Intelligent fault-tolerant landing decision method and system for carrier based on multi-dimensional physical field fusion prediction
By employing multi-dimensional physics field fusion prediction and graceful degradation survival assurance technology, the fault tolerance problem of reusable launch vehicles in progressive performance degradation and extreme environments has been solved, enabling real-time dynamic adaptation and safe landing, thereby improving the fault tolerance capability and mission success rate of the launch vehicle.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- GUANGDONG BOYIDA INTELLIGENT PARKING EQUIP CO LTD
- Filing Date
- 2026-04-24
- Publication Date
- 2026-07-24
AI Technical Summary
Existing technologies are insufficient to effectively address the fault tolerance issues of flight control systems in reusable launch vehicles under conditions of gradual performance degradation and extreme environments. Traditional methods are slow to react and cannot adapt to real-time changes, leading to mission failures or resource waste.
An intelligent fault-tolerant method based on multi-dimensional physical field fusion prediction is adopted. By monitoring propulsion and structural data in real time through an independent system, dynamic performance envelopes are generated, conflicts are predicted and the trajectory is reconstructed online. Combined with graceful degradation survival assurance technology, non-intrusive fault-tolerant control is achieved.
It achieves predictive fault tolerance, generates optimal adaptation strategies, ensures that the flight control system provides early warning before performance degradation and ensures safe landing, reduces hardware modification costs, and improves the safety of the launch vehicle and the success rate of missions.
Smart Images
Figure CN122449929A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the interdisciplinary technical field of aerospace vehicle guidance, navigation and control and aircraft health management, and specifically relates to an intelligent fault-tolerant decision-making method and system for reusable vertical take-off and landing vehicles in the terminal landing phase. Background Technology
[0002] Vertical precision landing of reusable launch vehicles is a key technology for achieving commercial viability. Current mainstream technologies rely on highly reliable main flight control systems and their hardware redundancy. However, these systems are inherently reactive, and their fault-tolerant logic is typically based on mode switching after fault diagnosis, which has inherent limitations.
[0003] First, for progressive performance degradation (such as engine thrust slowly decreasing due to turbine efficiency decline, or rocket body structure stiffness slowly changing due to fatigue or thermal load), traditional threshold fault detection methods are slow to react, often triggering a response only when performance has significantly deteriorated and safety margins have been exhausted, which may lead to the closure of the recovery window.
[0004] Secondly, existing fault-tolerance solutions mostly rely on a few pre-stored backup tracks. These tracks cannot adapt to the ever-changing real-time decay state, remaining fuel, and environmental disturbances. They may waste the remaining performance that could have completed the task due to being too conservative, or cause the task to fail due to being insufficiently safe.
[0005] As reusable launch vehicles evolve towards higher reuse rates and lower costs, their technological approaches are exhibiting new characteristics: propellants are increasingly shifting towards liquid oxygen and methane to enable in-situ manufacturing and clean reuse on Mars; and the use of stainless steel and advanced composite materials for rocket bodies is actively exploring ways to balance cost, heat resistance, and lightweighting. These new characteristics also introduce new challenges: the complex cycle of methane engines makes their failure modes more gradual and insidious; and the performance degradation of stainless steel and composite materials under repeated thermo-mechanical loads directly affects flight control boundaries. Traditional fault-tolerant methods based on fixed thresholds and preset templates are insufficient to effectively address this problem of continuously changing performance boundary contraction stemming from the slow evolution of the physical system's health state.
[0006] Furthermore, existing launch vehicle flight control systems generally neglect the survivability of core hardware in extreme environments: high heat flux during reentry and engine radiant heat during landing can cause localized overheating failures of the flight control computer, sensor interfaces, and actuator drivers. Traditional solutions rely on hardware redundancy, but this approach is costly, heavy, and cannot cope with simultaneous failures at multiple points. Summary of the Invention
[0007] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method and system for intelligent fault-tolerant landing decision-making of launch vehicles based on multi-dimensional physical field fusion prediction, so as to realize the paradigm shift from "post-failure handling" to "pre-failure prevention".
[0008] To achieve the above objectives, the present invention adopts the following technical solution: Firstly, this invention provides an intelligent fault-tolerant landing decision-making method for launch vehicles based on multi-dimensional physical field fusion prediction. While the main flight control system executes its mission, an independent system performs a closed-loop process in parallel: it fuses propulsion and structural physical field data through a dedicated channel to continuously predict the time-varying dynamic performance envelope; it anticipates future conflicts between this envelope and the nominal trajectory; once triggered, it reconstructs an adaptive landing strategy online using this envelope as a constraint; it verifies the strategy and converts it into high-level parameters, injecting them into the main system to guide a safe landing; simultaneously, it monitors the flight control hardware temperature in real time through graceful degradation survival assurance steps, triggering thermal control reconfiguration when the temperature is abnormal.
[0009] Secondly, the present invention provides a system for implementing the above-described method.
[0010] Thirdly, the present invention provides a reusable vehicle integrating the above-described system.
