Inertial platform stable loop fault diagnosis method based on deep neural network
Through the RNN and GRU network based on deep neural network combined with the cross attention mechanism for inertial platform stable loop fault diagnosis, the existing methods are solved, and fast and accurate fault detection and diagnosis are achieved, which is suitable for complex application scenarios.
Patent Information
- Application Number
- CN202510386013.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-30
- Publication Date
- 2025-08-01
AI Technical Summary
The existing inertial platform stable loop fault diagnosis method requires manual design of complex models and parameters, with slow speed and low accuracy, making it difficult to meet the requirements of real-time and high precision.
Using a deep neural network-based approach, the advanced features of the time series are extracted using RNN and GRU networks, and the features are fused and enhanced through the cross attention mechanism, combining the fully connected layer and the classification layer for troubleshooting.
It realizes fast and accurate inertial platform fault detection, reduces the difficulty of system development and maintenance, has high accuracy and robustness, and is suitable for complex industrial and academic application scenarios.
Smart Images
Figure CN120408291A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of automatic control, and particularly to a fault diagnosis method for an inertial platform stabilization loop based on a deep neural network. Background Art
[0002] The inertial platform is a core component of the navigation system and increasingly becomes a key device for various complex systems. The inertial platform has been widely used in the fields of aerospace, aviation, navigation, etc., and these application scenarios have relatively high requirements for its reliability. The stabilization loop of the inertial platform is the main structure to ensure the stability of the system in the inertial space. The research on the fault diagnosis of the inertial platform stabilization loop can provide accurate fault information for the platform and provide a prerequisite for active fault-tolerant control (FTC). Fault diagnosis can help the system quickly detect faults and develop FTC methods to ensure the safety and reliability of the system, and has been widely applied in the industrial community and academia.
[0003] Among the existing fault diagnosis methods, the model-based fault diagnosis method describes the state of the system through mathematical expressions. This type of method has been developed earlier and is highly complex, and has been successfully applied in engineering. The available models include qualitative models, analytical models, etc. However, the structure of the inertial platform system is relatively complex, and it is difficult to establish an accurate system model. Expert systems have also been widely used in the fault diagnosis of the industrial field, and complete fault detection and diagnosis through a knowledge base system and an inference mechanism, and provide explanatory information. However, the expert method requires a large amount of professional domain knowledge, and the expert system is unable to handle errors lacking practical experience or not occurring before. The data-driven method uses a neural network model to learn the internal relationship of data changes in fault diagnosis. The deep neural network can directly extract high-dimensional fault features from the data, with the characteristics of fast speed and high accuracy. In addition, the data-driven method does not require a large amount of professional domain knowledge, only simple model construction and training for fault diagnosis, and has a wide range of applications.
[0004] It should be noted that the information disclosed in the above background art section is only used to enhance the understanding of the background of the present invention, and thus may include information that does not constitute the prior art known to those of ordinary skill in the art. Summary of the Invention
[0005] In order to solve the problems in the existing fault diagnosis methods that require manual design of complex models and parameters, slow speed and low accuracy, the present invention provides a fault diagnosis method for an inertial platform stabilization loop based on a deep neural network.
[0006] Other features and advantages of the present invention will become apparent through the following detailed description, or will be partially learned through the practice of the present invention.
[0007] According to the first aspect of the present invention, there is provided a fault diagnosis method for an inertial platform stabilization loop based on a deep neural network, the method comprising:
[0008] Obtain the state signal, gyroscope signal, control circuit signal, and motor current signal of the inertial platform as input data;
[0009] Input the input data into an RNN network and a GRU network connected in parallel for feature extraction, and obtain RNN high-level features and GRU high-level features respectively;
[0010] Fuse the RNN high-level features and the GRU high-level features in a cross-attention network, and enhance the feature expression through a cross-attention mechanism;
[0011] After fusing the attention features calculated by the cross-attention network with the current time data, input them into a fully connected layer for feature extraction;
[0012] Send the fused features into a classification layer for fault classification, and output a fault label corresponding to the data.
[0013] In some exemplary embodiments, the input data is obtained by the following method:
[0014] Inject a zero signal at the gyroscope and motor positions and accumulate the signals to simulate a stabilization loop fault.
