Rotary inertial navigation state monitoring method based on IMU information fusion and dual-channel CNN

Abstract

The invention relates to a rotary inertial navigation state monitoring method based on IMU (Inertial Measurement Unit) information fusion and a dual-channel CNN (Convolutional Neural Network), which belongs to the technical field of inertial navigation and comprises the following steps: constructing a dual-channel CNN model; dividing the working state of the rotary inertial navigation; collecting IMU information of the rotary inertial navigation in different working states as a data set, performing fusion preprocessing on the IMU information in the training set and the verification set, and inputting the IMU information in the training set after fusion preprocessing into a dual-channel CNN model for repeated training to obtain a trained dual-channel CNN model; and acquiring IMU information of the rotary inertial navigation in real time, performing fusion preprocessing on the IMU information of the rotary inertial navigation acquired in real time, and inputting the IMU information of the rotary inertial navigation acquired in real time after fusion preprocessing into the trained dual-channel CNN model to obtain the working state of the rotary inertial navigation monitored in real time. According to the invention, the working state of the rotary inertial navigation can be automatically, simply, accurately and quickly obtained.

Description

Technical Field

[0001] The present invention belongs to the technical field of inertial navigation, and in particular relates to a rotational inertial navigation state monitoring method based on IMU information fusion and dual-channel CNN. Background Art

[0002] With the rapid development of inertial navigation technology, rotating inertial navigation (RI) has become a research hotspot due to its unique advantages in high-precision, long-duration passive navigation applications. By introducing rotational modulation technology, rotating RI effectively suppresses the accumulation and divergence of inertial navigation errors, showing broad prospects in high-precision, long-duration passive navigation applications, as well as medium- to high-precision, low-cost inertial navigation applications.

[0003] The introduction of an indexing mechanism inevitably increases the likelihood of system failure. Key components within the indexing mechanism (such as motors, bearings, and encoders) are susceptible to wear, overload, and aging over time, leading to degraded system performance and even potential safety hazards. Therefore, real-time monitoring of the operating status of the rotating inertial navigation system and early identification of potential failures are crucial.

[0004] Currently, fault detection in rotating machinery and equipment usually relies on various types of sensors to monitor the operating status of the equipment, including information such as vibration, acceleration, temperature, and sound field. By combining signal analysis, empirical knowledge, and machine learning, condition monitoring and fault diagnosis can be achieved without disassembling the equipment. For example, the invention patent with publication number CN117407784A proposes a method and system for intelligent fault diagnosis of rotating machinery for sensor data anomalies, the invention patent with publication number CN113076834A discloses a method, processing system, processing terminal, and medium for processing fault information of rotating machinery, and the invention patent with publication number CN113008583A proposes a method and device for rotating machinery condition monitoring and automatic abnormal alarm. However, there is currently no mature condition monitoring method for rotating inertial navigation. Summary of the Invention

[0005] In view of the shortcomings of the existing technology, the purpose of the present invention is to provide a rotation inertial navigation state monitoring method based on IMU information fusion and dual-channel CNN to solve or improve the defects in the existing technology.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a rotation inertial navigation state monitoring method based on IMU information fusion and dual-channel CNN, comprising the following steps: S1. Build a dual-channel CNN model; S2. Divide the working state of the rotating inertial navigation system; S3. Collect IMU information under different working states of the rotating inertial navigation system as a data set, and randomly divide the data set into a training set and a validation set; perform fusion preprocessing on the IMU information in the training set and the validation set, and input the fused preprocessed IMU information in the training set into the dual-channel CNN model for repeated training until the accuracy of verifying the fused preprocessed IMU information in the validation set using the trained model reaches a predetermined value, and then stop training to obtain a trained dual-channel CNN model; S4. Real-time acquisition of the IMU information of the rotating inertial navigation system, fusion preprocessing of the real-time acquired IMU information of the rotating inertial navigation system, and input of the fusion preprocessed real-time acquired IMU information of the rotating inertial navigation system into the trained dual-channel CNN model to obtain the working status of the rotating inertial navigation system monitored in real time.

[0007] Preferably, in step S3 and step S4, the IMU information includes a three-axis gyroscope signal and a three-axis accelerometer signal.

