UAV health assessment method and system based on multi-channel model

Abstract

The present invention relates to the technical field of predictive maintenance of unmanned aerial vehicles, and provides a method and system for health assessment of unmanned aerial vehicles based on a multi-channel model, the method comprising: S1, determining the input parameters of the health assessment model of the unmanned aerial vehicle engine; S2, collecting sensor data under the working state of the unmanned aerial vehicle engine, performing standardized and structured processing, and forming a data set; S3, establishing a multi-channel neural network model; S4, training the multi-channel neural network model; S5, using the multi-channel neural network model to evaluate the health status of the unmanned aerial vehicle. The present invention is based on a multi-channel neural network model, and comprehensively considers the timing parameters collected by the sensor and the operating environment when evaluating the health status of the unmanned aerial vehicle engine. Compared with the health assessment method that only establishes a deep learning model based on time series data, it can effectively improve the classification accuracy of various health states, thereby improving the effectiveness of predictive maintenance of unmanned aerial vehicle engines and reducing maintenance costs.

Description

Technical Field

[0001] The present invention relates to the technical field of predictive maintenance of unmanned aerial vehicles (UAVs), and in particular to a health assessment method and system for unmanned aerial vehicles (UAVs) based on a multi-channel model. Background Art

[0002] With the increasing application of drones in transportation, agriculture, emergency response and other fields, how to ensure the reliability of drone working status has become a research hotspot in drone technology in recent years. Predictive maintenance is an important method to ensure the reliability of drones. It refers to real-time monitoring of the operating status of equipment based on various sensors installed on the equipment, identifying equipment failures and hidden dangers at work based on expert knowledge or deep learning models, and pushing decisions and suggestions such as maintenance, replacement of parts, and shutdown detection to equipment maintenance personnel. Compared with other maintenance methods, predictive maintenance has the characteristics of intelligence, reduced maintenance costs, and increased non-stop operation time.

[0003] Obviously, the accuracy and effectiveness of predictive maintenance of UAV engines depends on the prediction accuracy of the health status identification model. The deep learning models used in existing methods are mostly based on time series data generated by the working state of the UAV engine, such as engine speed, cylinder head temperature, exhaust temperature, lubricating oil pressure, and lubricating oil temperature, to identify and classify the health status. This method is difficult to consider the operating environment of the UAV, such as the difference in health status standards of UAVs at different altitudes. To this end, the present invention proposes a UAV health status assessment method based on a multi-channel model, which realizes the health status assessment of the UAV under the premise of considering the operating environment of the UAV, and then realizes predictive maintenance. Summary of the invention

[0004] The purpose of the present invention is to overcome the shortcomings of the prior art: the deep learning model in the current UAV engine health assessment method is difficult to consider the impact of the UAV operating environment on the health assessment status, and a UAV health assessment method and system based on a multi-channel model is provided, which can effectively improve the classification accuracy of various health statuses, thereby improving the effectiveness of predictive maintenance of UAV engines.

[0005] The present invention adopts the following technical solution:

[0006] On the one hand, the present invention provides a method for evaluating the health of a drone based on a multi-channel model, characterized in that the method comprises:

[0007] S1. Determine the input parameters of the UAV engine health assessment model;

[0008] S2. Collect sensor data of the UAV engine in working state, perform standardization and structured processing to form a data set, and assign a label to each set of data in the data set, that is, the output of the data set, that is, the prediction result of the multi-channel model;

[0009] S3. According to the input parameters determined in step S1, a long short-term memory network LSTM branch for extracting time series data features, a fully connected neural network branch for extracting operating environment data features, a fully connected network and a Softmax layer for merging features and classifying are established, and connected in series to form a multi-channel neural network model. Such a model can extract features from flight data and environmental data respectively, and will not infect each other before the features are merged and input into the fully connected network, thereby affecting the accuracy of the results;

[0010] S4, training the multi-channel neural network model established in step S3 based on the data set in step S2 to obtain a multi-channel neural network model for evaluating the health status of the UAV engine;

[0011] S5. Evaluate the health status of the UAV using the multi-channel neural network model for evaluating the health status of the UAV engine obtained in step S4.

[0012] As any possible implementation described above, a further implementation is provided, in step S1, the input parameters include time series data collected by the sensor, and operating environment data of the drone in a working state.

[0013] Any possible implementation as described above further provides an implementation, wherein the time series data collected by the sensor includes engine speed, cylinder head temperature, exhaust temperature, lubricating oil pressure, and lubricating oil temperature; and the operating environment data includes flight altitude, altitude, ambient temperature, and wind speed.