[0015] In some exemplary embodiments, the forward connection of the RNN network is the propagation of a recurrent neural network, where the data of each time series is interconnected, that is, the hidden layer information at the current time can be transmitted to the next time point; the RNN hidden layer h is determined by the current input x and the previous hidden layer h; at time t, the output value o t and the hidden layer h t are defined as follows:
[0016] o t = g(Vh t ) (1)
[0017] h t = f(Ux t + Wh t-1 ) (2)
[0018] where V represents the connection weight matrix between the RNN hidden layer and the output layer, U represents the connection weight matrix between the RNN input layer and the hidden layer, and W represents the connection weight matrix between the RNN hidden layer and the hidden layer.
[0019] In some exemplary embodiments, the calculation formula of the GRU network is as follows:
[0020] ψrt = σ(W r · [h t-1 , x t ) (3)
[0021] ψz t = σ(W z · [h t-1 , x t ) (4)
[0022]
[0023] where r t is the reset gate, z t is the update gate, x t is the current input, represents the summary of the current input and the previous hidden layer state, h t is the output of the hidden layer after updating the memory, W r , W h and W z are the weight matrices of the reset gate, hidden layer, and update gate respectively, and σ and tanh are the Sigmoid and hyperbolic tangent activation functions respectively.
[0024] In some exemplary embodiments, fusing the RNN high-level features and the GRU high-level features in the cross-attention network includes:
[0025] Using the high-level features extracted by the RNN as the query, and the high-level features extracted by the GRU as the key and value.
[0026] In some exemplary embodiments, the cross-attention is calculated as follows:
[0027]
[0028] where Q = X1W Q , K = X2W K , V = X2W V , W Q , W K and W V are the weight matrices for training, represents the dimension of the query and the key, and is called the scaling factor.
[0029] In some exemplary embodiments, the fully connected layer selects Sigmod as the activation function.
[0030] In some exemplary embodiments, the classification layer selects softmax as the activation function and Adam as the optimizer; the error used for training is the mean squared error.
[0031] According to a second aspect of the present invention, there is provided a storage medium having stored thereon a computer program, which when executed by a processor implements the method for diagnosing faults in an inertial platform stabilization loop based on a deep neural network described in the first aspect above.
[0032] According to a third aspect of the present invention, there is provided a computer program product having stored thereon a computer program, which when executed by a processor implements the method for diagnosing faults in an inertial platform stabilization loop based on a deep neural network described in the first aspect above.
[0033] According to a fourth aspect of the present invention, there is provided an electronic device, comprising:
[0034] a processor; and
[0035] a memory for storing executable instructions of the processor;
[0036] wherein the processor is configured to implement the method for diagnosing faults in an inertial platform stabilization loop based on a deep neural network described in the first aspect above when executing the executable instructions.
[0037] The method for diagnosing faults in an inertial platform stabilization loop based on a deep neural network provided by the embodiments of the present invention utilizes RNN and GRU to extract high-level features in time series, and then fuses and enhances these features through a cross-attention mechanism, which can more accurately detect and diagnose faults in the inertial platform; the algorithm based on a neural network has a fast calculation speed, can meet the real-time requirement, realizes fast detection and diagnosis of faults in the inertial platform, and reduces the fault response time; the data-driven method does not require a large amount of professional domain knowledge, only simple model construction and training for fault diagnosis, and reduces the difficulty of system development and maintenance; this method has high precision, high robustness and real-time performance, is applicable to a variety of complex industrial and academic application scenarios, and has broad application prospects.
[0038] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit the present invention. BRIEF DESCRIPTION OF THE DRAWINGS
[0039] The accompanying drawings herein are incorporated into the specification and form a part of the specification, showing embodiments consistent with the present invention, and are used together with the specification to explain the principles of the present invention. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention, and those of ordinary skill in the art can obtain other drawings without creative efforts based on these drawings.
