Fault detection and diagnosis method based on unmanned aerial vehicle data

Abstract

The invention belongs to the technical field of unmanned aerial vehicle fault diagnosis, and particularly relates to a fault detection and diagnosis method based on unmanned aerial vehicle data, which comprises the following steps: S1, initializing a system model, defining a state space model 1 and a state space model 2, including no fault, fault 1 and fault 2, and analyzing the specific influence of each fault on the flight performance of an unmanned aerial vehicle; s2, constructing a Hankel matrix, collecting input and output data in normal flight and various fault states of the unmanned aerial vehicle, determining the length and QdF degree of the Hankel matrix, and respectively constructing input and output Hankel matrixes of normal flight and fault flight; s3, dissipation shaping is carried out, it is ensured that the system has dissipation performance during normal flight, the dissipation performance is designed so that the dissipation performance of the system can be damaged when a specific fault occurs, and the supply rate and a coefficient matrix of the storage function are calculated. The method can solve the problem that a large amount of data and complex calculation are needed in the prior art, and can quickly and accurately detect and diagnose potential faults in an unmanned aerial vehicle system.

Description

Technical Field

[0001] The present invention belongs to the technical field of unmanned aerial vehicle fault diagnosis, and in particular relates to a fault detection and diagnosis method based on unmanned aerial vehicle data. Background Art

[0002] UAVs may encounter various faults during missions, which may cause the UAV to stop flying or change its flight attitude. Therefore, fault detection of UAVs is an important measure to ensure their safe operation. Existing fault detection and diagnosis methods for UAVs are mostly divided into model-based fault detection and diagnosis and data-based fault detection and diagnosis. Model-based fault detection and diagnosis methods identify faults by monitoring changes in system characteristics. This method does not require additional dynamics knowledge and has low computational cost, but it is difficult to establish an accurate mathematical model in the actual system. It is only applicable to situations where the mathematical model of the system is known, but due to the complexity and uncertainty of the actual system, it is difficult to accurately model. Data-based fault detection and diagnosis methods include principal component analysis (PCA), support vector machine (SVM) and subspace methods. These methods rely on a large number of data samples for training, but in the case of UAV flight, it is difficult to obtain sufficient training data, and a large amount of data is required for training. The uncertain faults in UAV flight make it difficult to obtain data. These methods have high real-time requirements, are complex to implement and have high costs. Summary of the invention

[0003] The purpose of the present invention is to provide a fault detection and diagnosis method based on drone data, which can solve the problem of requiring a large amount of data and complex calculations in the prior art, and can quickly and accurately detect and diagnose potential faults in drone systems.

[0004] The technical solution adopted by the present invention is as follows: A method for fault detection and diagnosis based on drone data, comprising the following steps: S1, initialize the system model, define state space model 1 and state space model 2, including no fault, fault 1 and fault 2, and analyze the specific impact of each fault on the flight performance of the UAV; S2, construct the Hankel matrix, collect input and output data under normal flight and various fault conditions of the UAV, determine the length and QdF degree of the Hankel matrix, and construct the input and output Hankel matrices of normal flight and flight with faults respectively; S3, dissipative shaping, ensuring that the system is dissipative during normal flight, designing the dissipativeness so that when a specific fault occurs, the dissipativeness of the system is destroyed, and calculating the coefficient matrix of the supply rate and storage function; S4, online diagnosis, set the time step, introduce fault 1 or fault 2, obtain the dissipation rate through the coefficient matrix calculated by S3, and judge whether the drone has no fault, has fault 1 or has fault 2 according to the dissipation rate.

[0005] In a preferred embodiment, in the step of initializing the system model, the state space model 1 is: x k+1 =A q x k +B q u k ; The state space model 2 is: y k =C q x k ; Where q=0, 1, 2 represent the situations of no fault, fault 1 and fault 2 respectively. represents the state variable of the drone at time k; θ is the rotation angle of the fuselage; u k and k They represent the upward lift and rotation speed generated by the UAV at time k, i.e., the input variable and output variable respectively.

