A large eVTOL motor electronic speed controller fault tracing diagnosis method

Abstract

This invention discloses a method for tracing and diagnosing faults in large-scale eVTOL motors and electronic speed controllers (ESCs), belonging to the field of eVTOL fault diagnosis technology. This method collects operational data of different fault types from both the motor and ESC sides, and constructs a coupled fault feature library through preprocessing, feature extraction, and sample annotation. Based on this feature library sample set, a fault tracing and diagnosis model is constructed and trained, with parameters optimized to achieve the required recognition accuracy. Real-time operational data is simultaneously collected, preprocessed, and input into the model to obtain fault tracing and diagnosis results. Finally, a preset fault handling rule library is matched to generate corresponding handling suggestions, which are then transmitted to the flight control system and the ground maintenance platform. This method achieves precise tracing of the fault source, type, and severity, overcoming the limitations of traditional methods that can only detect the existence of faults, improving fault investigation efficiency and diagnostic accuracy, and providing technical support for eVTOL flight safety.

Description

Technical Field

[0001] This invention relates to the field of eVTOL fault diagnosis technology, and in particular to a method for tracing and diagnosing faults in large eVTOL motor electronic speed controllers. Background Technology

[0002] In recent years, electric vertical takeoff and landing (eVTOL) aircraft have become an important development direction in the field of low-altitude aviation equipment. Their distributed electric propulsion system is a key component for achieving stable flight. The motor and electronic speed controller (ESC), as the core actuators of this system, directly affect the overall operational status of the eVTOL through their coordinated operation. Currently, the aerospace and low-altitude equipment fields are continuously deepening their research and development of electric propulsion systems. Condition monitoring and fault diagnosis technologies for electrical components such as motors and ESCs have also become research hotspots. Various airborne data acquisition devices, data preprocessing technologies, and machine learning diagnostic models have been gradually applied to fault detection scenarios in industrial equipment and aviation components. Simultaneously, the development of airborne communication technology has enabled bidirectional transmission of equipment operation data to ground maintenance platforms. In the eVTOL field, the industry has achieved basic data acquisition for individual components such as motors and ESCs and has begun to explore the application of deep learning algorithms to fault identification. Related fault handling strategies are also being gradually improved in conjunction with the operational characteristics of aviation equipment. The overall technological development trend is moving towards real-time equipment condition monitoring and intelligent fault diagnosis.

[0003] Current fault diagnosis technologies for large eVTOL motors and ESCs lack a complete diagnostic process adapted to the strongly coupled operating characteristics of the two. Existing technologies have not built a motor-ESC coupled fault feature library, lack targeted data collection, and lack unified standards for preprocessing and feature extraction processes. Furthermore, standardized sample labeling based on fault sources and types is not provided, failing to offer sample support that closely reflects actual operating conditions for subsequent fault diagnosis. Simultaneously, the construction and training of existing diagnostic models are not based on coupled fault data from the motor and ESC, and parameter optimization methods lack specificity. They can only achieve basic fault detection, failing to accurately trace the fault's source and making it difficult to pinpoint the specific cause of the fault. Sources and Types: During the online diagnostic phase, the real-time operational data acquisition from the motor and ESC sides lacks synchronization standards, the preprocessing process is simplistic, and the data input into the diagnostic model cannot accurately reflect the coupled operational status of the two, resulting in inaccurate diagnostic results. Furthermore, existing technologies do not have a pre-defined fault handling rule base that matches the fault diagnosis results, making it impossible to quickly generate targeted fault handling suggestions based on the diagnostic results. There is also no mechanism for synchronously transmitting diagnostic suggestions to the flight control system and the ground maintenance platform, resulting in a lack of clear guidance for fault troubleshooting, low efficiency in maintenance operations, and difficulty in meeting the real-time and practical requirements for fault diagnosis during large eVTOL flights. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method for tracing and diagnosing faults in large eVTOL motors.

[0005] The objective of this invention is achieved through the following technical solution:

[0006] A method for tracing and diagnosing faults in the electronic speed controllers of large eVTOL motors is provided, which includes the following steps:

[0007] S1. Collect operating data under different fault types on the motor side and ESC side, preprocess the collected operating data, extract fault feature parameters from the operating data, label the fault feature parameters according to the fault source and fault type, and construct a motor-ESC coupling fault feature library.

[0008] S2. Based on the sample set of the motor-electronic controller coupling fault feature library, construct and train the fault source diagnosis model, optimize the parameters of the fault source diagnosis model, until the recognition accuracy of the fault source diagnosis model reaches the set value.

[0009] S3. Synchronously collect real-time operating data from the motor side and the ESC side, preprocess the collected real-time operating data, input the preprocessed real-time operating data into the trained fault tracing and diagnosis model, and obtain the fault tracing and diagnosis results.

[0010] S4. Based on the fault tracing and diagnosis results, match the preset fault handling rule library, generate corresponding fault handling suggestions, and transmit the fault handling suggestions to the flight control system and the ground operation and maintenance platform.

[0011] Furthermore, step S1 specifically includes the following sub-steps:

[0012] S1.1. Classify fault types, including motor-side faults and ESC-side faults, define the boundary between motor-side faults and ESC-side faults, and distinguish the signal transmission paths corresponding to motor-side faults and ESC-side faults.

[0013] S1.2. Build a fault simulation test platform. Through the motor fault injection device and the ESC fault simulation module, simulate different fault types and collect multi-dimensional operating data of the motor side and the ESC side under the corresponding fault types.

[0014] S1.3. Perform detrending, filtering and normalization processing on the collected operating data in sequence, and extract the fault characteristic parameters in the processed operating data;

[0015] S1.4. Label the fault feature parameters according to the fault source and fault type, label the fault source, fault type and fault feature mapping relationship corresponding to each sample, and construct the motor-electronic controller coupling fault feature library.

[0016] Furthermore, step S2 specifically includes the following sub-steps:

[0017] S2.1. Divide the sample set of the motor-electromechanical controller coupling fault feature library according to a set ratio to generate a training set, a validation set and a test set, and set the sample coverage and usage rules of the training set, validation set and test set.

[0018] S2.2. Construct a fault tracing and diagnosis model. The fault tracing and diagnosis model includes a convolutional neural network layer, a Transformer encoder layer, and two-level output layers. The output of the convolutional neural network layer is connected to the input of the Transformer encoder layer, and the output of the Transformer encoder layer is connected to the input of the two-level output layers.

[0019] S2.3. Input the training set into the fault tracing and diagnosis model, use the cross-entropy loss function to calculate the output error of the fault tracing and diagnosis model, optimize the parameters of the fault tracing and diagnosis model based on the output error, and introduce a dropout layer;

[0020] S2.4. Verify the recognition accuracy of the fault tracing and diagnosis model using the validation set and test set, and iterate the training process until the recognition accuracy of the fault tracing and diagnosis model reaches the set value.

[0021] Furthermore, step S3 specifically includes the following sub-steps:

[0022] S3.1. Deploy the airborne data acquisition unit, set the acquisition frequency and synchronous acquisition rules of the airborne data acquisition unit, and synchronously acquire real-time operating data from the motor side and the ESC side through the airborne data acquisition unit;

[0023] S3.2. Perform sliding window filtering on the collected real-time running data, calculate the characteristic parameters corresponding to the filtered real-time running data, and complete the preprocessing of the real-time running data;

[0024] S3.3. Simultaneously input the preprocessed real-time running data and corresponding feature parameters into the trained fault tracing and diagnosis model;

[0025] S3.4. First, determine the source of the fault through the fault tracing and diagnosis model, then output the specific fault type, fault severity and related operating parameters, and integrate the output to generate the fault tracing and diagnosis results.

