Multi-module collaborative flight simulator control panel and method

Abstract

This invention belongs to the field of aviation flight training technology, specifically involving a multi-module collaborative flight simulator control panel and method, solving the problems of low interaction efficiency, subjective evaluation, delayed warnings, and time-consuming debriefing in existing systems. This invention relies on traditional iOS hardware and includes: an intelligent voice interaction module that recognizes instructor voice via ASR, generates commands through intent recognition and three-level slot completion, and ensures accurate interaction with a two-layer flight phase recognition logic; a flight quality monitoring module that runs a rule engine and an enhanced LSTM neural network in parallel, combining an adaptive fusion algorithm to output a comprehensive flight quality score; a real-time reminder module that dynamically adjusts warning priorities according to event level and flight phase, providing multi-mode warnings; and an intelligent debriefing module that integrates multi-source data and generates personalized structured reports according to template priorities. This invention achieves intelligent interaction, quantitative evaluation, and automatic debriefing during training, improving training efficiency and evaluation objectivity.

Description

Technical Field

[0001] This invention belongs to the fields of aviation flight training and artificial intelligence technology, and specifically relates to a multi-module collaborative flight simulator control panel and method. Background Technology

[0002] In the field of aviation flight training, full-motion flight simulators are the core equipment for pilot training. The instructor control panel (IOS), as the core of the simulator's control and evaluation, directly affects training efficiency, safety, and the objectivity of assessment. As aviation training transforms towards refinement and intelligence, artificial intelligence-optimized operating systems have become a key support for overcoming traditional technological bottlenecks. However, existing related technologies have not yet fully realized their core value and have many limitations:

[0003] Traditional flight simulator instructor console systems (such as products from CAE and FlightSafety International) are based on physical buttons and on-screen menus. While they have basic functions such as flight parameter monitoring, freezing, resetting, and fault injection, they lack intelligent voice interaction, automated evaluation, and commentary generation capabilities. They rely entirely on manual operation and subjective judgment by instructors, resulting in low interaction efficiency and inconsistent evaluation standards.

[0004] The speech recognition-based aviation training system only supports predefined command control, lacks context understanding and multimodal interaction capabilities, and has insufficient command recognition accuracy in complex scenarios.

[0005] Flight operation quality monitoring systems (such as FOQA or FDM systems used by airlines) are mainly used for post-flight data analysis of actual flights. They cannot achieve real-time evaluation and debriefing generation during simulator training and are not deeply integrated with the training process.

[0006] Intelligent commentary and training feedback systems mostly use simple rules or basic machine learning models, lacking voice interaction, real-time reminders, and LSTM time series modeling capabilities, and the commentary content lacks specificity and personalization.

[0007] While multimodal interaction and context-aware systems have been applied in scenarios such as smart cockpits and virtual assistants, they have not yet been fully implemented in flight simulation training, especially lacking deep integration with flight phases and operational quality assessment.

[0008] The aforementioned technical deficiencies result in problems such as low interactive efficiency, strong subjectivity in assessment, delayed real-time intervention, and time-consuming preparation for debriefing during training. Instructors are distracted by operating complex menus during intense training, making it difficult to focus on monitoring trainees and flight status. The assessment of trainees' operations relies on personal experience, resulting in unconvincing results and inconsistent standards. Monitoring massive amounts of flight parameters can easily lead to missed risks and untimely intervention. After training, a lot of time needs to be spent reviewing data and organizing debriefing points, resulting in a heavy workload.

[0009] Therefore, there is an urgent need for a full-motion flight simulator instructor control panel and method that integrates intelligent interaction, real-time monitoring, quantitative evaluation and automatic commentary, and forms an artificial intelligence optimized operating system through its multi-module collaborative architecture, thereby solving the shortcomings of existing technologies. Summary of the Invention

[0010] To address the aforementioned problems in existing technologies, namely low interaction efficiency, high subjectivity in evaluation, delayed real-time intervention, time-consuming debriefing preparation, and lack of deep integration capabilities for contextual understanding, multimodal interaction, real-time evaluation, and flight phase and operational quality assessment, the first aspect of this invention proposes a multi-module collaborative flight simulator control panel, comprising:

[0011] The intelligent voice interaction module is configured to convert instructor voice data into text data, perform intent recognition on the text data, and fill parameter slots; when parameter slots are incomplete, it integrates the current flight phase, interaction context, and historical operation data to complete the missing slots; and generate and execute control commands based on the complete intent and slots; wherein, the flight phase recognition adopts a two-layer logic of main rule judgment and decision tree verification, and sets a transition window to ensure the stability of phase switching.