[0040] Figure 1 is a flowchart of the method for diagnosing faults in an inertial platform stabilization loop based on a deep neural network according to an exemplary embodiment of the present invention;
[0041] Figure 2 The architecture diagram of the RGCNN model for an exemplary embodiment of the present invention;
[0042] Figure 3 The structural diagram of the inertial platform stabilization loop for an exemplary embodiment of the present invention;
[0043] Figure 4 The structural diagram of the RNN network for an exemplary embodiment of the present invention;
[0044] Figure 5 The structural diagram of the GRU network for an exemplary embodiment of the present invention;
[0045] Figure 6 The confusion matrix of the RGCNN model test for an exemplary embodiment of the present invention. Detailed implementation manners
[0046] Example embodiments will now be described more fully with reference to the accompanying drawings. However, the example embodiments can be implemented in various forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the concept of example embodiments to those skilled in the art. The features, structures, or characteristics described may be combined in any suitable manner in one or more embodiments.
[0047] In addition, the drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and thus their repeated description will be omitted. Some of the block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities may be implemented in software, or in one or more hardware modules or integrated circuits, or in different networks and / or processor devices and / or microcontroller devices.
[0048] In view of the disadvantages and deficiencies of the prior art, in this example embodiment, a fault diagnosis method for an inertial platform stabilization loop based on RNN, GRU, and cross-attention neural network (RGCNN) is provided. Refer to Figure 1 As shown, it is the architecture diagram of the RGCNN model of the present invention. Refer to Figure 2 As shown, the fault diagnosis method for the inertial platform stabilization loop based on a deep neural network may specifically include the following steps:
[0049] Step S1, collect the status of the platform, gyroscope, control circuit signal, and motor current signal as input data;
[0050] Exemplarily, such as Figure 3As shown, it is the structural diagram of the inertial platform stabilization loop. Zero signals are injected and accumulated at positions such as gyroscopes and motors to simulate stabilization loop faults. In this embodiment, the stabilization loop faults include gyroscope jamming faults, gyroscope offset faults, frame angle jamming faults, frame angle offset faults, motor jamming faults, and motor offset faults. Datasets of these six faults and the normal state are collected as the input data for deep neural network training.
[0051] Step S2: Input the input data into the RNN and GRU networks connected in parallel for feature extraction.
[0052] Exemplarily, as Figure 4 shown, it is the structural diagram of the RNN network. RNN is a neural network structure that can be used for feature extraction of time series data and is widely applied in fields such as acoustic signal processing, natural language recognition, and time series prediction. Different from the feedforward neural network, RNN has connections related to the time series between neurons. Its basic structure is a sequence composed of multiple identical neural network layers, and each neural network layer has a recurrent neuron, representing the information transmission and state recording between adjacent time steps. In RNN, the output at the current moment depends on the current input and all previous inputs. Therefore, RNN can effectively extract high-level features from time series data. The forward connection of RNN can be understood as the propagation of a recurrent neural network, where the data of each time series are interconnected, that is, the hidden layer information at the current time can be transmitted to the next time point. The RNN hidden layer h is determined by the current input x and the previous hidden layer h. At time t, the output value o t and the hidden layer h t are defined as follows:
[0053] o t = g * Vh t ) (1)
[0054] h t = f(Ux t + Wh t-1 ) (2)
[0055] where, V represents the connection weight matrix between the RNN hidden layer and the output layer, U represents the connection weight matrix between the RNN input layer and the hidden layer, and W represents the connection weight matrix between the RNN hidden layer and the hidden layer.
[0056] Each unit in the RNN hidden layer can affect the value of the output layer, similar to a fully connected layer, where the value of the output layer can be calculated based on all hidden units.
[0057] Figure 5It is the structure diagram of the GRU network. The GRU network can enhance the network's ability to extract global features and historical information. A complementary method of GRU and RNN is selected, which can ensure the long-term and short-term high-dimensional features of time series data. The calculation formula of the GRU network is as follows:
[0058] ψr t =σ(W r ·[h t-1 ,x t ) (3)
[0059] ψz t =σ(W z ·[h t-1 ,x t ) (4)
[0060]
[0061] Among them, r t is the reset gate, z t is the update gate, x t is the current input, represents the summary of the current input and the previous hidden layer state, h t is the hidden layer output after updating the memory, W r , W h and W z are the weight matrices of the reset gate, hidden layer, and update gate respectively. σ and tanh are the Sigmoid and hyperbolic tangent activation functions respectively.