[0014] According to any possible implementation described above, a further implementation is provided, wherein step S2 specifically includes:

[0015] S2.1 collects sensor timing data under the working state of the UAV engine and records the operating environment data;

[0016] S2.2 normalizes the data obtained in step S2.1. The specific method is shown in formula (1):

[0017]

[0018] Among them, x′0 is the standardized data, x0 is the original data, is the average value of all data of the input variable to which the data belongs, and S is the standard deviation of all data of the input variable to which the data belongs;

[0019] S2.3 determines the step length of a single time series data, slices the standardized data (including the test data or simulation data) according to the step length, and assigns a health status label to each slice;

[0020] S2.4 combines the slice time series data, operating environment data, and health status labels to form a data set. The data structure is shown in formula (2):

[0021]

[0022] where x0…x n For operating environment data, is the time series data of the mth dimension at the tth time, and y is the health status label;

[0023] S2.5 divides the data set obtained in step S2.4 into an experimental data training set and a test set according to a certain ratio.

[0024] As any possible implementation described above, an implementation is further provided, in step S2.3, the health status label includes but is not limited to spark plug dirtiness, single magneto failure, lubricating oil system failure, lubricating oil shortage, and engine knock.

[0025] As any possible implementation described above, a further implementation is provided, in step S2.5, the data ratio of the experimental data training set to the test set is 8:2.

[0026] Any possible implementation described above further provides an implementation, in which in step S3, a multi-channel neural network model is established, specifically including:

[0027] S3.1 establish a long short-term memory network LSTM branch for extracting time series data features, and determine the hyperparameters of the network, which include: the feature dimension of the input data, the dimension of the hidden layer, the number of neural network layers, and the activation function;

[0028] S3.2 establish a fully connected neural network branch for extracting the characteristics of the operating environment data, and determine the hyperparameters of the network, wherein the hyperparameters include the number of layers of the network, the number of neurons in each layer, and the activation function of each layer;

[0029] S3.3 establishes a fully connected network and a Softmax layer for merging features and classifying, and determines the hyperparameters of the network, which include the number of layers of the network, the number of neurons in each layer, and the activation function of each layer.

[0030] On the other hand, the present invention also provides a UAV health assessment system based on a multi-channel model, comprising:

[0031] The data collection and processing module is used to determine the input parameters of the UAV engine health assessment model, collect the sensor data of the UAV engine under working conditions, perform standardization and structured processing, and form a data set;

[0032] The multi-channel neural network model building module builds a long short-term memory network LSTM branch for extracting time series data features, a fully connected neural network branch for extracting operating environment data features, a fully connected network and a Softmax layer for merging features and classifying, and connects them in series to form a multi-channel neural network model;

[0033] A multi-channel neural network model training module is used to train the multi-channel neural network model with the data set processed by the data collection and processing module to obtain a multi-channel neural network model for evaluating the health status of the UAV engine;

[0034] The UAV health status assessment module uses the multi-channel neural network model for UAV engine health status assessment to assess the UAV health status.

[0035] The beneficial effects of the present invention are:

[0036] The present invention is based on a multi-channel neural network model, which can comprehensively consider the timing parameters collected by sensors and the operating environment when evaluating the health status of UAV engines. Compared with the health assessment method that only establishes a deep learning model based on time series data, it can effectively improve the classification accuracy of various health statuses, thereby improving the effectiveness of predictive maintenance of UAV engines, reducing maintenance costs, and improving maintenance efficiency and UAV use efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 The figure is a flow chart of a method for evaluating the health of a UAV based on a multi-channel model according to an embodiment of the present invention.

[0038] Figure 2 Shown is a network architecture diagram of a multi-channel neural network model in an embodiment.

[0039] Figure 3 Shown is the LSTM network neuron structure. DETAILED DESCRIPTION

[0040] The specific embodiments of the present invention will be described in detail below in conjunction with specific drawings. It should be noted that the technical features or combinations of technical features described in the following embodiments should not be considered isolated, and they can be combined with each other to achieve better technical effects.