[0026] Furthermore, step S4 specifically includes the following sub-steps:

[0027] S4.1. Preset fault handling rule base. The fault handling rule base contains a one-to-one correspondence between fault source, fault type, fault severity and fault handling suggestions. Set the handling actions and execution logic corresponding to different fault combinations.

[0028] S4.2. The obtained fault tracing and diagnosis results are broken down into three dimensions: fault source, fault type, and fault severity. These are then matched against the corresponding content in the fault handling rule base to obtain corresponding fault handling suggestions.

[0029] S4.3. Standardize the format of the fault handling suggestions obtained from the matching, and transmit the standardized fault handling suggestions synchronously to the flight control system and the ground operation and maintenance platform through the airborne communication module.

[0030] Furthermore, in step S1.1, motor-side faults include inter-turn short circuits in windings, bearing wear, rotor eccentricity, and abnormal back EMF; ESC-side faults include open circuits in power devices, short circuits in power devices, aging of DC bus capacitors, distortion of drive signals, and abnormal dead time. According to the physical location of the fault and the source of the signal, the boundary between motor-side faults and ESC-side faults is defined, the location of the original signal generation corresponding to each type of fault is clarified, and the original fault signal is distinguished from the abnormal signal generated by coupling and transmission, so as to standardize the classification and definition of fault types.

[0031] Furthermore, in step S2.2, the convolutional neural network layer extracts the frequency domain features of the input data, the Transformer encoder layer captures the temporal correlation features of the input features, the first-level output of the two-level output layer outputs the fault source, and the second-level output of the two-level output layer outputs the fault type and fault severity. The frequency domain features output by the convolutional neural network layer are input into the Transformer encoder layer, and the temporal correlation features output by the Transformer encoder layer are input into the two-level output layer for hierarchical transfer and processing of features, and hierarchical output of fault source and fault type.

[0032] Furthermore, in step S3.1, the real-time operating data on the motor side includes three-phase stator current, back electromotive force, bearing temperature, and rotor speed, while the real-time operating data on the electronic speed controller side includes DC bus voltage, switching device case temperature, drive pulse duty cycle, dead time, and output voltage. According to the set acquisition frequency, the real-time operating data on the motor side and the electronic speed controller side are synchronously acquired with the same timestamp. The acquired real-time operating data is transmitted to the airborne diagnostic unit via Ethernet for real-time operating data acquisition and transmission.

[0033] Furthermore, in step S4.2, based on the three dimensions of fault source, fault type, and fault severity, the corresponding fault handling suggestions in the fault handling rule base are matched. The fault handling suggestions include load reduction operation, emergency shutdown, switching redundant equipment, and operation and maintenance troubleshooting. For motor-side faults, ESC-side faults, and coupling faults, corresponding hierarchical handling actions are matched respectively, and the execution conditions and execution order of each handling action are set to match and generate fault handling suggestions.

[0034] Furthermore, in step S3, after the preprocessed real-time running data is input into the trained fault tracing and diagnosis model, the inference delay of the fault tracing and diagnosis model is controlled within a set value; the fault tracing and diagnosis results output by the fault tracing and diagnosis model include the fault occurrence time, fault duration, fault source, fault type, fault severity and associated operating parameters, to trace and output fault information in all dimensions.

[0035] The beneficial effects of this invention are:

[0036] (1) By constructing a motor ESC coupling fault feature library, training a diagnostic model, conducting online fault diagnosis and generating transmission fault handling suggestions, the complete process of the motor ESC fault traceability diagnosis is realized, breaking through the limitation of the traditional method of only being able to determine the existence of faults, clarifying the key information related to faults, and making fault investigation a standardized process with clear direction.

[0037] (2) Combine the coupling characteristics of motor ESC to build a dedicated fault feature library and train a fault source diagnosis model in a targeted manner, so that the sample data of fault diagnosis and model analysis are more in line with the actual operating status of the equipment, effectively reducing the deviation of fault identification and greatly improving the accuracy of fault-related information judgment.

[0038] (3) By collecting real-time operating data of motor ESCs in a differentiated manner and performing rapid preprocessing and model diagnosis, and generating appropriate fault handling suggestions based on the diagnosis results and transmitting them synchronously to relevant platforms, the real-time requirements of fault diagnosis during equipment operation are met, the fault diagnosis and emergency response links are connected, the equipment operation and maintenance efficiency is improved, and technical support is provided for the safe operation of equipment. Attached Figure Description

[0039] Figure 1 A flowchart illustrating the steps of a fault tracing and diagnosis method for a large eVTOL motor ESC;

[0040] Figure 2 The flowchart illustrates the specific steps of a method for tracing and diagnosing faults in a large eVTOL motor electronic speed controller, as provided in this embodiment. Detailed Implementation

[0041] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0042] Example 1

[0043] See Figure 1This embodiment provides a method for tracing and diagnosing faults in the electronic speed controller of a large eVTOL motor. The method includes the following steps:

[0044] S1. Collect operating data under different fault types on the motor side and ESC side, preprocess the collected operating data, extract fault feature parameters from the operating data, label the fault feature parameters according to the fault source and fault type, and construct a motor-ESC coupling fault feature library.

[0045] S2. Based on the sample set of the motor-electronic controller coupling fault feature library, construct and train the fault source diagnosis model, optimize the parameters of the fault source diagnosis model, until the recognition accuracy of the fault source diagnosis model reaches the set value.

[0046] S3. Synchronously collect real-time operating data from the motor side and the ESC side, preprocess the collected real-time operating data, input the preprocessed real-time operating data into the trained fault tracing and diagnosis model, and obtain the fault tracing and diagnosis results.

[0047] S4. Based on the fault tracing and diagnosis results, match the preset fault handling rule library, generate corresponding fault handling suggestions, and transmit the fault handling suggestions to the flight control system and the ground operation and maintenance platform.

[0048] In some embodiments, step S1 specifically includes the following sub-steps:

[0049] S1.1. Classify fault types, including motor-side faults and ESC-side faults, define the boundary between motor-side faults and ESC-side faults, and distinguish the signal transmission paths corresponding to motor-side faults and ESC-side faults.

[0050] S1.2. Build a fault simulation test platform. Through the motor fault injection device and the ESC fault simulation module, simulate different fault types and collect multi-dimensional operating data of the motor side and the ESC side under the corresponding fault types.

[0051] S1.3. Perform detrending, filtering and normalization processing on the collected operating data in sequence, and extract the fault characteristic parameters in the processed operating data;

[0052] S1.4. Label the fault feature parameters according to the fault source and fault type, label the fault source, fault type and fault feature mapping relationship corresponding to each sample, and construct the motor-electronic controller coupling fault feature library.

[0053] In some embodiments, step S2 specifically includes the following sub-steps:

[0054] S2.1. Divide the sample set of the motor-electromechanical controller coupling fault feature library according to a set ratio to generate a training set, a validation set and a test set, and set the sample coverage and usage rules of the training set, validation set and test set.

