A method for diagnosing abnormal flight status of an electric vertical take-off and landing aircraft

Abstract

A method for diagnosing abnormal flight status of an electric vertical take-off and landing aircraft belongs to the technical field of flight management of electric vertical take-off and landing aircraft. In order to solve the detection and processing of abnormal status of an electric vertical take-off and landing aircraft, the present invention includes constructing flight monitoring indicators of an electric vertical take-off and landing aircraft; constructing influencing factors of flight monitoring indicators of an electric vertical take-off and landing aircraft including ambient temperature, wind speed, wind direction, load, flight time, battery voltage, and battery current; constructing a diagnostic data set for abnormal flight status of an electric vertical take-off and landing aircraft, including a training set, a validation set, and a test set; constructing a prediction model for flight monitoring indicators of an electric vertical take-off and landing aircraft based on an improved LSTM neural network; training the prediction model for flight monitoring indicators of an electric vertical take-off and landing aircraft using the training set to obtain a trained prediction model for flight monitoring indicators of an electric vertical take-off and landing aircraft; and diagnosing abnormal flight status of an electric vertical take-off and landing aircraft based on the prediction data.

Description

Technical Field

[0001] The invention belongs to the technical field of flight management of electric vertical take-off and landing aircraft, and in particular relates to a method for diagnosing abnormal flight status of an electric vertical take-off and landing aircraft. Background Art

[0002] eVTOL aircraft (electric vertical take-off and landing aircraft) have many significant features and advantages. First, they can take off and land vertically in a very small take-off and landing space, making it possible to operate in cities and crowded areas, greatly saving the space required for traditional aircraft that require long runways. Secondly, due to the use of electric propulsion systems, eVTOL aircraft are more environmentally friendly than traditional internal combustion engine aircraft, with lower emissions and significantly reduced noise, which helps to reduce the impact on the environment and improve the quality of life of residents. In addition, eVTOL aircraft have relatively low maintenance costs, flexible operations, and can quickly respond to short-distance transportation needs in cities. These characteristics make eVTOL an ideal choice for future fields such as urban air traffic and rapid rescue.

[0003] During the flight of an eVTOL aircraft, it is crucial to monitor the battery charge, motor temperature, and speed. Monitoring the battery charge ensures that the aircraft can return or land safely before the power is exhausted, and also helps maintain the battery and extend its service life. Real-time monitoring of the motor temperature can prevent equipment damage or performance degradation caused by motor overheating, while providing important data about the health of the motor. Accurate monitoring of the speed directly affects the lift and thrust of the aircraft, and is crucial for flight control and energy consumption optimization. Through continuous monitoring of this data, the safety and efficiency of flight can be greatly improved.

[0004] During the flight of eVTOL (electric vertical take-off and landing vehicle), the lack of abnormal state detection of battery power, motor temperature and speed will face a series of serious problems. First of all, battery power is the basic energy source for eVTOL flight. Abnormal power may cause the aircraft to fail to complete the scheduled flight mission, or even suddenly lose power in the air, causing a flight safety accident. Secondly, the motor is the core component that drives the eVTOL flight. The abnormal increase in motor temperature may indicate an overload or failure. If it is not detected and handled in time, it may cause motor damage, and in severe cases, it may cause a fire. At the same time, abnormal changes in motor speed are directly related to the control stability and operational responsiveness of the aircraft. Too high or too low speed may cause flight control difficulties and affect flight safety. The lack of effective abnormality detection methods not only increases the safety risks during the flight, but also may increase the maintenance cost of the aircraft and reduce flight efficiency. In addition, this will also affect the trust of passengers and operators and have a negative impact on commercial operations.

[0005] In the invention patent with application number 201810790066.3 and invention name “A method and system for predicting the flight state of a UAV based on LSTM”, a method and system for predicting the flight state based on LSTM is proposed to solve the problem that the existing UAV control model cannot predict the flight action based on the flight state data. By constructing an action label dictionary, collecting and preprocessing data, and using an improved LSTM model for training and verification, this method can accurately predict the flight action of the UAV and improve the accuracy and safety of the UAV operation. However, the above invention is dedicated to action prediction under normal operation conditions and has not yet involved abnormal state detection or processing. Summary of the invention

[0006] The problem to be solved by the present invention is to detect and process the abnormal state of an electric vertical take-off and landing aircraft, and a method for diagnosing the abnormal flight state of an electric vertical take-off and landing aircraft is proposed.

[0007] To achieve the above object, the present invention is implemented through the following technical solutions:

[0008] A method for diagnosing abnormal flight status of an electric vertical take-off and landing aircraft comprises the following steps:

[0009] S1. Construct the flight monitoring indicators of electric vertical take-off and landing aircraft as battery power, motor temperature and speed;

[0010] S2. Construct the influencing factors of the flight monitoring indicators of electric vertical take-off and landing aircraft, including ambient temperature, wind speed, wind direction, load, flight time, battery voltage, and battery current;

[0011] S3. Collect the flight monitoring index data of the electric vertical take-off and landing aircraft and the influencing factor data of the flight monitoring index of the electric vertical take-off and landing aircraft at the same time, and construct the abnormal flight state diagnosis data set of the electric vertical take-off and landing aircraft, including a training set, a validation set, and a test set;

[0012] S4. Constructing a flight monitoring index prediction model for electric vertical take-off and landing aircraft based on an improved LSTM neural network;

[0013] S5. Using the training set obtained in step S3, the electric vertical take-off and landing aircraft flight monitoring index prediction model is trained to obtain a trained electric vertical take-off and landing aircraft flight monitoring index prediction model;

[0014] S6. Using the trained electric vertical take-off and landing aircraft flight monitoring index prediction model obtained in step S5, the electric vertical take-off and landing aircraft flight monitoring index is predicted, and the abnormal flight state of the electric vertical take-off and landing aircraft is diagnosed based on the predicted data.