[0006] In a preferred embodiment, the input variables are: ; Among them, m>0 is a constant aerodynamic parameter, Ω i is the angular velocity of each propeller about the z-axis (the axis perpendicular to the ground); A q ,B q ,C q are three parameter matrices, and the drone is flying without any faults , where l is the distance between the motor and the center of gravity, and J is the constant moment of inertia.

[0007] In a preferred embodiment, in the step of constructing a Hankel matrix, the flight time span of the UAV is set, and flight status data of no fault, fault 1 and fault 2 are collected during the flight time span. The size of the Hankel matrix is ​​confirmed according to the flight time span and the amount of flight status data, and an input vector and an output vector are constructed from the flight status data, and the input vector and the output vector are used to fill the Hankel matrix.

[0008] In a preferred embodiment, in the dissipative shaping step, the dissipativeness is driven by combining QdF and finite-step data to separate normal flight and flight with faults, so that the dissipativeness is present during normal flight, but not during a fault. When the system fails, the dissipative property is destroyed, and the fault can be detected and diagnosed quickly and accurately, that is, the dissipative property is shaped to be fault-sensitive, i.e. ; ; in, express is a positive definite matrix, express is a positive semidefinite matrix, and ; pass ,q=0 and ,q=1,2, find the supply rate and stored functions The coefficient matrix of and .

[0009] In a preferred embodiment, in the step of online diagnosis, the time step is set to k, and fault 1 or fault 2 is introduced in k, and the coefficient matrix and The corresponding dissipation rate J(ω k ), and its calculation formula is: ; By examining the dissipation rate J(ω k ), if there exists k such that ; The fault is detected.

[0010] In a preferred embodiment, in the step of initializing the system model, the fault 1 is the aging of the UAV rotor, and the fault 2 is the jumping voltage.

[0011] In a preferred solution, the constructed Hankel matrix can be used to further analyze the dynamic behavior of the system.

[0012] The technical effects achieved by the present invention are: The present invention introduces the quadratic difference method into the traditional dissipative condition, which avoids the problem that in the data-based process, the traditional dissipative condition cannot be directly used in the data-driven situation because only the input-output signal can be accessed and the storage function depends on the unavailable state variable. Since the method only needs input and output data, it greatly reduces the amount of information required to obtain and the human, material and financial resources. The new fault detection and diagnosis method of the present invention is divided into two stages, offline and online. The offline stage is to form dissipative conditions and find a set of supply rates and storage function coefficient matrices so that the input-output trajectory is sensitive to specific faults and insensitive to other faults, which helps to identify faults more accurately, that is, the dissipative characteristics are designed for fault selection. Online fault detection, that is, using associated dissipative conditions to evaluate faults, can help the system discover and solve faults more quickly, improving the reliability and stability of the system. These two stages mean that it can perform fault detection in different situations and improve the flexibility of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 is a flow chart of the method of the present invention; Figure 2 This is a flight force analysis diagram of the UAV of the present invention; Figure 3 This is a flight effect diagram of the UAV of the present invention with fault 1; Figure 4 It is a schematic diagram of the lift change of the UAV of the present invention with numerical simulation of fault 1; Figure 5 This is a numerical simulation effect diagram of the UAV of the present invention with fault 1; Figure 6 This is a flight effect diagram of the drone of the present invention with Fault 2; Figure 7 It is a schematic diagram of lift change of the UAV with fault 2 numerical simulation of the present invention; Figure 8 This is a numerical simulation effect diagram of the UAV of the present invention with fault 2; Fig. 9 It is a flow chart of the fault offline diagnosis and online detection of the present invention. DETAILED DESCRIPTION

[0014] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0015] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0016] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure or characteristic that may be included in at least one implementation of the present invention. The phrase "in a preferred embodiment" that appears in different places in this specification does not refer to the same embodiment, nor is it a separate or selective embodiment that is mutually exclusive with other embodiments.

[0017] Secondly, the present invention is described in detail with reference to the schematic diagram. When describing the embodiments of the present invention in detail, for the sake of convenience, the cross-sectional diagrams showing the device structure will not be partially enlarged according to the general scale, and the schematic diagrams are only examples, which should not limit the scope of protection of the present invention. In addition, in actual production, the three-dimensional dimensions of length, width and depth should be included.