[0041] like Figure 1 As shown, an embodiment of the present invention provides a drone health assessment method based on a multi-channel model, including:

[0042] S1. Determine the input parameters of the UAV engine health assessment model;

[0043] S2. Collect sensor data of the UAV engine in working state, perform standardization and structured processing to form a data set, and assign a label to each set of data in the data set, that is, the output of the data set, that is, the prediction result of the multi-channel model;

[0044] S3. According to the input parameters determined in step S1, a long short-term memory network LSTM branch for extracting time series data features, a fully connected neural network branch for extracting operating environment data features, a fully connected network and a Softmax layer for merging features and classifying are established, and they are connected in series to form a multi-channel neural network model. Such a model can extract features from flight data and environmental data respectively, and will not infect each other before the features are merged and input into the fully connected network, thereby affecting the accuracy of the results; the network architecture of the multi-channel neural network model is as follows Figure 2 As shown;

[0045] S4, training the multi-channel neural network model established in step S3 based on the data set in step S2 to obtain a multi-channel neural network model for evaluating the health status of the UAV engine;

[0046] S5. Evaluate the health status of the UAV using the multi-channel neural network model for evaluating the health status of the UAV engine obtained in step S4.

[0047] In a specific embodiment, in step S1, the input parameters include time series data collected by sensors and operating environment data of the drone in working state.

[0048] In a specific embodiment, the time series data collected by the sensor includes engine speed, cylinder head temperature, exhaust temperature, lubricating oil pressure, and lubricating oil temperature; the operating environment data includes flight altitude, altitude, ambient temperature, and wind speed.

[0049] In a specific embodiment, step S2 specifically includes:

[0050] S2.1 collects sensor timing data under the working state of the UAV engine and records the operating environment data;

[0051] S2.2 normalizes the data obtained in step S2.1. The specific method is shown in formula (1):

[0052]

[0053] Among them, x′0 is the standardized data, x0 is the original data, is the average value of all data of the input variable to which the data belongs, and S is the standard deviation of all data of the input variable to which the data belongs;

[0054] S2.3 determines the step length of a single time series data, slices the standardized data (including the stress test data and the simulation data) according to the step length, and assigns a health status label to each slice;

[0055] S2.4 combines the slice time series data, operating environment data, and health status labels to form a data set. The data structure is shown in formula (2):

[0056]

[0057] where x0…x n For operating environment data, is the time series data of the mth dimension at the tth time, and y is the health status label;

[0058] S2.5 divides the data set obtained in step S2.4 into an experimental data training set and a test set according to a certain ratio.

[0059] In a specific embodiment, in step S2.3, the health status tag includes but is not limited to spark plug dirtiness, single magneto failure, oil system failure, oil shortage, and engine knock.

[0060] In a specific embodiment, in step S2.5, the data ratio of the experimental data training set to the test set is 8:2.

[0061] In a specific embodiment, in step S3, establishing a multi-channel neural network model specifically includes:

[0062] S3.1 establish a long short-term memory network LSTM branch for extracting time series data features, and determine the hyperparameters of the network, which include: the feature dimension of the input data, the dimension of the hidden layer, the number of neural network layers, and the activation function;

[0063] S3.2 establish a fully connected neural network branch for extracting the characteristics of the operating environment data, and determine the hyperparameters of the network, wherein the hyperparameters include the number of layers of the network, the number of neurons in each layer, and the activation function of each layer;

[0064] S3.3 establishes a fully connected network and a Softmax layer for merging features and classifying, and determines the hyperparameters of the network, which include the number of layers of the network, the number of neurons in each layer, and the activation function of each layer.

[0065] The neuron structure of LSTM network is as follows Figure 3 shown.

[0066] An embodiment of the present invention provides a drone health assessment system based on a multi-channel model, comprising:

[0067] The data collection and processing module is used to determine the input parameters of the UAV engine health assessment model, collect the sensor data of the UAV engine under working conditions, perform standardization and structured processing, and form a data set;

[0068] The multi-channel neural network model building module builds a long short-term memory network LSTM branch for extracting time series data features, a fully connected neural network branch for extracting operating environment data features, a fully connected network and a Softmax layer for merging features and classifying, and connects them in series to form a multi-channel neural network model;

[0069] A multi-channel neural network model training module is used to train the multi-channel neural network model with the data set processed by the data collection and processing module to obtain a multi-channel neural network model for evaluating the health status of the UAV engine;

[0070] The UAV health status assessment module uses the multi-channel neural network model for UAV engine health status assessment to assess the UAV health status.

[0071] The present invention is compared with the health assessment method that does not consider environmental data under different environments (10°C and 0°C), and the results are shown in Table 1:

[0072] Table 1 Comparison of health assessment methods

[0073]

[0074]

[0075] This verifies the beneficial effects of the present invention: based on a multi-channel neural network model, the present invention can comprehensively consider the timing parameters collected by sensors and the operating environment when evaluating the health status of UAV engines. Compared with the health assessment method that only establishes a deep learning model based on time series data, it can effectively improve the classification accuracy of various health statuses, thereby improving the effectiveness of predictive maintenance of UAV engines, reducing maintenance costs, and improving maintenance efficiency and UAV use efficiency.