[0015] Furthermore, in step S1, a coulomb meter equipped on the eVTOL aircraft is used to measure the battery charge, an infrared sensor is used to measure the motor temperature, and an optical encoder is used to detect the rotation speed.

[0016] Furthermore, the specific implementation method of step S3 includes the following steps:

[0017] S3.1. Collect the flight monitoring index data of the electric vertical take-off and landing aircraft and the influencing factor data of the flight monitoring index of the electric vertical take-off and landing aircraft at the same time;

[0018] Collecting drones during flight The battery power corresponding to the time is , the corresponding motor temperatures are The corresponding motor speeds are ;

[0019] Using infrared thermal imager to collect data during the flight of drone The corresponding ambient temperatures at the time are ;

[0020] Hot wire anemometer is used to collect the anemometer data of the drone during flight. The wind speeds at the time are ;

[0021] Ultrasonic wind vane is used to collect the wind speed of UAV during flight. The wind directions corresponding to the time are: ;

[0022] The piezoelectric weighing sensor is used to collect the weight of the drone during flight. The loads corresponding to the time are ;

[0023] The GPS module is used to collect data during the flight of the drone. The flight time corresponding to the time is ;

[0024] The voltage sensor is used to collect the voltage of the drone during flight. The battery voltages corresponding to the time are ;

[0025] The current sensor is used to collect the current of the drone during flight. The battery currents corresponding to the time are ;

[0026] S3.2. The result obtained in step S3.1 t 1~ tnThe data at each moment is divided into multiple continuous time series, the length of each time series is j, and the number of divided time series is K; that is, the time corresponding to the first time series is: t1~tj; the time corresponding to the second time series is: t2~t(j+1); the time corresponding to the third time series is: t3~t(j+2); ...; the time corresponding to the Kth time series is: tK~t(j+K-1);

[0027] Divide all K time series data into training set, validation set, and test set in a ratio of 8:1:1.

[0028] Furthermore, the specific implementation method of step S4 includes the following steps:

[0029] S4.1. Set the activation value calculation expression of the improved forget gate in the improved LSTM neural network as:

[0030] ;

[0031] in, is the activation value of the forget gate, is the sigmoid activation function, is layer normalization, is Dropout regularization, is the weight of the forget gate, is the final output of the previous time step, is the input of the current time step, For the connection operation, Forgotten Gate The weight of is the cell state at the previous time step, is the bias of the forget gate;

[0032] S4.2. In the improved LSTM neural network, the activation value calculation method of the improved input gate is as follows:

[0033] ;

[0034] in, is the activation value of the input gate, is the weight of the input gate, For the input gate The weight of is the bias of the input gate;

[0035] S4.3. In the improved LSTM neural network, the improved cell candidate value calculation method is as follows:

[0036] ;

[0037] in, is the cell candidate value, is the hyperbolic tangent activation function, is the weight of the cell candidate value, For cell candidate values The weight of is the bias of the cell candidate value;

[0038] S4.4. In the improved LSTM neural network, the calculation method of the improved cell state update is as follows:

[0039] ;

[0040] in, is the attention weight, is the softmax activation function, is the coefficient of attention weight, is the attention weight The weight of is the bias of the attention weight;

[0041] ;

[0042] in, is the cell state; is the dot product operation;

[0043] S4.5. In the improved LSTM neural network, the improved output gate and final output calculation method are as follows:

[0044] ;

[0045] in, is the activation value of the output gate, is the weight of the output gate, is the weight of the output gate, is the bias of the output gate;

[0046] ;

[0047] in, is the final output value;

[0048] S4.6. Update the improved LSTM model calculation formula to the algorithm corresponding to the LSTM model and deploy it to the computer to complete the construction of the flight monitoring index prediction model of the electric vertical take-off and landing aircraft.

[0049] Furthermore, the specific implementation method of step S5 includes the following steps:

[0050] S5.1. The training set is input into the flight monitoring index prediction model of the electric vertical take-off and landing aircraft for training. During the training process, the loss function adopts the mean square error MSE, and the calculation method is as follows:

[0051] ;

[0052] in, is the sample size, is the true value, is the predicted value;

[0053] S5.2. Set the mean square error corresponding to battery power, motor temperature, and motor speed to be recorded as , , , then the total loss function for:

[0054] ;

[0055] in, is the weight coefficient corresponding to the mean square error of battery power, is the weight coefficient corresponding to the mean square error of motor temperature, is the weight coefficient corresponding to the mean square error of the motor speed;

[0056] S5.3. Error correction is achieved through back propagation and stochastic gradient descent (SGD) methods. When the loss function value tends to be stable or reaches the preset training rounds, the training ends. The training of the flight monitoring indicator prediction model of the electric vertical take-off and landing aircraft is completed, which is used to predict the battery power, motor temperature, and motor speed of the eVTOL aircraft during flight.

