Six-blade linked coaxial twin-rotor unmanned helicopter system and adaptive control method

Abstract

The present invention discloses a six-blade coaxial twin-rotor unmanned helicopter system and adaptive control method, comprising the following steps: S1, collecting multi-source flight operating status data and preprocessing to generate a standardized data set; S2, screening stable operating condition flight data and training a variational autoencoder model; S3, inputting the standardized data set into the model, performing distribution reconstruction inference and health discrimination, and extracting abnormal disturbance features; S4, constructing an integral sliding mode surface, designing an adaptive sliding mode control law, adjusting parameters and fault-tolerant compensation; S5, combining the mission type and operating conditions, and intelligently switching control parameters based on the health discrimination results. The present invention implements intelligent health discrimination and adaptive robust optimization control of a six-blade coaxial twin-rotor unmanned helicopter under complex operating conditions.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent control and fault tolerance of aircraft, and in particular to a six-propeller linked coaxial dual-rotor unmanned helicopter system and an adaptive control method. Background Art

[0002] In recent years, with the continuous expansion of intelligent manufacturing, aerospace, and complex operational scenarios, unmanned helicopter systems, especially six-blade coaxial twin-rotor unmanned helicopters, have shown great application potential in urban air traffic, logistics distribution, disaster relief, and special inspections due to their excellent hovering capabilities, strong power output, and high redundancy. To ensure flight safety and efficient mission execution, the intelligent control and fault tolerance capabilities of unmanned helicopters have become the focus of industry attention. Most existing flight control systems are based on traditional PID control, classical sliding mode control, or adaptive control algorithms, which improve the system's steady-state performance and anti-interference capabilities by closed-loop adjustment of the aircraft's attitude, speed, and position. However, with the increasing complexity of the system and the changing mission environment, traditional methods have gradually exposed a number of limitations in practical applications.

[0003] Existing flight control methods often rely heavily on system modeling and disturbance characteristics, making it difficult to accurately respond to multi-source complex disturbances and unknown abnormal conditions, resulting in insufficient robustness of flight control under extreme conditions. Most methods rely on threshold settings and manual experience to determine the health of the flight state, and are unable to achieve adaptive anomaly detection and accurate early warning of multi-dimensional complex states, which can easily lead to missed faults or misjudgments. Existing sliding mode control methods have limited parameter adaptive adjustment capabilities when dealing with strong disturbances and sensor anomalies, and the room for improving fault tolerance performance is limited, especially in high-dimensional multi-propeller redundant systems. It is difficult to achieve optimal thrust distribution and real-time adaptive adjustment in all scenarios. Existing systems generally have response lags and unintelligent switching in terms of control parameter switching and mission scenario adaptation, making it difficult to meet the optimization control requirements under actual complex flight missions.

[0004] Therefore, how to provide a six-blade linked coaxial twin-rotor unmanned helicopter system and an adaptive control method is a problem that technical personnel in this field urgently need to solve. Summary of the Invention

[0005] One objective of the present invention is to propose a six-blade, coaxial, dual-rotor unmanned helicopter system and adaptive control method. This invention leverages deep anomaly detection technology driven by multi-source flight status data and the robust adaptive characteristics of sliding mode control. It describes in detail a technical solution for achieving full-condition adaptive optimization control of the unmanned helicopter through intelligent health status identification, abnormal disturbance feature extraction, adaptive adjustment of sliding mode gains, and intelligent switching of control parameters. This invention offers advantages such as intelligent sensing of abnormal conditions, real-time compensation for strong disturbances, adaptive optimization of flight control parameters, and high reliability in multi-mission scenarios. It can significantly enhance the safety, intelligence, and mission adaptability of the unmanned helicopter system.

[0006] The adaptive control method of a six-blade linked coaxial dual-rotor unmanned helicopter according to an embodiment of the present invention includes the following steps:

[0007] S1. Collect multi-source operating status data of a six-propeller linked coaxial twin-rotor unmanned helicopter during flight, pre-process the multi-source operating status data, and generate a standardized data set;

[0008] S2. Filter flight data of unmanned helicopters in a fault-free, disturbance-free state with stable indicators from a standardized dataset, construct a normal operating condition dataset, train the variational autoencoder model, and establish the potential distribution structure of the flight system.

[0009] S3. Input the standardized dataset into the trained variational autoencoder model to perform distribution reconstruction inference, determine the current health status of the flight system, and determine whether the reconstruction error or latent variable distribution exceeds a preset threshold. If so, mark the current operating condition as abnormal and extract the abnormal disturbance feature vector.

[0010] S4. Based on the standardized data set and the abnormal disturbance eigenvector, an integral sliding mode surface is constructed, a variable structure integral sliding mode control law is designed, and the sliding mode control gain parameters related to the integral sliding mode surface are adjusted. When abnormal operating conditions or extreme disturbances are detected, the controller parameters are automatically adjusted to trigger torque compensation and fault-tolerant control mechanisms in real time.

[0011] S5. Based on the actual flight mission type and environmental conditions, and according to the health status judgment results, the control parameters of the unmanned helicopter are intelligently switched to achieve adaptive optimization control under different flight missions.

[0012] Optionally, the multi-source operating status data specifically includes each blade speed, thrust, motor current, aircraft attitude angle, mission type, flight phase and operating condition label.

[0013] Optionally, the preprocessing of multi-source operating status data specifically includes data cleaning, missing value filling, outlier removal and normalization.

[0014] Optionally, the S2 specifically includes:

[0015] S21. Extracting blade speed, thrust, motor current, aircraft attitude angle, mission type, flight phase, and operating condition labels from the generated standardized data set;

[0016] S22. Based on the unmanned helicopter operation log and system status labels, filter the flight data under stable, fault-free, and disturbance-free conditions from the standardized data set to construct a normal operating condition data set. ;

[0017] S23, based on normal working condition data set , design a variational autoencoder model, which includes a multi-channel encoder, a joint latent space, a conditional fusion mechanism, a temporal recursive structure and a decoder. The multi-channel encoder uses independent neural network branches to encode different types of features. The output of each encoding branch is spliced ​​into a comprehensive feature vector, which is mapped to the latent space through a fully connected layer. The conditional fusion mechanism uses the mission type, flight phase or working condition label as a conditional variable Combined with the comprehensive feature vector as input, the temporal recursive structure uses a gated recurrent unit network at both the encoding and decoding ends to achieve temporal modeling of multi-time data sequences. The decoder consists of a fully connected neural network and a recursive structure.