[0055] S2.2. Construct a fault tracing and diagnosis model. The fault tracing and diagnosis model includes a convolutional neural network layer, a Transformer encoder layer, and two-level output layers. The output of the convolutional neural network layer is connected to the input of the Transformer encoder layer, and the output of the Transformer encoder layer is connected to the input of the two-level output layers.

[0056] S2.3. Input the training set into the fault tracing and diagnosis model, use the cross-entropy loss function to calculate the output error of the fault tracing and diagnosis model, optimize the parameters of the fault tracing and diagnosis model based on the output error, and introduce a dropout layer;

[0057] S2.4. Verify the recognition accuracy of the fault tracing and diagnosis model using the validation set and test set, and iterate the training process until the recognition accuracy of the fault tracing and diagnosis model reaches the set value.

[0058] In some embodiments, step S3 specifically includes the following sub-steps:

[0059] S3.1. Deploy the airborne data acquisition unit, set the acquisition frequency and synchronous acquisition rules of the airborne data acquisition unit, and synchronously acquire real-time operating data from the motor side and the ESC side through the airborne data acquisition unit;

[0060] S3.2. Perform sliding window filtering on the collected real-time running data, calculate the characteristic parameters corresponding to the filtered real-time running data, and complete the preprocessing of the real-time running data;

[0061] S3.3. Simultaneously input the preprocessed real-time running data and corresponding feature parameters into the trained fault tracing and diagnosis model;

[0062] S3.4. First, determine the source of the fault through the fault tracing and diagnosis model, then output the specific fault type, fault severity and related operating parameters, and integrate the output to generate the fault tracing and diagnosis results.

[0063] In some embodiments, step S4 specifically includes the following sub-steps:

[0064] S4.1. Preset fault handling rule base. The fault handling rule base contains a one-to-one correspondence between fault source, fault type, fault severity and fault handling suggestions. Set the handling actions and execution logic corresponding to different fault combinations.

[0065] S4.2. The obtained fault tracing and diagnosis results are broken down into three dimensions: fault source, fault type, and fault severity. These are then matched against the corresponding content in the fault handling rule base to obtain corresponding fault handling suggestions.

[0066] S4.3. Standardize the format of the fault handling suggestions obtained from the matching, and transmit the standardized fault handling suggestions synchronously to the flight control system and the ground operation and maintenance platform through the airborne communication module.

[0067] In some embodiments, in step S1.1, motor-side faults include inter-turn short circuits in windings, bearing wear, rotor eccentricity, and abnormal back EMF; ESC-side faults include open circuits in power devices, short circuits in power devices, aging of DC bus capacitors, distortion of drive signals, and abnormal dead time. According to the physical location of the fault and the source of the signal, the boundary between motor-side faults and ESC-side faults is defined, the location of the original signal generation corresponding to each type of fault is clarified, the original fault signal is distinguished from the abnormal signal generated by coupling and transmission, and the fault types are standardized and defined.

[0068] In some embodiments, in step S2.2, the convolutional neural network layer extracts the frequency domain features of the input data, the Transformer encoder layer captures the temporal correlation features of the input features, the first-level output of the two-level output layer outputs the fault source, and the second-level output of the two-level output layer outputs the fault type and fault severity. The frequency domain features output by the convolutional neural network layer are input into the Transformer encoder layer, and the temporal correlation features output by the Transformer encoder layer are input into the two-level output layer for hierarchical transfer and processing of features, and hierarchical output of fault source and fault type.

[0069] In some embodiments, in step S3.1, the real-time operating data on the motor side includes three-phase stator current, back electromotive force, bearing temperature, and rotor speed, while the real-time operating data on the electronic speed controller side includes DC bus voltage, switching device case temperature, drive pulse duty cycle, dead time, and output voltage. The real-time operating data on the motor side and the electronic speed controller side are synchronously collected with the same timestamp according to the set collection frequency. The collected real-time operating data is transmitted to the airborne diagnostic unit via Ethernet for real-time operating data collection and transmission.

[0070] In some embodiments, in step S4.2, based on three dimensions—fault source, fault type, and fault severity—fault handling suggestions are matched with corresponding fault handling suggestions in the fault handling rule base. The fault handling suggestions include load reduction operation, emergency shutdown, switching redundant equipment, and operation and maintenance troubleshooting. For motor-side faults, ESC-side faults, and coupling faults, corresponding hierarchical handling actions are matched respectively, and the execution conditions and execution order of each handling action are set to match and generate fault handling suggestions.

[0071] In some embodiments, in step S3, after the preprocessed real-time running data is input into the trained fault tracing and diagnosis model, the inference delay of the fault tracing and diagnosis model is controlled within a set value; the fault tracing and diagnosis results output by the fault tracing and diagnosis model include the fault occurrence time, fault duration, fault source, fault type, fault severity and associated operating parameters, and perform fault tracing and output of full-dimensional information.

[0072] Example 2

[0073] This embodiment proposes a specific implementation process for a fault tracing and diagnosis method for large-scale eVTOL motor ESCs. This process, considering the strong coupling operation characteristics of the motor and ESC in large-scale eVTOLs, forms a complete motor ESC fault tracing and diagnosis process through the systematic construction of a fault feature database, targeted training of the fault tracing and diagnosis model, online diagnosis of real-time operating data, and accurate generation and transmission of fault handling suggestions. This solves the problem of difficulty in distinguishing the source and specific type of motor ESC faults. Figure 2 As shown, the specific implementation process is as follows:

[0074] S1. Construct a coupled fault feature library:

[0075] This step focuses on the fault characteristics of the motor and ESC sides, completing the entire process of fault type classification, fault data collection, data preprocessing, feature annotation, and feature library construction. It provides standardized and targeted sample data for the subsequent training of the fault tracing and diagnostic model, and is a fundamental step in achieving accurate fault tracing. Specifically, it is implemented through the following sub-steps:

[0076] S1.1. Classify fault types and set classification rules:

[0077] First, the overall classification dimensions of fault types are clarified, dividing eVTOL motor and ESC faults into two categories: motor-side faults and ESC-side faults. This classification is based on the physical operating boundaries and signal transmission characteristics of the motor and ESC in the eVTOL distributed electric propulsion system. In this embodiment, the boundary classification rules between motor-side faults and ESC-side faults are first set, with the physical location of the fault as the primary basis, supplemented by the source of the fault signal. Then, the signal transmission paths corresponding to motor-side faults and ESC-side faults are distinguished. The signal transmission path is the path along which the abnormal signal is transmitted from the location of the fault to the coupling component after the fault occurs. Clarifying the signal transmission paths of different faults allows for accurate capture of the original fault signal and the coupled abnormal signal during the data acquisition stage, avoiding signal confusion.

[0078] Motor-side faults refer to various faults occurring within the motor itself, specifically including inter-turn short circuits in the windings, bearing wear, rotor eccentricity, and abnormal back EMF. Inter-turn short circuits occur when adjacent coils in the stator windings of the motor experience a short circuit, leading to a regular abnormality in the motor's operating current signal. In this embodiment, it is considered a typical electrical fault on the motor side and is included as a core type in fault simulation and data acquisition. Bearing wear is a mechanical wear problem caused by long-term operation of the motor bearings, directly affecting the stability of the motor's rotor speed, and is classified as a mechanical fault on the motor side. Rotor eccentricity is a misalignment between the center of the motor rotor and the center of the stator, leading to an imbalance in the three-phase current of the motor. Abnormal back EMF occurs when the amplitude and frequency of the back EMF generated during motor operation deviate from the normal range, often caused by abnormal rotor operation.