[0057] Furthermore, the specific implementation method of step S6 includes the following steps:

[0058] S6.1. Use the trained electric vertical take-off and landing aircraft flight monitoring index prediction model obtained in step S5 to predict the electric vertical take-off and landing aircraft flight monitoring index, and predict the first Predicted battery level at the moment , No. Predicted motor temperature at time , No. Predicted motor speed at time ;

[0059] S6.2. For the first Predicted battery level at the moment , the coulomb meter is used to monitor the actual The actual battery charge at this moment is ;

[0060] Compare and ,when When The allowable battery charge error determined by the experts indicates that the battery charge value predicted by the model is consistent with the actual monitored value;

[0061] when When tp +1 predicted battery level and tp +1 Actual battery level at the moment For further comparison, when When , it is considered that the deviation between the predicted value and the actual monitored value of the battery power is caused by the sensor deviation, and there is no need to process the eVTOL aircraft equipment;

[0062] when When tp +2 predicted battery level and tp +2 actual battery level For further comparison, when When , it is considered that the deviation between the predicted value and the actual monitored value of the battery power is caused by the sensor deviation, and there is no need to process the eVTOL aircraft equipment; and when When , it indicates that the battery power value predicted by the model is significantly different from the actual monitored value. It is believed that the deviation between the predicted battery power and the actual monitored value is a problem with the battery itself, and the battery needs to be inspected and repaired;

[0063] S6.3. For the first Predicted motor temperature at time , the actual monitoring result obtained by infrared sensor The actual motor temperature at the time is ;

[0064] Compare and ,when When The allowable motor temperature error determined by the experts indicates that the motor temperature predicted by the model is consistent with the actual monitored value;

[0065] when When tp +1 predicted motor temperature and tp +1 Actual motor temperature at time For further comparison, when When , it is considered that the deviation between the motor temperature prediction value and the monitoring value is caused by the sensor deviation, and there is no need to process the eVTOL aircraft equipment; and when When , it indicates that the motor temperature value predicted by the model is significantly different from the actual monitored value. It is believed that the deviation between the motor temperature prediction value and the actual monitored value is a problem with the motor itself, and the motor needs to be inspected and repaired;

[0066] S6.4. For the first Predicted motor speed at time ; The actual monitoring result obtained by using the optical encoder The actual motor speed at the moment is ;

[0067] Compare and ,when When The allowable motor speed error determined by the experts indicates that the motor speed value predicted by the model is consistent with the actual monitored value;

[0068] when When , it indicates that the motor speed value predicted by the model is significantly different from the actual monitored value. At this time, the UAV motor needs to be inspected and repaired immediately;

[0069] S6.5. When the battery power, motor temperature, and motor speed mentioned above are abnormal at the same time, the motor speed problem needs to be dealt with first, followed by the motor temperature problem, and finally the battery power problem.

[0070] Beneficial effects of the present invention:

[0071] The method for diagnosing abnormal flight status of an electric vertical take-off and landing aircraft described in the present invention takes into account the problems of large data fluctuations, data time complexity, important feature mining in data, overfitting, etc., and proposes an improved LSTM prediction model to improve the accuracy of predictions for battery power, motor temperature, and motor speed. It is of great value in ensuring the flight safety of eVTOL aircraft, optimizing motor performance management, extending battery life, improving energy efficiency, and reducing maintenance costs, making energy utilization more efficient and economical.

[0072] The electric vertical take-off and landing aircraft flight abnormal state diagnosis method described in the present invention can timely discover potential eVTOL aircraft equipment failures or performance degradation by accurately identifying data anomalies, so as to take preventive measures or perform necessary maintenance. This not only helps to ensure the safe operation of eVTOL aircraft equipment and reduce accident risks, but also effectively prolongs the service life of the equipment, reduces operating costs, and ensures production efficiency and quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] Figure 1 The present invention provides a flow chart of a method for diagnosing abnormal flight status of an electric vertical take-off and landing aircraft. DETAILED DESCRIPTION

[0074] In order to make the purpose, technical solution and advantages of the present invention more clear, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention, that is, the specific embodiments described are only part of the embodiments of the present invention, rather than all of the specific embodiments. The components of the specific embodiments of the present invention described and shown in the drawings herein can be arranged and designed in various different configurations, and the present invention can also have other embodiments.

[0075] Therefore, the following detailed description of the specific embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents the selected specific embodiments of the present invention. Based on the specific embodiments of the present invention, all other specific embodiments obtained by those skilled in the art without making creative work are within the scope of protection of the present invention.

[0076] In order to further understand the content, features and effects of the present invention, the following specific implementation methods are given as examples, and the attached Figure 1 The detailed instructions are as follows:

[0077] Embodiment 1:

[0078] A method for diagnosing abnormal flight status of an electric vertical take-off and landing aircraft comprises the following steps:

[0079] S1. Construct the flight monitoring indicators of electric vertical take-off and landing aircraft as battery power, motor temperature and speed;

[0080] Furthermore, in step S1, a coulomb meter equipped on the eVTOL aircraft is used to measure the battery charge, an infrared sensor is used to measure the motor temperature, and an optical encoder is used to detect the rotation speed;

[0081] S2. Construct the influencing factors of the flight monitoring indicators of electric vertical take-off and landing aircraft, including ambient temperature, wind speed, wind direction, load, flight time, battery voltage, and battery current;

[0082] Furthermore, when the ambient temperature is low, the battery chemical reaction rate will be reduced, the internal resistance will increase, the available power will be reduced, and the power consumption will be accelerated; low temperature makes the motor take longer to reach the optimal operating temperature. At the same time, the friction between the mechanical parts of the motor increases, and the motor speed needs to be increased to overcome the additional resistance.