[0012] The flight quality monitoring and analysis module is configured to process real-time flight parameters in parallel. One path outputs rule-triggered events and penalty scores through a rule engine, while the other path outputs a quantitative score of operational quality through an enhanced LSTM neural network. An adaptive fusion algorithm is used to fuse the rule-triggered events, penalty scores, and quantitative scores of operational quality to generate a comprehensive flight quality score and event stream. The enhanced LSTM neural network adopts an enhanced architecture that includes a global attention module and a local attention module.

[0013] The real-time intelligent alert module is configured to receive the event stream, dynamically prioritize the alerts based on the severity level of the events in the event stream and the current flight phase, assign multi-mode alerts to the events based on the alert priority ranking, and generate alert records.

[0014] The intelligent evaluation generation module is configured to receive multi-source data and perform correlation and fusion processing, wherein the multi-source data includes the text data, the comprehensive flight quality score and event stream, and the early warning record;

[0015] Based on the multi-source data after correlation and fusion, a structured and personalized review report is generated through intelligent template matching and semantic filling.

[0016] The beneficial effects of this invention are:

[0017] 1) Intelligent voice interaction enables "hands-free operation". Instructors do not need to operate complex menus. They can directly issue commands or query status through natural language, which greatly reduces operation time and allows them to focus more on monitoring students and flight status.

[0018] 2) The “rule engine + LSTM” hybrid model transforms soft skills such as “manipulation softness” and “state awareness” into quantifiable scores, unifying the evaluation criteria and overcoming the evaluation differences caused by different instructors’ personal experience and style, thereby improving training quality and certification credibility.

[0019] 3) The real-time intelligent reminder module realizes proactive early warning of "risk finding people". The system continuously monitors flight parameters and can capture subtle deviations that instructors may overlook. Through multi-mode early warning, it promptly reminds potential risks (such as speeding, excessive bank angle, and hard landing risk), adding a reliable "electronic safety officer" to training safety.

[0020] 4) The intelligent review generation module quickly and automatically generates a structured review report with detailed data and rich graphics after the training is completed, freeing instructors from tedious data sorting and report writing. The report can accurately point out the problems and is linked to the instructor's voice annotations, making the review preparation time shorter, the content more focused, and the feedback more efficient. Attached Figure Description

[0021] Other features, objects, and advantages of this application will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0022] Figure 1 This is the overall system architecture diagram of the multi-module collaborative flight simulator control panel of the present invention.

[0023] Figure 2 This is a flowchart illustrating the implementation of the intelligent voice interaction module of the multi-module collaborative flight simulator control panel of the present invention.

[0024] Figure 3 This is a flowchart of the flight phase identification process of the multi-module collaborative flight simulator control panel of the present invention.

[0025] Figure 4 This is a flowchart illustrating the implementation of the flight quality monitoring and analysis module of the multi-module collaborative flight simulator control panel of the present invention. Detailed Implementation

[0026] The present application will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention. Furthermore, it should be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0027] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0028] To more clearly explain the multi-module collaborative control panel of a fully dynamic flight simulator based on an intelligent system according to the present invention, the following description is in conjunction with... Figures 1 to 4 The steps in the embodiments of the present invention will be described in detail below.

