fault sources, performing fault location based on the multiple fault sources to determine a fault location information set; According to the fault location information set, operation feedback is provided to the power devices in the electric tricycle controller to generate maintenance suggestions.

2. The automatic fault detection method for power devices in an electric tricycle controller as claimed in claim 1, characterized in that: Based on the multiple real-time operation data, deep learning is performed to build a fault detection channel, and the method includes: Performing characteristic analysis of the power device based on the plurality of real-time operation data to generate a plurality of operation characteristics; Retrieving a historical fault data set of the power device, integrating the multiple real-time operation data with the historical fault data set based on the multiple operation characteristics, and obtaining a data training set and a data test set according to the integration result; The data training set is input into the autoencoder for unsupervised learning training to generate data training results, a test accuracy threshold is set according to the data test set, and the data training result is tested and verified by the data test set until the data training result reaches the test accuracy threshold, thereby obtaining the fault detection channel.

3. The automatic fault detection method for power devices in an electric tricycle controller as claimed in claim 2, characterized in that: The multiple real-time operation data are synchronized to the fault detection channel for dynamic fault detection to obtain a fault detection parameter set, the method comprising: Segmenting the plurality of real-time operation data according to the historical fault data set to generate multiple batches of real-time operation data; Synchronizing the multiple batches of real-time operation data to the fault detection channel and combining with the historical fault data set for dynamic analysis to obtain fault-generated dynamic data; Using the fault-generated dynamic data as an index, searching and identifying the plurality of real-time operation data to obtain data retrieval results; Extracting a plurality of operation fault data according to the data retrieval result to perform data confidence analysis and generate a data confidence value; The plurality of operating fault data are screened and integrated according to the data confidence value to obtain the fault detection parameter set.

4. The automatic fault detection method for power devices in an electric tricycle controller as claimed in claim 2, characterized in that: Performing coverage traversal according to the fault detection parameter set, classifying and identifying the fault detection parameter set according to the coverage traversal result, and obtaining multiple fault types, the method includes: Introducing operation scenario parameters of the power device, performing multi-dimensional coverage analysis on the fault detection parameter set according to the operation scenario parameters, and determining a multi-dimensional coverage mode; Mapping the fault detection parameter set to the multi-dimensional coverage mode for traversal analysis to generate a coverage traversal result; Performing data association analysis on the fault detection parameter set according to the coverage traversal result to generate multiple association factors; Cluster analysis is performed on the fault detection parameter set according to the multiple correlation factors to obtain the multiple fault types.

5. The automatic fault detection method for power devices in an electric tricycle controller as claimed in claim 4, characterized in that: Performing cluster analysis on the fault detection parameter set according to the multiple correlation factors to obtain the multiple fault types, the method comprising: Perform density analysis based on the multiple association factors combined with the coverage traversal result to obtain an association density coefficient; Performing cluster analysis on the fault detection parameter set according to the association density coefficient in combination with the multiple association factors to generate multiple data clusters; Perform fault identification according to the multiple data clusters in combination with the historical fault data set to generate multiple fault labels; The multiple fault labels are matched with the fault detection parameter set to determine the multiple fault types.

6. The automatic fault detection method for power devices in an electric tricycle controller as claimed in claim 1, characterized in that: Based on the multiple fault types and the fault detection parameter set, the power device is diagnosed and traced to generate multiple fault sources, the fault is located based on the multiple fault sources, and the fault location information set is determined, the method comprising: Analyzing data correlations between the multiple fault types and the fault detection parameter set to obtain multiple parameter correlation coefficients; According to the multiple parameter correlation coefficients and the multiple fault types, a data tracing rule is set, wherein the data tracing rule includes multiple tracing paths, and the multiple tracing paths correspond to the multiple fault types; By using the data tracing rule, tracing the multiple fault types to the power device according to the multiple tracing paths to generate the multiple fault sources; The position of the power device is estimated according to the multiple fault sources in combination with the multiple parameter correlation coefficients to determine the fault area information, and the fault location information set is determined by analyzing the fault area information in combination with the spatial location information of the power device.

7. The automatic fault detection method for power devices in an electric tricycle controller as claimed in claim 6, characterized in that: According to the fault location information set, operation feedback is provided to the power device in the electric tricycle controller to generate maintenance suggestions, the method comprising: Based on the fault location information set, real-time operation feedback is performed on the power devices in the electric tricycle controller to generate feedback response data; Performing a fault impact analysis on the operation control strategy according to the feedback response data combined with the multiple parameter correlation coefficients to generate an operation impact data set; Determining whether the operation impact data set is greater than or equal to an expected impact threshold; When the operation impact data set is greater than or equal to the expected impact threshold, a self-recovery instruction is activated, data security recovery of the power device is performed according to the fault location information set through the self-recovery instruction, and the maintenance suggestion is generated according to the recovery result; When the operation impact data set is less than the expected impact threshold, the remote data terminal is started, the fault location information set is transmitted to the remote data terminal for analysis, and the maintenance suggestion is generated.

8. The automatic fault detection method for power devices in an electric tricycle controller as claimed in claim 7, characterized in that: Performing a fault impact analysis on an operation control strategy according to the feedback response data combined with the multiple parameter correlation coefficients to generate an operation impact data set, the method comprising: Performing weight analysis on the feedback response data according to the multiple parameter correlation coefficients to determine multiple weight coefficients; Performing a fault response analysis on the operation control strategy based on the multiple weight coefficients to determine multiple fault feedback response data; Setting an impact analysis target according to the operation control strategy and the plurality of parameter correlation coefficients; The operation impact data set is generated by performing a regression analysis on the operation control strategy fault impact according to the impact analysis target and in combination with the multiple fault reaction response data.

Images