[0083] When the wind speed is high and in the direction of the tailwind, the battery power consumption can be reduced. At the same time, the cooling effect of the motor can be increased. The motor needs to reduce its speed to maintain the current flight speed.

[0084] The heavier the load of the eVTOL aircraft, the greater the battery consumption, the higher the motor temperature, and the faster the motor speed. A piezoelectric load cell is used to measure the load weight of the aircraft.

[0085] The longer the eVTOL aircraft flies, the greater the power consumption, the higher the motor temperature will be, and the motor efficiency may decrease slightly, requiring the speed to be increased to maintain the same performance.

[0086] There is a nonlinear relationship between voltage and power. At the same time, when the voltage decreases, the motor temperature will increase while maintaining the same power; and when the voltage decreases, the maximum achievable speed of the motor will decrease. Use a voltage sensor to measure the battery terminal voltage.

[0087] Current directly determines power output and energy consumption. An increase in current will lead to an increase in power consumption, an increase in motor temperature, and an increase in motor speed. A current sensor is used to measure the current in the circuit.

[0088] S3. Collect the flight monitoring index data of the electric vertical take-off and landing aircraft and the influencing factor data of the flight monitoring index of the electric vertical take-off and landing aircraft at the same time, and construct the abnormal flight state diagnosis data set of the electric vertical take-off and landing aircraft, including a training set, a validation set, and a test set;

[0089] Furthermore, the specific implementation method of step S3 includes the following steps:

[0090] S3.1. Collect the flight monitoring index data of the electric vertical take-off and landing aircraft and the influencing factor data of the flight monitoring index of the electric vertical take-off and landing aircraft at the same time;

[0091] Collecting drones during flight The battery power corresponding to the time is , the corresponding motor temperatures are The corresponding motor speeds are ;

[0092] Using infrared thermal imager to collect data during the flight of drone The corresponding ambient temperatures at the time are ;

[0093] Hot wire anemometer is used to collect the anemometer data of the drone during flight. The wind speeds at the time are ;

[0094] Ultrasonic wind vane is used to collect the wind speed of UAV during flight. The wind directions corresponding to the time are: ;

[0095] The piezoelectric weighing sensor is used to collect the weight of the drone during flight. The loads corresponding to the time are ;

[0096] The GPS module is used to collect data during the flight of the drone. The flight time corresponding to the time is ;

[0097] The voltage sensor is used to collect the voltage of the drone during flight. The battery voltages corresponding to the time are ;

[0098] The current sensor is used to collect the current of the drone during flight. The battery currents corresponding to the time are ;

[0099] S3.2. The result obtained in step S3.1 t 1~ tn The data at each moment is divided into multiple continuous time series, the length of each time series is j, and the number of divided time series is K; that is, the time corresponding to the first time series is: t1~tj; the time corresponding to the second time series is: t2~t(j+1); the time corresponding to the third time series is: t3~t(j+2); ...; the time corresponding to the Kth time series is: tK~t(j+K-1);

[0100] Divide all K time series data into training set, validation set, and test set in a ratio of 8:1:1.

[0101] There are two problems in building a prediction model: (1) It is difficult to capture the long-term dependencies between data in traditional models, that is, it is difficult for the model to remember and use the battery power, motor temperature, and motor speed at earlier times to predict the battery power, motor temperature, and motor speed at future times. (2) When predicting the battery power, motor temperature, and motor speed, not all information at earlier times is equally important, and it is necessary to distinguish the importance of the information at existing times.

[0102] To address the above issues, scholars proposed the LSTM (Long Short-Term Memory) model, which introduced a gating mechanism, including input gate, forget gate, and output gate. By controlling the long-term preservation and forgetting of information, it effectively solved the long-term problem of data. At the same time, the LSTM gating mechanism allows the model to learn the importance of information at different time nodes.

[0103] However, there are still the following problems in LSTM: (1) When training LSTM, the performance of the model may be affected due to the large fluctuations in the data in the time series, especially when the data is unevenly distributed. (2) Traditional LSTM models may not be accurate enough when processing data with complex time dependencies, such as motor speed or battery charge changes, because this method usually only relies on short-term input and hidden states. (3) When processing long sequence data, traditional LSTM may ignore information that is particularly important for prediction. This method is usually used to process information of the entire sequence. (4) When training complex LSTM networks, the model will overfit the training data, resulting in poor performance in practical applications.

[0104] The above problems will make it difficult to accurately predict the battery power, motor temperature, and motor speed based on monitoring data during the flight of the eVTOL aircraft.

[0105] The LSTM model has been improved. The details are as follows:

[0106] (1) Introducing layer normalization technology. Layer normalization can stabilize the learning process of LSTM and improve the model's adaptability under different data conditions, which is very important for drones flying in different environments and load conditions. This technology can help the model better adapt to environmental changes, such as the impact of temperature changes on motor temperature, or the impact of load changes on battery consumption and motor performance.