[0079] ESC-side faults refer to various faults occurring in the ESC drive unit, specifically including open circuits in power devices, short circuits in power devices, aging of DC bus capacitors, drive signal distortion, and abnormal dead time. Among these, power devices are the main components in the ESC that realize power conversion and control. Open circuits and short circuits in power devices are typical electrical faults of the ESC, which will directly lead to abnormalities in the output voltage and current of the ESC. Aging of DC bus capacitors refers to the performance degradation of the capacitors at the DC bus of the ESC due to the use time or operating conditions, which will cause abnormal ripple in the DC bus voltage. Drive signal distortion refers to the deviation of the drive pulse signal output by the ESC from the set value in amplitude, frequency, or duty cycle, which will affect the normal drive of the motor. Abnormal dead time refers to the deviation of the set dead time during the switching process of the ESC power devices from the set range, which will cause abnormal harmonics in the ESC output.

[0080] In this embodiment, the original signal generation location corresponding to each type of fault is further clarified according to the physical location of the fault and the source of the signal generation. The original signal generation location is the first signal acquisition point where an anomaly occurs after the fault occurs. At the same time, the original fault signal and the abnormal signal generated by coupling are strictly distinguished. The original fault signal is the signal anomaly directly caused by the fault, while the abnormal signal generated by coupling is the signal anomaly generated after the original fault signal is transmitted to another component through the coupling relationship of the motor ESC. Based on the above, the standardized classification and definition of fault types are completed, providing a clear basis for subsequent fault simulation and data acquisition.

[0081] In some embodiments, based on the operating conditions of the eVTOL, a classification of coupled fault types can be added on the basis of the basic fault types. Coupled fault types refer to fault types caused by motor-side faults leading to ESC-side abnormalities or ESC-side faults leading to motor-side abnormalities. The classified coupled fault types include two categories: motor rotor eccentricity causing ESC current imbalance and ESC drive signal distortion causing motor back EMF abnormalities. There is no need to repeatedly classify a single fault type; the fault type dimension is supplemented only by the coupling relationship.

[0082] S1.2. Build an experimental platform and collect multi-dimensional operational data:

[0083] Based on the standardized classification of fault types completed in S1.1, a fault simulation test platform was built. This test platform is a physical test platform, which includes two core parts: a motor fault injection device and an ESC fault simulation module. The motor fault injection device is a special device for simulating various faults on the motor side. It can specifically simulate motor-side faults such as inter-turn short circuits in windings and rotor eccentricity according to test requirements. The ESC fault simulation module is a functional module for simulating various faults on the ESC side. It can simulate ESC-side faults such as open circuits in power devices, short circuits in power devices, and aging of DC bus capacitors through circuit adjustment.

[0084] In this embodiment, different fault types are simulated using a motor fault injection device and an ESC fault simulation module. The simulation is conducted according to the principle of single fault type simulation, that is, only one fault type is simulated at a time to avoid signal mixing problems caused by simulating multiple fault types at the same time. During the simulation of each fault type, multi-dimensional operating data of the motor side and ESC side under the corresponding fault type are collected. During the collection process, it is ensured that each set of operating data corresponds to a unique fault type. At the same time, relevant parameters during the fault simulation process are recorded to ensure the traceability of the operating data. The collected multi-dimensional operating data covers electrical signals, mechanical operating parameters, and temperature parameters of the motor side and ESC side, so as to achieve comprehensive capture of fault signals.

[0085] S1.3. Preprocess the operating data and extract fault characteristic parameters:

[0086] The multi-dimensional operating data from the motor and ESC sides collected in S1.2 are systematically preprocessed. The preprocessing operations are performed in the order of detrending, filtering, and normalization. Detrending eliminates the bias caused by system trend changes in the operating data, making the data more accurately reflect the anomalies caused by faults. Filtering eliminates noise interference in the operating data, which mainly comes from the circuits and equipment operation of the test platform. Filtering improves the effectiveness of the data. Normalization converts operating data with different dimensions and numerical ranges to the same numerical range, avoiding the impact of differences in data dimensions and ranges on the subsequent extraction of fault feature parameters and model training.

[0087] After completing the above preprocessing operations, fault feature parameters are extracted from the processed running data. Fault feature parameters are the main parameters that can characterize the fault type and fault state. Different fault types correspond to different fault feature parameters. During the extraction process, according to the fault types divided in S1.1, typical feature parameters of each type of fault are extracted in a targeted manner to ensure that the extracted fault feature parameters can accurately reflect the running state of the corresponding fault, and provide core data support for subsequent sample labeling and feature library construction.

[0088] S1.4. Label the samples and construct a motor-electromechanical controller coupling fault feature library:

[0089] The fault feature parameters extracted in S1.3 are sample-labeled. The labeling process takes the fault source and fault type as the core labeling dimensions. The fault sources are divided into two categories: motor side and ESC side. The fault types are the specific fault types standardized in S1.1. During labeling, the corresponding fault source and fault type are labeled for each fault feature parameter sample. At the same time, the fault feature mapping relationship corresponding to each sample is clarified. The fault feature mapping relationship is the correspondence between the fault feature parameters and the fault source and fault type. Through this relationship, the basic information of the fault can be directly judged from the fault feature parameters.

[0090] After labeling all fault feature parameter samples, a motor-electronic speed controller (ESC) coupling fault feature library is constructed. The labeled samples are entered into the feature library in a unified format. At the same time, a sample retrieval mechanism is established, which can quickly retrieve samples in the feature library based on dimensions such as fault source and fault type. This ensures the practicality and convenience of the motor-electronic speed controller coupling fault feature library and provides a standardized sample set for the training of subsequent fault tracing and diagnosis models.

[0091] In some specific implementations, to address the problem of insufficient sample size in the fault feature library leading to weak model training generalization ability and large fault identification bias, when constructing the motor-electronic controller coupling fault feature library, a total of 50,000 sets of valid samples are collected and labeled. The sample distribution is matched according to the actual occurrence probability of the fault type. Among them, the proportion of motor-side fault samples and electronic controller-side fault samples is basically equal, and the proportion of coupling fault-related samples is about one-fifth of the total sample size, ensuring that the sample library can cover various fault scenarios in actual operation. During the sample collection phase, for each type of fault, the state of different fault development stages is simulated, and at least two thousand sets of corresponding data are collected. For example, for inter-turn short circuits in motor windings, data are collected from minor short circuits, moderate short circuits to severe short circuits, and characteristic parameters such as zero-sequence current harmonics at different stages are extracted. For short circuits in ESC power devices, characteristic parameters such as bus current mutation rate under different short circuit degrees are collected. Then, all collected and preprocessed samples are accurately labeled according to the fault source and fault type, ultimately forming a motor-ESC coupling fault feature library of 50,000 sets of valid samples. This provides sufficient and realistic sample support for the training of subsequent fault tracing and diagnosis models, effectively solving the problem of inaccurate model recognition caused by single samples.