[0008] Preferably, in step S3 and step S4, the specific method of the fusion preprocessing is: S21. Perform signal fusion on the three-axis gyroscope signal and the three-axis accelerometer signal respectively to obtain a three-axis gyroscope fusion signal and a three-axis accelerometer fusion signal: Where ω(t) is the three-axis gyroscope fusion signal at time t, ω ox (t) is the X-axis gyroscope signal at time t, ω oy (t) is the Y-axis gyroscope signal at time t, ω oz (t) is the Z-axis gyroscope signal at time t, f(t) is the three-axis accelerometer fusion signal at time t, and f ox (t) is the X-axis accelerometer signal at time t, f oy (t) is the Y-axis accelerometer signal at time t, f oz (t) is the Z-axis accelerometer signal at time t; S22. Perform continuous wavelet transform on the three-axis gyroscope fusion signal and the three-axis accelerometer fusion signal, respectively, to convert the one-dimensional three-axis gyroscope fusion signal and the one-dimensional three-axis accelerometer fusion signal into a two-dimensional time-frequency diagram of the three-axis gyroscope fusion signal and a two-dimensional time-frequency diagram of the three-axis accelerometer fusion signal, respectively.

[0009] Preferably, in step S1, the dual-channel CNN model consists of a first network branch and a second network branch, the first network branch includes a first input layer, a first convolutional layer, a first batch normalization layer, a first activation layer, a first pooling layer, a second convolutional layer, a second batch normalization layer, a second activation layer, and a second pooling layer connected in sequence from front to back; the second network branch includes a second input layer, a third convolutional layer, a third batch normalization layer, a third activation layer, a third pooling layer, a fourth convolutional layer, a fourth batch normalization layer, a fourth activation layer, and a fourth pooling layer connected in sequence from front to back, the second pooling layer of the first network branch and the fourth pooling layer of the second network branch converge and output to the fully connected layer, and the fully connected layer outputs to the output layer.

[0010] Preferably, the size of each input layer is 156 in height×190 in width×3 in depth.

[0011] Preferably, the weights of the convolution kernels of each convolutional layer are randomly initialized and gradually adjusted according to the optimization goal during the training process to extract key features in the data.

[0012] Preferably, batch normalization layers are used to normalize the output of each layer, reducing the internal covariance shift of the activations by subtracting the mean and dividing by the standard deviation.

[0013] Preferably, each activation layer uses the ReLU function as the activation function and maps the input value to a non-negative number, which is expressed as follows: f(x) ReLU =max(0,x); Where f(x) ReLU Indicates the activation of the linear correction unit function for the variable, and max(0,x) indicates the selection of the maximum value between 0 and x.

[0014] Preferably, each pooling layer adopts a maximum pooling operation to reduce the feature dimension while retaining significant feature information.

[0015] Preferably, the fully connected layer is used to integrate and extract the features of the previous layer, and its activation function adopts the ReLU function to improve the nonlinear expression ability of the model.

[0016] Preferably, the output layer converts the features into probability distribution through the Softmax function to estimate the probability that the sample belongs to each category, and takes the category with the highest confidence as the final classification result.

[0017] Preferably, the specific method of step S2 is: dividing the working state of the rotary inertial navigation into five working states, namely, control vibration state, flying state, rotation stop vibration fault state, bearing fault state and normal working state.