[0107] (2) Introducing peak connection technology. Peak connection allows the model to directly use long-term memory to adjust the opening and closing of the gate, which is very useful for predicting long-term battery consumption patterns or the temperature and speed of a motor under continuous operation.

[0108] (3) Introducing the attention mechanism. The attention mechanism can help the model identify the flight phases that are most critical to battery consumption, as well as specific operating conditions that affect motor temperature and speed. For example, the model can learn the fastest battery consumption patterns during takeoff and landing, as well as the changes in motor temperature and speed during high-speed flight.

[0109] (4) Introducing Dropout regularization technology. In the prediction model of the drone’s battery power, motor temperature, and speed, using Dropout can increase the generalization ability of the model and reduce the prediction error under new or unseen flight conditions. This helps maintain the stability and accuracy of the prediction when the drone faces different environmental conditions or flight modes.

[0110] The above four technologies are used to improve the forget gate, input gate, cell candidate value, cell state update value, output gate, and final output in the traditional LSTM.

[0111] The cell state is the core part of the network memory, which transmits information through a series of time steps. The cell state can be thought of as a flowing "information highway" that carries all the information that may be important for future decisions. In LSTM, this cell state can be transmitted in an almost unchanged form between different time steps, with only slight adjustments made through gate control.

[0112] S4. Constructing a flight monitoring index prediction model for electric vertical take-off and landing aircraft based on an improved LSTM neural network;

[0113] Furthermore, the specific implementation method of step S4 includes the following steps:

[0114] S4.1. Set the activation value calculation expression of the improved forget gate in the improved LSTM neural network as:

[0115] ;

[0116] in, is the activation value of the forget gate, is the sigmoid activation function, is layer normalization, is Dropout regularization, is the weight of the forget gate, is the final output of the previous time step, is the input of the current time step, For the connection operation, Forgotten Gate The weight of is the cell state at the previous time step, is the bias of the forget gate;

[0117] Furthermore, the forget gate determines the information to be discarded from the cell state. It calculates a value between 0 and 1 through a sigmoid activation function, which is multiplied by the cell state element by element, thereby achieving "forgetfulness" of the state.

[0118] For battery charge prediction during the flight of eVTOL aircraft, if the battery discharge rate does not change much under certain conditions, the forget gate can help the model ignore these unimportant changes; for motor temperature prediction, the forget gate can help the model discard historical temperature information that is irrelevant to the current temperature prediction, such as abnormal temperature fluctuations in the past; for motor speed prediction, when the speed changes rapidly, the forget gate can make the model ignore outdated speed information and focus on more recent data.

[0119] S4.2. In the improved LSTM neural network, the activation value calculation method of the improved input gate is as follows:

[0120] ;

[0121] in, is the activation value of the input gate, is the weight of the input gate, For the input gate The weight of is the bias of the input gate;

[0122] Furthermore, the input gate controls the importance of the new information and determines the new information that will be added to the cell state. The cell candidate value is a transformed form of the new information that may be added to the cell state.

[0123] During the flight of the eVTOL aircraft, for battery power prediction, the input gate controls the importance of new power input information (such as charging status, usage time, etc.), and the cell candidate value updates the power status based on this information; for motor temperature prediction, new environmental or machine operation information updates the temperature prediction model through the input gate and cell candidate value; for motor speed prediction, the input gate and cell candidate value help the model integrate the latest operating commands and environmental factors to adjust the speed prediction.

[0124] S4.3. In the improved LSTM neural network, the improved cell candidate value calculation method is as follows:

[0125] ;

[0126] in, is the cell candidate value, is the hyperbolic tangent activation function, is the weight of the cell candidate value, For cell candidate values The weight of is the bias of the cell candidate value;

[0127] S4.4. In the improved LSTM neural network, the calculation method of the improved cell state update is as follows:

[0128] ;

[0129] in, is the attention weight, is the softmax activation function, is the coefficient of attention weight, is the attention weight The weight of is the bias of the attention weight;

[0130] ;

[0131] in, is the cell state; is the dot product operation;

[0132] Furthermore, the cell state is where LSTM stores past information, and it is updated at each time step to maintain long-term dependent information.

[0133] During the flight of the eVTOL aircraft, when predicting the battery power, motor temperature and motor speed, the update of the cell state allows the model to retain and forget information, ensuring that only useful information is retained, thereby improving the accuracy of the prediction.

[0134] S4.5. In the improved LSTM neural network, the improved output gate and final output calculation method are as follows:

[0135] ;

[0136] in, is the activation value of the output gate, is the weight of the output gate, is the weight of the output gate, is the bias of the output gate;

[0137] ;

[0138] in, is the final output value;

[0139] The output gate controls the amount of information output from the LSTM unit. The final output is based on the current cell state and the output of the output gate, which is the output passed to the next time step and the final output.

[0140] For battery power prediction during the flight of eVTOL aircraft, the output gate and final output help determine the power output at each step, providing data support for battery management; for motor temperature prediction, by adjusting the output gate and final output, the model can accurately predict the instantaneous temperature of the motor to help prevent overheating; for motor speed prediction, the update of the output gate and final output enables the model to reflect current and possible future speed changes and optimize machine performance.

[0141] S4.6. Update the improved LSTM model calculation formula to the algorithm corresponding to the LSTM model and deploy it to the computer to complete the construction of the flight monitoring index prediction model of the electric vertical take-off and landing aircraft.