[0092] S2. Training the fault tracing and diagnosis model:

[0093] This step, based on the motor-electromechanical controller coupling fault feature library built in S1, completes the sample set partitioning, diagnostic model construction, model parameter optimization, and model validation iteration. Through targeted training, the fault tracing and diagnostic model acquires the ability to identify fault sources and fault types. This is implemented through the following sub-steps:

[0094] S2.1. Divide the sample set and set the usage rules:

[0095] The sample set in the motor-electronic controller coupling fault feature library is divided into three categories according to a set ratio: training set, validation set, and test set. The training set is used for basic training of the fault tracing and diagnosis model; the validation set is used for parameter verification and optimization during model training; and the test set is used for overall performance verification after model training is completed. In this embodiment, the sample coverage of the training set, validation set, and test set is first defined. The sample coverage requires that all three sample sets cover all fault types and fault sources defined in S1, ensuring the comprehensiveness of model training, validation, and testing. Then, usage rules for the three sample sets are defined, clarifying the stages and methods of use of the training set, validation set, and test set during model training. Simultaneously, the three sample sets must not have duplicate coverage to avoid model training bias caused by sample duplication, ensuring the effectiveness of model training.

[0096] S2.2. Construct a fault tracing and diagnosis model and design hierarchical connection relationships:

[0097] A fault tracing and diagnosis model is constructed. This model is a dedicated model for tracing and diagnosing faults in eVTOL motor ESCs. It consists of three main layers: a convolutional neural network layer, a Transformer encoder layer, and a two-stage output layer. Each layer is connected end-to-end, meaning that the output of the convolutional neural network layer is directly connected to the input of the Transformer encoder layer, and the output of the Transformer encoder layer is directly connected to the input of the two-stage output layer. There are no intermediate conversion links in the signal transmission between layers, ensuring the complete transmission of fault characteristics.

[0098] The convolutional neural network (CNN) layer is a neural network layer based on convolution operations, possessing powerful local feature extraction capabilities. In this embodiment, the main function of the CNN layer is to extract the frequency domain features of the input data. Frequency domain features reflect abnormal features of the operating data in the frequency dimension, such as current harmonics and voltage ripple, which are important features characterizing motor ESC faults. The Transformer encoder layer is a neural network layer based on the self-attention mechanism, possessing the ability to capture the temporal correlation of data. In this embodiment, the main function of the Transformer encoder layer is to capture the temporal correlation features of the input features. Temporal correlation features reflect the trend of parameter changes during the fault development process, and can reflect the entire process of fault occurrence and development.

[0099] The two-level output layer is the result output part of the fault tracing and diagnosis model, divided into a primary output end and a secondary output end. The primary output end outputs the fault source, i.e., the motor side or the ESC side, while the secondary output end outputs the fault type and fault severity. The fault severity is divided into different levels based on the degree of abnormality of the fault feature parameters. In this embodiment, the frequency domain features extracted by the convolutional neural network layer are directly input into the Transformer encoder layer. After performing temporal correlation analysis on the input frequency domain features, the Transformer encoder layer outputs a comprehensive feature that integrates the frequency domain features and the temporal correlation features. This comprehensive feature is directly input into the two-level output layer, completing the hierarchical transfer and processing of features and outputting the fault source and fault type hierarchically.

[0100] In some embodiments, the convolutional neural network layer of the fault tracing and diagnosis model can adopt a stacked structure of multiple convolutional kernels, extracting frequency domain features of different dimensions through convolutional kernels of different sizes. The number of attention heads in the Transformer encoder layer is set to a set value, capturing multi-dimensional temporal correlation features through multiple attention heads. The two-level output layer can adopt a fully connected layer structure, realizing hierarchical output of fault information through the fully connected layer. There is no need to change the overall hierarchical structure of the model, and the feature processing capability of the model can be improved only by adjusting the internal parameters.

[0101] S2.3. Input the training set and optimize the model parameters:

[0102] The training set divided in S2.1 is input into the fault tracing and diagnosis model to start the basic training process of the model. During the training process, the cross-entropy loss function is used to calculate the output error of the fault tracing and diagnosis model. The cross-entropy loss function is a loss function used for classification problems. It can accurately calculate the deviation between the model's prediction results and the actual sample labeling results. In this embodiment, the identification of fault source and fault type is taken as a classification problem, and the output error of the model during the training process is calculated by the cross-entropy loss function.

[0103] Based on the calculated output error, the parameters of the fault tracing and diagnosis model are optimized. The parameter optimization process follows the direction of gradient descent, gradually adjusting the parameters of each level of the model to gradually reduce the output error. At the same time, a dropout layer is introduced during model training. The dropout layer is a technique used to prevent overfitting of neural networks. By randomly discarding the connections of some neurons during training, the model avoids overfitting the sample features of the training set, improves the generalization ability of the model, and ensures that the model can still maintain good recognition performance when facing new real-time running data.

[0104] S2.4. Verify the model accuracy and iteratively train it:

[0105] The validation and test sets defined in S2.1 are sequentially input into the fault tracing and diagnosis model during training. The recognition accuracy of the fault tracing and diagnosis model is verified using the validation and test sets. The verification process focuses on the model's accuracy in identifying the fault source and the fault type, while also verifying the model's accuracy in judging the severity of the fault. In this embodiment, if the model's recognition accuracy does not reach the set value, the model parameters are optimized and adjusted again based on the output errors calculated during the verification and testing processes. Then, the training set is re-inputted into the model for training. This training, verification, and optimization process is iterated until the recognition accuracy of the fault tracing and diagnosis model reaches the set value. At this point, the overall training process of the fault tracing and diagnosis model is complete, and the trained fault tracing and diagnosis model possesses stable fault tracing and recognition capabilities.

[0106] In some specific implementations, to address the problem of poor model training and substandard recognition accuracy caused by unreasonable sample set division ratios, when dividing the sample set of the motor-electronic speed controller coupling fault feature library, the 50,000 effective samples are strictly divided into training, validation, and test sets according to a 7:2:1 ratio. That is, the training set consists of 35,000 samples, the validation set consists of 10,000 samples, and the test set consists of 5,000 samples. Furthermore, the distribution of fault types and fault sources in each sample set is consistent with the total sample set, avoiding the excessively high or low proportion of a certain type of fault in a single sample set. Meanwhile, clear model recognition accuracy targets were set, with the accuracy target for fault source identification set at no less than 99% and the accuracy target for fault type identification set at no less than 97%. During the model training iteration process, the accuracy was tested with a validation set after each iteration. If the accuracy target for fault source identification was lower than 99% or the accuracy target for fault type identification was lower than 97%, the model parameters were further optimized based on the error calculated by the cross-entropy loss function. When introducing the dropout layer, the probability of dropping neurons was set to a predetermined value to effectively prevent model overfitting. The model training was completed only when the recognition accuracy of the model on both the validation set and the test set reached the above-mentioned predetermined values. This solved the problem of poor model training effect caused by unclear sample division and accuracy standards.