[0029] This invention proposes a multi-module collaborative control panel for a full-motion flight simulator based on an intelligent system. This panel includes traditional iOS hardware (touchscreen display, physical buttons, communication interfaces, etc.) and an intelligent system. See [link to relevant documentation]. Figure 1 ,include:

[0030] Intelligent voice interaction module, see Figure 2 The system is configured to convert instructor speech data into text data (e.g., through Automatic Speech Recognition, ASR), perform intent recognition on the text data (e.g., SetParameter, QueryStatus, SetReminder) and fill parameter slots (e.g., ParameterName, TargetValue); verify the integrity of the slots corresponding to the intents; when a parameter slot is incomplete, it integrates the current flight phase, interaction context, and historical operation data to complete the missing slots; generate and execute control commands based on the complete intents and slots; wherein, the flight phase recognition adopts a two-layer logic of main rule judgment and decision tree verification, and sets a transition window to ensure the stability of phase switching;

[0031] In this embodiment, missing slots are filled by integrating the current flight phase, interaction context, and historical operation data. Specifically, the missing slots are filled by sequentially calling the corresponding data sources in the order of key parameters of the flight phase, recent operation history, and common command preferences. Among them, the current flight phase (such as Taxi, Takeoff, Approach, etc.) is obtained by querying the flight phase recognizer; the recent operation history is obtained by retrieving the recent dialogue topics and incomplete commands recorded by the dialogue context tracker; and common command preferences are obtained by retrieving the instructor's historical common command preference data.

[0032] See Figure 3 The flight phase is identified through a two-layer logic of "main rule judgment + decision tree verification". The main rule judgment is: set judgment conditions based on wheel load status, speed, altitude, throttle position parameters, and output the initial flight phase.

[0033] The decision tree verification is as follows: when the rule determination is ambiguous, take the wheel load status as the root node and perform secondary confirmation according to the order of ground clearance, vertical speed, throttle position, and the previous stage.

[0034] Smooth transition between stages: Set a 2-second transition window. Update the system after the new stage's criteria are met for 2 seconds to ensure a smooth transition and avoid frequent transitions.

[0035] Finally, based on the complete intent and slot, simulator control commands or data query requests are generated and executed, and the execution results are fed back to the instructor through text-to-speech (TTS) technology.

[0036] Flight quality monitoring and analysis module, see Figure 4 The system is configured to process real-time flight parameters in parallel. One path outputs rule-triggered events and penalty scores through a rule engine, while the other path outputs a quantitative score for operational quality through an enhanced LSTM neural network. An adaptive fusion algorithm is used to fuse the rule-triggered events, penalty scores, and quantitative scores for operational quality to generate a comprehensive flight quality score and event stream. The enhanced LSTM neural network adopts an enhanced architecture that includes a global attention module and a local attention module.

[0037] In this embodiment, before parallel processing, the real-time flight parameters are preprocessed. Twelve core parameters (covering three dimensions: control input, flight attitude, and dynamic response) collected at a frequency of 25Hz in the past 15 seconds are selected to form a time-series feature sequence. Outliers are removed by the 3σ criterion, and Z-score standardization (mean μ=0, standard deviation σ=1) is used to unify the dimensions. A continuous input sequence is generated through a sliding window (step size 5 seconds) as time-series data.

[0038] Specifically, the rule engine outputs rule trigger events and penalty scores. The rule engine runs on a dynamic and reconfigurable rule base, which allows instructors to customize rule thresholds or inject new rules through the control panel. For example, the dynamic and reconfigurable rule base is built based on the Civil Aviation Administration's "Basic Flight Rules" and crew training specifications. Each rule contains four elements: trigger conditions, threshold range, event level, and associated parameters. The system outputs rule trigger events and penalty scores, and allows instructors to customize rule thresholds or add new rules. The system automatically iterates the rule base every quarter.

[0039] The rules trigger events include: Takeoff Phase - Overspeed Pitch Rule: When the flight phase is takeoff, wheel load is off the ground, radio altitude < 1000 feet, airspeed > V2 + 20 knots, and pitch angle > 15°, an overspeed pitch event is triggered; Cruise Phase - Altitude Deviation Rule: When the flight phase is cruise, altitude > 10000 feet, and the absolute value of the deviation between actual altitude and target altitude > 300 feet for more than 10 seconds, an altitude deviation event is triggered; Approach Phase - Course Deviation Rule: When the flight phase is approach, ground load is < 1500 feet, and the absolute value of the heading deviation > 0.7 or the absolute value of the glide slope deviation > 0.6 and the glide slope has been captured, a course deviation event is triggered; Landing Phase - Enhanced Hard Landing Rule: When the flight phase is landing, wheel load is on the ground, vertical speed < -550 feet / minute, and vertical acceleration < -1.8g, a hard landing event is triggered; Special Situation Phase - Engine failure handling rules: Any flight phase, single engine failure, bank angle > 25°, altitude loss > 500 feet, handling time > 15 seconds triggers an improper engine failure handling event; Taxi phase - runway deviation rules: Taxi phase, wheel touchdown status, deviation from runway centerline > 15 meters, speed > 10 knots triggers a taxi deviation event;