[0142] S5. Using the training set obtained in step S3, the electric vertical take-off and landing aircraft flight monitoring index prediction model is trained to obtain a trained electric vertical take-off and landing aircraft flight monitoring index prediction model;

[0143] Furthermore, the specific implementation method of step S5 includes the following steps:

[0144] S5.1. The training set is input into the flight monitoring index prediction model of the electric vertical take-off and landing aircraft for training. During the training process, the loss function adopts the mean square error MSE, and the calculation method is as follows:

[0145] ;

[0146] in, is the sample size, is the true value, is the predicted value;

[0147] S5.2. Set the mean square error corresponding to battery power, motor temperature, and motor speed to be recorded as , , , then the total loss function for:

[0148] ;

[0149] in, is the weight coefficient corresponding to the mean square error of battery power, is the weight coefficient corresponding to the mean square error of motor temperature, is the weight coefficient corresponding to the mean square error of the motor speed;

[0150] S5.3. Error correction is achieved through back propagation and stochastic gradient descent (SGD) methods. When the loss function value tends to be stable or reaches the preset training rounds, the training ends. The training of the flight monitoring index prediction model of the electric vertical take-off and landing aircraft is completed, which is used to predict the battery power, motor temperature, and motor speed of the eVTOL aircraft during flight.

[0151] S6. Using the trained electric vertical take-off and landing aircraft flight monitoring index prediction model obtained in step S5, the electric vertical take-off and landing aircraft flight monitoring index is predicted, and the abnormal flight state of the electric vertical take-off and landing aircraft is diagnosed based on the predicted data.

[0152] Furthermore, the specific implementation method of step S6 includes the following steps:

[0153] S6.1. Use the trained electric vertical take-off and landing aircraft flight monitoring index prediction model obtained in step S5 to predict the electric vertical take-off and landing aircraft flight monitoring index, and predict the first Predicted battery level at the moment , No. Predicted motor temperature at time , No. Predicted motor speed at time ;

[0154] S6.2. For the first Predicted battery level at the moment , the coulomb meter is used to monitor the actual The actual battery charge at this moment is ;

[0155] Compare and ,when When The allowable battery charge error determined by the experts indicates that the battery charge value predicted by the model is consistent with the actual monitored value;

[0156] when When tp +1 predicted battery level and tp +1 Actual battery level at the moment For further comparison, when When , it is considered that the deviation between the predicted value and the actual monitored value of the battery power is caused by the sensor deviation, and there is no need to process the eVTOL aircraft equipment;

[0157] when When tp +2 predicted battery level and tp +2 The actual battery level at the moment For further comparison, when When , it is considered that the deviation between the predicted value and the actual monitored value of the battery power is caused by the sensor deviation, and no processing is required for the eVTOL aircraft equipment; and when When , it indicates that the battery power value predicted by the model is significantly different from the actual monitored value. It is believed that the deviation between the predicted battery power and the actual monitored value is a problem with the battery itself, and the battery needs to be inspected and repaired;

[0158] S6.3. For the first Predicted motor temperature at time , the actual monitoring result obtained by infrared sensor The actual motor temperature at the time is ;

[0159] Compare and ,when When The allowable motor temperature error determined by the experts indicates that the motor temperature predicted by the model is consistent with the actual monitored value;

[0160] when When tp +1 predicted motor temperature and tp +1 Actual motor temperature at time For further comparison, when When , it is considered that the deviation between the motor temperature prediction value and the monitoring value is caused by the sensor deviation, and there is no need to process the eVTOL aircraft equipment; and when When , it indicates that the motor temperature value predicted by the model is significantly different from the actual monitored value. It is believed that the deviation between the motor temperature prediction value and the actual monitored value is a problem with the motor itself, and the motor needs to be inspected and repaired;

[0161] S6.4. For the first Predicted motor speed at time ; The actual monitoring result obtained by using the optical encoder The actual motor speed at the moment is ;

[0162] Compare and ,when When The allowable motor speed error determined by the experts indicates that the motor speed value predicted by the model is consistent with the actual monitored value;

[0163] when When , it indicates that the motor speed value predicted by the model is significantly different from the actual monitored value. At this time, the UAV motor needs to be inspected and repaired immediately;

[0164] S6.5. When the battery power, motor temperature, and motor speed mentioned above are abnormal at the same time, the motor speed problem needs to be dealt with first, followed by the motor temperature problem, and finally the battery power problem.

[0165] The key points and intended protection points of the present invention are:

[0166] (1) A prediction model for battery charge, motor temperature, and motor speed.

[0167] (2) A method for determining abnormal data of battery power, motor temperature, and motor speed.

[0168] It should be noted that relational terms such as "first" and "second" are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0169] Although the present application has been described above with reference to specific embodiments, various modifications may be made thereto and parts thereof may be replaced with equivalents without departing from the scope of the present application. In particular, as long as there is no structural conflict, the various features in the specific embodiments disclosed in the present application may be used in combination with each other in any manner, and the fact that these combinations are not exhaustively described in this specification is only for the sake of omitting space and saving resources. Therefore, the present application is not limited to the specific embodiments disclosed herein, but includes all technical solutions falling within the scope of the claims.