[0062] Step 3: Fuse the obtained high-level features in the cross-attention network and improve the accuracy of the algorithm through feature enhancement of the cross-attention network;
[0063] Traditional feature fusion methods usually simply connect these two features in a direct connection manner, which cannot fully promote feature exchange and information fusion. Therefore, we adopt a cross-attention network to fuse the extracted high-level features. The cross-attention network can calculate the attention on two different sequences and process the feature relationship between the two sequences. Use the high-level features extracted by the RNN as the query, and use the high-level features extracted by the GRU as the key and value. The cross-attention network can fuse the weights of the two high-level features to express their importance. Therefore, the information exchange between the two high-level feature sequences is improved through the cross-attention network, and the features are further enhanced. The calculation of the cross-attention network is as follows:
[0064]
[0065] Among them, Q = X1W Q , K = X2W K , V = X2WV , W Q , W K and W V are trainable weight matrices, representing the dimensions of queries and keys, known as the scaling factor.
[0066] Step 4: After fusing the attention features calculated by the cross-attention network with the current time data, input them into a fully connected layer for feature extraction;
[0067] Introduce the current time data into the fully connected layer for learning. The purpose of the fully connected layer is to integrate the cross-attention features trained in parallel and the current state information, and input them into the classification layer. Select Sigmod as the activation function of the fully connected layer.
[0068] Step 5: Send the fused features into the Softmax classification layer for fault classification, and finally output the fault label corresponding to the data;
[0069] The output of the classification layer is the fault label corresponding to this data set. Select softmax as the activation function of the classification layer, and select Adam as the optimizer. The error used in training is the mean square error. Its calculation method is as follows:
[0070]
[0071] Next, evaluate and analyze the performance of this method through simulation experiments.
[0072] The configuration of the computer used in the experiment is: the CPU is i9-13900, and the graphics processor is NVIDIA GeForce RTX4090.
[0073] Figure 6 is the confusion matrix for testing the RGCNN model, clearly showing the classification accuracy of different types of faults. Among them, the experimental results show that: this method can complete the fault diagnosis task of the inertial platform stabilization loop.
[0074] Compare the training results of this model with those of three different methods: the RNN model, the GRU model, and the RNN-GRU model. For better comparison, summarize the accuracy, precision, recall rate, F1 score, MCC, and AUC of each classifier as follows in the table:
[0075]
[0076]
[0077] As can be seen from the above table, the RGCNN classification method is superior to other methods in all metrics. The method of the present invention has good performance.
[0078] It should be noted that the above-mentioned accompanying drawings are only schematic illustrations of the processes included in the method according to the exemplary embodiments of the present invention, rather than for restrictive purposes. It is easy to understand that the processes shown in the above-mentioned accompanying drawings do not indicate or limit the chronological order of these processes. Additionally, it is also easy to understand that these processes can be executed synchronously or asynchronously, for example, in multiple modules.
[0079] It should be noted that although several modules or units of the device for action execution are mentioned in the above detailed description, such a division is not mandatory. In fact, according to the embodiments of the present invention, the features and functions of the two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.
[0080] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architectures, functions, and operations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in the flowchart or block diagram may represent a module, a program segment, or a part of code, and the above-mentioned module, program segment, or part of code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than that marked in the accompanying drawings. For example, two consecutive blocks shown may actually be executed substantially in parallel, and they may sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the block diagram or flowchart, and the combination of blocks in the block diagram or flowchart, can be implemented by a dedicated hardware-based system for performing the specified functions or operations, or can be implemented by a combination of dedicated hardware and computer instructions.
[0081] The units involved in the embodiments of the present invention can be implemented in software or in hardware, and the described units can also be provided in a processor. Among them, the names of these units do not constitute a limitation to the unit itself in some cases.
[0082] It should be noted that on the other hand, the present application also provides a storage medium, which can be included in an electronic device; or can exist separately without being assembled into the electronic device. The above storage medium carries one or more programs, and when the above one or more programs are executed by an electronic device, the electronic device is caused to implement the method described in the following embodiments. For example, the electronic device can implement each step of the method as Figure 1 shown.
[0083] In one embodiment, the present application provides a computer program product, including a computer program which, when executed by a processor, implements the steps in the above-mentioned method embodiments.