[0018] Preferably, the control jitter state is: the rotating inertial guidance is modulated by regular forward and reverse rotation, and the modulation angular velocity is ω mc When the control system is unstable, due to the integral effect of the controller, the mean value of the control angular velocity and the control angle will fluctuate around the true value. At this time, the output of the celestial gyroscope will jitter. In this state, the rotational angular velocity of the indexing mechanism during the rotation modulation process is: Where, ω mo (t) is the angular velocity at time t, ω mc (t) is the modulation angular velocity at time t, ω mc (t) is abbreviated as ω mc , A c is the jitter amplitude, f c is the jitter frequency, t is the time, is the jitter phase, n(t) is the control angular velocity noise signal; The flying state is when the control system finally loses stability, and the rotating axis of the rotating inertial guidance gradually increases from the initial indexing speed to the maximum indexing speed, which may be accompanied by control jitter. In this state, the rotational angular velocity of the indexing mechanism during the rotation modulation process is: Where, ω mo (t) is the angular velocity at time t, ω mc is the modulation angular velocity at time t, k is the flying angular acceleration, t is time, ω max is the theoretically stable angular velocity of the full control quantity, ω mo (t) reaches ω max time, n(t) is the control angular velocity noise signal; The rotation-stop jitter fault state is: the rotating inertial guidance system rotates alternately in forward and reverse directions during the rotation modulation process. During the forward and reverse switching process, if the control parameters are improperly adjusted or there is a significant external interference, a jitter fault may occur during the angular velocity switching. In this state, the rotational angular velocity of the indexing mechanism during the rotation modulation process is: Where, ω mo (t) is the angular velocity at time t, ω mc is the modulation angular velocity, n(t) is the control angular velocity noise signal, h(t) is the jitter signal, t∈t forwawd Indicates that the rotating axis of the rotating inertial navigation is rotating in the positive direction, t∈t forwawd →t reverse Indicates that the rotation axis of the rotating inertial navigation switches from forward rotation to reverse rotation, t∈t reverse Indicates that the rotation axis of the rotating inertial navigation is reversed, t∈t reverse→t forwawd Indicates that the rotation axis of the rotary inertial navigation switches from reverse rotation to forward rotation; The bearing fault state is: when a bearing fault occurs, the fault location will generate an impact signal with exponential decay characteristics; in this state, the vibration response of the rotating inertial guidance caused by the impact is: Where x(t) is a single vibration shock signal contaminated by noise, s(t) is an exponentially decaying sinusoidal signal, n(t) is the control angular velocity noise signal, A0 is the initial amplitude of the vibration shock signal, c is the attenuation coefficient, t is the time, and f is the n is the natural frequency of the system, is the initial phase of interference.

[0019] Preferably, the specific method of step S3 is: S31, respectively collecting the three-axis gyroscope signals and the three-axis accelerometer signals under five groups of working states, namely, the control jitter state, the flying state, the stop jitter fault state, the bearing fault state, and the normal working state of the rotating inertial navigation system to construct a data set, wherein the amount of data for each group of working states is greater than or equal to 1500; S32. Randomly select 70% of the samples in each data set as a training set and 30% of the samples as a validation set; perform fusion preprocessing on the three-axis gyroscope signals and the three-axis accelerometer signals in the training set and the validation set to obtain the three-axis gyroscope signal data and the three-axis accelerometer signal data in the training set after fusion preprocessing, and the three-axis gyroscope signal data and the three-axis accelerometer signal data in the validation set after fusion preprocessing; S33. Input the three-axis gyroscope signal data and the three-axis accelerometer signal data in the training set after fusion preprocessing into the dual-channel CNN model for repeated training. Stop training when the accuracy of verifying the three-axis gyroscope signal data and the three-axis accelerometer signal data in the verification set after fusion preprocessing using the trained model reaches 99%, and obtain a trained dual-channel CNN model.

[0020] Preferably, the specific method of step S4 is: S41, collecting three-axis gyroscope signals and three-axis accelerometer signals of the rotating inertial navigation in real time, to obtain three-axis gyroscope signals and three-axis accelerometer signals collected in real time; S42, performing fusion preprocessing on the three-axis gyroscope signal and the three-axis accelerometer signal collected in real time to obtain fusion preprocessed three-axis gyroscope signal data and three-axis accelerometer signal data; S43. Input the fused pre-processed three-axis gyroscope signal data and the three-axis accelerometer signal data into the trained dual-channel CNN model to obtain the real-time monitoring of the working status of the rotational inertial navigation.

[0021] Compared with the existing technology, the present invention has the following beneficial effects: the present invention monitors the state of the rotating inertial navigation system based on IMU (Inertial Measurement Unit) information fusion and a dual-channel CNN (Convolutional Neural Network) model, cleverly utilizes the accelerometer and gyroscope built into the IMU in the rotating inertial navigation system, and performs signal fusion and transformation on the three-axis gyroscope signal and the three-axis accelerometer signal respectively. The working state of the rotating inertial navigation system can be automatically, simply, highly accurately and quickly obtained, effectively improving the reliability and safety of the inertial navigation system and having good practicality. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, for ordinary technicians in this field, other drawings can be obtained based on the drawings in the following description without any creative work.