[0018] S24, the normal working condition data set Input the multi-channel encoder of the variational autoencoder model to encode each type of feature separately to obtain the feature channel encoding result , extract continuous historical multi-step time series features from standardized data sets , using the temporal recursive structure of the variational autoencoder model to perform temporal recursive modeling on historical multi-step temporal features and extract temporal hidden states ,in, is the number of historical time series steps;

[0019] S25, encoding results of each feature channel Splicing to form a comprehensive feature vector , the comprehensive feature vector , conditional variables and temporal hidden states After weighted fusion of multi-head self-attention and conditional fusion mechanism, adaptive attention weights and gating coefficients are assigned to different feature channels, historical moments and conditional variables to obtain the fusion representation , the fusion representation Input the joint latent space of the variational autoencoder model to obtain dynamically adaptive latent variables ;

[0020] S26. The latent variables obtained and condition variables Input the decoder of the variational autoencoder model, and restore the reconstruction results of each source feature by the fully connected neural network and recursive structure to obtain the reconstructed sequence of multi-source features ;

[0021] S27. In the decoder, the long short-term memory network is based on the latent variables , conditional variables and temporal hidden states Recursively generate reconstruction sequences of multi-source features ;

[0022] S28. Reconstruction sequence based on multi-source features , construct the loss function of the variational autoencoder model ;

[0023] S29. Using the normal operating data set Train the variational autoencoder model to minimize the loss function , optimize the parameters of the multi-channel encoder, joint latent space, conditional fusion mechanism, temporal recursive structure and decoder.

[0024] Optionally, the S3 specifically includes:

[0025] S31, using the trained variational autoencoder model, the standardized data set is The input data below , conditional variables And the historical multi-step time series hidden state is distributed reconstructed and inferred to obtain the reconstructed sequence and latent variables ;

[0026] S32. According to input data and reconstruction sequence , using multi-head self-attention mechanism and historical error trend modulation to calculate the dynamic weighted reconstruction error of the fusion time series trend ;

[0027] S33, Mahalanobis distance calculation using adaptive covariance weights and difference measurement between normal operating conditions and statistical distribution ;

[0028] S34. According to the unmanned helicopter mission type, flight phase and working condition label, a multi-scale dynamic threshold strategy is adopted to set adaptive thresholds for the dynamic weighted reconstruction error and difference metric of the fusion time series trend. 、 ;

[0029] S35, if it exists or , then the current working condition is marked as abnormal, and the disturbance feature tracing function is called Extract abnormal disturbance feature vector , where the disturbance feature tracing function is a function that combines the latent variable distribution with the dynamic weighted reconstruction error, inputs it into the multi-layer perceptron for nonlinear transformation and feature extraction, and outputs the current moment The health status judgment results and abnormal disturbance feature vector , Dynamic weighted reconstruction error integrating time series trends and latent variables .

[0030] Optionally, the S4 specifically includes:

[0031] S41, based on standardized data set and abnormal disturbance feature vector , constructing adaptive nonlinear integral sliding surface ;

[0032] S42, according to the adaptive nonlinear integral sliding surface and dynamic change characteristics, design a multi-channel adaptive sliding mode control law ;

[0033] S43, using the historical normalized disturbance weight mechanism, the abnormal disturbance feature vector Decompose;

[0034] S44. Dynamic weighted reconstruction error based on health status judgment results and integration of time series trends and Mahalanobis distance metric , adaptively adjust the sliding mode controller structure;

[0035] S45, integrated AI fault tolerance compensation, using neural network prediction model Feedforward compensation is performed on the disturbance trend that may occur in the future, and the predicted compensation vector is ,in, is the feature splicing operation, are neural network parameters;

[0036] S46, predicting the compensation vector Multi-channel adaptive sliding mode control law Fusion, automatically distributes multi-propeller thrust and compensation torque, and generates specific thrust distribution control instructions;

[0037] S47, real-time update of adaptive nonlinear integral sliding surface With multi-channel adaptive sliding mode control law parameters, closed-loop adaptive adjustment and intelligent optimization are achieved;

[0038] S48 outputs the final multi-channel adaptive sliding mode control law, AI fault-tolerant compensation and thrust distribution control instructions to drive the robust adaptive optimization operation of the six-propeller linked coaxial twin-rotor unmanned helicopter under multi-source disturbances and extreme working conditions.

[0039] Optionally, the S5 specifically includes:

[0040] S51. Determine the actual flight mission type and target parameters of the unmanned helicopter according to the flight scheduling plan and current mission requirements, and collect mission-related environmental conditions;

[0041] S52. Analyze the health status determination results in real time to determine whether the current flight system is in a normal, abnormal, or extreme state;

[0042] S53. Based on the flight mission type, environmental conditions, and health status judgment results, query the control parameter switching strategy library to match the optimal control parameters;

[0043] S54, automatically switching flight control parameters, and loading the optimal control parameters into the flight control system in real time;

[0044] S55. After the control parameters are switched, continuously monitor the flight system's response and operational performance. If performance degradation or abnormal signals occur, dynamically adjust the control parameter group.

[0045] S56. Record the optimized control parameters, flight mission status, and adjustment logs in real time, and simultaneously push them to the upper-level task management and operation and maintenance platform.

[0046] The six-blade linked coaxial twin-rotor unmanned helicopter system according to an embodiment of the present invention includes the following modules:

[0047] The data acquisition and preprocessing module is used to collect and preprocess multi-source operating status data during the flight of the unmanned helicopter and generate a standardized data set;

[0048] Normal operating condition modeling module, used to filter normal operating condition data in the standardized dataset and train the variational autoencoder model;

[0049] The health status identification and feature extraction module is used to input a standardized data set into the variational autoencoder model, identify the health status, and extract abnormal disturbance features when abnormal;

[0050] Adaptive sliding mode control and fault tolerance module, which is used to construct the integral sliding mode surface, design the sliding mode control law, and trigger torque compensation and fault tolerance control in real time based on the standardized data set and abnormal disturbance characteristics;

[0051] The parameter switching and optimization control module is used to combine multi-source operating status data and health judgment results, intelligently switch control parameters, and achieve adaptive optimization control.

