A fast fault location method for onboard neural network intelligent algorithm
By constructing a fault location model with standard input/output interfaces and encoding/decoding functions, the problem of difficulty in fault location of airborne neural network intelligent algorithms in real air flight tests was solved, achieving rapid fault location, saving resources and shortening the test cycle.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SHENYANG AIRCRAFT DESIGN INST AVIATION IND CORP OF CHINA
- Filing Date
- 2023-09-27
- Publication Date
- 2026-06-12
AI Technical Summary
In real-world flight tests of airborne neural network intelligent algorithms, existing technologies struggle to quickly reproduce fault phenomena in ground environments, leading to difficulties in fault location and impacting the progress of test flights and the completion of model development tasks.
By constructing standard input/output interfaces and encoding/decoding functions, a fault location model is formed. Fault location is achieved in an environment where physical testing is not required by using software debugging. By reading flight test data and recording output results, the branch modules and key variables at the time of the problem are determined, and fault location is performed.
It enables rapid fault location without relying on physical testing environments, saving testing resources, shortening testing and verification cycles, and improving fault diagnosis efficiency.
Smart Images
Figure CN117312793B_ABST
Abstract
Description
Technical Field
[0001] This application belongs to the field of aircraft software design, and specifically relates to a rapid fault location method for airborne neural network intelligent algorithms. Background Technology
[0002] Airborne neural network intelligent algorithms are algorithmic systems that replace pilots in flight and mission execution. The quality of these algorithms directly impacts flight and mission performance. Because these intelligent algorithms are developed through large-scale adversarial training in a high-fidelity ground environment, and because neural network algorithms are inherently sensitive to input data, rapid data changes can lead to different algorithm outputs. If the algorithm's performance deviates from expectations during actual airborne flight testing, achieving rapid fault location in a ground environment becomes a critical issue. On one hand, the complex airspace environment of field test flights cannot be realistically simulated on the ground. On the other hand, manually inputting data cannot simulate the rapid changes in data during actual flight testing, making it difficult to reproduce fault phenomena and pinpoint the problem location. Under conditions of heavy testing workloads and limited flight test resources, the inability to locate faults delays troubleshooting progress, further impacting the test flight schedule and the completion of the model development mission. Summary of the Invention
[0003] To address the aforementioned issues, this application provides a rapid fault location method for airborne neural network intelligent algorithms. This method enables rapid fault location of intelligent algorithms through pure software debugging without the need for a physical testing environment.
[0004] This application provides a rapid fault location method based on airborne neural network intelligent algorithms, mainly including:
[0005] Step S1: Extract all input information that will affect the state of the airborne neural network intelligent algorithm, and form a standard input / output interface for inputting to the airborne neural network intelligent algorithm and receiving its output;
[0006] Step S2: Add encoding and decoding functions for the standard input / output interface to construct a fault location model that can read flight test data into the airborne neural network intelligent algorithm and output reconstructed data.
[0007] Step S3: In the fault location model, read the input flight test data and record the output results during the time period in which the problem occurred;
[0008] Step S4: Determine the branch module that the fault location model enters at the moment the problem occurs, and find key variables based on the entry conditions of the branch module to complete fault location.
[0009] Preferably, step S1, when forming the standard input / output interface, further includes: unifying the timing of periodic and event-type input interfaces.
[0010] Preferably, step S2 further includes inputting flight test data into the airborne neural network intelligent algorithm based on the standard input / output interface, and adjusting the input / output interface and the airborne neural network intelligent algorithm according to the difference between the output result of the airborne neural network intelligent algorithm and the reconstructed data actually generated on the aircraft, so that its output is consistent with the airborne reconstructed data.
[0011] Preferably, step S4 further includes determining the time when the problem occurred, mapping the time of occurrence to a specific time point on the machine, inserting a breakpoint at the specified time when loading the reconstructed data, running the fault location model to the specified time, tracking and determining the branch module that the fault location model entered, determining the entry conditions of the branch module, and finding key variables.
[0012] Preferably, after step S4, the method further includes:
[0013] Step S5: Reconstruct the fault code to form a new airborne neural network intelligent algorithm, and test the reconstructed code through multiple sets of experimental data.
[0014] This application has the following advantages:
[0015] 1) Enables rapid fault location without the need for a physical testing environment, saving testing resources;
[0016] 2) It can locate the time of fault occurrence without frequent serial port printing and breakpoint debugging;
[0017] 3) Effectively solves the problem of troubleshooting intelligent algorithms and shortens the testing and verification cycle. Attached Figure Description
[0018] Figure 1 This is a flowchart of a preferred embodiment of the rapid fault location method based on airborne neural network intelligent algorithms in this application. Detailed Implementation
[0019] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions in the embodiments of this application will be described in more detail below with reference to the accompanying drawings. In the drawings, the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The described embodiments are only some, not all, of the embodiments of this application. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application. The embodiments of this application will be described in detail below with reference to the accompanying drawings.
[0020] This application provides a rapid fault location method based on airborne neural network intelligent algorithms, such as... Figure 1 As shown, it mainly includes:
[0021] Step S1: Extract all input information that will affect the state of the airborne neural network intelligent algorithm, and form a standard input / output interface for inputting to the airborne neural network intelligent algorithm and receiving its output;
[0022] Step S2: Add encoding and decoding functions for the standard input / output interface to construct a fault location model that can read flight test data into the airborne neural network intelligent algorithm and output reconstructed data.
