Unmanned aerial vehicle flight log data analysis method and device, computing equipment and storage medium

Abstract

The invention discloses an unmanned aerial vehicle flight log data analysis method and device, computing equipment and a storage medium, and the method comprises the steps: obtaining a question input by a user, and determining a prompt word according to the question; acquiring corresponding time stream data from the flight log data according to the cue word, respectively determining calling functions from a local first function set and a second function set built in a large model according to the cue word, and respectively operating the time stream data by utilizing the calling functions, pre-processing result information after mutual verification is obtained; and inputting the question, the preprocessing result information and / or the cue word into a large model, performing analysis by using the large model, and obtaining and outputting an analysis result. According to the scheme, the local function set and the log data are utilized to accurately guide and train the large model, the dimension and depth of data analysis are expanded, the robustness and practicability of the system are enhanced, the scene adaptation capability is enhanced, and the use cost is reduced.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) flight control technology, specifically to a method, apparatus, computing device, and storage medium for analyzing UAV flight log data. Background Technology

[0002] Drone flight control log analysis is of great significance to flight safety. Like a "black box," it records key data throughout the flight and provides objective evidence for analyzing accidents such as crashes and loss of control. It can accurately pinpoint the root cause of faults. For example, a sudden drop in barometer data when the actual altitude drops can identify sensor malfunctions. At the same time, analyzing daily logs can uncover "hidden anomalies," such as PID oscillation on a certain axis or occasional GPS signal loss. This prevents small problems from accumulating into major accidents, identifies potential risks in advance, and strengthens the safety defense line.

[0003] Current UAV flight control log analysis suffers from the following shortcomings: First, log analysis tools struggle to balance compatibility, ease of use, and specificity. While some tools support multi-format log parsing and offer high customization, they have high operational barriers, requiring manual selection of key parameters and lacking "one-click" diagnostic functions for flight control faults. Novices or users without development backgrounds (such as frontline maintenance personnel) must spend considerable time learning configuration, and misdiagnosis due to missed parameter selection is likely. Some tools are adapted to mission scenario review, but compatibility is limited. Some require manual script writing, restricting the scope of use for non-programming users, leading to different user groups needing to switch between multiple tools, resulting in low analysis efficiency. Second, log analysis lacks scenario-based intelligent adaptation capabilities. Existing analysis relies on manual data mining (such as comparing parameter differences under air density in different regions, and investigating the timing relationship between airspeed anomalies and throttle commands), and cannot automatically associate scene features to output targeted solutions. For example, it cannot intelligently recommend TECS parameter adjustment direction based on real-time regional meteorological data, nor can it automatically optimize flight path and battery strategy in combination with mission requirements. This results in parameter calibration and fault diagnosis relying too much on manual experience, making it difficult to guarantee accuracy and efficiency. Summary of the Invention

[0004] In view of the above problems, the present invention is proposed to provide a method, apparatus, computing device and storage medium for analyzing UAV flight log data that overcomes or at least partially solves the above problems.

[0005] According to one aspect of the present invention, a method for analyzing unmanned aerial vehicle (UAV) flight log data is provided, the method comprising: Obtain the user's input question and determine prompt words based on the question; According to the prompt words, the corresponding time stream data is obtained from the flight log data, and the calling functions are determined from the first function set in the local area and the second function set built into the large model, respectively, based on the prompt words. The calling functions are used to perform calculations on the time stream data to obtain preprocessed result information after mutual verification. The problem, preprocessing results, and / or prompts are input into the large model, analyzed using the large model, and the analysis results are obtained and output.

[0006] In some implementations, calling functions are determined from a local first function set and a second function set built into the large model based on prompt words. These calling functions are then used to perform calculations on the time-stream data to obtain mutually verified preprocessed result information, including: The time stream data is preprocessed using the functions called in the first function set of the local machine to obtain the first processing result; The time stream data is preprocessed using the calling functions of the second function set in the large model to obtain the second processing result; If the first processing result and the second processing result are the same or similar, the preprocessing result information is output based on the first processing result and the second processing result.

[0007] In some implementations, retrieving the corresponding time-stream data from the flight log data based on the prompt word includes: Load flight log data and save it locally; Extract and preprocess the corresponding time stream data from the data unit according to the preset mode type; The flight log data is stored in multiple data units, each of which is constructed using the mapping relationship between log path, mode type, and time stream data.

[0008] In some implementations, the first function set and the second function set include the following call functions: Rotor mode speed error analysis module, fixed-wing mode speed relationship analysis module, IMU sensor data Fourier transform analysis module, GPS data accuracy analysis module, UAV angular velocity analysis module, UAV attitude analysis module, log data correlation and delay analysis module, log data time delay and sampling interval analysis module, or log parameter extraction and filtering module.

[0009] In some implementations, the problem, preprocessing result information, and / or prompt words are input into a large model, analyzed using the large model, and the analysis results are obtained and output, including: The large model receives the question, prompt words, and preprocessing results. Combined with the calling function information of the second function set or the parsing results of the calling functions, semantic understanding, knowledge reasoning, comprehensive judgment, and / or association analysis are performed to obtain and output the analysis results.

[0010] In some embodiments, the method further includes: The system receives user questions through a web interface and displays the analysis results to the user.

[0011] In some implementations, before obtaining a user-input question and determining a prompt word based on the question, the method further includes: Preset question examples are displayed on the receiving interface, making it easy for users to edit the questions they want to ask based on the question examples; The problem, preprocessing results, and / or prompts are input into the large model, analyzed using the large model, and the analysis results are obtained and output, including: Output an HTML report containing the following information: flight status overview, performance index analysis, anomaly detection, and resolution recommendations.

[0012] According to another aspect of the present invention, a drone flight log data analysis device is provided, which is executed by each drone in a formation, the device comprising: The question acquisition module is adapted to acquire the question input by the user and determine the prompt words based on the question; The preprocessing module is adapted to obtain the corresponding time stream data from the flight log data according to the prompt words, and to determine the calling functions from the local first function set and the second function set built into the large model according to the prompt words, and to use the calling functions to perform calculations on the time stream data to obtain mutually verified preprocessing result information; The model analysis module is suitable for inputting the problem, preprocessing result information and / or prompt words into the large model, using the large model for analysis, and obtaining and outputting the analysis results.

[0013] According to another aspect of the present invention, a computing device is provided, comprising: a processor and a memory arranged to store computer-executable instructions, which, when executed, cause the processor to perform a drone flight log data analysis method according to any one of the above embodiments.

[0014] According to another aspect of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores one or more programs, which, when executed by a processor, implement the UAV flight log data analysis method according to any one of the above.

[0015] As can be seen from the above, according to the embodiments of the present invention, a flight control log analysis solution with low threshold and targeted diagnostic capabilities is provided. Users do not need to manually write scripts or configure complex parameters. They can operate the log analysis tool of the present invention through natural language interaction, quickly initiate log parsing, fault diagnosis and other needs, and significantly reduce the usage threshold for users without development background (such as front-line maintenance personnel and flight control debugging personnel), and solve the problems of complex operation and reliance on professional skills of existing tools.