[0052] The beneficial effects of the present invention are:

[0053] By integrating a variational autoencoder model with an adaptive sliding mode control method, this paper proposes a comprehensive adaptive robust control solution for intelligent health status identification, abnormal disturbance feature extraction, and adaptive optimization of flight control parameters for a six-blade coaxial twin-rotor unmanned helicopter under complex and variable operating conditions. Compared to existing technologies, this paper utilizes multi-source flight data to deeply model the unmanned helicopter's state. Using a variational autoencoder, it accurately detects and reconstructs abnormal operating conditions, significantly improving the system's sensitivity to minor disturbances and latent faults and its early warning capabilities. Incorporating abnormal disturbance feature vectors, the proposed adaptive sliding mode control law dynamically adjusts control gains and structural parameters based on actual flight states and operating conditions, activating torque compensation and fault tolerance mechanisms in real time to effectively suppress flight deviations and abnormal responses under multi-source disturbances. In scenarios involving mission switching and environmental changes, this paper enables intelligent switching of flight control parameters and full-scenario adaptive optimization based on health status identification results, ensuring the system maintains high robustness, reliability, and safety across various flight missions. This invention not only improves the intelligent control level and fault self-healing capability of the unmanned helicopter system in complex environments, but also lays a solid technical foundation for the practical and industrial application of a new generation of high-performance unmanned helicopters. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0055] Figure 1 This is a flow chart of the adaptive control method for the six-propeller linked coaxial twin-rotor unmanned helicopter proposed in the present invention;

[0056] Figure 2 This is a schematic structural diagram of the six-blade linked coaxial twin-rotor unmanned helicopter system proposed in the present invention. DETAILED DESCRIPTION

[0057] The present invention will now be described in further detail with reference to the accompanying drawings, which are simplified schematic diagrams that illustrate the basic structure of the present invention in a schematic manner.

[0058] refer to Figure 1 The adaptive control method of a six-propeller linked coaxial twin-rotor unmanned helicopter includes the following steps:

[0059] S1. Collect multi-source operating status data of a six-propeller linked coaxial twin-rotor unmanned helicopter during flight, pre-process the multi-source operating status data, and generate a standardized data set;

[0060] S2. Filter flight data of unmanned helicopters in a fault-free, disturbance-free state with stable indicators from a standardized dataset, construct a normal operating condition dataset, train the variational autoencoder model, and establish the potential distribution structure of the flight system.

[0061] S3. Input the standardized dataset into the trained variational autoencoder model to perform distribution reconstruction inference, determine the current health status of the flight system, and determine whether the reconstruction error or latent variable distribution exceeds a preset threshold. If so, mark the current operating condition as abnormal and extract the abnormal disturbance feature vector.

[0062] S4. Based on the standardized data set and the abnormal disturbance eigenvector, an integral sliding mode surface is constructed, a variable structure integral sliding mode control law is designed, and the sliding mode control gain parameters related to the integral sliding mode surface are adjusted. When abnormal operating conditions or extreme disturbances are detected, the controller parameters are automatically adjusted to trigger torque compensation and fault-tolerant control mechanisms in real time.

[0063] S5. Based on the actual flight mission type and environmental conditions, and according to the health status judgment results, the control parameters of the unmanned helicopter are intelligently switched to achieve adaptive optimization control under different flight missions.

[0064] The present invention realizes intelligent adaptive optimization control of a six-blade linked coaxial twin-rotor unmanned helicopter under complex working conditions by combining the intelligent collection and preprocessing of multi-source flight status data, deep distribution modeling and health judgment of variational autoencoders, and adaptive robust adjustment of integral sliding mode control. Compared with traditional methods, the present invention can not only accurately distinguish between normal and abnormal working conditions, but also dynamically adjust the sliding mode control parameters and automatically activate the fault-tolerant compensation mechanism when disturbances and faults occur, ensuring that the unmanned helicopter system still has excellent flight stability and safety in multi-source disturbances and extreme environments. The present invention can intelligently switch and optimize control parameters according to the type of flight mission and environmental conditions, thereby improving the system's adaptability and intelligence level to different mission scenarios. It significantly enhances the unmanned helicopter's abnormal response capability and full-scene robustness, and improves the system's reliability and mission completion efficiency.

[0065] In this embodiment, the multi-source operating status data specifically includes the rotational speed of each blade, thrust, motor current, aircraft attitude angle, mission type, flight phase and operating condition label.

[0066] In this embodiment, the preprocessing of multi-source operating status data specifically includes data cleaning, missing value filling, outlier removal and normalization.

[0067] In this embodiment, S2 specifically includes:

[0068] S21. Extracting blade speed, thrust, motor current, aircraft attitude angle, mission type, flight phase, and operating condition labels from the generated standardized data set;

[0069] S22. Based on the unmanned helicopter operation log and system status labels, filter the flight data under stable, fault-free, and disturbance-free conditions from the standardized data set to construct a normal operating condition data set. ;

[0070] S23, based on normal working condition data set , design a variational autoencoder model, which includes a multi-channel encoder, a joint latent space, a conditional fusion mechanism, a temporal recursive structure and a decoder. The multi-channel encoder uses independent neural network branches to encode different types of features. The output of each encoding branch is spliced ​​into a comprehensive feature vector, which is mapped to the latent space through a fully connected layer. The conditional fusion mechanism uses the mission type, flight phase or working condition label as a conditional variable Combined with the comprehensive feature vector as input, the temporal recursive structure uses a gated recurrent unit network at both the encoding and decoding ends to achieve temporal modeling of multi-time data sequences. The decoder consists of a fully connected neural network and a recursive structure.