[0018] See also Figures 1 to 9 , a fault detection and diagnosis method based on drone data, comprising the following steps: S1, initialize the system model, define state space model 1 and state space model 2, including no fault, fault 1 and fault 2, and analyze the specific impact of each fault on the flight performance of the UAV; S2, construct the Hankel matrix, collect input and output data under normal flight and various fault conditions of the UAV, determine the length and QdF degree of the Hankel matrix, and construct the input and output Hankel matrices of normal flight and flight with faults respectively; S3, dissipative shaping, ensuring that the system is dissipative during normal flight, designing the dissipativeness so that when a specific fault occurs, the dissipativeness of the system is destroyed, and calculating the coefficient matrix of the supply rate and storage function; S4, online diagnosis, setting the time step, introducing fault 1 or fault 2, obtaining the dissipation rate through the coefficient matrix calculated by S3, and judging whether the drone has no fault, has fault 1 or has fault 2 according to the dissipation rate; In the above, the state space model 1 and state space model 2 that describe the dynamic behavior of the normal flight of the UAV established when the system model is initialized include the input (such as control commands) and output (such as flight data) of the system, as well as the state variables of the system (such as speed, position, etc.), where Fault 1 modifies the state space model to reflect the abnormalities caused by rotor aging, such as reduced lift, unstable flight, etc. This will be achieved by adjusting the parameters in the model, such as reducing the lift coefficient or increasing air resistance, and Fault 2 adjusts the state space model to simulate control problems caused by voltage fluctuations, such as control signal distortion, response delay, etc. This may involve introducing additional noise terms or time-varying parameters in the model; Specifically, Fault 1: The aging of the drone rotor and abnormal friction can cause minor faults. Lift is the main force that supports the drone to fly in the air. It is affected by the aerodynamic effect generated by the rotor. However, abnormal friction and aging may cause minor faults or irregularities on the rotor surface, which will affect the generation and distribution of lift. Therefore, the balance between lift and resultant force may be affected, which may lead to a decrease in the drone's flight performance or stability problems. The drone will not be able to maintain the original altitude, but will gradually descend. See Figure 3 , Figure 4 As shown; Fault 2: During the flight of the UAV, external airflow disturbances may cause actuator failures, making it impossible for the UAV drive voltage to stabilize at the preset benchmark, resulting in voltage jumps or failures. In this case, the UAV may not be able to respond to flight commands normally, and the flight control performance will be significantly reduced, and it may even cause loss of control or emergency landing. Especially when flying at high altitudes, strong wind disturbances are more likely to affect the stability and flight altitude of the UAV, and increase the workload of the actuator. In addition, voltage instability will directly affect the working stability of actuators (such as thrusters or rotors), thereby affecting the generation and control of lift. The instability of the UAV's lift will cause the flight altitude to fluctuate rapidly or fail to maintain the expected altitude, further exacerbating the complexity and risk of flight control, see. Figure 6 , Figure 7 As shown; When constructing the Hankel matrix, input (such as control commands) and output (such as flight data) data are collected during normal flight and various fault conditions of the drone. These data will be used in subsequent analysis steps to identify the normal operating mode of the system and potential signs of faults. Ensure that the data covers all relevant operating conditions and environmental changes so that the dynamic behavior of the system can be fully understood. Determine the parameters and set the length and QdF degree of the Hankel matrix. These parameters should be determined based on the dynamic characteristics of the system and the required diagnostic accuracy. The length determines the time span of the matrix, while the QdF degree affects the degree of data compression and the effect of feature extraction. Construct the Hankel matrix and use the data from normal flight to construct a benchmark Hankel matrix. This will provide a reference point for comparing the differences in the behavior of the system under different fault conditions. Use the data with faults to construct the corresponding fault Hankel matrices. These matrices will reveal subtle changes in the behavior of the system under fault conditions and provide key information for fault detection. Furthermore, the Hankel matrix is ​​constructed using persistent excitation (PE), which defines a vector signal sequence The composed Hankel matrix is: ; Where L is the length of the Hankel matrix, and ; if , then the sequence is L-order continuous excitation, where , n is the dimension of the signal space; Furthermore, the continuous incentive theorem set up ,but In the interval The above is defined as: ; in is the McMillan degree, that is, the system The dimension of .