[0107] S3. Conduct online fault tracing and diagnosis:

[0108] This step, based on the fault tracing and diagnosis model trained in S2, completes the acquisition, preprocessing, model input, and diagnostic result generation of real-time operating data of the motor ESC, realizing online fault tracing and diagnosis of the eVTOL motor ESC. Specifically, it is implemented through the following sub-steps:

[0109] S3.1. Deploy the data acquisition unit and synchronously collect real-time operational data:

[0110] An airborne data acquisition unit is deployed. This unit is a dedicated unit for collecting real-time operating data of the motor and ESC during eVTOL flight. It is adaptable to the airborne operating environment of eVTOL. In this embodiment, the acquisition frequency and synchronous acquisition rules of the airborne data acquisition unit are first set. The acquisition frequency is set according to the characteristics of different types of real-time operating data. The synchronous acquisition rules require that the real-time operating data of the motor side and the ESC side be acquired with the same timestamp to ensure the time consistency of the data on the motor side and the ESC side. This ensures that the acquired real-time operating data can truly reflect the coupled operating state of the motor and the ESC. At the same time, the acquisition time deviation between the motor side and the ESC side is controlled within the set value to avoid fault diagnosis errors caused by time deviation.

[0111] The onboard data acquisition unit synchronously collects real-time operating data from both the motor and electronic speed controller sides. The real-time operating data from the motor side includes three-phase stator current, back electromotive force (EMF), bearing temperature, and rotor speed. The three-phase stator current is the real-time current signal in the motor stator windings and is a key parameter reflecting the electrical operating status of the motor. The back EMF is the induced electromotive force generated during motor operation and reflects the rotor operating status. The bearing temperature is the real-time temperature of the motor bearings and reflects the mechanical operating status of the motor. The rotor speed is the real-time operating speed of the motor rotor and is an important parameter reflecting the overall operating status of the motor.

[0112] The real-time operating data on the ESC side includes DC bus voltage, switching device case temperature, drive pulse duty cycle, dead time, and output voltage. The DC bus voltage is the real-time voltage at the DC bus of the ESC, which is the main parameter reflecting the power supply status of the ESC. The switching device case temperature is the real-time temperature of the ESC power device's casing, which reflects the operating status of the ESC power device. The drive pulse duty cycle is the duty cycle of the ESC's output drive pulse, which is an important parameter reflecting the ESC's drive control status. The dead time is the real-time dead time during the switching process of the ESC power device, and the output voltage is the real-time voltage output by the ESC to the motor. Both of these can reflect the output control status of the ESC.

[0113] In this embodiment, the real-time operating data collected is transmitted to the airborne diagnostic unit via Ethernet. Ethernet transmission has the characteristics of high transmission speed and high stability, which can meet the real-time requirements of eVTOL online fault diagnosis. After the real-time operating data is collected and transmitted, the data is temporarily stored in the airborne diagnostic unit to prepare for subsequent preprocessing operations.

[0114] In some embodiments, a data cache module can be added to the airborne data acquisition unit. The cache module can temporarily store real-time operating data for a set duration. If the Ethernet transmission is briefly interrupted, the temporarily stored data can be retrieved through the cache module to avoid data loss. The acquired real-time operating data retains only three core data types: current, voltage, and temperature, simplifying the data transmission and processing flow.

[0115] In some specific implementations, to address the problem of inaccurate fault feature capture and insufficient real-time performance caused by mismatched acquisition frequencies for different types of operational data, the acquisition frequency of the airborne data acquisition unit is set differently based on the characteristics of the data and the needs of fault diagnosis. For example, the acquisition frequency for current and voltage parameters on the motor side and ESC side is set to 20 kHz, which can accurately capture fault features in these high-frequency changing electrical signals, such as harmonic changes in three-phase stator current and abnormal ripple in DC bus voltage. The acquisition frequency for temperature parameters is set to 1 kHz, which can reflect temperature change trends in a timely manner without causing data redundancy due to excessively high acquisition frequencies. The acquisition frequency for drive signal parameters is set to 10 kHz, which can accurately capture subtle deviations in drive pulse duty cycle and dead time. Meanwhile, the time deviation of all operating data acquisition between the motor side and the ESC side is controlled within microseconds to ensure the consistency of data acquisition at the same timestamp. The acquired data stream is transmitted to the airborne diagnostic unit via 100 Mbps Ethernet, and the transmission delay is controlled within the set value. This effectively solves the problem of inaccurate fault signal capture caused by unreasonable acquisition frequency and time deviation, allowing subsequent fault diagnosis to be carried out based on real and synchronous real-time operating data.

[0116] S3.2. Preprocess real-time running data and calculate characteristic parameters:

[0117] The real-time operating data collected and transmitted to the airborne diagnostic unit in S3.1 is preprocessed. The core operation of the preprocessing is sliding window filtering. Sliding window filtering refers to the method of selecting a time window of a set length and continuously filtering the real-time operating data within the window. This method can effectively eliminate random fluctuation interference in the real-time operating data. In this embodiment, by using sliding window filtering, various types of real-time operating data from the motor side and the ESC side are filtered one by one to improve the smoothness and effectiveness of the data.

[0118] After the sliding window filtering process is completed, the corresponding feature parameters are calculated based on the filtered real-time running data. The calculated feature parameters are of the same type as the fault feature parameters extracted in S1.3. The calculation process is carried out according to the set algorithm to ensure the accuracy and standardization of the calculation results. After the feature parameters are calculated, the overall preprocessing operation of the real-time running data is completed. The preprocessed real-time running data and the calculated feature parameters are used together as input data for fault tracing and diagnosis.

[0119] S3.3. Input the preprocessed data into the fault tracing and diagnosis model:

[0120] The preprocessed real-time operating data and corresponding feature parameters from step S3.2 are synchronously input into the fault tracing and diagnosis model trained in step S2, according to the input format requirements of the fault tracing and diagnosis model. During the input process, it is ensured that the data from the motor side and the ESC side are input synchronously without any sequential deviation, while also ensuring the integrity of the input data to avoid diagnostic errors due to missing data. In this embodiment, the airborne diagnostic unit and the fault tracing and diagnosis model use a direct data connection, eliminating intermediate conversion steps in the transmission of input data and ensuring real-time data transmission to meet the requirements of eVTOL online fault diagnosis.

[0121] S3.4. Determine fault information and generate fault tracing and diagnosis results:

[0122] The fault tracing and diagnosis model analyzes and processes the input preprocessed data. The model first determines the source of the fault based on the input data, that is, whether the fault occurs on the motor side or the ESC side. Then, based on the determination of the source of the fault, it further outputs the specific fault type and fault severity. At the same time, it extracts the associated operating parameters related to the fault. The associated operating parameters refer to the motor and ESC operating parameters related to the occurrence and development of the fault, which can reflect the specific operating status of the fault.

[0123] In this embodiment, the fault source, specific fault type, fault severity, and associated operating parameters output by the fault tracing and diagnosis model are integrated. The occurrence time and duration of the fault are also recorded. Fault tracing and diagnosis results are generated according to a unified format, providing the primary basis for generating subsequent fault handling suggestions. Furthermore, after inputting preprocessed real-time operating data into the trained fault tracing and diagnosis model, the inference latency of the model is controlled within a set value to ensure the real-time performance of fault tracing and diagnosis, meeting the requirements for rapid fault diagnosis during eVTOL flight.

[0124] In some embodiments, the fault tracing and diagnosis results can include the prediction of fault development trends. By continuously analyzing the input real-time operating data through the model, the development trend of the fault within a set time period can be predicted. The prediction content only includes the changing trend of fault severity, without the need for complex fault consequence prediction. The generated fault tracing and diagnosis results contain three core types of information: fault source, fault type, and development trend, simplifying the presentation of the diagnosis results.