[0040] The cumulative penalty score Penalty_total is the cumulative penalty for rule-triggered events output by the rule engine, Penalty_total = Σ[Severity (event_i) × Occur_Count (event_i)]; the severity of events is graded as follows: Level 1 event (25 points / event), Level 2 event (15 points / event), Level 3 event (8 points / event), and Level 4 event (3 points / event); the penalty doubles for each event triggered more than 3 times.

[0041] The enhanced LSTM neural network outputs a quantitative score for operational quality. This enhanced LSTM neural network employs an enhanced architecture including a global attention module and a local attention module. Specifically, the enhanced LSTM neural network includes: an input layer configured to receive and process temporal data of a fixed length (375 time steps, 12 features, corresponding to 15 seconds × 25Hz), covering the core dimensions of manipulation, posture, and dynamic response; a first LSTM layer, directly receiving data from the input layer, consisting of 160 neurons, using a tanh activation function, setting a dropout ratio of 0.2, and configured to return the complete sequence; a global attention module (Bahdanau): concatenated with the first LSTM layer, globally weighting the output sequence to capture critical time periods crucial to the entire evaluation task; a local attention module: concatenated with the global attention module, using a 3-second sliding window to traverse the sequence, focusing on identifying and enhancing feature representations within short local time periods where inputs such as joystick input or throttle change drastically; and a second LSTM layer consisting of 80 neurons, using a tanh activation function and a dropout ratio of 0.2. The opout ratio receives the feature sequence refined by the local attention module, performs higher-level temporal integration, and outputs a comprehensive feature vector to the subsequent fully connected layers. The first fully connected layer receives the output of the second LSTM layer, contains 64 neurons, uses the ReLU activation function, and applies L2 regularization (λ=0.001) to prevent overfitting. The second fully connected layer is sequentially connected to the first fully connected layer, contains 32 neurons, uses the ReLU activation function, and is used for further feature compression and nonlinear transformation. The output layer, directly connected to the second fully connected layer, consists of 2 neurons and uses the Sigmoid activation function to constrain the final output to the 0~1 range, corresponding to the quantitative scores of "manipulation smoothness" (Score_smooth) and "manipulation predictability" (Score_pred), respectively.

[0042] The enhanced LSTM neural network parameters are optimized through training using a dynamic adaptive batch size and combined learning rate strategy. Specifically, the dataset consists of 800 hours of full-motion flight simulator data, covering 12 typical training scenarios for mainstream aircraft such as the Boeing 737 and Airbus A320, with a total data size of approximately 200GB. Label generation involves eight senior instructors with an average of 12 years of teaching experience and holding Civil Aviation Administration inspector qualifications, who score the data back-to-back using a 10-point scale. After outlier removal and weighted averaging, the labels are normalized to the continuous labels required for model training. Training optimization employs a dynamic... An adaptive batch size (adjusted between 32 / 64 / 128 based on sequence complexity) combined with a CosineAnnealing warm-up learning rate strategy was employed. The AdamW optimizer (weight decay of 0.0001) was used to minimize the weighted mean squared error loss (softness weight 0.55, predictive weight 0.45). The training process employed transfer learning, first pre-training with 600 hours of general-purpose data, then fine-tuning with 200 hours of data from a specific device model. Model validation and selection were performed using 5-fold cross-validation to ensure a validation set accuracy ≥94% and a test set F1 score ≥0.90.

[0043] Therefore, the enhanced LSTM neural network adopts a "global attention (Bahdanau) + local attention" concatenated architecture. Local attention focuses on key features during the manipulation mutation period through a 3-second sliding window. The network is configured with a dynamically adaptive batch size (32 / 64 / 128) and a combined learning rate strategy of CosineAnnealing + Warmup. After transfer learning pre-training (600 hours of general data) and fine-tuning with model data (200 hours of model-specific data), it outputs two types of operation quality quantitative scores: "manipulation smoothness" (Score_smooth) and "manipulation predictability" (Score_pred).