[0016] Furthermore, the aforementioned technical solution enables scenario-based intelligent analysis and adaptation of flight control logs. First, the raw log data undergoes standardized preprocessing (such as data cleaning, key parameter extraction, and format conversion). Then, the preprocessed data is input into a large model. Leveraging the large model's comprehensive analysis capabilities across multiple dimensions, it outputs preliminary suggestions for flight status assessment, parameter calibration direction, and mission efficiency optimization. Simultaneously, to mitigate the potential "illusion" risk when the large model handles real-world problems, preset code rules, such as flight control fault judgment logic, parameter calibration constraints, and data correlation verification algorithms, are used to verify and correct the large model's output. Ultimately, this ensures the reliability and accuracy of the analysis conclusions, addressing the problem of existing analyses relying on human experience and struggling to guarantee accuracy and efficiency.

[0017] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0018] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 A flowchart illustrating a method for analyzing UAV flight log data according to an embodiment of the present invention is shown. Figure 2 A schematic diagram of the framework of a flight control system log analysis tool according to an embodiment of the present invention is shown; Figure 3 A flowchart illustrating a function key information extraction algorithm according to an embodiment of the present invention is shown; Figure 4 A flowchart illustrating a large model analysis algorithm according to an embodiment of the present invention is shown; Figure 5A schematic diagram of a user interface for a flight control system log analysis tool according to an embodiment of the present invention is shown. Figure 6 A schematic diagram of a user dialogue for a flight control system log analysis tool for a large model according to an embodiment of the present invention is shown. Figure 7 A schematic diagram of a drone flight log data analysis device according to an embodiment of the present invention is shown; Figure 8 A schematic diagram of a computing device according to an embodiment of the present invention is shown. Detailed Implementation

[0019] Exemplary embodiments of the invention will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the invention and to fully convey the scope of the invention to those skilled in the art.

[0020] Figure 1 A flowchart illustrating a method for analyzing drone flight log data according to an embodiment of the present invention is shown. The method includes the following steps: Step S110: Obtain the question input by the user, and determine the prompt words (or keywords) based on the question; wherein, the prompt words, the mapping relationship between the prompt words and functions, and the mapping relationship between the prompt words and time stream data need to be determined in advance; Step S120: Obtain the corresponding time stream data from the flight log data according to the prompt words, and determine the calling functions from the local first function set and the second function set built into the large model according to the prompt words. Use the calling functions to perform calculations on the time stream data to obtain preprocessed result information after mutual verification. Step S130: Input the problem, preprocessing result information and prompt words into the large model, use the large model to analyze, and obtain and output the analysis results.

[0021] The analysis method flow disclosed in the embodiments of this invention can be found in [reference needed]. Figure 2 , 3 As shown, this embodiment does not require users to manually write scripts or configure complex parameters. Users can operate the log analysis tool in this invention simply through natural language interaction, quickly initiate log parsing, troubleshooting and other needs, greatly reduce the usage threshold for users without a development background (such as front-line maintenance personnel and flight control debugging personnel), and solve the problems of existing tools being complex to operate and relying on professional skills.

[0022] The above technical solution complements each other through two paths. It relies on traditional AST parsing to achieve efficient and accurate syntactic-level information extraction, while leveraging large models to overcome the limitations of document strings and generate descriptions with greater semantic depth. This overcomes the problem of auditory hallucinations in large models and supports accurate backtracking of source code, providing comprehensive data support for the transmission of function call information and the understanding of code functions in large models.

[0023] Specifically, according to Figure 2 The overall framework shown is as follows: On the webpage interaction side, users configure log paths and source code paths, and relying on the natural language dialogue module, complete the reading of question prompts and submit questions, triggering the analysis process. The backend analysis chain is based on Px4.ulog logs. First, the logs are loaded through pyulog and data preprocessing is performed to accurately extract the time streams of Offboard mode and VTOL mode. Then, it enters the large model logic distribution stage, integrating user question information, built-in callable functions, and preset prompts, linking the cloud-based large model and built-in functions to generate preliminary analysis results. In specific implementation, the callable functions can be pre-saved and parsed in the large model, and the large model can be trained using the callable functions, so that the callable functions are built into the large model. The large model can also obtain the preprocessed structure, which facilitates subsequent cross-verification and avoids illusions. Finally, after a second deep analysis by the large model, the conclusions are output by integrating multi-dimensional information and sent back to the webpage to provide feedback to the user, realizing a closed loop from interaction initiation to intelligent analysis.

[0024] The method disclosed in this embodiment innovatively constructs a multi-stage collaborative intelligent analysis system. A natural language interactive entry point is created on the web interface, lowering the user's operational threshold and enabling convenient configuration and questioning. In the background data preprocessing stage, the flight control log timeline is precisely broken down, laying a data foundation for in-depth analysis. Key innovations focus on the large-scale model logic distribution and analysis process: on the one hand, it gathers user needs, tool resources (built-in functions), and guidance information (preset prompts) to build an analysis "arsenal"; on the other hand, through a two-stage operation of "initial call to the cloud-based large model + obtaining basic conclusions through built-in functions → secondary in-depth analysis and iterative optimization," it links model intelligence with the accuracy of tool functions, efficiently interpreting complex flight control log data. This breaks through the reliance of traditional log analysis on human experience and single tools, significantly improving analysis efficiency and conclusion accuracy, and providing intelligent support for flight control system fault diagnosis and performance optimization.

[0025] Combination Figure 3 The flowchart shown is for the algorithm to extract key information from function calls and time stream data: I. Starting Point of the Process: "Key Function Information Acquisition Module" and Initialization The function key information acquisition module is the trigger entry point for the entire process. It is responsible for starting the code analysis process and clarifying that the target to be processed is the "code file".

[0026] Initialization: Reading the target code file: The first step in the process is to load the contents of the code file to provide raw material for subsequent analysis.

[0027] II. Branch Selection: "Select Analysis Path" This is the core branching point of the process, which utilizes two paths: traditional code parsing and large-scale intelligent analysis, covering different scenarios, such as manual in-depth code parsing and rapid understanding of logic with the help of large-scale models.

[0028] 3. Path 1: Parsing using local function code Focusing on manually parsing the code structure using the AST (Abstract Syntax Tree) to extract key elements such as classes and functions: AST (Abstract Syntax Tree) parsing of code structure: Transforming code into an AST, breaking down code logic into "tree-like nodes" that computers can understand, making it easier to accurately locate classes, functions, and variables.

[0029] Identify element type by traversing nodes: Traverse the AST nodes to distinguish between classes (ClassDef) and functions (FunctionDef), because the parsing logic for the two is different.

[0030] IV. Path 2: Intelligent Analysis of Large Models Leveraging the reasoning capabilities of large-scale models, this process supplements code logic interpretation, emphasizing "understanding the code's intent through the large-scale model." The process involves: AST (Abstract Syntax Tree) parsing of the code structure: Similar to traditional methods, the code is first converted into an AST to obtain the basic structure. Filtering elements to be analyzed: The classes or functions to be analyzed are selected from the AST. Extracting element type + name + complete code snippets: The code of the class / function is precisely extracted, including type, identifier name, and complete implementation, providing context for the large-scale model. Constructing a unified request content: The information extracted from the AST is packaged into a prompt containing an element list and special instructions, enabling the large-scale model to understand the task. Calling the large-scale model's interface: The prepared prompt is sent to the large-scale model, allowing the AI ​​to analyze the code logic. Receiving the large-scale model's response: The analysis results from the large-scale model are obtained. Understanding and completing the analysis according to requirements, and associating corresponding code elements: The large-scale model's response is associated with the original code elements, creating a dictionary of the large-scale model's analysis results.