Claims

1. A method for diagnosing abnormal flight status of an electric vertical take-off and landing aircraft, characterized in that: The steps include: S1. Constructing flight monitoring indicators of electric vertical take-off and landing aircraft as battery power, motor temperature and speed; S2. The factors affecting the flight monitoring indicators of electric vertical take-off and landing aircraft include ambient temperature, wind speed, wind direction, load, flight time, battery voltage, and battery current; S3. Collect the flight monitoring index data of the electric vertical take-off and landing aircraft and the influencing factor data of the flight monitoring index of the electric vertical take-off and landing aircraft at the same time, and construct the abnormal flight state diagnosis data set of the electric vertical take-off and landing aircraft, including a training set, a validation set, and a test set; S4. Construct a flight monitoring index prediction model for electric vertical take-off and landing aircraft based on the improved LSTM neural network; The specific implementation method of step S4 includes the following steps: S4.

1. Set the activation value calculation expression of the improved forget gate in the improved LSTM neural network as: f t =σ(LayerNorm(Dropout(W f )·[h t-1 ,x t ]+P f ·C t-1 +b f )) Among them, f t is the activation value of the forget gate, σ is the sigmoid activation function, LayerNorm is the layer normalization, Dropout is the Dropout regularization, W f is the weight of the forget gate, h t-1 is the final output of the previous time step, x t is the input of the current time step, [h t-1 ,x t ] is the connection operation, P f For the Forgotten Gate t-1 The weight of C t-1 is the cell state at the previous time step, b f is the bias of the forget gate; S4.

2. In the improved LSTM neural network, the activation value calculation method of the improved input gate is as follows: i t =σ(LayerNorm(Dropout(W i )·[h t-1 ,x t ]+P i ·C t-1 +b i )) Among them, i t is the activation value of the input gate, W i is the weight of the input gate, P i C is the input gate t-1 The weight of b i is the bias of the input gate; S4.

3. In the improved LSTM neural network, the improved cell candidate value calculation method is as follows: in, is the cell candidate value, tanh is the hyperbolic tangent activation function, W C is the weight of the cell candidate value, P C C is the candidate value of the cell t-1 The weight of b C is the bias of the cell candidate value; S4.

4. In the improved LSTM neural network, the calculation method of the improved cell state update is as follows: a t =softmax(V a ·tanh(W a [h t-1 ,x t ]+b a )) Among them, a t is the attention weight, softmax is the softmax activation function, V a is the coefficient of attention weight, W a is the attention weight [h t-1 ,x t ] weight, b a is the bias of the attention weight; Among them, C t is the cell state; ⊙ is the dot product operation; S4.

5. In the improved LSTM neural network, the improved output gate and final output calculation method are as follows: o t =σ(LayerNorm(Dropout(W o )·[h t-1 ,x t ]+P o ·C t-1 +b o )) Among them, t is the activation value of the output gate, W o is the weight of the output gate, P o C in the output gate t-1 The weight of b o is the bias of the output gate; h t =o t *tanh(C t ) Among them, h t is the final output value; S4.

6. Update the improved LSTM model calculation formula to the algorithm corresponding to the LSTM model and deploy it to the computer to complete the construction of the flight monitoring index prediction model of the electric vertical take-off and landing aircraft; S5. Using the training set obtained in step S3 to train the electric vertical take-off and landing aircraft flight monitoring index prediction model, a trained electric vertical take-off and landing aircraft flight monitoring index prediction model is obtained; S6. Use the trained electric vertical take-off and landing aircraft flight monitoring index prediction model obtained in step S5 to predict the electric vertical take-off and landing aircraft flight monitoring index, and diagnose the abnormal flight state of the electric vertical take-off and landing aircraft based on the predicted data.

2. The method for diagnosing abnormal flight status of an electric vertical take-off and landing aircraft according to claim 1, characterized in that: In step S1, the coulomb meter equipped on the eVTOL aircraft is used to measure the battery power, the infrared sensor is used to measure the motor temperature, and the optical encoder is used to detect the rotation speed.

3. The method for diagnosing abnormal flight status of an electric vertical take-off and landing aircraft according to claim 2, characterized in that: The specific implementation method of step S3 includes the following steps: S3.

1. Collect the flight monitoring index data of the electric vertical take-off and landing aircraft and the influencing factor data of the flight monitoring index of the electric vertical take-off and landing aircraft at the same time; The battery power corresponding to the time t1, t2, ..., tn during the flight of the drone is a t1 、a t2 , …, a tn , the corresponding motor temperatures are b t1 , b t2 , …, b tn , the corresponding motor speed is c t1 、c t2 ,…,c tn ; The ambient temperatures at the time t1, t2, ..., tn collected by the infrared thermal imager during the flight of the UAV are D t1 , D t2 ,…,D tn ; The hot wire anemometer is used to collect the wind speeds at t1, t2, ..., tn during the flight of the UAV, which are E t1 、E t2 ,…,E tn ; The ultrasonic wind vane is used to collect the wind directions at t1, t2, ..., tn during the flight of the UAV, which are F t1 、F t2 ,…,F tn ; The piezoelectric weighing sensor is used to collect the load corresponding to the time t1, t2, ..., tn during the flight of the UAV, which is G t1 , G t2 , …, G tn ; The GPS module is used to collect the flight time corresponding to the moments t1, t2, ..., tn during the flight of the UAV, which are H t1 , H t2 ,…,H tn ; The voltage sensor is used to collect the battery voltage corresponding to the time t1, t2, ..., tn during the flight of the drone, which is J t1 , J t2 , …, J tn ; The current sensor is used to collect the battery current corresponding to the time t1, t2, ..., tn during the flight of the drone, which is I t1 ,I t2 ,…,I tn ; S3.