[0023] Figure 1 The figure is a flow chart of a rotational inertial navigation state monitoring method based on IMU information fusion and dual-channel CNN according to an embodiment of the present invention.

[0024] Figure 2 Schematic diagram of the structure of the dual-channel CNN model in an embodiment of the present invention. DETAILED DESCRIPTION

[0025] In order to make the purpose, technical solutions and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with the drawings in the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. In order to make the above-mentioned features and advantages of the present invention more obvious and easy to understand, the following embodiments are specifically cited and described in detail with reference to the drawings.

[0026] like Figure 1 and Figure 2 As shown, an embodiment of the present invention provides a rotation inertial navigation state monitoring method based on IMU information fusion and dual-channel CNN, comprising the following steps: S1. Build a dual-channel CNN model; S2. Divide the working state of the rotating inertial navigation system; S3. Collect IMU information under different working states of the rotating inertial navigation system as a data set, and randomly divide the data set into a training set and a validation set; perform fusion preprocessing on the IMU information in the training set and the validation set, and input the fused preprocessed IMU information in the training set into the dual-channel CNN model for repeated training until the accuracy of verifying the fused preprocessed IMU information in the validation set using the trained model reaches a predetermined value, and then stop training to obtain a trained dual-channel CNN model; S4. Real-time acquisition of the IMU information of the rotating inertial navigation system, fusion preprocessing of the real-time acquired IMU information of the rotating inertial navigation system, and input of the fusion preprocessed real-time acquired IMU information of the rotating inertial navigation system into the trained dual-channel CNN model to obtain the working status of the rotating inertial navigation system monitored in real time.

[0027] In this embodiment, in step S3 and step S4, the IMU information includes a three-axis gyroscope signal (i.e., gyroscope information) and a three-axis accelerometer signal (i.e., accelerometer information); the specific method of the fusion preprocessing is: S21. Perform signal fusion on the three-axis gyroscope signal and the three-axis accelerometer signal respectively to obtain a three-axis gyroscope fusion signal and a three-axis accelerometer fusion signal: Where ω(t) is the three-axis gyroscope fusion signal at time t, ω ox (t) is the X-axis gyroscope signal at time t, ω oy (t) is the Y-axis gyroscope signal at time t, ω oz (t) is the Z-axis gyroscope signal at time t, f(t) is the three-axis accelerometer fusion signal at time t, and f ox (t) is the X-axis accelerometer signal at time t, f oy (t) is the Y-axis accelerometer signal at time t, f oz (t) is the Z-axis accelerometer signal at time t; S22. Perform continuous wavelet transform on the three-axis gyroscope fusion signal and the three-axis accelerometer fusion signal, respectively, to convert the one-dimensional three-axis gyroscope fusion signal and the one-dimensional three-axis accelerometer fusion signal into a two-dimensional time-frequency diagram of the three-axis gyroscope fusion signal and a two-dimensional time-frequency diagram of the three-axis accelerometer fusion signal, respectively.

[0028] In this embodiment, in step S1, the dual-channel CNN model consists of a first network branch and a second network branch, the first network branch includes a first input layer, a first convolutional layer, a first batch normalization layer, a first activation layer, a first pooling layer, a second convolutional layer, a second batch normalization layer, a second activation layer, and a second pooling layer connected in sequence from front to back; the second network branch includes a second input layer, a third convolutional layer, a third batch normalization layer, a third activation layer, a third pooling layer, a fourth convolutional layer, a fourth batch normalization layer, a fourth activation layer, and a fourth pooling layer connected in sequence from front to back, the second pooling layer of the first network branch and the fourth pooling layer of the second network branch are merged and output to a fully connected layer, and the fully connected layer outputs to an output layer. Wherein, the first input layer takes a two-dimensional time-frequency map of a three-axis gyroscope fusion signal as input, and the second input layer takes a two-dimensional time-frequency map of a three-axis accelerometer fusion signal as input.