[0031] V. Final Output: Integrate the results and enhance the information of the large model. After both paths are completed, the traditional results, code, and large model information will be integrated to output a more complete analysis dictionary, which can provide both the "objective structure" of the code (class and function definitions) and the "subjective understanding" of the large model.

[0032] In some embodiments, the large model pre-stores and parses a second function set, and the second function set corresponds to the first function set. Then step S120 specifically includes: The time stream data is preprocessed using the functions called in the first function set of the local machine to obtain the first processing result; The second processing result is obtained by preprocessing the time stream data using the functions called in the second function set. If the first processing result and the second processing result are the same or similar, the preprocessing result information is output according to the first processing result and the second processing result. If the two are not the same and are not similar, the process is repeated until the same or similar result is obtained, or an error is returned and the large model is retrained.

[0033] Combination Figure 4 As shown, the large model analysis process according to an embodiment of the present invention is as follows: I. Process Start Point: The des_data method is initiated. As the entry point for the entire process, it triggers the log analysis report generation task, receiving two core inputs: user messages and log data. All subsequent steps revolve around these two inputs.

[0034] II. Input Preprocessing and Validation Receive user messages and data: Clearly define the "user requirements" and "log statistics (such as GPS error, angular velocity, etc.)" to be processed. Data validation: Determine the validity of the data; it acts as the "gatekeeper" of the process, preventing invalid data from entering subsequent stages and causing errors.

[0035] III. Data Serialization: Custom Encoding Adaptation for Special Values Log data may contain special floating-point numbers (NaN, ±Infinity) or non-serializable objects, and directly converting it to JSON will result in an error.

[0036] IV. Building the Large Model Prompt: Clarifying Tasks and Format Requirements To generate a suitable HTML report from a large model, a detailed Prompt (task instructions) needs to be constructed: Integrate information: Combine user_message (user request) and data_str (serialized log data) to allow the large model to understand the "analysis context".

[0037] The task is clearly defined: the large model is required to output an HTML report, which should include the following information: flight status overview, performance index analysis, anomaly detection, and solution suggestions, and the format should be specified.

[0038] Role setting: Through the system role, the large model is set as "Professional PX4 Flight Control Log Analyst", and the output style is constrained.

[0039] V. Large Model Call and Response Processing Constructing dialogue messages: Package the system role and user role into messages, which serve as input to the larger model.

[0040] Call the `generate` method of the large model: send messages to the large model and retrieve the returned analysis content response.

[0041] Clean up Markdown code blocks: Call the _clean_markdown_code_blocks method to remove Markdown tags (such as `html`) that may be present in the large model return, ensuring that the output is clean HTML content.

[0042] VI. Response Verification and Formatting Repair The content returned by the large model may not conform to HTML specifications and requires secondary processing. Verify the model response: If the response is empty, return an error message. Error: Unable to generate analysis description... ).

[0043] Check HTML tag integrity: If the returned content does not contain <h2> < / h2> <h3> The presence of such key tags indicates that the large model was not output according to the specified format; it will be automatically replaced. The content enclosed in the tag ensures that the output is a valid HTML fragment.

[0044] VII. Exception Handling: Capturing Errors Throughout the Process Any exception thrown at any stage of the process (such as data serialization failure or large model call error) will be caught by try-except: the error message will be wrapped into HTML with the error class ( An error occurred during the analysis... It provides a unified error message to facilitate problem identification.

[0045] VIII. Summary of Core Logic The entire process revolves around "input (user messages + log data) → verification → serialization → large model call → response processing → output HTML". Through data preprocessing (solving special value serialization), precise prompt guidance, and response repair, a flight control log analysis report that meets the requirements is finally generated, allowing for seamless integration of technical processes and business needs.

[0046] Compared with conventional analysis tools, the advantages and differences of the analysis tools disclosed in this invention are mainly reflected in the innovation of analysis paradigm, data structure design, large model collaboration mechanism, and multi-dimensional correlation capabilities, the specific features of which are as follows: I. Analysis Paradigm: Shifting from "pure rule-driven" or "single tool analysis" to collaborative analysis of "large model + dedicated module".

[0047] II. Pattern Information Processing: Shifting from "unstructured storage" to "dictionary-based three-level mapping" to achieve rapid time-pattern positioning. Flight patterns are the core basis for interpreting UAV behavior.

[0048] 3. Large-scale model collaboration: Shifting from "large-scale models directly processing raw logs" to "structured data input + precise logical distribution".

[0049] IV. Parameter Analysis: Shifting from "isolated parameter parsing" to "pattern correlation + multi-dimensional verification" to improve the depth of analysis. Existing flight log analysis solutions mostly focus on "numerical statistics of single parameters", lacking correlation and pattern matching between parameters, resulting in insufficient analysis depth.

[0050] In some embodiments, retrieving corresponding time-stream data from flight log data based on the prompt words includes: Load flight log data and save it locally; Extract and preprocess the corresponding time stream data from the data unit according to the preset pattern type; it should be noted that the preprocessing step can also be completed on the local server or cloud server.

[0051] The flight log data is stored in multiple data units, each of which is constructed using the mapping relationship between log path, mode type, and time stream data.

[0052] Specifically, the original format of UAV flight control logs is mostly binary files. Therefore, the first step before conducting analysis is to parse and process the binary files. By adapting to the dedicated parsing logic of flight control logs (such as decoding algorithms for PX4 ulog and ArduPilot bin formats), the logs are converted into readable structured data, ultimately generating a time-stream dataset of various UAV status information, covering the complete sequence of changes in key parameters such as UAV attitude (such as pitch angle and roll angle), position (such as latitude and longitude and relative altitude), airspeed, and throttle commands over time.

[0053] Log loading uses pyulog, a library that converts binary data into a message stream, which involves the following three steps: 1) Path normalization and status verification The module first converts the input log file path to an absolute path to standardize the path identifier format. It then checks whether the log file has already been loaded into the system (existing in the log storage collection). If it has, it returns a success status directly; otherwise, if the log file is currently being loaded (existing in the loading task collection), it waits for the loading task to complete before returning a final status to avoid duplicate loading or conflicts.

[0054] 2) Creation and execution of asynchronous loading tasks For log files that are not loaded or not in a loading state, the module creates an asynchronous loading task, which includes: calling a log parsing tool (such as pyulog.ULog) to load the log file, storing the parsing results in the log storage collection, and establishing a mapping relationship between the absolute path and the log data; after successful loading, automatically triggering time stream extraction preprocessing for two specific flight modes (Offboard mode and VTOL mode); if an exception occurs during loading (such as file corruption or format incompatibility), the error information is captured and a detailed stack trace is recorded to facilitate troubleshooting.

[0055] 3) Dynamic management of loading status During asynchronous task execution, the task instance is stored in the loading task collection to indicate that the log is being loaded; after the task is completed (whether successfully or not), the task is automatically removed from the loading task collection to ensure state consistency.