2. Divide the data at time t1 to tn obtained in step S3.1 into multiple continuous time series, each time series has a length of j, and the number of divided time series is K; that is, the time corresponding to the first time series is: t1 to tj; the time corresponding to the second time series is: t2 to t(j+1); the time corresponding to the third time series is: t3 to t(j+2); ...; the time corresponding to the Kth time series is: tK to t(j+K-1); Divide all K time series data into training set, validation set, and test set in a ratio of 8:1:

1.

4. The method for diagnosing abnormal flight status of an electric vertical take-off and landing aircraft according to claim 3, characterized in that: The specific implementation method of step S5 includes the following steps: S5.

1. Input the training set into the electric vertical take-off and landing aircraft flight monitoring index prediction model for training. During the training process, the loss function uses the mean square error MSE, and the calculation method is as follows: Among them, m is the number of samples, y tr is the true value, y pr is the predicted value; S5.

2. Set the mean square error corresponding to battery power, motor temperature, and motor speed to be MSE respectively a , MSE b , MSE c , then the total loss function L is: L=ω1MSE a +ω2MSE b +ω3MSE c Among them, ω1 is the weight coefficient corresponding to the mean square error of battery power, ω2 is the weight coefficient corresponding to the mean square error of motor temperature, and ω3 is the weight coefficient corresponding to the mean square error of motor speed; S5.

3. Error correction is achieved through back propagation and stochastic gradient descent SGD methods. When the loss function value tends to stabilize or reaches the preset training rounds, the training ends; the training of the flight monitoring index prediction model of the electric vertical take-off and landing aircraft is completed, which is used to predict the battery power, motor temperature, and motor speed of the eVTOL aircraft during flight.

5. The method for diagnosing abnormal flight status of an electric vertical take-off and landing aircraft according to claim 4, characterized in that: The specific implementation method of step S6 includes the following steps: S6.

1. Use the trained electric vertical take-off and landing aircraft flight monitoring index prediction model obtained in step S5 to predict the electric vertical take-off and landing aircraft flight monitoring index, and predict the battery predicted power a at the time tp pr-tp , the predicted motor temperature b at time tp pr-tp , the predicted motor speed c at time tp pr-tp ; S6.

2. The predicted battery power a at time tp obtained in step S6.1 pr-tp The actual battery charge at the time tp is a, which is monitored by a coulomb meter. tp ; Compare pr-tp with a tp , when |a pr-tp -a tp |≤a tr When a tr The allowable battery charge error determined by the experts indicates that the battery charge value predicted by the model is consistent with the actual monitored value; when|a pr-tp -a tp |>a tr When tp+1, the battery power a is predicted pr-(tp+1) and the actual battery charge a at time tp+1 tp+1 For further comparison, when |a pr-(tp+1) -a tp+1 |≤a tr When , it is considered that the deviation between the predicted value and the actual monitored value of the battery power is caused by the sensor deviation, and there is no need to process the eVTOL aircraft equipment; when|a pr-(tp+1) -a tp+1 |>a tr When the battery power a is predicted at the time tp+2 pr-(tp+2) and the actual battery charge a at time tp+2 tp+2 For further comparison, when |a pr-(tp+2) -a tp+2 |≤a tr When |a pr-(tp+2) -a tp+2 |>a tr When , it indicates that the battery power value predicted by the model is significantly different from the actual monitored value. It is believed that the deviation between the predicted battery power and the actual monitored value is a problem with the battery itself, and the battery needs to be inspected and repaired; S6.

3. The predicted motor temperature b at time tp obtained in step S6.1 pr-tp The actual motor temperature at time tp obtained by infrared sensor monitoring is b tp ; Compare b pr-tp With b tp , when |b pr-tp -b tp |≤b tr When b tr The allowable motor temperature error determined by the experts indicates that the motor temperature predicted by the model is consistent with the actual monitored value; When | pr-tp -b tp |>b tr When the motor temperature b is predicted at the time tp+1 pr-(tp+1) and the actual motor temperature b at time tp+1 tp+1 For further comparison, when |b pr-(tp+1) -b tp+1 |≤b tr When |b pr-(tp+1) -b tp+1 |>b tr When , it indicates that the motor temperature value predicted by the model is significantly different from the actual monitored value. It is believed that the deviation between the motor temperature prediction value and the actual monitored value is a problem with the motor itself, and the motor needs to be inspected and repaired; S6.

4. The predicted motor speed c at time tp obtained in step S6.1 pr-tp ; The actual motor speed at time tp obtained by actual monitoring using the optical encoder is c tp ; Compare pr-tp With c tp , when |c pr-tp -c tp |≤c tr When c tr The allowable motor speed error determined by the experts indicates that the motor speed value predicted by the model is consistent with the actual monitored value; When |c pr-tp -c tp |>c tr When , it indicates that the motor speed value predicted by the model is significantly different from the actual monitored value. At this time, the UAV motor needs to be inspected and repaired immediately; S6.

5. When the above-mentioned battery power, motor temperature, and motor speed are abnormal at the same time, the motor speed problem needs to be dealt with first, followed by the motor temperature problem, and finally, the battery power problem.

Images