[0029] In this embodiment, the size of each input layer is 156×190×3, that is, the height is 156 pixels, the width is 190 pixels, and the depth is the RGB color channels; The weights of the convolution kernels of each convolutional layer are randomly initialized and gradually adjusted according to the optimization objectives during training to extract key features from the data; Each batch normalization layer standardizes the output of each layer, reducing the internal covariance shift (ICS) of the activation layer by subtracting the mean and dividing by the standard deviation. ICS refers to the change in the distribution of activations in each layer due to continuous weight updates during training, which can be particularly significant when training data is collected from multiple sources. Batch normalization can alleviate the vanishing gradient problem, optimize weight initialization, reduce network convergence time, and reduce the risk of overfitting. Each activation layer uses the ReLU (Rectified Linear Units) function as the activation function and maps the input value to a non-negative number. Its expression is as follows: f(x) ReLU =max(0,x); Where f(x) ReLU Indicates activation of the linear correction unit function for the variable, max(0,x) means selecting the maximum value of 0 and x; Each pooling layer uses the maximum pooling operation to reduce the feature dimension while retaining significant feature information; The fully connected layer is used to integrate and extract the features of the previous layer. Its activation function uses the ReLU function to improve the nonlinear expression ability of the model. The output layer converts the features into probability distribution through the Softmax function, which is used to estimate the possibility that the sample belongs to each category, and takes the category with the highest confidence as the final classification result.

[0030] In this embodiment, the specific method of step S2 is: dividing the working state of the rotating inertial navigation into five working states, namely, a control jitter state, a flying state, a stop jitter fault state, a bearing fault state, and a normal working state. Among them, the control jitter state, the flying state, the stop jitter fault state, and the bearing fault state are abnormal working states. The five divided working states can be labeled: the control jitter state is labeled 1, the flying state is labeled 2, the stop jitter fault state is labeled 3, the bearing fault state is labeled 4, and the normal working state is labeled 0.

[0031] In this embodiment, the control jitter state is: the rotating inertial guidance is modulated by regular forward and reverse rotation, and the modulation angular velocity is ω mc When the control system is unstable, due to the integral effect of the controller, the mean value of the control angular velocity and the control angle will fluctuate around the true value. At this time, the output of the celestial gyroscope will have a certain degree of jitter. In this state, the rotational angular velocity of the indexing mechanism during the rotation modulation process is: Where, ω mo (t) is the angular velocity at time t, ω mc (t) is the modulation angular velocity at time t, ω mc (t) is abbreviated as ω mc , A c is the jitter amplitude, f c is the jitter frequency, t is the time, is the jitter phase, n(t) is the control angular velocity noise signal; The flying state is when the control system finally loses stability, and the rotating axis of the rotating inertial guidance gradually increases from the initial indexing speed to the maximum indexing speed, which may be accompanied by control jitter. In this state, the rotational angular velocity of the indexing mechanism during the rotation modulation process is: Where, ω mo (t) is the angular velocity at time t, ω mc is the modulation angular velocity at time t, k is the flying angular acceleration, t is time, ω max is the theoretically stable angular velocity of the full control quantity, ω mo (t) reaches ω max time, n(t) is the control angular velocity noise signal; The rotation-stop jitter fault state is: the rotating inertial guidance system rotates alternately in forward and reverse directions during the rotation modulation process. During the forward and reverse switching process, if the control parameters are improperly adjusted or there is a significant external interference, a jitter fault may occur during the angular velocity switching. In this state, the rotational angular velocity of the indexing mechanism during the rotation modulation process is: Where, ω mo (t) is the angular velocity at time t, ω mc is the modulation angular velocity, n(t) is the control angular velocity noise signal, h(t) is the jitter signal, t∈t forwawd Indicates that the rotating axis of the rotating inertial navigation is rotating in the positive direction, t∈t forwawd →t reverse Indicates that the rotation axis of the rotating inertial navigation switches from forward rotation to reverse rotation, t∈t reverse Indicates that the rotation axis of the rotating inertial navigation is reversed, t∈t reverse →t forwawd Indicates that the rotation axis of the rotary inertial navigation switches from reverse rotation to forward rotation; The bearing fault state is: when a bearing fault occurs, the fault location will generate an impact signal with exponential decay characteristics; in this state, the vibration response of the rotating inertial guidance caused by the impact is: Where x(t) is a single vibration shock signal contaminated by noise, s(t) is an exponentially decaying sinusoidal signal, n(t) is the control angular velocity noise signal, A0 is the initial amplitude of the vibration shock signal, c is the attenuation coefficient, t is the time, and f is the n is the natural frequency of the system, is the initial phase of interference; Since different working states will cause the three-axis accelerometer and three-axis gyroscope to present different output characteristics, classification of different states is achieved through discrimination.