[0019] Similarly, for a matrix , represents the truncated matrix selected from row i to row j.

[0020] Assumptions ,definition , , yes A trajectory of yes If the excitation is continuous, then for all ,exist , so that , and ; in In signal space The set of all subspaces that satisfy the linear, closed, translation-invariant properties and controllability conditions.

[0021] Specifically, assume that the drone has been flying for 30 minutes. During this time span, generate input and output vectors for the normal flight process and the flight process when each fault occurs. Set the length L and degree d of the Hankel matrix QdF to ) are set to 3 and 2 respectively, and then the input and output Hankel matrices of normal flight and flight with fault 1 and fault 2 are constructed respectively.

[0022] During dissipative shaping, a positive definite storage function and a semi-positive definite supply rate are designed so that the system meets the dissipative condition when there is no fault, so that the energy of the system will not increase spontaneously, thereby ensuring the stability and controllability of the system. The coefficient matrix of the storage function and the supply rate is adjusted so that when a specific fault occurs, the dissipativeness of the system is destroyed, so that the fault state can be quickly detected through the change of dissipativeness. The coefficient matrix of the supply rate and the storage function is calculated using the formula to ensure that these matrices meet the required mathematical properties (positive definite or semi-positive definite). These matrices will be used for real-time calculations in the online diagnosis process to evaluate the current state of the system. When a specific fault occurs in the system, the fault can be quickly and accurately detected and diagnosed, that is, the dissipativeness is shaped to be fault-sensitive. The three steps of initializing the system model, constructing the Hankel matrix, and dissipative shaping are the offline detection part. In this stage, the system will form a characteristic that makes the system sensitive to specific faults and insensitive to other faults, and this characteristic is used to evaluate faults in the online stage. Since this method only requires input and output data, it greatly reduces the amount of information required and the human, material, and financial resources. During online diagnosis, a suitable time step is selected for real-time data collection and processing. This time step needs to be small enough to capture the rapid dynamic changes of the system, but not too small to generate too much computational burden. Faults are introduced and the dissipation rate is calculated. Fault 1 or Fault 2 is introduced in each time step, and the coefficient matrix calculated in S3 is used to calculate the current dissipation rate. This process will monitor the dissipation of the system in real time to promptly detect any signs of deviation from normal behavior. According to the change in the dissipation rate, it is determined whether the drone is in a fault-free state, fault 1 or fault 2. If the dissipation rate exceeds the preset threshold, it is considered that a fault is detected, and the fault type is further identified based on the specific pattern of the dissipation rate. This method allows the system to automatically adapt to new fault conditions without prior knowledge, improving the accuracy of diagnosis.

[0023] Furthermore, in the step of initializing the system model, the state space model 1 is: x k+1 =A q x k +B q u k ; The state space model 2 is: y k =C q x k ; Where q=0, 1, 2 represent the situations of no fault, fault 1 and fault 2 respectively. represents the state variable of the drone at time k; θ is the rotation angle of the fuselage; uk and k They represent the upward lift and rotation speed generated by the UAV at time k, i.e., the input variable and output variable respectively.

[0024] Furthermore, the input variables are: ; Among them, m>0 is a constant aerodynamic parameter, Ω i is the angular velocity of each propeller about the z-axis (the axis perpendicular to the ground); A q ,B q ,C q are three parameter matrices, and the drone is flying without any faults , where l is the distance between the motor and the center of gravity, and J is the constant moment of inertia; Specific reference Figure 2 As shown in the figure, when a drone flies normally in the air, it will generate longitudinal forces, including lift and resultant force. Lift is composed of self-generated lift and longitudinal updraft lift. The resultant force is generated by gravity and longitudinal wind disturbance force. The longitudinal wind disturbance force is a force generated by other factors (such as air resistance, inertia and airflow changes).