[0071] S24, the normal working condition data set Input the multi-channel encoder of the variational autoencoder model to encode each type of feature separately to obtain the feature channel encoding result , extract continuous historical multi-step time series features from standardized data sets , using the temporal recursive structure of the variational autoencoder model to perform temporal recursive modeling on historical multi-step temporal features and extract temporal hidden states ,in, is the number of historical time series steps;

[0072] S25, encoding results of each feature channel Splicing to form a comprehensive feature vector , the comprehensive feature vector , conditional variables and temporal hidden states After weighted fusion of multi-head self-attention and conditional fusion mechanism, adaptive attention weights and gating coefficients are assigned to different feature channels, historical moments and conditional variables to obtain the fusion representation , the fusion representation Input the joint latent space of the variational autoencoder model to obtain dynamically adaptive latent variables :

[0073] ;

[0074] in, 、 、 is the weighted coefficient obtained by adaptive learning of multi-head self-attention and conditional fusion mechanism, is the weighted fusion function, is the total number of feature channels, is the weighted coefficient of the i-th feature channel and the l-th historical time sequence step obtained by adaptive learning of the multi-head self-attention mechanism, is the encoding result of the i-th feature channel, is the temporal hidden state at time t−l;

[0075] Dynamically adaptive latent variables The system achieves deep fusion and adaptive weighting of multi-source flight status characteristics, historical time series information, mission types, and operating condition labels of unmanned helicopters, breaking through the traditional single feature or static weighting method. It autonomously allocates fusion weights based on the actual contribution of different characteristics, historical moments, and operating condition information, significantly improving the dynamic modeling capability of flight status. Through the joint fusion of multi-dimensional, global, and local features, the system can more accurately characterize the health status and potential risks of unmanned helicopters when dealing with changing environments, complex disturbances, and multi-task switching, achieving sensitive identification and early warning of abnormal conditions, providing a high-quality feature foundation for adaptive adjustment of control parameters and intelligent decision-making, effectively enhancing the robustness, intelligence, and mission completion rate of unmanned helicopters in extreme working conditions and unknown scenarios, significantly improving the system's adaptability and health monitoring accuracy, and laying a solid foundation for achieving highly reliable intelligent control of unmanned helicopters.

[0076] S26. The latent variables obtained and condition variables Input the decoder of the variational autoencoder model, and restore the reconstruction results of each source feature by the fully connected neural network and recursive structure to obtain the reconstructed sequence of multi-source features ;

[0077] S27. In the decoder, the long short-term memory network is based on the latent variables , conditional variables and temporal hidden states Recursively generate reconstruction sequences of multi-source features ;

[0078] S28. Reconstruction sequence based on multi-source features , construct the loss function of the variational autoencoder model :

[0079] ;

[0080] in, is the actual input data, is the reconstructed sequence of multi-source features, is the KL divergence, is the prior distribution of the latent variable, is the conditional probability distribution, is the mathematical expectation;

[0081] Loss Function By simultaneously minimizing the data reconstruction error and the distance between the potential distribution and the prior distribution, the accuracy and generalization ability of the health status modeling of the unmanned helicopter system are effectively improved. The first term in the loss function ensures that the model can accurately restore the original input data, improving the system's sensitivity and response speed to changes in flight status and abnormal working conditions. The second term constrains the distribution of potential features learned by the model to be consistent with the expected standard distribution, thereby enhancing the stability and adaptability of the model in the face of different tasks and environmental changes. The loss function can not only effectively prevent overfitting and improve the reliability of anomaly detection, but also provide a high-quality and reasonably distributed feature basis for health judgment, anomaly warning and adaptive control. The loss function design of the present invention significantly improves the adaptability, robustness and safety assurance level of the unmanned helicopter intelligent control system under complex working conditions.

[0082] S29. Using the normal operating data set Train the variational autoencoder model to minimize the loss function , optimize the parameters of the multi-channel encoder, joint latent space, conditional fusion mechanism, temporal recursive structure and decoder.

[0083] The present invention achieves in-depth characterization and potential distribution modeling of multi-source flight data of a six-blade linked coaxial twin-rotor unmanned helicopter by constructing a variational autoencoder model with a multi-channel encoder, a conditional fusion mechanism, and a time-series recursive structure on a normal working condition data set. It can not only fully tap the intrinsic correlation between multiple types of features such as blade speed, thrust, motor current, attitude angle, etc., but also dynamically integrate prior conditions such as mission type, flight phase, and working condition label to achieve accurate modeling of complex time series features and states under different flight mission scenarios. Through multi-head self-attention and conditional fusion mechanisms, the model's sensitivity to abnormal working conditions and small disturbances and its feature discrimination ability are effectively improved. The reconstruction error and potential distribution of the variational autoencoder are used as the loss function optimization target, so that the training process takes into account both global distribution consistency and local feature accuracy. The present invention greatly improves the state modeling accuracy and adaptability of the unmanned helicopter system to healthy working conditions.

[0084] In this embodiment, S3 specifically includes:

[0085] S31, using the trained variational autoencoder model, the standardized data set is The input data below , conditional variables And the historical multi-step time series hidden state is distributed reconstructed and inferred to obtain the reconstructed sequence and latent variables ;

[0086] S32. According to input data and reconstruction sequence , using multi-head self-attention mechanism and historical error trend modulation to calculate the dynamic weighted reconstruction error of the fusion time series trend :

[0087] ;

[0088] in, is the total number of feature channels, is the number of self-attention heads, For the Attention head at time For the first The adaptive weights learned by feature channels, and Respectively The actual value and reconstructed value of the feature, Under normal working conditions The variance of the features, For the The reconstruction error increment of a feature at the current and previous moments, is an adjustable parameter for modulating the trend effect;

[0089] Dynamically weighted reconstruction error It fully integrates the multi-head self-attention mechanism and historical error trend modulation to achieve adaptive sensitivity adjustment and anomaly identification for the multi-source state features of unmanned helicopters. By introducing self-attention weights, the system dynamically assigns weights based on the importance of each moment and each feature, highlighting the influence of key features in flight health judgment, effectively suppressing the interference of noise and non-critical information, and combined with the modulation of historical error trends, it reflects the abnormal trends of state changes in real time, significantly enhancing the early detection capabilities of sudden anomalies and slow drift faults. Compared with traditional static error measurement methods, dynamic weighted reconstruction error can automatically adapt to changes in feature distribution in different scenarios when facing multiple working conditions, multi-task switching, and complex disturbances, improving the accuracy and real-time performance of unmanned helicopter health monitoring.