[0125] In some specific implementations, to address the issue of insufficient real-time online diagnostics due to excessively long inference latency in fault tracing and diagnosis models, which fails to meet the emergency response requirements during eVTOL flights, the inference latency of the fault tracing and diagnosis model is strictly controlled to within five milliseconds after the preprocessed real-time operational data is input into the trained model. This is achieved through model lightweighting optimization and hardware computing power adaptation. Specifically, convolutional neural network layers in the fault tracing and diagnosis model undergo kernel pruning to remove redundant kernels, and the self-attention mechanism of the Transformer encoder layer is simplified to reduce computational load. Furthermore, the trained model is deployed on a high-performance airborne embedded chip with high-speed matrix operation capabilities, enabling rapid processing of feature extraction and analysis calculations. During model inference, the preprocessed data is input into the model in the form of small batches to avoid increased inference latency due to excessive data volume. This ensures that the entire process from data input to output of fault tracing and diagnosis results is completed within five milliseconds, effectively solving the problem of excessively long model inference latency. This allows pilots and ground maintenance personnel to obtain fault diagnosis information as soon as possible, buying time for emergency response.

[0126] S4. Generate and transmit fault handling suggestions:

[0127] This step, based on the fault tracing and diagnosis results generated in S3, completes the operations of presetting the fault handling rule base, matching diagnostic results, generating and transmitting handling suggestions, thereby achieving effective connection between fault diagnosis and fault handling. It provides targeted fault handling guidance for eVTOL flight operations and maintenance, and is implemented through the following sub-steps:

[0128] S4.1. Preset the fault handling rule base and set the execution logic:

[0129] A pre-defined fault handling rule base is established. This rule base is a database used to store the correspondence between fault tracing and diagnosis results and fault handling suggestions. In this embodiment, the fault handling rule base contains a one-to-one correspondence between fault source, fault type, fault severity, and fault handling suggestions. That is, for different combinations of fault sources, different fault types, and different fault severity, corresponding fault handling suggestions are pre-defined. At the same time, processing actions and execution logic corresponding to different fault combinations are set. Processing actions refer to the specific operations that need to be taken for the fault, and execution logic refers to the execution order and execution requirements of the processing actions. Fault combinations include single fault combinations and multi-fault coupling combinations. A single fault combination refers to the case where only one type of fault exists, while a multi-fault coupling combination refers to the case where multiple types of faults exist simultaneously or one type of fault triggers another type of fault, ensuring that the fault handling rule base can cover various fault scenarios of eVTOL motor ESCs.

[0130] S4.2. Match diagnostic results and generate fault handling suggestions:

[0131] The fault tracing and diagnosis results obtained in S3 are broken down into three dimensions: fault source, fault type, and fault severity. These three dimensions are then matched against the corresponding content in the fault handling rule base. Full-dimensional matching requires that the content of each dimension precisely corresponds to the content in the rule base to avoid matching bias caused by missing dimensions. In this embodiment, fault handling suggestions include load reduction operation, emergency shutdown, switching to redundant equipment, and maintenance troubleshooting. Based on the matching results of the three dimensions—fault source, fault type, and fault severity—the corresponding fault handling suggestions are retrieved from the fault handling rule base.

[0132] For motor-side faults, ESC-side faults, and coupling faults, corresponding graded handling actions are matched. Graded handling actions refer to different levels of handling actions based on the severity of the fault. Different fault severity corresponds to different levels of handling actions. At the same time, the execution conditions and execution order of each handling action are set. The execution conditions refer to the prerequisites for taking the handling action, and the execution order refers to the order in which the actions are executed when there are multiple handling actions. Based on the above, fault handling suggestions are matched and generated to ensure that the generated fault handling suggestions are targeted and operable.

[0133] S4.3. Standardized processing recommendations and simultaneous transmission to relevant platforms:

[0134] The fault handling suggestions obtained in S4.2 are standardized in format. Standardization involves converting these suggestions into a unified, standardized format according to the receiving format requirements of the eVTOL flight control system and the ground maintenance platform. This ensures that the flight control system and the ground maintenance platform can accurately identify and receive the fault handling suggestions. In this embodiment, the standardized fault handling suggestions are synchronously transmitted to the flight control system and the ground maintenance platform via an airborne communication module. This airborne communication module is a dedicated communication module for eVTOL aircraft, enabling wireless data transmission between airborne equipment and the flight control system and ground maintenance platform. This ensures that fault handling suggestions can be transmitted quickly and accurately to the relevant platforms, providing timely guidance for pilots' emergency response and ground maintenance personnel's troubleshooting.

[0135] In some embodiments, fault handling suggestions can be graded and labeled. Based on the severity of the fault, the fault handling suggestions are divided into three levels: general prompts, key reminders, and emergency alarms. Different levels of suggestions adopt different transmission priorities, and emergency alarm suggestions adopt the highest transmission priority to ensure that pilots can receive them as soon as possible. The transmitted fault handling suggestions only contain core processing actions and do not need to be accompanied by complex parameter descriptions, thereby improving the readability and execution efficiency of the suggestions.

[0136] This embodiment achieves accurate source tracing and diagnosis of large-scale eVTOL motor ESC faults through a complete process involving constructing a coupled fault feature library, training a fault tracing and diagnosis model, conducting online fault tracing and diagnosis, and generating transmission fault handling suggestions. From the core process perspective, by constructing a dedicated motor-ESC coupled fault feature library, the fault diagnosis sample data better reflects the coupled operating characteristics of the motor and ESC, effectively reducing diagnostic bias caused by discrepancies between sample data and actual operating scenarios, thus providing a more solid foundation for subsequent model training. The fault tracing and diagnosis model employs a combination of convolutional neural network layers and Transformer encoder layers, which can extract frequency domain features from the data and capture the temporal correlation of features, allowing the model to process fault features more comprehensively and improving the accuracy of fault source and fault type identification. During online fault diagnosis, real-time operating data from both the motor and ESC sides is collected synchronously, ensuring data temporal consistency. This allows the model to perform diagnosis based on real coupled operating data, reducing diagnostic errors caused by data deviations, and keeping the model's inference latency within a set value, meeting the real-time requirements of eVTOL online fault diagnosis. The generation of fault handling recommendations is based on a comprehensive matching of diagnostic results, ensuring that the recommendations accurately correspond to the actual situation of the fault. The standardized format and synchronous transmission method allow the flight control system and ground maintenance platform to quickly receive and execute the recommendations, achieving seamless integration of fault diagnosis and handling. Furthermore, the various steps of the entire method are logically interconnected, with the results of each step providing direct evidence for the implementation of the next, making the entire diagnostic process more systematic and coherent. This effectively improves the efficiency of eVTOL motor and ESC fault diagnosis, reduces ineffective operations during maintenance, and provides technical support for eVTOL flight safety, reducing flight risks caused by motor and ESC failures. This method, through standardized process design, overcomes the limitations of traditional fault diagnosis, which can only determine the existence of faults. It achieves precise tracing of the fault source, specific type, and severity, upgrading the fault diagnosis of eVTOL motor ESCs from a single alarm to a comprehensive source analysis. It provides a standardized methodological reference for the operation and maintenance of eVTOL distributed electric propulsion systems. At the same time, each step of this method can be flexibly adjusted according to different operating conditions of eVTOLs, possessing strong adaptability and practicality, and can be applied to different types of large eVTOL motor ESC fault diagnosis scenarios.