[0084] In addition, the above-mentioned drawings are only schematic illustrations of the processes included in the method according to the exemplary embodiments of the present invention, rather than for limiting purposes. It is easy to understand that the processes shown in the above-mentioned drawings do not indicate or limit the chronological order of these processes. Additionally, it is also easy to understand that these processes can be executed, for example, synchronously or asynchronously in multiple modules.
[0085] Those skilled in the art will readily think of other embodiments of the present invention after considering the specification and practicing the invention herein. The present application aims to cover any variations, uses, or adaptations of the present invention, which follow the general principles of the present invention and include the common general knowledge or conventional technical means in the technical field not disclosed in the present invention. The specification and embodiments are only regarded as exemplary, and the true scope and spirit of the present invention are pointed out by the claims.
[0086] It should be understood that the present invention is not limited to the exact structures already described and shown in the drawings, and various modifications and changes can be made without departing from its scope. The scope of the present invention is only defined by the appended claims.
Claims
1. A fault diagnosis method for the inertial platform stabilization loop based on a deep neural network, characterized in that The method includes: Obtaining the state signal, gyroscope signal, control circuit signal, and motor current signal of the inertial platform as input data; Inputting the input data into a parallel-connected RNN network and GRU network for feature extraction to obtain RNN high-level features and GRU high-level features respectively; Fusing the RNN high-level features and the GRU high-level features in a cross-attention network to enhance feature expression through a cross-attention mechanism; After fusing the attention features calculated by the cross-attention network with the current time data, inputting them into a fully connected layer for feature extraction; Feeding the fused features into a classification layer for fault classification and outputting a fault label corresponding to the data.
2. The method according to claim 1, characterized in that, The input data is obtained by the following method: Injecting zero signals at the gyroscope and motor positions and accumulating the signals to simulate a stable loop fault.
3. The method according to claim 1 or 2, characterized in that, The forward connection of the RNN network is the propagation of the recurrent neural network, where the data of each time series are interconnected, that is, the hidden layer information at the current time can be transmitted to the next time point; the RNN hidden layer h is determined by the current input x and the previous hidden layer h; at time t, the output value o of the RNN t and the hidden layer h t are defined as follows: o t = g(Vh t ) (1) h t = f(Ux t + Wh t-1 ) (2) Wherein, V represents the connection weight matrix between the RNN hidden layer and the output layer, U represents the connection weight matrix between the RNN input layer and the hidden layer, and W represents the connection weight matrix between the RNN hidden layers.
4. The method according to claim 3, characterized in that, The calculation formula of the GRU network is as follows: ψr t = σ(W r ·[h t-1 ,x t ) (3) ψz t = σ(W z · [h t-1 , x t ) (4) where r t is the reset gate, z t is the update gate, x t is the current input, represents the summary of the current input and the previous hidden layer state, h t is the output of the hidden layer after updating the memory, W r 、W h and W z are the weight matrices of the reset gate, the hidden layer, and the update gate respectively, and σ and tanh are the Sigmoid and hyperbolic tangent activation functions.
5. The method according to claim 1, characterized in that, The fusing of the RNN high-level features and the GRU high-level features in the cross-attention network includes: Using the high-level features extracted by the RNN as queries, and the high-level features extracted by the GRU as keys and values.
6. The method according to claim 1 or 5, characterized in that The calculation of the cross-attention is as follows: where Q = X1W Q , K = X2W K , V = X2W V , W Q , W K and W V are weight matrices for training, represents the dimension of the query and the key, and is called the scaling factor.
7. The method according to claim 1, wherein The fully connected layer selects Sigmod as the activation function.
8. The method according to claim 1, wherein The classification layer selects softmax as the activation function and Adam as the optimizer; the error used in training is the mean square error.
9. A computer program product comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method for diagnosing faults in the stable loop of an inertial platform based on a deep neural network according to any one of claims 1 to 7.
10. An electronic device, characterized in that, It includes: A processor; And A memory for storing executable instructions of the processor; Wherein, the processor is configured to execute the method for diagnosing faults in the stable loop of an inertial platform based on a deep neural network according to any one of claims 1 to 7 by executing the executable instructions.