[0090] S33. Use the standard Mahalanobis distance metric to calculate the difference between the mean of the latent variable at the current moment and the mean of the normal working condition :

[0091] ;

[0092] in, is the mean of the latent variable at the current moment, is the covariance matrix of the latent variables at the current moment, and are the mean and covariance matrix of the latent variables under normal working conditions;

[0093] difference The definition of standard Mahalanobis distance is adopted to measure the statistical distribution difference between the current health state of the unmanned helicopter and the normal working condition. By calculating the distance between the mean of the latent variable at the current moment and the mean of the normal working condition, and normalizing it with the covariance matrix of the normal working condition, it comprehensively reflects the correlation and actual variation between multidimensional features. It can not only accurately quantify the degree of deviation between the system state and the health distribution center, but also effectively filter out the influence caused by different feature variances and coordinated changes, making the judgment results more physically meaningful and engineering reliable.

[0094] S34. According to the unmanned helicopter mission type, flight phase and working condition label, a multi-scale dynamic threshold strategy is adopted to set adaptive thresholds for the dynamic weighted reconstruction error and difference metric of the fusion time series trend. 、 ;

[0095] S35, if it exists or , then the current working condition is marked as abnormal, and the disturbance feature tracing function is called Extract abnormal disturbance feature vector , where the disturbance feature tracing function is a function that combines the latent variable distribution with the dynamic weighted reconstruction error, inputs it into the multi-layer perceptron for nonlinear transformation and feature extraction, and outputs the current moment The health status judgment results and abnormal disturbance feature vector , Dynamic weighted reconstruction error integrating time series trends and latent variables .

[0096] The present invention introduces a multi-head self-attention mechanism and historical error trend modulation, combined with the Mahalanobis distance of adaptive covariance weights and a multi-scale dynamic threshold strategy, to achieve efficient distribution reconstruction inference and anomaly detection of the flight state of a six-propeller linked coaxial twin-rotor unmanned helicopter. It can calculate the dynamic weighted reconstruction error in real time based on standardized flight data and prior condition information, and integrate the Mahalanobis distance to quantify the degree of deviation of the latent variable distribution, greatly improving the sensitivity and robustness of health status discrimination. The use of a multi-layer perceptron to perform deep feature extraction on abnormal disturbance features significantly enhances the ability to identify abnormal types and trace disturbance sources. Compared with traditional static threshold discrimination or single error measurement methods, the present invention can adaptively respond to different mission types, flight phases and working conditions, and achieve accurate health monitoring and early warning of abnormalities in multiple scenarios of unmanned helicopters.

[0097] In this embodiment, the S4 specifically includes:

[0098] S41, based on standardized data set and abnormal disturbance feature vector , constructing adaptive nonlinear integral sliding surface :

[0099] ;

[0100] in, 、 、 、 is the sliding surface coefficient matrix, is the time series feature fusion function, is a high-order perturbation nonlinear mapping;

[0101] Adaptive nonlinear integral sliding surface The real-time operating status, historical time series characteristics, abnormal disturbance information, and high-order nonlinear dynamics of the unmanned helicopter are organically integrated to achieve accurate characterization and efficient response to the overall dynamics of the system. Compared with the traditional linear sliding surface method, the adaptive nonlinear integral sliding surface can adaptively adjust the sliding surface structure and weights according to changes in flight status and disturbance characteristics, effectively enhancing the control system's adaptability to complex multi-source disturbances and extreme operating conditions. By introducing integral, time series, and nonlinear disturbance terms, the sliding surface has stronger robustness and self-healing properties, which not only improves the suppression effect of external disturbances and internal anomalies, but also significantly reduces the impact and error accumulation caused by controller switching, significantly improving the flight stability and safety of the six-propeller linked coaxial twin-rotor unmanned helicopter in changing environments.

[0102] S42, according to the adaptive nonlinear integral sliding surface And its dynamic change characteristics, design an adaptive sliding mode control law, and define the control input as a vector , define each component as :

[0103] ;

[0104] in, For the Channel adaptive gain, For the The channel's auto-tuning function, For the Channel error signal, For the Adaptive gain of the channel, is a symbolic function, For the Fusion mapping of channel sliding mode variables, disturbances, and memory information, is the historical perturbation memory vector;

[0105] By adopting a vectorized sliding mode control law with independent control of each channel, the six-propeller linked coaxial twin-rotor unmanned helicopter system can output precise control instructions according to the actual operating status of each propeller or actuator, completely solving the practical problem that traditional scalar sliding mode control is difficult to adapt to multi-input and multi-output systems, enabling the control system to achieve more detailed and efficient dynamic adjustment when facing complex flight conditions and multi-source disturbances. The gain, self-tuning function and error signal of each channel can be independently and adaptively adjusted, improving the system's robustness and response speed to various abnormal and sudden disturbances. Independent control simplifies the calculation structure of the control law, facilitates embedded implementation and engineering application, and effectively avoids parameter tuning difficulties and system decoupling problems caused by coupling between channels.

[0106] S43, using the historical normalized disturbance weight mechanism, the abnormal disturbance feature vector To break it down:

[0107] ;

[0108] in, For historical windows The mean of the internal disturbance characteristics, is the short-term surge disturbance component, is the long-term stationary disturbance component, is the Euclidean norm;

[0109] The abnormal perturbation eigenvector Decomposition is performed by adaptively normalizing the difference between the current disturbance characteristics and the historical disturbance mean, achieving dynamic decentralization and integration of short-term surge disturbances and long-term stable disturbances. This enables the control system to automatically distinguish the impact of different disturbance components when faced with sudden anomalies and continuous small disturbances, and adjust the response weights of the control law based on the actual change trend of the disturbance. Compared with traditional methods that rely only on fixed thresholds or single anomaly measurements, this significantly improves the system's sensitivity and fine-tuning capabilities to complex disturbance conditions, avoids misjudgments and missed judgments, and improves the intelligence of fault-tolerant control and compensation mechanisms.