[0023] Step S3: In the fault location model, read the input flight test data and record the output results during the time period in which the problem occurred;
[0024] Step S4: Determine the branch module that the fault location model enters at the moment the problem occurs, and find key variables based on the entry conditions of the branch module to complete fault location.
[0025] It should be noted that airborne neural network intelligent algorithms typically calculate the next action based on the current state information of the aircraft or UAV, and determine the action at the next moment after that based on the newly reconstructed data after the action is executed, repeating this cycle until the aircraft completes the maneuver. It can be seen that airborne neural network intelligent algorithms interact with other aircraft data systems through input / output interfaces. To quickly locate faults, the airborne neural network intelligent algorithm needs to be separated from the overall system. Therefore, firstly, in step S1, the input and output of the entire airborne neural network intelligent algorithm need to be determined. Then, in step S2, encoding and decoding functions are added so that the airborne neural network intelligent algorithm can read the airborne information input by the tester and output data in the format expected by the tester. The airborne neural network intelligent algorithm with added standard input / output interfaces is then encapsulated as a fault location model.
[0026] Then, in step S3, the fault location model is run. When the problem of the original aircraft or drone is reproduced, the fault location model can be debugged by setting breakpoints. The program is executed step by step to find the root cause of the problem.
[0027] In some alternative implementations, step S1, when forming a standard input / output interface, further includes: unifying the timing of periodic and event-based input interfaces.
[0028] In this embodiment, all input information that affects the algorithm state is summarized, all input interface data required in the neural network algorithm is extracted, periodic and event-based data are rearranged and their timing is unified, and a standard input / output interface is written.
[0029] In some optional implementations, step S2 further includes inputting flight test data into an airborne neural network intelligent algorithm based on a standard input / output interface, and adjusting the input / output interface and the airborne neural network intelligent algorithm according to the difference between the output of the airborne neural network intelligent algorithm and the reconstructed data actually generated on board, so that its output is consistent with the airborne reconstructed data.
[0030] In step S2, encoding and decoding functions for the standard input / output interface are added. Standard input is fed into the algorithm program, and the difference between the output and the reconstructed data actually generated on-board is observed. The post-processing program is debugged to gradually approximate the on-board data, completing the model establishment and synchronization. In step S3, input data is read, and the software model is run until one set of experimental data is processed. The difference between the output data and the actual output is compared, especially the period when the problem occurs in the actual output. It is observed whether the phenomenon can be reproduced. If it can be reproduced, this set of data can be used for repeated verification. For faults that are difficult to reproduce, this method can greatly save time.
[0031] In some optional implementations, step S4 further includes determining the time when the problem occurred, mapping the time of occurrence to a specific time point on the machine, inserting a breakpoint at the specified time when loading the reconstructed data, running the fault location model to the specified time, tracking and determining the branch module that the fault location model entered, determining the entry conditions of the branch module, and finding key variables.
[0032] In this step, observe the time when the problem occurs and map it to a specific time point on the machine; when loading the reconstructed data, insert a breakpoint at the specified time of the problem, run the fault location model to the specified time, track and observe which algorithm branches the fault location model enters, determine the branch entry conditions, and then find the key variables.
[0033] In some alternative implementations, after step S4, the following is further included:
[0034] Step S5: Reconstruct the fault code to form a new airborne neural network intelligent algorithm, and test the reconstructed code through multiple sets of experimental data.
[0035] In this embodiment, the reconstructed code is loaded into the model for execution, and the reconstructed code is tested through multiple sets of experimental data, covering all branches of the algorithm. During the testing process, the intelligent algorithm can be quickly located without relying on a physical experimental environment, and the test verification cycle can be shortened by using pure software debugging.
[0036] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.
Claims
1. A rapid fault location method based on airborne neural network intelligent algorithms, characterized in that, include: Step S1: Extract all input information that will affect the state of the airborne neural network intelligent algorithm, and form a standard input / output interface for inputting to the airborne neural network intelligent algorithm and receiving its output; Step S2: Add functions for encoding and decoding standard input / output interfaces to construct a fault location model that can read flight test data into the airborne neural network intelligent algorithm and output reconstructed data. Step S3: In the fault location model, read the input flight test data and record the output results during the time period in which the problem occurred; Step S4: Determine the branch module that the fault location model enters at the moment the problem occurs, and find key variables based on the entry conditions of the branch module to complete the fault location. Step S2 further includes inputting flight test data into the airborne neural network intelligent algorithm based on the standard input / output interface, and adjusting the input / output interface and the airborne neural network intelligent algorithm according to the difference between the output result of the airborne neural network intelligent algorithm and the reconstructed data actually generated on the aircraft, so that the output result of the airborne neural network intelligent algorithm is consistent with the reconstructed data actually generated on the aircraft.
2. The rapid fault location method based on airborne neural network intelligent algorithms as described in claim 1, characterized in that, In step S1, the process of forming a standard input / output interface further includes: unifying the timing of periodic input interface data and event-type input interface data.
3. The rapid fault location method based on airborne neural network intelligent algorithms as described in claim 1, characterized in that, Step S4 further includes determining the time when the problem occurred, mapping the time of occurrence to a specific time point on the machine, inserting a breakpoint at the specified time when loading the reconstructed data, running the fault location model to the specified time, tracing and determining the branch module that the fault location model entered, determining the conditions for entering the branch module, and finding key variables.
4. The rapid fault location method based on airborne neural network intelligent algorithms as described in claim 1, characterized in that, Following step S4, the process further includes: Step S5: Reconstruct the fault code to form a new airborne neural network intelligent algorithm, and test the reconstructed code through multiple sets of experimental data.