[0056] When a user asks multiple questions, each log only needs to be loaded once. This is significantly valuable for improving system efficiency and optimizing the user experience. From an efficiency perspective, avoiding repeated loading of the same log significantly reduces unnecessary file I / O operations and computational resource consumption. This is especially true for large flight control logs (such as the Px4.ulog file containing long-duration flight data), where repeated loading can significantly consume memory and processor resources. The mechanism of reusing data after a single load allows the system to focus its computing power on the analysis of user questions, improving response speed. From a user experience perspective, users do not need to wait for the repeated loading of the same log when asking multiple questions, reducing interaction waiting time. Especially in multi-turn dialogue scenarios, this maintains the continuity and smoothness of the analysis process, avoiding operational stutters or delays caused by repeated loading, and enhancing the tool's usability and convenience.

[0057] Regarding data preprocessing, time-stream data often has the characteristics of "high dimensionality and large capacity": On the one hand, a single flight log may contain tens of thousands to hundreds of thousands of time-series data. If it is directly uploaded to a large model for analysis, it will generate extremely high computational resource consumption and time costs, significantly increasing the analysis threshold; on the other hand, large models have a token limit, and it is difficult to achieve joint correlation analysis of multiple time-stream information by directly inputting complete data (such as synchronously matching "attitude anomaly time series" and "sensor data jump time series"), which can easily lead to the break of key data correlation and affect the completeness and accuracy of the analysis.

[0058] In this embodiment of the invention, the remote-controlled Offboard mode analysis module is used to extract and standardize the time distribution information of Offboard modes from the flight control log, for analyzing data during the remote manual control phase of the UAV. First, it verifies whether the target log file has been loaded. After confirming the log is available, it locates the flight status record by searching for topic data containing the keyword "vehicle_status" in the log. It extracts the timestamp and navigation status parameter (nav_state) from this data, converting the timestamp into a relative time in seconds relative to the log's start time to standardize the time measurement. It generates the original mode state sequence by determining whether nav_state equals a preset threshold. To improve subsequent analysis efficiency, the original state sequence is resampled at 1-second intervals, using the majority of states within each time window as the mode determination result for that second. Finally, the standardized Offboard mode time sequence (0 / 1 states in seconds) is stored, providing structured mode time period data for large model analysis.

[0059] The Vertical Take-Off (VTOL) mode analysis module targets the mode characteristics of VTOL fixed-wing aircraft, achieving multi-state identification and time standardization. The module prioritizes searching the logs for data under the "vtol_vehicle_status" topic. If found, it directly extracts the multirotor mode identifier `vtol_in_rw_mode` and the transition mode identifier `vtol_in_trans_mode`, classifying the state into three categories—multirotor, fixed-wing, and transitioning—through logical operations. If this topic data is missing, an alternative approach is used: the mixed parameters of the "actuator_controls_0" topic are used to distinguish the basic states of multirotors and fixed-wing aircraft using a threshold method, and the transitioning state is supplemented using a state transition detection algorithm (marking the state 1 second before and after as a transition state). Finally, resampling is performed at 1-second intervals, and the duration of each mode is calculated, forming structured data containing subdivided state sequences, VTOL aircraft type identifiers, and mode durations. This provides a precise time dimension reference for large-scale model analysis of VTOL aircraft mode switching logic.

[0060] In the analysis of UAV flight control system logs, mode information such as Offboard mode and VTOL mode is the key basis for interpreting flight behavior and locating abnormal problems. For example, when troubleshooting, it is necessary to first determine whether the UAV is in autonomous control or manual control mode when the fault occurs, and when analyzing performance, it is necessary to distinguish the parameter differences between multi-rotor and fixed-wing modes.

[0061] To address this, this invention stores pattern analysis results in a dictionary structure, constructing a three-level mapping relationship of "log path - pattern type - time series". This design overcomes the limitations of traditional log analysis, where pattern data is stored in a scattered manner and queries require repeated traversal of the original logs. It enables the pattern information of each log to form an independent, structured data unit, providing an efficient data foundation for "quickly locating patterns by inputting time".

[0062] This dictionary structure directly supports fast reverse lookups based on "time-pattern": users or subsequent analysis modules only need to input the absolute path and specific time of the target log, and the system can locate the corresponding pattern data list through the dictionary index. Combining the correspondence between time and sequence indexes, the system can instantly determine the UAV's pattern status at that moment. This query method eliminates the need to re-parse logs or process raw data, significantly reducing pattern localization time. Simultaneously, the structured dictionary facilitates integration with large model analysis modules—large models can directly read the pattern time series from the dictionary, quickly associating it with flight parameters (such as attitude and power data) for the corresponding time period, enabling linked analysis of "pattern-parameter-problem," providing accurate and efficient technical support for UAV fault diagnosis and flight process review.

[0063] In some embodiments, the first function set and the second function set include the following calling functions (modules): Rotor mode speed error analysis module, fixed-wing mode speed relationship analysis module, IMU sensor data Fourier transform analysis module, GPS data accuracy analysis module, UAV angular velocity analysis module, UAV attitude analysis module, log data correlation and delay analysis module, log data time delay and sampling interval analysis module, or log parameter extraction and filtering module.

[0064] Specifically, the rotor mode speed error analysis module is used to accurately analyze the speed control performance of UAVs in rotor mode. The specific process is as follows: First, verify the loading status of the target log file and the integrity of the VTOL mode data. By retrieving the "vehicle_local_position" and "vehicle_local_position_setpoint" topic data in the log, the actual speed (vx, vy, vz) and expected speed parameters are obtained respectively. Then, through abnormal time filtering, rotor mode period filtering, and timestamp synchronization, a standardized speed data sequence in rotor mode is obtained. Based on this sequence, the speed error is calculated, and a multi-subplot visualization chart containing the comparison between the actual speed and expected speed on the X / Y / Z axes is generated. At the same time, key indicators such as the mean and standard deviation of the speed error are statistically analyzed. Finally, the analysis conclusion description, statistical data, and visualization charts are stored in the result dictionary, providing structured data and intuitive chart support for the output speed control performance evaluation conclusion of the large model, helping to quickly locate speed control anomalies in rotor mode.

[0065] The fixed-wing mode speed relationship analysis module focuses on the correlation analysis of multi-dimensional speed parameters in fixed-wing mode. The specific process is as follows: First, verify the target log loading status and VTOL mode data. By retrieving the "vehicle_local_position", "airspeed", and "wind" topic data from the log, extract the core parameters of ground speed (vx, vy, and composite ground speed), airspeed (e.g., true_airspeed_m_s), and wind speed (east / north wind speed and composite wind speed). After abnormal time filtering, fixed-wing mode time period selection, and timestamp synchronization (interpolation processing to match the time axes of airspeed, wind speed, and ground speed), a standardized speed data sequence is obtained. Based on the fluid dynamics principle that "airspeed = ground speed + wind speed", the sequence is then analyzed. The theoretical relationship between airspeed and ground speed is calculated, and a multi-subplot visualization chart containing ground speed-airspeed comparison, wind speed changes, and speed differences is generated. At the same time, key indicators such as the mean, standard deviation, and extreme values ​​of each speed parameter are statistically analyzed. Finally, the theoretical explanation of the speed relationship, structured statistical data, and visualization charts are stored in the results dictionary, providing data support for the analysis of the rationality of speed control and the impact of wind field on flight performance in fixed-wing mode under large-scale model analysis, and helping to accurately judge the anomalies of fixed-wing speed systems.