[0110] S44. Dynamic weighted reconstruction error based on health status judgment results and integration of time series trends and Mahalanobis distance metric , adaptively adjust the sliding mode controller structure;

[0111] S45, integrated AI fault tolerance compensation, using neural network prediction model Feedforward compensation is performed on the disturbance trend that may occur in the future, and the predicted compensation vector is ,in, is the feature splicing operation, are neural network parameters;

[0112] S46, predicting the compensation vector Multi-channel adaptive sliding mode control law Fusion, automatically distributes multi-propeller thrust and compensation torque, and generates specific thrust distribution control instructions;

[0113] S47, real-time update of adaptive nonlinear integral sliding surface With multi-channel adaptive sliding mode control law parameters, closed-loop adaptive adjustment and intelligent optimization are achieved;

[0114] S48 outputs the final multi-channel adaptive sliding mode control law, AI fault-tolerant compensation and thrust distribution control instructions to drive the robust adaptive optimization operation of the six-propeller linked coaxial twin-rotor unmanned helicopter under multi-source disturbances and extreme working conditions.

[0115] This invention achieves highly robust adaptive control of a six-propeller, coaxial, twin-rotor unmanned helicopter under complex operating conditions by integrating multiple advanced technologies, including an adaptive nonlinear integral sliding surface, a multi-channel adaptive sliding mode control law, AI fault-tolerant compensation, and historical disturbance weight decomposition. This design not only fully exploits standardized flight data and abnormal disturbance characteristics to construct an adaptive sliding mode control structure for multi-source disturbances, but also achieves rapid response and intelligent fault tolerance to different types of disturbances and abnormal conditions through multi-channel gain self-tuning, historical disturbance decomposition, and AI feedforward compensation. Compared with traditional fixed-parameter sliding mode control or single fault-tolerant schemes, this invention optimizes the controller structure and thrust distribution instructions in real time based on metrics such as health judgment results, dynamic error, and Mahalanobis distance, significantly improving the safety and control accuracy of the unmanned helicopter under extreme disturbances and highly dynamic environments. This invention provides an intelligent, real-time, and adaptive flight control solution for multi-propeller, highly redundant unmanned helicopter systems, promoting the innovative development of robust intelligent control technology for unmanned helicopters.

[0116] In this embodiment, the S5 specifically includes:

[0117] S51. Determine the actual flight mission type and target parameters of the unmanned helicopter according to the flight scheduling plan and current mission requirements, and collect mission-related environmental conditions;

[0118] S52. Analyze the health status determination results in real time to determine whether the current flight system is in a normal, abnormal, or extreme state;

[0119] S53. Based on the flight mission type, environmental conditions, and health status judgment results, query the control parameter switching strategy library to match the optimal control parameters;

[0120] S54, automatically switching flight control parameters, and loading the optimal control parameters into the flight control system in real time;

[0121] S55. After the control parameters are switched, continuously monitor the flight system's response and operational performance. If performance degradation or abnormal signals occur, dynamically adjust the control parameter group.

[0122] S56. Record the optimized control parameters, flight mission status, and adjustment logs in real time, and simultaneously push them to the upper-level task management and operation and maintenance platform.

[0123] The present invention realizes the adaptive optimization configuration of the control parameters of the six-propeller linked coaxial twin-rotor unmanned helicopter through the intelligent calling of the task type identification, environmental condition collection, real-time analysis of health status and parameter switching strategy library. Compared with the traditional control method that relies on manual experience or fixed switching logic, the present invention can automatically query and load the optimal control parameters according to the flight mission requirements, environmental changes and health judgment results, which significantly improves the accuracy and response speed of parameter switching. After the control parameters are switched, the system can continuously monitor the flight response and operating performance, promptly identify performance degradation or abnormal signals, and dynamically adjust the parameter group to achieve closed-loop optimization and self-healing capabilities. All optimization processes and adjustment logs will be recorded and reported in real time, which is convenient for upper-level operation and maintenance management and task decision-making, and realizes data traceability and intelligent management throughout the life cycle. The present invention effectively improves the adaptability, operating efficiency and system security of unmanned helicopters in multi-task and multi-working condition scenarios.

[0124] refer to Figure 2 The six-propeller linked coaxial twin-rotor unmanned helicopter system includes the following modules:

[0125] The data acquisition and preprocessing module is used to collect and preprocess multi-source operating status data during the flight of the unmanned helicopter and generate a standardized data set;

[0126] Normal operating condition modeling module, used to filter normal operating condition data in the standardized dataset and train the variational autoencoder model;

[0127] The health status identification and feature extraction module is used to input a standardized data set into the variational autoencoder model, identify the health status, and extract abnormal disturbance features when abnormal;

[0128] Adaptive sliding mode control and fault tolerance module, which is used to construct the integral sliding mode surface, design the sliding mode control law, and trigger torque compensation and fault tolerance control in real time based on the standardized data set and abnormal disturbance characteristics;

[0129] The parameter switching and optimization control module is used to combine multi-source operating status data and health judgment results, intelligently switch control parameters, and achieve adaptive optimization control.

[0130] Example 1:

[0131] To verify the feasibility of this invention, the project team applied it to a power inspection project in a mountainous area. Using the proposed six-blade, coaxial, twin-rotor unmanned helicopter system and adaptive control method, they conducted large-scale field flight tests. The area's terrain is highly undulating and subject to frequent and significant wind speed fluctuations. Conventional unmanned helicopters are prone to flight attitude deviations, mission interruptions, and increased energy consumption in similar environments, severely impacting operational efficiency and equipment safety.

[0132] In the inspection scenario, the unmanned helicopter departed from the substation and inspected 10 towers along a predetermined route, covering a total distance of approximately 12 kilometers and an estimated operation duration of 52 minutes. Wind speeds during takeoff were 4.5 meters per second, reaching a maximum of 13.2 meters per second during cruising. The average temperature at the site was 17°C, with a humidity of 63%. The aircraft collected real-time data on multiple sources, including propeller speed, motor current, three-axis attitude angles, acceleration, and wind speed and direction. This data was standardized using the data acquisition and preprocessing module. Based on previous low-disturbance normal operating conditions, the system used a variational autoencoder to model the health status distribution and monitor flight status in real time. During one actual flight, when inspecting the fourth tower, a sudden crosswind gust suddenly increased the wind speed to 12.6 meters per second. The motor current fluctuation jumped from a normal 0.5 amps to 2.2 amps, and the roll angle deviation increased from 1.6 degrees to 8.9 degrees. The reconstruction error of the system variational autoencoder increased from 0.022 to 0.108, the Mahalanobis distance increased from 1.7 to 4.2, and the anomaly discrimination module responded accurately within 0.3 seconds.