[0032] In this embodiment, the specific method of step S3 is: S31, respectively collecting the three-axis gyroscope signals and the three-axis accelerometer signals under five groups of working states, namely, the control jitter state, the flying state, the stop jitter fault state, the bearing fault state, and the normal working state of the rotating inertial navigation system to construct a data set, wherein the amount of data for each group of working states is greater than or equal to 1500; S32. Randomly select 70% of the samples in each data set as a training set and 30% of the samples as a validation set; perform fusion preprocessing on the three-axis gyroscope signals and the three-axis accelerometer signals in the training set and the validation set to obtain the three-axis gyroscope signal data and the three-axis accelerometer signal data in the training set after fusion preprocessing, and the three-axis gyroscope signal data and the three-axis accelerometer signal data in the validation set after fusion preprocessing; S33. Input the three-axis gyroscope signal data and the three-axis accelerometer signal data in the training set after fusion preprocessing into the dual-channel CNN model for repeated training. Stop training when the accuracy of verifying the three-axis gyroscope signal data and the three-axis accelerometer signal data in the verification set after fusion preprocessing using the trained model reaches 99%, and obtain a trained dual-channel CNN model.

[0033] In this embodiment, the specific method of step S4 is: S41, collecting three-axis gyroscope signals and three-axis accelerometer signals of the rotating inertial navigation in real time, to obtain three-axis gyroscope signals and three-axis accelerometer signals collected in real time; S42, performing fusion preprocessing on the three-axis gyroscope signal and the three-axis accelerometer signal collected in real time to obtain fusion preprocessed three-axis gyroscope signal data and three-axis accelerometer signal data; S43. Input the fused pre-processed three-axis gyroscope signal data and the three-axis accelerometer signal data into the trained dual-channel CNN model to obtain the real-time monitoring of the working status of the rotational inertial navigation.

[0034] In order to verify the effectiveness of the present invention, the working state of a certain rotating inertial navigation system was monitored using the method provided by the present invention, and the accuracy of the working state monitoring was 99.71%, which verified the effectiveness of the present invention.

[0035] The present invention monitors the state of the rotating inertial navigation system based on IMU information fusion and a dual-channel CNN model. It cleverly utilizes the accelerometer and gyroscope built into the IMU in the rotating inertial navigation system, and performs signal fusion and transformation on the three-axis gyroscope signal and the three-axis accelerometer signal respectively. It can automatically, simply, highly accurately and quickly obtain the working state of the rotating inertial navigation system, effectively improving the reliability and safety of the inertial navigation system and having good practicality.

[0036] Parts of the present invention that are not disclosed in detail belong to the common knowledge in the art.

[0037] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A rotation inertial navigation state monitoring method based on IMU information fusion and dual-channel CNN, characterized in that: The following steps are involved: S1. Build a dual-channel CNN model; S2. Divide the working state of the rotating inertial navigation system; S3. Collect IMU information under different working states of the rotating inertial navigation system as a data set, and randomly divide the data set into a training set and a validation set; perform fusion preprocessing on the IMU information in the training set and the validation set, and input the fused preprocessed IMU information in the training set into the dual-channel CNN model for repeated training until the accuracy of verifying the fused preprocessed IMU information in the validation set using the trained model reaches a predetermined value, and then stop training to obtain a trained dual-channel CNN model; S4. Real-time acquisition of the IMU information of the rotating inertial navigation system, fusion preprocessing of the real-time acquired IMU information of the rotating inertial navigation system, and input of the fusion preprocessed real-time acquired IMU information of the rotating inertial navigation system into the trained dual-channel CNN model to obtain the working status of the rotating inertial navigation system monitored in real time.

2. The rotation inertial navigation state monitoring method based on IMU information fusion and dual-channel CNN according to claim 1 is characterized in that: In step S3 and step S4, the IMU information includes a three-axis gyroscope signal and a three-axis accelerometer signal.