[0066] The IMU sensor data Fourier transform analysis module is used to analyze the accelerometer and gyroscope data characteristics of the UAV IMU (Inertial Measurement Unit) from the frequency domain. The specific process is as follows: First, verify the target log loading status. By retrieving the "sensor_combined" topic data in the log, extract the original measurement data and corresponding timestamps of the accelerometer and gyroscope. Calculate the relative time difference based on the timestamps to deduce the sensor sampling frequency, providing basic parameters for subsequent frequency domain analysis. Perform frequency domain transformation on the sensor data of each axis using a custom Fourier transform function (calculate_fft), filter the positive frequency intervals and calculate the corresponding amplitude values ​​to obtain the frequency-amplitude characteristics of the data for each axis. Generate Fourier transform visualization charts for each axis of the accelerometer and gyroscope, and store the frequency domain data (frequency sequence, amplitude sequence) and sampling frequency parameters for each axis. Finally, output structured frequency domain data and intuitive charts to provide frequency domain data support for large-scale model analysis of IMU sensor noise distribution and vibration frequency anomalies, helping to accurately locate IMU sensor performance problems or potential structural vibration hazards in the UAV.

[0067] The GPS data accuracy analysis module, based on the reasonable assumption that "the actual positions of the UAV takeoff and landing points are consistent," achieves a quantitative assessment of GPS altitude and horizontal position accuracy. The specific process is as follows: First, verify the target log loading status. By retrieving the "vehicle_gps_position" topic data from the log, extract the raw latitude, longitude, and altitude data and convert them to standard units (degrees, meters). Simultaneously, calculate the relative time (in seconds) with respect to the log's start time. Using the first and last log data as stable position benchmarks before takeoff and after landing, respectively, convert latitude and longitude differences to horizontal distances (meters) using the Earth's radius to latitude / longitude conversion formula, and directly calculate the altitude difference. Generate a visual chart (converted to SVG format) containing the "GPS altitude changing over time" horizontal position trajectory (marking takeoff and landing points and distances), simultaneously statistically analyzing key parameters such as takeoff and landing point coordinates, altitude / horizontal difference, altitude extremes and standard deviations, and flight duration. Based on preset thresholds (altitude difference 1 meter / 3 meters, horizontal distance 5 meters / 10 meters), classify altitude and horizontal position accuracy into "good / average / poor." Level 3 forms an overall accuracy assessment conclusion; finally, the assessment description, structured statistical data, hierarchical assessment results and visualization charts are stored in the results dictionary, providing a quantitative basis for the large model to judge the accuracy of the GPS system and the reliability of positioning data, and helping to investigate flight control problems caused by GPS errors.

[0068] The UAV angular velocity analysis module focuses on the correlation analysis between angular velocity data and flight modes. The specific process is as follows: First, the target log loading status and the integrity of VTOL mode data are verified. By retrieving the "vehicle_angular_velocity" and "vehicle_rates_setpoint" topic data from the log, the actual angular velocity (roll rate, pitch rate, yaw rate) and expected angular velocity parameters are extracted respectively. After abnormal time filtering and timestamp standardization, a synchronized angular velocity data sequence is obtained. A multi-subplot visualization chart is generated, which includes a comparison of actual and expected angular velocities on the X / Y / Z axes and flight mode bars (fixed-wing / rotor / conversion status are marked with different colors). At the same time, combined with the flight mode data, the statistical characteristics of angular velocity under different modes are calculated. Finally, the structured statistical data and visualization charts are stored in the results dictionary, providing data support for large-scale model analysis of angular velocity control accuracy under various modes and the impact of mode switching on angular velocity, helping to accurately locate anomalies in the angular velocity control system.

[0069] The UAV attitude analysis module enables accurate comparison and mode correlation analysis between actual and desired attitudes. The specific process is as follows: First, the target log loading status and VTOL mode data integrity are verified. Quaternion parameters for actual and desired attitudes are extracted by retrieving "vehicle_attitude" and "vehicle_attitude_setpoint" topic data from the logs. After anomaly filtering and timestamp standardization, the quaternions are converted into roll, pitch, and yaw Euler angles, forming an intuitive attitude data sequence. Multi-subgraph visualization charts are generated, including comparisons of actual and desired attitudes along the X / Y / Z axes and flight mode stripes. Attitude statistical characteristics under different modes are calculated simultaneously. Finally, the structured statistical data and visualization charts are stored in the results dictionary, providing data support for large-scale model analysis of attitude control accuracy, attitude tracking error, and the impact of mode switching on attitude stability under various modes.

[0070] The log data correlation and latency analysis module is used to quantify the correlation and time lag between any two types of log messages. The specific process is as follows: First, the target log loading status is verified. The system receives the two types of message names and analysis parameters (ignore time before and after, window size, etc.) specified by the user. The corresponding topic data is extracted by parsing the message name (e.g., "message_name_1" is split into topic name and field name). After time range filtering (removing invalid time periods at the beginning and end), moving average filtering (smoothing noise), extreme value pruning (suppressing outliers), and normalization, a standardized data sequence is obtained. The two types of data are unified to a common time axis through linear interpolation. The original correlation coefficient is calculated, and then cross-correlation analysis is used to locate the time offset corresponding to the maximum correlation peak, which is converted into the actual latency in seconds. A visual chart comparing the two types of data sequences is generated, and the correlation coefficient, latency time, and interpolated data are output synchronously. Finally, the structured analysis results and charts are stored, providing a quantitative basis for interpreting the multi-parameter correlation patterns of large models and helping to troubleshoot system timing anomalies or parameter coupling problems.

[0071] The log data time delay and sampling interval analysis module is used to quantify the temporal characteristics of log data for a specified topic, compatible with scenarios without message identifiers (msg_id). The specific process is as follows: First, verify the target log loading status and topic validity (supporting preset topics such as vehicle_attitude and sensor_combined). Retrieve the corresponding data by topic name and classify it by msg_id or automatically generated grouping key. Extract timestamps from each group of data and sort them. Calculate the relative time and adjacent time intervals, and statistically analyze key indicators such as the mean, standard deviation, extreme values, median, and expected frequency of the intervals. Generate visualization charts containing "time interval variation curves" and "interval distribution histograms" (labeled with 50Hz / 100Hz reference lines), and synchronously output the statistical characteristics of each group of data. Finally, store the analysis description, structured statistical data, and charts in the results dictionary, providing quantitative basis for large-scale model evaluation of sensor data timeliness and system sampling stability, and helping to troubleshoot flight control response problems caused by abnormal time delays.

[0072] The log parameter extraction and filtering module is used to extract specific parameters from PX4 logs on demand, avoiding resource redundancy caused by full extraction. The specific process is as follows: First, verify the target log loading status and receive the parameter filtering string specified by the user; obtain the initial parameter set from the log, filter parameters whose names contain the target string through string matching (case-insensitive), and sort them by parameter name; store the filtered parameters in the form of "name-value" key-value pairs in the result dictionary, without generating visualization charts; if no parameter is matched, return a prompt message. This module allows users to accurately obtain the required parameters, providing targeted data support for interpreting the correlation between parameter configuration and flight performance in large models, improving parameter query efficiency and analysis accuracy.