[0133] The sliding mode control parameters were then adaptively adjusted, activating the torque compensation and thrust distribution mechanisms. The AI ​​fault-tolerant compensation module predicted wind field trends based on disturbance characteristics, enabling early correction of control variables. This shortened the aircraft's attitude recovery time to 2.7 seconds, significantly reducing the risk of flight deviation and secondary disturbances. The system intelligently switched to the optimal control parameter set based on different mission phases, such as inspection, hovering, and fixed-point positioning, maintaining excellent control performance even during subsequent high-wind disturbance phases.

[0134] During the entire testing period, unmanned helicopters using the system developed in this invention completed 28 inspections, totaling 1,220 minutes of flight time. Compared to a traditional PID + conventional sliding mode control scheme, the number of flight deviations exceeded standards was reduced by 86%, the mission completion rate increased by 8.1%, and energy consumption under high wind disturbances was reduced by 9.6%. All health checks, anomaly detections, parameter switching, and other logs are recorded in real time, supporting operational review and decision-making.

[0135] Table 1 Comparative data of wind farm inspection flights between the present invention and traditional methods

[0136]

[0137] As can be seen from Table 1, the six-blade linked coaxial twin-rotor unmanned helicopter system and adaptive control method proposed by the present invention have shown significantly better performance than traditional control schemes in complex wind field inspection tasks. In the maximum wind speed environment, whether the wind speed is as high as 12.6 meters per second or 13.2 meters per second, the unmanned helicopter of the present invention can still stably complete the flight mission. Although the current fluctuation peak and roll angle deviation in some batches are comparable to or even slightly higher than those of the traditional scheme, the system of the present invention greatly shortens the attitude recovery time through intelligent health judgment, adaptive adjustment of sliding mode control parameters and AI fault-tolerant compensation. For example, in batches 1 and 2 of the present invention, the attitude recovery time was 2.7 seconds and 2.8 seconds respectively, which is only about one-third of that of traditional schemes 1 and 2. The unmanned helicopter can recover to a stable state more quickly when encountering strong winds and disturbances.

[0138] The number of flight deviations exceeding the standard using the system of the present invention was zero, while the traditional solution still experienced deviations exceeding the standard, indicating that the present invention has significantly improved the robustness and safety margin against extreme disturbances. In terms of mission completion rate, each batch of the present invention approached or reached 100%, while the traditional solution was slightly lower, at 90.2% and 92.7%, respectively, demonstrating the new system's significant reliability advantage in complex environments. From the perspective of average energy consumption, the energy consumption of all batches of the present invention's solution was lower than that of the traditional solution, with the energy consumption of batch 2 of the present invention being the lowest, at only 20.7Wh, demonstrating outstanding energy-saving effects.

[0139] In summary, the data in Table 1 fully demonstrates that the present invention not only improves the stability and safety of the unmanned helicopter in challenging scenarios such as strong wind disturbances and complex task switching, but also effectively reduces energy consumption and operation and maintenance costs, significantly enhancing the system's intelligent adaptability and practical engineering application value.

[0140] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. An adaptive control method for a six-propeller linked coaxial twin-rotor unmanned helicopter, characterized in that: The steps include: S1. Collect multi-source operating status data of a six-propeller linked coaxial twin-rotor unmanned helicopter during flight, pre-process the multi-source operating status data, and generate a standardized data set; S2. Filter flight data of unmanned helicopters in a fault-free, disturbance-free state with stable indicators from a standardized dataset, construct a normal operating condition dataset, train the variational autoencoder model, and establish the potential distribution structure of the flight system. S3. Input the standardized dataset into the trained variational autoencoder model to perform distribution reconstruction inference, determine the current health status of the flight system, and determine whether the reconstruction error or latent variable distribution exceeds a preset threshold. If so, mark the current operating condition as abnormal and extract the abnormal disturbance feature vector. S4. Based on the standardized data set and the abnormal disturbance eigenvector, an integral sliding mode surface is constructed, a variable structure integral sliding mode control law is designed, and the sliding mode control gain parameters related to the integral sliding mode surface are adjusted. When abnormal operating conditions or extreme disturbances are detected, the controller parameters are automatically adjusted to trigger torque compensation and fault-tolerant control mechanisms in real time. S5. Combine the actual flight mission type and environmental conditions, and according to the health status judgment results, intelligently switch the control parameters of the unmanned helicopter to achieve adaptive optimization control under different flight missions; The S2 specifically includes: S21. Extracting blade speed, thrust, motor current, aircraft attitude angle, mission type, flight phase, and operating condition labels from the generated standardized data set; S22. Based on the unmanned helicopter operation log and system status labels, filter the flight data under stable, fault-free, and disturbance-free conditions from the standardized data set to construct a normal operating condition data set. ; S23, based on normal working condition data set , design a variational autoencoder model, which includes a multi-channel encoder, a joint latent space, a conditional fusion mechanism, a temporal recursive structure and a decoder. The multi-channel encoder uses independent neural network branches to encode different types of features. The output of each encoding branch is spliced ​​into a comprehensive feature vector, which is mapped to the latent space through a fully connected layer. The conditional fusion mechanism uses the mission type, flight phase or working condition label as a conditional variable Combined with the comprehensive feature vector as input, the temporal recursive structure uses a gated recurrent unit network at both the encoding and decoding ends to achieve temporal modeling of multi-time data sequences. The decoder consists of a fully connected neural network and a recursive structure. S24, the normal working condition data set Input the multi-channel encoder of the variational autoencoder model to encode each type of feature separately to obtain the feature channel encoding result , extract continuous historical multi-step time series features from the standardized data set , using the temporal recursive structure of the variational autoencoder model to perform temporal recursive modeling on the historical multi-step temporal features and extract the temporal hidden state ,in, is the number of historical time series steps; S25, encoding results of each feature channel Splicing to form a comprehensive feature vector , the comprehensive feature vector , condition variables and temporal hidden states After weighted fusion of multi-head self-attention and conditional fusion mechanism, adaptive attention weights and gating coefficients are assigned to different feature channels, historical moments and conditional variables to obtain the fusion representation , the fusion representation Input the joint latent space of the variational autoencoder model to obtain dynamically adaptive latent variables ; S26. The latent variables obtained and condition variables Input the decoder of the variational autoencoder model, and restore the reconstruction results of each source feature by the fully connected neural network and recursive structure to obtain the reconstructed sequence of multi-source features ; S27. In the decoder, the long short-term memory network is based on the latent variables , condition variables and temporal hidden states Recursively generate reconstruction sequences of multi-source features ; S28. Reconstruction sequence based on multi-source features , construct the loss function of the variational autoencoder model ; S29. Using the normal operating data set Train the variational autoencoder model to minimize the loss function , optimize the parameters of the multi-channel encoder, joint latent space, conditional fusion mechanism, temporal recursive structure and decoder.