3. The rotation inertial navigation state monitoring method based on IMU information fusion and dual-channel CNN according to claim 2 is characterized in that: In step S3 and step S4, the specific method of the fusion preprocessing is: S21. Perform signal fusion on the three-axis gyroscope signal and the three-axis accelerometer signal respectively to obtain a three-axis gyroscope fusion signal and a three-axis accelerometer fusion signal: Where ω(t) is the three-axis gyroscope fusion signal at time t, ω ox (t) is the X-axis gyroscope signal at time t, ω oy (t) is the Y-axis gyroscope signal at time t, ω oz (t) is the Z-axis gyroscope signal at time t, f(t) is the three-axis accelerometer fusion signal at time t, and f ox (t) is the X-axis accelerometer signal at time t, f oy (t) is the Y-axis accelerometer signal at time t, f oz (t) is the Z-axis accelerometer signal at time t; S22. Perform continuous wavelet transform on the three-axis gyroscope fusion signal and the three-axis accelerometer fusion signal, respectively, to convert the one-dimensional three-axis gyroscope fusion signal and the one-dimensional three-axis accelerometer fusion signal into a two-dimensional time-frequency diagram of the three-axis gyroscope fusion signal and a two-dimensional time-frequency diagram of the three-axis accelerometer fusion signal, respectively.

4. The rotation inertial navigation state monitoring method based on IMU information fusion and dual-channel CNN according to claim 1 is characterized in that: In step S1, the dual-channel CNN model consists of a first network branch and a second network branch, the first network branch includes a first input layer, a first convolutional layer, a first batch normalization layer, a first activation layer, a first pooling layer, a second convolutional layer, a second batch normalization layer, a second activation layer, and a second pooling layer connected in sequence from front to back; the second network branch includes a second input layer, a third convolutional layer, a third batch normalization layer, a third activation layer, a third pooling layer, a fourth convolutional layer, a fourth batch normalization layer, a fourth activation layer, and a fourth pooling layer connected in sequence from front to back, the second pooling layer of the first network branch and the fourth pooling layer of the second network branch converge and output to the fully connected layer, and the fully connected layer outputs to the output layer.

5. The method for monitoring the rotational inertial navigation state based on IMU information fusion and dual-channel CNN according to claim 4 is characterized in that: The size of each input layer is 156×190×3; The weights of the convolution kernels of each convolutional layer are randomly initialized and gradually adjusted according to the optimization goal during training to extract key features from the data; Each batch normalization layer is used to standardize the output of each layer, reducing the internal covariance shift of the activation layer by subtracting the mean and dividing by the standard deviation; Each activation layer uses the ReLU function as the activation function and maps the input value to a non-negative number. The expression is as follows: f(x) ReLU =max(0,x); Where f(x) ReLU Indicates activation of the linear correction unit function for the variable, max(0,x) means selecting the maximum value of 0 and x; Each pooling layer uses the maximum pooling operation to reduce the feature dimension while retaining significant feature information; The fully connected layer is used to integrate and extract the features of the previous layer. Its activation function uses the ReLU function to improve the nonlinear expression ability of the model. The output layer converts the features into probability distribution through the Softmax function, which is used to estimate the possibility that the sample belongs to each category, and takes the category with the highest confidence as the final classification result.

6. The method for monitoring the rotational inertial navigation state based on IMU information fusion and dual-channel CNN according to claim 1, characterized in that: The specific method of step S2 is: dividing the working state of the rotating inertial navigation into five working states, namely, control vibration state, flying state, rotation stop vibration fault state, bearing fault state and normal working state.