[0073] After obtaining the aforementioned callable function modules, a structured function information transmission system needs to be built to provide a clear "tool list" for the large model, ensuring that it can accurately schedule modules according to user needs and the actual situation of log data. Specifically, the core information of each callable module is first organized into a standardized dictionary. For example, for the "ana_offboard_mode" module, its function is clearly marked as "extracting the Offboard mode time stream", the input is "the loaded log path", the output is "second-level mode state sequence", and the applicable scenario is "mode analysis of VTOL model logs"; for the log data correlation and latency analysis module, it is stated that it supports "correlation and latency calculation of any two types of log messages", and requires key information such as "log path, two types of message identifiers, and time filtering parameters" to be passed in.

[0074] Furthermore, regarding the steps for function calls and obtaining analysis result information, combined with... Figure 3 The function call information extraction module shown constructs a dual-path collaborative information extraction system combining "traditional code parsing" and "large-scale intelligent analysis," with the specific process as follows: Traditional analysis approach: By reading the target code file, using Abstract Syntax Tree (AST) parsing technology to traverse the code structure, identify class (ClassDef) and function (FunctionDef) nodes; extract the class name, docstring, and internal method list, and simultaneously obtain the function / method name and docstring, organizing them into a structured description according to the "class-method" hierarchy or independent function form. If the docstring is missing, a default suggestion is generated, realizing automated extraction of syntax-level information of key code elements.

[0075] Large model analysis path: Also based on AST parsing, the classes and functions to be analyzed are filtered, the type, name and complete code snippets of each element are extracted and integrated into a unified request content; by calling the large model interface, the model is guided to generate in-depth descriptions of classes and functions with a structured prompt, and then the model response is parsed in the order of the request, and the results are associated with and stored with the corresponding code elements.

[0076] Auxiliary functions: Supports locating the line number range of the target function / method in the code file by using the AST (Abstract Syntax Tree) in the format of "function name" or "class name.method name", accurately extracting the original source code and providing complete code evidence for subsequent information verification or secondary analysis.

[0077] This module employs a dual-path complementary approach: it leverages traditional AST parsing to achieve efficient and accurate syntactic-level information extraction, while also utilizing a large model to overcome the limitations of document strings and generate more semantically profound descriptions. Simultaneously, it supports precise source code backtracking, providing comprehensive data support for the transmission of function call information and the understanding of code functionality within the large model.

[0078] Furthermore, according to the embodiments of the present invention... Figure 4 As shown, the large model analysis and processing module is the core execution carrier that connects large model code generation with actual log analysis. It provides a deterministic runtime environment and context support for the large model generated code. The specific process is as follows: First, the system receives the target log file path and the code string generated by the large model. Basic validation is performed: if the code is empty, an empty result is returned with a warning; for valid code, the code to be executed is printed to ensure process traceability, and then the Python code is executed using the `exec()` function. During execution, the dictionary of the module's class serves as the context container, automatically receiving the structured analysis data and visualized SVG chart data output after code execution. If an exception occurs during execution, error information is captured and a detailed stack trace is printed for troubleshooting. Finally, the analysis results and chart data corresponding to the log path are extracted from the dictionary and returned to the upper-level calling stage.

[0079] This module ensures that the log analysis code generated by the large model can run stably in the preset environment by using a fixed context structure and a unified execution entry point, avoiding execution failures caused by environmental differences. At the same time, it enables the monitoring of the code execution process and the traceability of anomalies, providing a reliable guarantee for the closed loop of "large model decision-code execution-result output".

[0080] Therefore, in some embodiments, the problem, preprocessing result information, and / or prompt words are input into a large model, analyzed using the large model, and the analysis results are obtained and output, including: The large model receives the question, prompt words, and preprocessing results. Combined with the calling function information of the second function set or the parsing results of the calling functions, semantic understanding, knowledge reasoning, comprehensive judgment, and / or association analysis are performed to obtain and output the analysis results.

[0081] In summary, the core function of the large-scale model analysis method in this invention is to generate standardized and professional HTML-formatted analysis reports based on log statistics and user needs, providing an intuitive and structured display medium for flight control log analysis results. (Continue to see...) Figure 4 As shown, its specific implementation process can be divided into the following key steps: The first step is input validation and data preprocessing. After receiving user messages and log statistics, the method first checks the validity of the data. If the data is empty, it directly returns an HTML error message to avoid subsequent errors due to invalid data. For special floating-point numbers and non-serializable objects that may exist in the log data, a custom `CustomEncoder` class is used to implement JSON serialization adaptation. This converts special floating-point numbers into readable strings, while other objects retain information as strings. Finally, the processed data is converted into a formatted JSON string with indentation, ensuring that the large model can clearly identify the log data structure and key information.

[0082] Secondly, there's the large model prompt construction and invocation phase. To guide the large model in generating HTML reports that meet the requirements, the method constructs a prompt containing multi-dimensional information: it not only integrates the user-inputted analysis requirements and preprocessed log data, but also explicitly stipulates that the report must cover four core contents: "flight status overview, performance index analysis, anomaly detection, and problem-solving suggestions." Specific requirements are also set for the HTML format (such as heading hierarchy, key value highlighting, and list usage standards) to prevent the content generated by the large model from deviating from the actual application scenario. In terms of dialogue message design, the large model is set as a "professional PX4 flight control log analyst" through the "system role," emphasizing concise descriptions and definitive suggestions. Then, the complete prompt is delivered through the "user role," ensuring that the large model can accurately understand the mission objectives. Finally, the `generate` method is called to obtain the HTML report content generated by the model.

[0083] Finally, there's the response processing and format calibration stage. Considering that large models might return content with Markdown code block tags (such as ```html`), the method uses the `clean_markdown_code_blocks` method to clean up redundant tags, ensuring the purity of the HTML code. Then, the model response is double-validated: if the response is empty, an HTML error message "Unable to generate analysis description" is returned; if the response lacks basic HTML tags (such as ```html`),... `` <h2>If so, it will automatically wrap the content. Tags are used to ensure that the output conforms to HTML formatting specifications, avoiding formatting errors during front-end display. Furthermore, the entire process uses try-except to catch exceptions. If an error occurs, the error message is encapsulated as an HTML paragraph with the "error" class, making it easier for users to identify the problem and providing a clear basis for subsequent troubleshooting.

[0084] Overall, this method, through a complete workflow design of "data preprocessing - precise prompt guidance - response format calibration," not only solves the adaptation problem of the special nature of flight control log data and the understanding threshold of large models, but also ensures the professionalism and consistency of the analysis report through standardized format requirements. This allows the log analysis results to be directly applied to practical scenarios such as report output and problem communication, thereby improving the practicality and efficiency of flight control log analysis.

[0085] In some embodiments, combined with Figure 2 , 6 As shown, the method further includes: The system receives user questions through a web interface and displays the analysis results to the user.

[0086] In some specific embodiments, combined with Figure 5 As shown, before obtaining the user's input question and determining the prompt words based on the question, the method further includes: Preset question examples are displayed on the receiving interface, making it easy for users to edit the questions they want to ask based on the question examples.

[0087] Combination Figure 6 As shown, the problem, preprocessing results, and / or prompts are input into the large model, analyzed using the large model, and the analysis results are obtained and output, including: Output an HTML report containing the following information: flight status overview, performance index analysis, anomaly detection, and resolution recommendations.