2. The adaptive control method for a six-blade linked coaxial twin-rotor unmanned helicopter according to claim 1, characterized in that: The multi-source operating status data specifically includes each blade speed, thrust, motor current, aircraft attitude angle, mission type, flight phase and operating condition label.

3. The adaptive control method for a six-blade linked coaxial twin-rotor unmanned helicopter according to claim 1, characterized in that: The preprocessing of multi-source operating status data specifically includes data cleaning, missing value filling, outlier removal and normalization.

4. The adaptive control method for a six-blade linked coaxial twin-rotor unmanned helicopter according to claim 1, characterized in that: The S3 specifically includes: S31, using the trained variational autoencoder model, the standardized data set is The input data below , condition variables And the historical multi-step time series hidden state is distributed reconstructed and inferred to obtain the reconstructed sequence and latent variables ; S32. According to input data and reconstruction sequence , using multi-head self-attention mechanism and historical error trend modulation to calculate the dynamic weighted reconstruction error of the fusion time series trend ; S33, Mahalanobis distance calculation using adaptive covariance weights and difference measurement between normal operating conditions and statistical distribution : S34. According to the unmanned helicopter mission type, flight phase and working condition label, a multi-scale dynamic threshold strategy is adopted to set adaptive thresholds for the dynamic weighted reconstruction error and difference metric of the fusion time series trend. 、 ; S35, if it exists or , then the current working condition is marked as abnormal, and the disturbance feature tracing function is called Extract abnormal disturbance feature vector , where the disturbance feature tracing function is a function that combines the latent variable distribution with the dynamic weighted reconstruction error, inputs it into the multi-layer perceptron for nonlinear transformation and feature extraction, and outputs the current moment The health status judgment results and abnormal disturbance feature vector , Dynamic weighted reconstruction error integrating time series trends and latent variables .

5. The adaptive control method for a six-blade linked coaxial twin-rotor unmanned helicopter according to claim 1, characterized in that: The S4 specifically includes: S41, based on standardized data set and abnormal disturbance feature vector , constructing adaptive nonlinear integral sliding surface ; S42, according to the adaptive nonlinear integral sliding surface and dynamic change characteristics, design a multi-channel adaptive sliding mode control law ; S43, using the historical normalized disturbance weight mechanism, the abnormal disturbance feature vector to decompose; S44. Dynamic weighted reconstruction error based on health status judgment results and integration of time series trends and Mahalanobis distance metric , adaptively adjust the sliding mode controller structure; S45, integrated AI fault tolerance compensation, using neural network prediction model Feedforward compensation is performed on the disturbance trend that may occur in the future, and the predicted compensation vector is ,in, is the feature splicing operation, are neural network parameters; S46, predicting the compensation vector Multi-channel adaptive sliding mode control law Fusion, automatically distributes multi-propeller thrust and compensation torque, and generates specific thrust distribution control instructions; S47, real-time update of adaptive nonlinear integral sliding surface With multi-channel adaptive sliding mode control law parameters, closed-loop adaptive adjustment and intelligent optimization are achieved; S48 outputs the final multi-channel adaptive sliding mode control law, AI fault-tolerant compensation and thrust distribution control instructions to drive the robust adaptive optimization operation of the six-propeller linked coaxial twin-rotor unmanned helicopter under multi-source disturbances and extreme working conditions.

6. The adaptive control method for a six-blade linked coaxial twin-rotor unmanned helicopter according to claim 1, characterized in that: The S5 specifically includes: S51. Determine the actual flight mission type and target parameters of the unmanned helicopter according to the flight scheduling plan and current mission requirements, and collect mission-related environmental conditions; S52. Analyze the health status determination results in real time to determine whether the current flight system is in a normal, abnormal, or extreme state; S53. Based on the flight mission type, environmental conditions, and health status judgment results, query the control parameter switching strategy library to match the optimal control parameters; S54, automatically switching flight control parameters, and loading the optimal control parameters into the flight control system in real time; S55. After the control parameters are switched, continuously monitor the flight system's response and operational performance. If performance degradation or abnormal signals occur, dynamically adjust the control parameter group. S56. Record the optimized control parameters, flight mission status, and adjustment logs in real time, and simultaneously push them to the upper-level task management and operation and maintenance platform.

7. A six-blade linkage coaxial twin-rotor unmanned helicopter system, applied to the six-blade linkage coaxial twin-rotor unmanned helicopter adaptive control method according to any one of claims 1 to 6, characterized in that: Includes the following modules: The data acquisition and preprocessing module is used to collect and preprocess multi-source operating status data during the flight of the unmanned helicopter and generate a standardized data set; Normal operating condition modeling module, used to filter normal operating condition data in the standardized dataset and train the variational autoencoder model; The health status identification and feature extraction module is used to input a standardized data set into the variational autoencoder model, identify the health status, and extract abnormal disturbance features when abnormal; Adaptive sliding mode control and fault tolerance module, which is used to construct the integral sliding mode surface, design the sliding mode control law, and trigger torque compensation and fault tolerance control in real time based on the standardized data set and abnormal disturbance characteristics; The parameter switching and optimization control module is used to combine multi-source operating status data and health judgment results, intelligently switch control parameters, and achieve adaptive optimization control.

Images