[0076] Although several embodiments of the present invention have been given herein, those skilled in the art should understand that the embodiments of the present invention may be modified without departing from the spirit of the present invention. The above embodiments are merely exemplary and should not be used as a limitation on the scope of the present invention.

Claims

1. A UAV health assessment method based on a multi-channel model, characterized in that: The method comprises: S1. Determine the input parameters of the UAV engine health assessment model; the input parameters include time series data collected by sensors and operating environment data of the UAV in working state; the time series data collected by the sensors include engine speed, cylinder head temperature, exhaust temperature, lubricating oil pressure, and lubricating oil temperature; the operating environment data include flight altitude, altitude, ambient temperature, and wind speed; S2. Collect sensor data of the UAV engine in working state, perform standardization and structured processing to form a data set, and assign a label to each set of data in the data set; S3, according to the input parameters determined in step S1, establish a long short-term memory network LSTM branch for extracting time series data features, a fully connected neural network branch for extracting operating environment data features, a fully connected network and a Softmax layer for merging features and classifying, and connect them in series to form a multi-channel neural network model; S4, training the multi-channel neural network model established in step S3 based on the data set in step S2 to obtain a multi-channel neural network model for evaluating the health status of the UAV engine; S5, using the multi-channel neural network model for evaluating the health status of the UAV engine obtained in step S4 to evaluate the health status of the UAV; Step S2 specifically includes: S2.1 collects sensor timing data under the working state of the UAV engine and records the operating environment data; S2.2 Standardize the data obtained in step S2.

1. The specific method is shown in formula (1): （1） in is the standardized data, is the original data, is the average value of all the data of the input variable to which the data belongs. is the standard deviation of all data of the input variable; S2.3 Determine the step size of a single time series data, slice the standardized data according to the step size, and assign a health status label to each slice; S2.4 Merge the slice time series data, operating environment data, and health status labels to form a data set. The data structure is shown in formula (2): （2） in For operating environment data, For the The moment dimensional time series data, Label for health status; S2.5 dividing the data set obtained in step S2.4 into a test data training set and a test set according to a certain ratio; In step S3, a multi-channel neural network model is established, which specifically includes: S3.1 Establish a long short-term memory network LSTM branch for extracting time series data features, and determine the hyperparameters of the network, which include: the feature dimension of the input data, the dimension of the hidden layer, the number of neural network layers, and the activation function; S3.2 establish a fully connected neural network branch for extracting the characteristics of the operating environment data, and determine the hyperparameters of the network, wherein the hyperparameters include the number of layers of the network, the number of neurons in each layer, and the activation function of each layer; S3.3 Establish a fully connected network and a Softmax layer for merging features and classification, and determine the hyperparameters of the network, which include the number of layers of the network, the number of neurons in each layer, and the activation function of each layer.

2. The method for evaluating the health of a drone based on a multi-channel model as claimed in claim 1, characterized in that: In step S2.1, the timing data includes stress test data or simulation data.

3. The UAV health assessment method based on a multi-channel model as claimed in claim 1, characterized in that: In step S2.3, the health status tags include but are not limited to dirty spark plugs, single magneto failure, oil system failure, insufficient oil, and engine knock.

4. The method for evaluating the health of a drone based on a multi-channel model as claimed in claim 1, wherein: In step S2.5, the data ratio of the experimental data training set to the test set is 8:

2.

5. A drone health assessment system based on a multi-channel model, characterized in that: The system is used to implement the method according to any one of claims 1 to 4, and the system includes: The data collection and processing module is used to determine the input parameters of the UAV engine health assessment model, collect the sensor data of the UAV engine under working conditions, perform standardization and structured processing, and form a data set; The multi-channel neural network model building module builds a long short-term memory network LSTM branch for extracting time series data features, a fully connected neural network branch for extracting operating environment data features, a fully connected network and a Softmax layer for merging features and classifying, and connects them in series to form a multi-channel neural network model; A multi-channel neural network model training module is used to train the multi-channel neural network model with the data set processed by the data collection and processing module to obtain a multi-channel neural network model for evaluating the health status of the UAV engine; The UAV health status assessment module uses the multi-channel neural network model for UAV engine health status assessment to assess the UAV health status.

Images