[0088] In summary, the key points and beneficial effects of the embodiments of the present invention include: I. Enhancing the Collaboration and Analysis Efficiency of Large Models (I) Large-scale model for precise guidance In the large model invocation phase, precise guidance for the large model output is achieved by constructing a prompt containing user messages, formatted log data, and detailed analysis requirements, and by setting a system role of "Professional PX4 Flight Control Log Analyst." Compared to the traditional method of directly feeding raw data into the large model without clear constraints, the content of the generated HTML analysis report is significantly more relevant. For example, when analyzing flight status overviews and performance indicators, the output strictly follows the preset title hierarchy and tag specifications, with key values ​​highlighted and problem suggestions presented in a list format, making it directly usable as a professional analysis document without the need for manual secondary processing.

[0089] (II) Analytical Dimensions and Depth Expansion Traditional log analysis tools are limited by rules and algorithms, resulting in a single analytical dimension and a focus primarily on basic parameter statistics. This invention, leveraging a large-scale model, achieves a leap from single-parameter statistics to multi-dimensional correlation analysis. The large-scale model can combine multi-dimensional data such as flight status, performance indicators, and anomalies to uncover in-depth information, such as the correlation between abnormal angular velocity during a certain flight phase and subsequent attitude deviations. In actual testing, the accuracy of root cause analysis of flight failures is significantly improved, uncovering more than 10 potential failure modes missed by traditional tools, providing a more comprehensive basis for flight control system optimization and troubleshooting.

[0090] II. Results Output and Application Value Upgrade (a) HTML Output Standardization and Usability The HTML analysis results generated by this invention strictly adhere to formatting standards, with clear headings, highlighted key information, and a logical content structure. Compared to the plain text or disorganized HTML fragments output by traditional tools, its usability is significantly improved in professional report presentations and team collaborative communication. In practical applications by flight control R&D teams, communication efficiency based on this HTML report has been greatly enhanced, allowing R&D personnel to quickly grasp key information and shorten the fault analysis and optimization decision-making cycle.

[0091] (ii) Error handling and robustness enhancement Traditional log analysis workflows are prone to crashing or outputting meaningless messages when encountering errors. This invention addresses this by implementing unified exception capture and error handling, providing clear and localized HTML error messages at each stage, including data validation, large model invocation, and response processing. When encountering situations such as empty data, unresponsive large models, or abnormal HTML formatting, it accurately identifies the problem type and approximate location, shortening troubleshooting time. Simultaneously, it automatically repairs lightweight errors such as missing HTML tags, ensuring the analysis workflow continues to run under non-fatal errors, thus improving system robustness.

[0092] III. Overall Process Efficiency and Cost Optimization (a) Process automation and efficiency improvement From log data input to HTML report output, this invention achieves full-process automation. Compared to the traditional model of manual data preprocessing and result organization and analysis, taking a typical single flight mission (2 hours of flight, generating 500MB of log data) as an example, traditional analysis takes more than 1 hour, while the process of this invention can be completed in just 10 minutes, including data processing, large model analysis, and formatted result output, greatly improving the efficiency of flight control log analysis.

[0093] (ii) Reduced human and resource costs Thanks to process automation and improved analytical accuracy, manpower can be reduced in the research, development, and maintenance of flight control systems. Traditional solutions require professional personnel to manually process data and interpret analysis results; this invention can replace most basic manual operations. Simultaneously, the precise guidance of the large model and error handling mechanism reduce the resource consumption of repeated debugging and re-analysis. In terms of hardware resource usage, server computing power consumption is reduced, resulting in significant overall cost optimization and bringing considerable cost advantages to UAV R&D and maintenance companies.

[0094] Combination Figure 7 As shown, the present invention also protects a UAV flight log data analysis device 700, the UAV flight log data analysis device 700 comprising: The question acquisition module 710 is suitable for acquiring questions input by users and determining prompt words based on the questions; I. Analysis paradigm: from "pure rule-driven" or "single tool analysis" to collaborative analysis of "large model + dedicated module".

[0095] The preprocessing module 720 is adapted to obtain the corresponding time stream data from the flight log data according to the prompt words, and to determine the calling functions from the local first function set and the second function set built into the large model according to the prompt words, and to use the calling functions to perform calculations on the time stream data to obtain mutually verified preprocessing result information. The model analysis module 730 is adapted to input the problem, preprocessing result information and / or prompt words into the large model, use the large model for analysis, and obtain and output the analysis results.

[0096] According to the UAV flight log data analysis device in this embodiment, the large model is accurately guided and trained using local function sets and log data, which expands the dimensions and depth of data analysis, enhances the robustness and practicality of the system, improves the scene adaptability, and reduces the cost of use.

[0097] In some embodiments, the large model pre-stores and parses a second function set, the second function set corresponding to the first function set, and the preprocessing module 720 is further adapted to: The time stream data is preprocessed using the functions called in the first function set of the local machine to obtain the first processing result; The second processing result is obtained by preprocessing the time stream data using the functions called in the second function set. If the first processing result and the second processing result are the same or similar, the preprocessing result information is output according to the first processing result and the second processing result. If the two are neither the same nor similar, the process is repeated until the same or similar result is obtained, or an error is returned and retraining is performed.

[0098] In some embodiments, the preprocessing module 720 retrieves the corresponding time stream data from the flight log data based on the prompt words, including: Load flight log data and save it locally; Extract and preprocess the corresponding time stream data from the data unit according to the preset mode type; The flight log data is stored in multiple data units, each of which is constructed using the mapping relationship between log path, mode type, and time stream data.

[0099] In some embodiments, the first function set and the second function set include the following functions: Rotor mode speed error analysis module, fixed-wing mode speed relationship analysis module, IMU sensor data Fourier transform analysis module, GPS data accuracy analysis module, UAV angular velocity analysis module, UAV attitude analysis module, log data correlation and delay analysis module, log data time delay and sampling interval analysis module, or log parameter extraction and filtering module.

[0100] In some embodiments, the model analysis module 730 is adapted to: The large model receives the question, prompt words, and preprocessing results. Combined with the calling function information of the second function set or the parsing results of the calling functions, semantic understanding, knowledge reasoning, comprehensive judgment, and / or association analysis are performed to obtain and output the analysis results.

[0101] In some embodiments, the device 700 is further adapted to: The system receives user questions through a web interface and displays the analysis results to the user.

[0102] In some embodiments, the problem acquisition module 710 is further adapted to: Preset question examples are displayed on the receiving interface, making it easy for users to edit the questions they want to ask based on the question examples; Furthermore, the model analysis module 730 is also suitable for: Output an HTML report containing the following information: flight status overview, performance index analysis, anomaly detection, and resolution recommendations.

[0103] It should be noted that the specific implementation methods of the above-mentioned device embodiments can be referred to the specific implementation methods of the corresponding methods described above, and will not be repeated here.

[0104] It should be noted that: The algorithms and displays provided herein are not inherently related to any particular computer, virtual device, or other equipment. Various general-purpose devices can also be used in conjunction with the teachings herein. The required structure for constructing such devices is apparent from the above description. Furthermore, this invention is not directed to any particular programming language. It should be understood that the contents of the invention described herein can be implemented using various programming languages, and the above description of specific languages ​​is for the purpose of disclosing the best mode of implementation of the invention.

[0105] Numerous specific details are set forth in the specification provided herein. However, it will be understood that embodiments of the invention may be practiced without these specific details. In some instances, well-known methods, structures, and techniques have not been shown in detail so as not to obscure the understanding of this specification.