[0025] In a preferred embodiment, in the step of constructing the Hankel matrix, a flight time span of the UAV is set, flight status data of no fault, fault 1 and fault 2 are collected during the flight time span, the size of the Hankel matrix is ​​determined according to the flight time span and the amount of flight status data, an input vector and an output vector are constructed from the flight status data, and the input vector and the output vector are used to fill the Hankel matrix; The constructed Hankel matrix can be used to further analyze the dynamic behavior of the system; In the above, the stability and response characteristics of the system are understood by calculating the eigenvalues ​​and eigenvectors of the matrix. In fault detection and diagnosis based on drone data, this analysis helps to identify potential fault modes and provide timely warnings and maintenance. In this way, the Hankel matrix not only provides a powerful tool to analyze and understand the dynamic characteristics of complex systems, but also shows its unique value and effect in practical applications, such as drone fault detection and diagnosis.

[0026] Furthermore, in the dissipative shaping step, the dissipativeness is driven by combining QdF and finite-step data to separate normal flight and flight with faults, so that the dissipativeness is present during normal flight, but not during a fault. When the system fails, the dissipative property is destroyed, and the fault can be detected and diagnosed quickly and accurately, that is, the dissipative property is shaped to be fault-sensitive, i.e. ; ; in, express is a positive definite matrix, express is a positive semidefinite matrix, and ; pass ,q=0 and ,q=1,2, find the supply rate and stored functions The coefficient matrix of and ; In the above, the quadratic difference method (QdF) is: In a fully data-driven setting, only input-output data are used as the only source of information. Since the storage function involves unknown state variables, the traditional dissipative condition cannot be directly applied to the data-driven case. By introducing the QdF framework that conforms to the data-driven extended input and output, the instability of the system can be described in more detail. Therefore, by applying QdF to the supply rate and storage function, data-driven dissipativeness is developed. Typically, a (Q,S,R)-type supply rate is used, i.e. ; Definition col(a,b) is the matrix obtained by stacking vector a on vector b. Definition and are the input and output of the expansion of QdF, respectively, where and is finite. Definition , Therefore, the QdF supply rate and storage function can be expressed as: ; Among them, the coefficient matrix , ; The finite-step data-driven dissipativeness is: Consider a discrete-time dynamical system (a subset of the signal space) , if for all 0 <k<L( ), all satisfy the following inequality: ; Then the system is called finite-step data-driven dissipation, where For storage functions, is the supply rate, .

[0027] Furthermore, in the step of online diagnosis, the time step is set to k, and fault 1 or fault 2 is introduced, and the coefficient matrix and The corresponding dissipation rate J(ω k ), and its calculation formula is: ; By examining the dissipation rate J(ω k ), if there exists k such that ; The fault is detected.

[0028] Furthermore, Figure 5 and Figure 8 The feasibility of this method was verified. Figure 5 The three figures on the left, middle and right represent normal flight, flight with fault 1 and flight with fault 2 respectively. The dissipation rate of the middle figure is obviously less than 0, while the dissipation rate of the right figure is always greater than 0, which indicates that fault 1 has been diagnosed. Similarly, Figure 8 Since the dissipation rate of the right figure is obviously less than 0, while the dissipation rate of the middle figure is always greater than 0, fault 2 is diagnosed. This technical solution has two advantages over the new method of fault diagnosis and detection of linear process systems based on data dissipation theory: first, it introduces the supply rate and storage function in the form of quadratic differences, and expresses the dissipative condition as a quadratic function of the process input-output trajectory. Compared with the traditional dissipative condition, it captures more detailed dynamic characteristics, allowing us to capture the behavior of the system in more detail. Second, it proposes a new method for fault detection and diagnosis based on drone data, which is divided into two stages: offline and online. In the offline stage, the system will form a characteristic that makes the system sensitive to specific faults and insensitive to other faults, while the online stage uses this characteristic to evaluate faults. Since this method only requires input and output data, it greatly reduces the amount of information required and the human, material, and financial resources required.