7. The method for monitoring the rotational inertial navigation state based on IMU information fusion and dual-channel CNN according to claim 6, characterized in that: The control jitter state is: the rotating inertial guidance is modulated by regular forward and reverse rotation, and the modulation angular velocity is ω mc When the control system is unstable, due to the integral effect of the controller, the mean value of the control angular velocity and the control angle will fluctuate around the true value. At this time, the output of the celestial gyroscope will jitter. In this state, the rotational angular velocity of the indexing mechanism during the rotation modulation process is: Where, ω mo (t) is the angular velocity of rotation at time t, ω mc (t) is the modulation angular velocity at time t, ω mc (t) is abbreviated as ω mc , A c is the jitter amplitude, f c is the jitter frequency, t is the time, is the jitter phase, n(t) is the control angular velocity noise signal; The flying state is when the control system finally loses stability, and the rotating axis of the rotating inertial guidance gradually increases from the initial indexing speed to the maximum indexing speed, which may be accompanied by control jitter. In this state, the rotational angular velocity of the indexing mechanism during the rotation modulation process is: Where, ω mo (t) is the angular velocity of rotation at time t, ω mc is the modulation angular velocity at time t, k is the flying angular acceleration, t is time, ω max is the theoretically stable angular velocity of the full control quantity, ω mo (t) reaches ω max time, n(t) is the control angular velocity noise signal; The rotation-stop jitter fault state is: the rotating inertial guidance system rotates alternately in forward and reverse directions during the rotation modulation process. During the forward and reverse switching process, if the control parameters are improperly adjusted or there is a significant external interference, a jitter fault may occur during the angular velocity switching. In this state, the rotational angular velocity of the indexing mechanism during the rotation modulation process is: Where, ω mo (t) is the angular velocity of rotation at time t, ω mc is the modulation angular velocity, n(t) is the control angular velocity noise signal, h(t) is the jitter signal, t∈t forwawd Indicates that the rotating axis of the rotating inertial navigation is rotating in the positive direction, t∈t forwawd →t reverse Indicates that the rotation axis of the rotating inertial navigation switches from forward rotation to reverse rotation, t∈t reverse Indicates that the rotation axis of the rotating inertial navigation is reversed, t∈t reverse →t forwawd Indicates that the rotation axis of the rotary inertial navigation switches from reverse rotation to forward rotation; The bearing fault state is: when a bearing fault occurs, the fault location will generate an impact signal with exponential decay characteristics; in this state, the vibration response of the rotating inertial guidance caused by the impact is: Where x(t) is a single vibration shock signal contaminated by noise, s(t) is an exponentially decaying sinusoidal signal, n(t) is the control angular velocity noise signal, A0 is the initial amplitude of the vibration shock signal, c is the attenuation coefficient, t is the time, and f is the n is the natural frequency of the system, is the initial phase of interference.

8. The method for monitoring the rotational inertial navigation state based on IMU information fusion and dual-channel CNN according to claim 6, characterized in that: The specific method of step S3 is: S31, respectively collecting the three-axis gyroscope signals and the three-axis accelerometer signals under five groups of working states, namely, the control jitter state, the flying state, the stop jitter fault state, the bearing fault state, and the normal working state of the rotating inertial navigation system to construct a data set, wherein the amount of data for each group of working states is greater than or equal to 1500; S32, randomly select 70% of the samples in each data set as the training set and 30% of the samples as the validation set; Performing fusion preprocessing on the three-axis gyroscope signals and the three-axis accelerometer signals in the training set and the validation set to obtain the three-axis gyroscope signal data and the three-axis accelerometer signal data in the training set after fusion preprocessing and the three-axis gyroscope signal data and the three-axis accelerometer signal data in the validation set after fusion preprocessing; S33. Input the three-axis gyroscope signal data and the three-axis accelerometer signal data in the training set after fusion preprocessing into the dual-channel CNN model for repeated training. Stop training when the accuracy of verifying the three-axis gyroscope signal data and the three-axis accelerometer signal data in the verification set after fusion preprocessing using the trained model reaches 99%, and obtain a trained dual-channel CNN model.

9. The method for monitoring the rotational inertial navigation state based on IMU information fusion and dual-channel CNN according to claim 2, characterized in that: The specific method of step S4 is: S41, collecting three-axis gyroscope signals and three-axis accelerometer signals of the rotating inertial navigation in real time, to obtain three-axis gyroscope signals and three-axis accelerometer signals collected in real time; S42, performing fusion preprocessing on the three-axis gyroscope signal and the three-axis accelerometer signal collected in real time to obtain fusion preprocessed three-axis gyroscope signal data and three-axis accelerometer signal data; S43. Input the fused pre-processed three-axis gyroscope signal data and the three-axis accelerometer signal data into the trained dual-channel CNN model to obtain the real-time monitoring of the working status of the rotational inertial navigation.

Images