[0044] The adaptive fusion algorithm in the flight quality monitoring and analysis module executes the following formula:

[0045] Final_Score = {[W_smooth (t) × Score_smooth + W_pred (t) × Score_pred] × Phase_Coeff - Penalty_total - Penalty_overlap} × Ability_Coeff;

[0046] Wherein, W_smooth(t) and W_pred(t) are dynamic weights, W_smooth(t) + W_pred(t) = 1, dynamically adjusted according to the flight phase, for example, takeoff / landing phase (W_smooth = 0.65, W_pred = 0.35), cruise phase (W_smooth = 0.4, W_pred = 0.6), and emergency handling phase (W_smooth = 0.3, W_pred = 0.7); Phase_Coeff is the phase difficulty coefficient, dynamically adjusted according to the difficulty of the flight phase, with a value range of 0.9~1.1; 1.0 for normal scenarios (cruise, taxiing), 0.9 for high-difficulty scenarios (takeoff, landing, emergency), and 0.9 for basic training scenarios (straight flight, simple turn); Penalty_total is the cumulative penalty score; Penalty_overlap is the overlap penalty score, which is additionally added when more than 2 events are triggered in the same time window (within 3 seconds), Penalty_overlap = Σ[Severity [(event_j)×0.2] (j represents overlapping events), with a maximum cumulative score of 20 points; Ability_Coeff is the trainee's ability coefficient, dynamically calculated based on the trainee's historical training data, with a value range of 0.8~1.2; 0.8~0.9 for novice trainees (historical average score <70 points), 1.0 for intermediate trainees (70 points ≤ historical average score ≤ 85 points), and 1.1~1.2 for advanced trainees (historical average score > 85 points); Final_Score is the comprehensive flight quality score, with output constraints of a minimum of 0 points and a maximum of 100 points, below 60 points being "unqualified" and above 85 points being "excellent";

[0047] The real-time intelligent alert module is configured to receive the event stream (including event type, severity level, associated parameters, and trigger time), dynamically prioritize the alerts based on the severity level of the events in the event stream and the current flight phase, assign multi-mode alerts to the events based on the alert priority ranking, and generate alert records.

[0048] In this embodiment, the real-time intelligent alert module pre-stores an event priority mapping table. The module determines the final warning priority of the event by querying the current flight stage and the severity level of the event, according to the following sorting: The dynamic warning priority sorting includes:

[0049] Level 1 Priority: Level 1 events (such as hard landing, engine mishandling) will trigger an immediate warning regardless of the flight phase;

[0050] Level 2 Priority: Level 3 events (such as serious deviation from the flight path or exceeding altitude limits) + critical flight phases (takeoff, landing, and special situations) are upgraded to Level 2;

[0051] Level 3 priority: Level 3 events in non-critical flight phases + Level 4 events;

[0052] Early warning methods include:

[0053] First-level priority: prominent red pop-up window on the screen + voice alarm (repeated twice) + vibration alert;

[0054] Secondary priority: Yellow pop-up window on screen + voice alarm (1 time);

[0055] Level 3 Priority: Blue on-screen alert bar + text warning;

[0056] The warning content includes: generating concise and clear warning text data based on the event type and associated parameters (such as "Attention: Altitude deviation from target is 350 feet, has been ongoing for 12 seconds").

[0057] The early warning log includes: automatically recording the trigger time, processing status, and subsequent operations of all early warning events for post-training review and analysis;

[0058] The intelligent evaluation generation module is configured to receive multi-source data and perform correlation and fusion processing, wherein the multi-source data includes the text data, the comprehensive flight quality score and event stream, and the early warning record;

[0059] Based on the multi-source data after association and fusion, a structured and personalized review report is generated through intelligent template matching and semantic filling.