[0029] The above is only a preferred embodiment of the present invention. It should be noted that, for those skilled in the art, several improvements and modifications can be made without departing from the principles of the present invention, and these improvements and modifications should also be considered as the protection scope of the present invention. The structures, devices and operating methods not specifically described and explained in the present invention shall be implemented according to the conventional means in the art unless otherwise specified and limited.

Claims

1. A fault detection and diagnosis method based on drone data, characterized in that: The following steps are involved: S1, initialize the system model, define state space model 1 and state space model 2, including no fault, fault 1 and fault 2, and analyze the specific impact of each fault on the flight performance of the UAV; S2, construct the Hankel matrix, collect input and output data under normal flight and various fault conditions of the UAV, determine the length and QdF degree of the Hankel matrix, and construct the input and output Hankel matrices of normal flight and flight with faults respectively; S3, dissipative shaping, ensuring that the system is dissipative during normal flight, designing the dissipativeness so that when a specific fault occurs, the dissipativeness of the system is destroyed, and calculating the coefficient matrix of the supply rate and storage function; S4, online diagnosis, set the time step, introduce fault 1 or fault 2, obtain the dissipation rate through the coefficient matrix calculated by S3, and judge whether the drone has no fault, has fault 1 or has fault 2 according to the dissipation rate.

2. A method for fault detection and diagnosis based on drone data according to claim 1, characterized in that: In the step of initializing the system model, the state space model 1 is: x k+1 =A q x k +B q u k ; The state space model 2 is: y k =C q x k ; Where q=0, 1, 2 represent the situations of no fault, fault 1 and fault 2 respectively. represents the state variable of the drone at time k; θ is the rotation angle of the fuselage; u k and k They represent the upward lift and rotation speed generated by the UAV at time k, i.e., the input variable and output variable respectively.

3. A method for fault detection and diagnosis based on drone data according to claim 2, characterized in that: The input variables are: ； Among them, m>0 is a constant aerodynamic parameter, Ω i is the angular velocity of each propeller about the z-axis (the axis perpendicular to the ground); A q ,B q ,C q are three parameter matrices, and the drone is flying without any faults , where l is the distance between the motor and the center of gravity, and J is the constant moment of inertia.

4. The method for fault detection and diagnosis based on drone data according to claim 1, characterized in that: In the step of constructing the Hankel matrix, the flight time span of the UAV is set, and the flight status data of no fault, fault 1 and fault 2 are collected during the flight time span. The size of the Hankel matrix is ​​confirmed according to the flight time span and the amount of flight status data, and an input vector and an output vector are constructed from the flight status data. The input vector and the output vector are used to fill the Hankel matrix.

5. The method for fault detection and diagnosis based on drone data according to claim 1, characterized in that: In the dissipative shaping step, the dissipativeness is driven by combining QdF and finite-step data to separate normal flight and flight with faults, so that the dissipativeness is present during normal flight, but not during a fault. When the system fails, the dissipative property is destroyed, and the fault can be detected and diagnosed quickly and accurately, that is, the dissipative property is shaped to be fault-sensitive, i.e. ； ； in, express is a positive definite matrix, express is a positive semidefinite matrix, and ； pass ,q=0 and ,q=1,2, find the supply rate and stored functions The coefficient matrix of and .

6. A method for fault detection and diagnosis based on drone data according to claim 5, characterized in that: In the online diagnosis step, the time step is set to k, and fault 1 or fault 2 is introduced in k, and the coefficient matrix and The corresponding dissipation rate J(ω k ), and its calculation formula is: ； By examining the dissipation rate J(ω k ), if there exists k such that ； The fault is detected.

7. The method for fault detection and diagnosis based on drone data according to claim 1, characterized in that: In the step of initializing the system model, the fault 1 is the aging of the UAV rotor, and the fault 2 is the jumping voltage.

8. The method for fault detection and diagnosis based on drone data according to claim 4, characterized in that: The constructed Hankel matrix can be used to further analyze the dynamic behavior of the system.

Images