[0137] The above description is merely a preferred embodiment of the present invention. It should be understood that the present invention is not limited to the forms disclosed herein and should not be construed as excluding other embodiments. It can be used in various other combinations, modifications, and environments, and can be altered within the scope of the concept described herein through the above teachings or related technologies or knowledge. Modifications and variations made by those skilled in the art that do not depart from the spirit and scope of the present invention should be within the protection scope of the appended claims.

Claims

1. A large eVTOL motor electric governor fault tracing diagnosis method, characterized in that, Includes the following steps: S1. Collect operating data under different fault types on the motor side and ESC side, preprocess the collected operating data, extract fault feature parameters from the operating data, label the fault feature parameters according to the fault source and fault type, and construct a motor-ESC coupling fault feature library. S2. Based on the sample set of the motor-electronic controller coupling fault feature library, construct and train the fault source diagnosis model, optimize the parameters of the fault source diagnosis model, until the recognition accuracy of the fault source diagnosis model reaches the set value. S3. Synchronously collect real-time operating data from the motor side and the ESC side, preprocess the collected real-time operating data, input the preprocessed real-time operating data into the trained fault tracing and diagnosis model, and obtain the fault tracing and diagnosis results. S4. Based on the fault tracing and diagnosis results, match the preset fault handling rule library, generate corresponding fault handling suggestions, and transmit the fault handling suggestions to the flight control system and the ground operation and maintenance platform.

2. The method of claim 1, wherein, Step S1 specifically includes the following sub-steps: S1.

1. Classify fault types, including motor-side faults and ESC-side faults, define the boundary between motor-side faults and ESC-side faults, and distinguish the signal transmission paths corresponding to motor-side faults and ESC-side faults. S1.

2. Build a fault simulation test platform. Through the motor fault injection device and the ESC fault simulation module, simulate different fault types and collect multi-dimensional operating data of the motor side and the ESC side under the corresponding fault types. S1.

3. Perform detrending, filtering and normalization processing on the collected operating data in sequence, and extract the fault characteristic parameters in the processed operating data; S1.

4. Label the fault feature parameters according to the fault source and fault type, label the fault source, fault type and fault feature mapping relationship corresponding to each sample, and construct the motor-electronic controller coupling fault feature library.

3. The method of claim 1, wherein, Step S2 specifically includes the following sub-steps: S2.

1. Divide the sample set of the motor-electromechanical controller coupling fault feature library according to a set ratio to generate a training set, a validation set and a test set, and set the sample coverage and usage rules of the training set, validation set and test set. S2.

2. Construct a fault tracing and diagnosis model. The fault tracing and diagnosis model includes a convolutional neural network layer, a Transformer encoder layer, and two-level output layers. The output of the convolutional neural network layer is connected to the input of the Transformer encoder layer, and the output of the Transformer encoder layer is connected to the input of the two-level output layers. S2.

3. Input the training set into the fault tracing and diagnosis model, use the cross-entropy loss function to calculate the output error of the fault tracing and diagnosis model, optimize the parameters of the fault tracing and diagnosis model based on the output error, and introduce a dropout layer; S2.

4. Verify the recognition accuracy of the fault tracing and diagnosis model using the validation set and test set, and iterate the training process until the recognition accuracy of the fault tracing and diagnosis model reaches the set value.

4. The method of claim 1, wherein, Step S3 specifically includes the following sub-steps: S3.

1. Deploy the airborne data acquisition unit, set the acquisition frequency and synchronous acquisition rules of the airborne data acquisition unit, and synchronously acquire real-time operating data from the motor side and the ESC side through the airborne data acquisition unit; S3.

2. Perform sliding window filtering on the collected real-time running data, calculate the characteristic parameters corresponding to the filtered real-time running data, and complete the preprocessing of the real-time running data; S3.

3. Simultaneously input the preprocessed real-time running data and corresponding feature parameters into the trained fault tracing and diagnosis model; S3.

4. First, determine the source of the fault through the fault tracing and diagnosis model, then output the specific fault type, fault severity and related operating parameters, and integrate the output to generate the fault tracing and diagnosis results.

5. The method of claim 1, wherein, Step S4 specifically includes the following sub-steps: S4.

1. Preset fault handling rule base. The fault handling rule base contains a one-to-one correspondence between fault source, fault type, fault severity and fault handling suggestions. Set the handling actions and execution logic corresponding to different fault combinations. S4.

2. The obtained fault tracing and diagnosis results are broken down into three dimensions: fault source, fault type, and fault severity. These are then matched against the corresponding content in the fault handling rule base to obtain corresponding fault handling suggestions. S4.

3. Standardize the format of the fault handling suggestions obtained from the matching, and transmit the standardized fault handling suggestions synchronously to the flight control system and the ground operation and maintenance platform through the airborne communication module.

6. The method of claim 2, wherein, In step S1.1, motor-side faults include inter-turn short circuits in windings, bearing wear, rotor eccentricity, and abnormal back EMF; ESC-side faults include open circuits in power devices, short circuits in power devices, aging of DC bus capacitors, distortion of drive signals, and abnormal dead time. Based on the physical location of the fault and the source of the signal, the boundary between motor-side faults and ESC-side faults is defined, the location of the original signal corresponding to each type of fault is clarified, and the original fault signal is distinguished from the abnormal signal generated by coupling and transmission. The fault types are then standardized and defined.

7. The method of claim 3, wherein, In step S2.2, the convolutional neural network layer extracts the frequency domain features of the input data, the Transformer encoder layer captures the temporal correlation features of the input features, the first-level output of the two-level output layer outputs the fault source, and the second-level output of the two-level output layer outputs the fault type and fault severity. The frequency domain features output by the convolutional neural network layer are input into the Transformer encoder layer, and the temporal correlation features output by the Transformer encoder layer are input into the two-level output layer for hierarchical transfer and processing of features, and hierarchical output of fault source and fault type.

8. The method of claim 4, wherein, In step S3.1, the real-time operating data on the motor side includes three-phase stator current, back electromotive force, bearing temperature, and rotor speed. The real-time operating data on the electronic speed controller side includes DC bus voltage, switching device case temperature, drive pulse duty cycle, dead time, and output voltage. The real-time operating data on the motor side and the electronic speed controller side are synchronously collected with the same timestamp according to the set acquisition frequency. The collected real-time operating data is transmitted to the airborne diagnostic unit via Ethernet for real-time operating data acquisition and transmission.

9. The method according to claim 5, characterized in that, In step S4.2, based on the three dimensions of fault source, fault type and fault severity, the corresponding fault handling suggestions in the fault handling rule base are matched. The fault handling suggestions include load reduction operation, emergency shutdown, switching redundant equipment and operation and maintenance investigation. For motor-side faults, ESC-side faults, and coupling faults, corresponding hierarchical processing actions are matched, and the execution conditions and execution order of each processing action are set to match and generate fault handling suggestions.

10. The method according to claim 1, characterized in that, In step S3, after the preprocessed real-time running data is input into the trained fault tracing and diagnosis model, the inference delay of the fault tracing and diagnosis model is controlled within the set value. The fault tracing and diagnosis results output by the fault tracing and diagnosis model include the fault occurrence time, fault duration, fault source, fault type, fault severity and related operating parameters, and perform fault tracing and output of full-dimensional information.

Images