[0060] In this embodiment, receiving and fusing multi-source data involves the following steps: event log scanning (traversing the event logs output by the monitoring module in ascending order of timestamps, filtering events with a Severity score ≥ 8, and marking the event name, trigger time, severity level, and associated parameter ID); parameter snapshot extraction (extracting complete flight parameter snapshots within ±8 seconds before and after the event trigger time, and calculating parameter extreme values, average values, and trends); semantic association retrieval (extracting real-time voice annotations from instructors using ASR technology, matching voice and text data near the event timestamp based on the TF-IDF algorithm, and extracting keywords); and event correlation analysis (clustering multiple events triggered by the same time period or the same operation, and marking the main event and associated events).

[0061] In the intelligent review generation module, the priority of intelligent template matching is: personalized template > scenario template > basic template. The optimal template is matched according to the priority of event name + flight stage + trainee ability coefficient. Among them, personalized template is selected according to trainee ability coefficient, scenario template is selected according to flight stage and event type, and basic template is a general event review framework.

[0062] Semantic Filling: Event parameters (timestamp, speed, height, etc.) are filled into the template according to placeholder type (numerical, text, boolean). The results of voice keywords and event correlation analysis are integrated to optimize the presentation and enhance naturalness and relevance. The depth of suggestions is adjusted according to the student's ability coefficient. New students focus on basic operation norms, while advanced students focus on optimizing and improving skills.

[0063] The debriefing report is structured as follows: Training overview (duration, device type, scenario, overall score) → Summary of key events (sorted by severity level, including annotations of main and related events) → Operational quality analysis (quantified score + strengths / weaknesses) → Review of early warning events (early warnings and their impact due to untimely responses) → Personalized optimization suggestions (sorted by priority) → Key points for handling special situations (for events with a severity score ≥ 15); Report format: Supports three formats: text version, graphic version (including parameter change curves), and PDF export.

[0064] It should be noted that the multi-module collaborative flight simulator control panel provided in the above embodiments is only an example of the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the modules or steps in the embodiments of the present invention can be further decomposed or combined. For example, the modules in the above embodiments can be merged into one module, or further divided into multiple sub-modules to complete all or part of the functions described above. The names of the modules and steps involved in the embodiments of the present invention are only for distinguishing the various modules or steps and are not considered as an improper limitation of the present invention.

[0065] A second embodiment of the present invention provides a multi-module collaborative flight simulator control method, based on the aforementioned multi-module collaborative flight simulator control panel, comprising the following steps:

[0066] Collect and recognize instructors' voice data and convert it into text data to generate executable control commands;

[0067] Parallel processing of real-time flight parameters generates a comprehensive flight quality score and event stream;

[0068] Receive and process the event stream to generate multi-mode early warning signals and early warning records;

[0069] The text data, the comprehensive flight quality score, the event stream, and the early warning records are semantically correlated, and a structured review report is generated through data fusion.

[0070] The semantic association includes: using the event trigger timestamp as a benchmark, associating text data within the ΔT time window with the corresponding event, and storing the associated text data keywords as contextual information of the event in the review report.

[0071] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process and related explanations of the methods described above can be found in the corresponding processes in the foregoing system embodiments, and will not be repeated here.

[0072] Those skilled in the art will recognize that the modules and method steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. The programs corresponding to the software modules and method steps can be placed in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disks, removable disks, CD-ROMs, or any other form of storage medium known in the art. To clearly illustrate the interchangeability of electronic hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in electronic hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the invention.

[0073] The terms “first”, “second”, etc., are used to distinguish similar objects, not to describe or indicate a specific order or sequence.

[0074] The term "comprising" or any other similar term is intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus / device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent in such process, method, article, or apparatus / device.