[0106] Similarly, it should be understood that, in order to streamline the invention and aid in understanding one or more of the various aspects of the invention, features of the invention are sometimes grouped together in a single embodiment, figure, or description thereof in the above description of exemplary embodiments of the invention. However, this disclosure should not be construed as reflecting an intention that the claimed invention requires more features than are expressly recited in each claim.

[0107] Those skilled in the art will understand that modules in the device of the embodiments can be adaptively changed and placed in one or more devices different from that embodiment. Modules, units, or components in the embodiments can be combined into a single module, unit, or component, and further, they can be divided into multiple sub-modules, sub-units, or sub-components. Except where at least some of such features and / or processes or units are mutually exclusive, any combination can be used to combine all features disclosed in this specification (including the accompanying claims, abstract, and drawings) and all processes or units of any method or device so disclosed. Unless expressly stated otherwise, each feature disclosed in this specification (including the accompanying claims, abstract, and drawings) may be replaced by an alternative feature that serves the same, equivalent, or similar purpose.

[0108] Furthermore, those skilled in the art will understand that although some embodiments described herein include certain features included in other embodiments but not others, combinations of features from different embodiments are meant to be within the scope of the invention and form different embodiments.

[0109] The various component embodiments of the present invention can be implemented in hardware, or as software modules running on one or more processors, or a combination thereof. Those skilled in the art will understand that microprocessors or digital signal processors (DSPs) can be used in practice to implement some or all of the functions of some or all of the components in the UAV flight log data analysis device according to embodiments of the present invention. The present invention can also be implemented as a device or apparatus program (e.g., a computer program and computer program product) for performing part or all of the methods described herein. Such programs implementing the present invention can be stored on a computer-readable medium or can be in the form of one or more signals. Such signals can be downloaded from an Internet website, provided on a carrier signal, or provided in any other form.

[0110] This invention provides a non-volatile computer storage medium storing at least one executable instruction that can execute the above-described UAV flight log data analysis method in any of the above method embodiments.

[0111] Figure 8 The diagram shows a structural schematic of an embodiment of the computing device of the present invention. The computing device may be a computer, a local server, or a cloud server. The specific embodiments of the present invention do not limit the specific structure of the drone.

[0112] like Figure 8 As shown, the computing device may include: a processor 802, a communications interface 804, a memory 806, and a communications bus 808.

[0113] The processor 802, communication interface 804, and memory 806 communicate with each other via communication bus 808. Communication interface 804 is used to communicate with other network elements such as clients or other servers. Processor 802 executes program 810, specifically performing the relevant steps in the above-described embodiment of the UAV flight log data analysis method for UAVs.

[0114] Specifically, program 810 may include program code that includes computer operation instructions.

[0115] Processor 802 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present invention. The one or more processors included in the drone may be processors of the same type, such as one or more CPUs; or they may be processors of different types, such as one or more CPUs and one or more ASICs.

[0116] Memory 806 is used to store program 810. Memory 806 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0117] Specifically, program 810 can be used to enable processor 802 to perform the operations corresponding to the above-described UAV flight log data analysis method embodiment.

[0118] It should be noted that the above embodiments are illustrative of the invention and not restrictive, and that those skilled in the art can devise alternative embodiments without departing from the scope of the appended claims. In the claims, any reference signs placed between parentheses should not be construed as limiting the claims. The word "comprising" does not exclude the presence of elements or steps not listed in the claims. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. The invention can be implemented by means of hardware comprising several different elements and by means of a suitably programmed computer. In the unit claims enumerating several means, several of these means may be embodied by the same item of hardware. The use of the words first, second, and third, etc., does not indicate any order. These words can be interpreted as names. < / h2> < / h3>

Claims

1. A method for analyzing unmanned aerial vehicle flight log data, the method comprising: receiving a flight log data file; and analyzing the flight log data file to determine a flight log data file quality score. The method comprises: acquiring a question input by a user, determining a prompt word according to the question; acquiring corresponding time flow data from flight log data according to the prompt word, and respectively determining calling functions from a first function set locally and a second function set built in a large model according to the prompt word, performing operation on the time flow data by using the calling functions to obtain preprocessed result information verified mutually; inputting the question, the preprocessed result information and / or the prompt word into the large model, performing analysis by using the large model to obtain and output analysis results.

2. The method of claim 1, wherein, The operation on the time flow data by using the calling functions to obtain preprocessed result information verified mutually comprises: performing preprocessing on the time flow data by using the calling functions in the first function set locally to obtain first processing results; performing preprocessing on the time flow data by using the calling functions in the second function set in the large model to obtain second processing results; comparing the first processing results and the second processing results, if the first processing results and the second processing results are the same or similar, outputting the preprocessed result information according to the first processing results and the second processing results.

3. The method according to claim 1 or 2, characterized in that, The operation on the time flow data by using the calling functions to obtain preprocessed result information verified mutually comprises: loading and saving the flight log data locally; extracting and preprocessing corresponding time flow data from data units according to a preset mode type; The flight log data is saved in a plurality of data units, and each data unit is constructed by using a mapping relationship of a log path, a mode type and time flow data.

4. The method of claim 2, wherein, The first function set and the second function set comprise calling functions as follows: a rotor mode speed error analysis module, a fixed-wing mode speed relationship analysis module, an IMU sensor data Fourier transform analysis module, a GPS data accuracy analysis module, a UAV angular velocity analysis module, a UAV attitude analysis module, a log data correlation and delay analysis module, a log data time delay and sampling interval analysis module or a log parameter extraction and screening module.

5. The method according to claim 1 or 2, characterized in that, The operation on the time flow data by using the calling functions to obtain preprocessed result information verified mutually comprises: receiving the question, the prompt word and the preprocessed result information by using the large model, combining calling function information of the second function set or analysis results of the calling functions, performing semantic understanding, knowledge reasoning, comprehensive research and judgment and / or correlation analysis to obtain and output analysis results.

6. The method of claim 1 or 2, wherein, The method further comprises: receiving a question of a user through a webpage interface and showing analysis results to the user.

7. The method according to claim 1 or 2, characterized in that, Before acquiring a question input by a user, determining a prompt word according to the question, the method further comprises: displaying preset question examples on a receiving interface to facilitate the user to edit a question to be proposed according to the question examples; The operation on the time flow data by using the calling functions to obtain preprocessed result information verified mutually comprises: outputting a report in HTML format, comprising information as follows: flight state profile, performance index analysis, abnormality detection and solution suggestion. 8.A UAV flight log data analysis apparatus, characterized by, The device comprises: a question obtaining module adapted to obtain a question input by a user, and determine a prompt word according to the question; a preprocessing module adapted to obtain corresponding time flow data from flight log data according to the prompt word, and determine calling functions from a first function set locally and a second function set built in a large model respectively according to the prompt word, and perform operation on the time flow data by using the calling functions to obtain preprocessed result information verified with each other; a model analysis module adapted to input the question, the preprocessed result information and / or the prompt word into the large model, perform analysis by using the large model, and obtain and output an analysis result.

9. A computing device, comprising: A processor and a memory arranged to store computer executable instructions which, when executed, cause the processor to perform the method of analyzing the flight log data of the unmanned aerial vehicle according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium stores one or more programs which, when executed by a processor, implement the method of analyzing the flight log data of the unmanned aerial vehicle according to any one of claims 1-7.

Images