[0075] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. A multi-module collaborative flight simulator control panel, characterized in that, include: The intelligent voice interaction module is configured to convert the instructor's voice data into text data, perform intent recognition on the text data, and fill parameter slots. When parameter slots are incomplete, the missing slots are filled by integrating the current flight phase, interaction context, and historical operation data; control commands are generated and executed based on the complete intent and slots; among them, flight phase identification adopts a two-layer logic of main rule judgment and decision tree verification, and a transition window is set to ensure the stability of phase switching; The flight quality monitoring and analysis module is configured to process real-time flight parameters in parallel. One path outputs rule-triggered events and penalty scores through a rule engine, while the other path outputs a quantitative score of operational quality through an enhanced LSTM neural network. An adaptive fusion algorithm is used to fuse the rule-triggered events, penalty scores, and quantitative scores of operational quality to generate a comprehensive flight quality score and event stream. The enhanced LSTM neural network adopts an enhanced architecture that includes a global attention module and a local attention module. The adaptive fusion algorithm in the flight quality monitoring and analysis module executes the following formula: Final_Score = {[W_smooth (t) × Score_smooth + W_pred (t) × Score_pred]× Phase_Coeff - Penalty_total - Penalty_overlap} × Ability_Coeff; Where W_smooth(t) and W_pred(t) are dynamic weights, W_smooth(t) + W_pred(t) = 1; Score_smooth is the smoothness of control; Phase_Coeff is the phase difficulty coefficient; Penalty_total is the cumulative penalty score; Penalty_overlap is the overlap penalty score; Ability_Coeff is the trainee's ability coefficient; and Final_Score is the overall flight quality score. The real-time intelligent alert module is configured to receive the event stream, dynamically prioritize the alerts based on the severity level of the events in the event stream and the current flight phase, assign multi-mode alerts to the events based on the alert priority ranking, and generate alert records. The intelligent evaluation generation module is configured to receive multi-source data and perform correlation and fusion processing, wherein the multi-source data includes the text data, the comprehensive flight quality score and event stream, and the early warning record; Based on the multi-source data after correlation and fusion, a structured and personalized review report is generated through intelligent template matching and semantic filling.

2. The multi-module collaborative flight simulator control panel according to claim 1, characterized in that, In the intelligent voice interaction module, when filling in missing slots, the corresponding data sources are called sequentially in the order of key parameters of the flight phase, recent operation history, and commonly used command preferences to infer and fill in the missing slots.

3. The multi-module collaborative flight simulator control panel according to claim 1, characterized in that, The network parameters of the enhanced LSTM neural network are optimized through a dynamic adaptive batch_size and combined learning rate strategy. The enhanced LSTM neural network includes an input layer, a first LSTM layer, a global attention module, a local attention module, a second LSTM layer, a first fully connected layer, a second fully connected layer, and an output layer connected in sequence. The local attention module processes the output sequence of the global attention module through a sliding window and outputs the processing result to the second LSTM layer. The output layer contains two neurons that use the sigmoid activation function.

4. The multi-module collaborative flight simulator control panel according to claim 1, characterized in that, The rule engine runs on a dynamically reconfigurable rule base, allowing instructors to customize rule thresholds or inject new rules through the control panel.

5. The multi-module collaborative flight simulator control panel according to claim 1, characterized in that, In the real-time intelligent alert module, the dynamic early warning priority ranking includes: during critical flight phases such as takeoff, landing, or emergency handling, raising the early warning priority of Level 2 and Level 3 events by at least one level.

6. The multi-module collaborative flight simulator control panel according to claim 1, characterized in that, In the intelligent review generation module, the priority of intelligent template matching is: personalized template > scenario template > basic template; wherein, personalized templates are selected according to the trainee's ability coefficient, and scenario templates are selected according to the flight stage and event type.

7. The multi-module collaborative flight simulator control panel according to claim 1, characterized in that, When generating reports, the intelligent review and feedback generation module is configured to dynamically adjust the level of detail and technical depth of the review and feedback suggestions based on the trainees' ability coefficients.

8. A multi-module collaborative flight simulator control method, based on the multi-module collaborative flight simulator control panel according to any one of claims 1-7, characterized in that, Includes the following steps: Collect and recognize instructors' voice data and convert it into text data to generate executable control commands; Parallel processing of real-time flight parameters generates a comprehensive flight quality score and event stream; Receive and process the event stream to generate multi-mode early warning signals and early warning records; The text data, the comprehensive flight quality score, the event stream, and the early warning records are semantically correlated, and a structured review report is generated through data fusion.

9. The multi-module collaborative flight simulator control method according to claim 8, characterized in that, The semantic association includes: using the event trigger timestamp as a benchmark, associating text data within the ΔT time window with the corresponding event, and storing the associated text data keywords as contextual information of the event in the review report.

Images