[0011] Compared with the prior art, the present invention has the following beneficial effects: The beneficial effects of this invention are as follows: (1) Predictive fault tolerance is achieved: through physical field signal fusion analysis, early warning and decision-making are made before performance degradation leads to functional failure; (2) Dynamic optimal adaptation is achieved: the generated fault tolerance strategy is not a fixed template, but an optimal solution "tailored" to the real-time performance envelope; (3) Deep collaboration with next-generation launch vehicles: specifically designed for the health management needs of liquid oxygen methane engines and stainless steel / composite material rocket bodies; (4) Non-intrusive system integration: The main system is influenced by high-level guidance parameter injection without changing its core reliable control law; (5) Flight control hardware survivability assurance: By integrating graceful degradation survivability assurance technology, millisecond-level thermal control reconfiguration of flight control hardware is achieved. Attached Figure Description
[0012] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0013] Figure 1 This is a schematic diagram of the integrated architecture of the system of the present invention with the main flight control system and the independent sensor network; Figure 2 This is a flowchart illustrating the closed-loop decision-making logic of the method of the present invention; Figure 3 A schematic diagram illustrating the dynamic performance envelope contraction and adaptive strategy generation principle; Figure 4 A sequence diagram of the data flow between the strategy injection mechanism and the main system; Figure 5 As a specific embodiment, this is a spatial comparison diagram of the launch vehicle switching from the nominal trajectory to the fault-tolerant trajectory under thrust decay; Figure 6 This is a structural and operational flowchart of the graceful degradation survival assurance unit in this invention; Figure 7 This is a schematic diagram of the temperature monitoring and pre-reconfiguration triggering of the flight control computer in this invention; Figure 8 This is a schematic diagram of the closed-loop thermal-computational collaborative optimization in this invention. Detailed Implementation
[0014] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.
[0015] Example 1: System Architecture and Graceful Degradation Integration
[0016] like Figure 1 As shown, the decision-making system of this invention interacts with the main flight control system of the launch vehicle via a data bus. The system's input originates from a completely independent dedicated monitoring channel, which acquires direct physical field signals from sources other than the main system's navigation sensors. The system includes a multi-dimensional physical field fusion prediction unit, an envelope-trajectory conflict prediction unit, an online trajectory reconstruction unit, a strategy injection management unit, and a graceful degradation survival assurance unit, all connected in sequence.
[0017] Combination Figure 2 The closed-loop process is explained in detail, with each step described below: S1. Multidimensional physical field fusion prediction. The unit receives two types of core data from a dedicated monitoring channel: (1) Propulsion system physical field data: For example, for liquid oxygen methane engines, high-frequency vibration acceleration at the turbopump bearing is monitored, and the microscopic degradation of rotational mechanical efficiency is predicted by analyzing the energy changes of characteristic frequency bands related to rotor imbalance and blade passage frequency in the spectrum. At the same time, combustion chamber pressure oscillations are monitored to identify specific growth rates that indicate thrust instability.
[0018] (2) Structural system physical field data: For example, for a stainless steel rocket body, fiber optic grating sensors are arranged in a grid pattern on the back of the expected high-temperature zone to measure temperature and strain in real time, and the slight decrease in the local elastic modulus of the material is inverted through a thermo-structural coupling model. For a composite material tank, an array of acoustic emission sensors is arranged to assess the accumulation of internal damage through event count rate, energy, and location.
[0019] The unit runs a reduced-order digital twin model, mapping the aforementioned physical field signals into a unified and intuitive time-varying dynamic performance envelope. For example... Figure 3 As shown, the envelope core can be characterized by the evolution of two key boundaries over time: the maximum available thrust boundary and the maximum safe attitude angular rate boundary.
[0020] S2. Envelope-Trajectory Conflict Prediction. The unit continuously receives the expected thrust and angular velocity requirements for the nominal trajectory at future moments from the main system. Its core algorithm compares the requirements with the envelope boundary. If a future moment is predicted where the requirements exceed the envelope, a trigger command is immediately generated. The key to this step is to utilize the slow performance degradation to identify risks in advance before a conflict actually occurs. The calculation of the warning time window takes into account the real-time temperature status of the flight control computer; when the temperature is high and causes a decrease in computing performance, the warning window is dynamically extended.
[0021] S3. Online Adaptive Trajectory Reconstruction. The system starts immediately upon triggering. The mathematical description of its optimization problem is: Under the premise of satisfying the dynamic equations, find a trajectory starting from the current state, ensuring its terminal lands in a preset landing area, and meeting the hard constraints that thrust and angular velocity requirements do not exceed the envelope boundary throughout the entire trajectory, while simultaneously optimizing secondary objectives such as fuel consumption or landing accuracy. Using a convex optimization framework, this parameterized trajectory can be obtained within hundreds of milliseconds.
[0022] S4. Strategy Verification and Seamless Injection. The unit first verifies the safety of the new trajectory. Then, it sends correction commands to the main system's trajectory tracker via the bus. These commands are not low-level control variables, but rather high-level guidance parameters such as "update the target landing point to the offset position, allow vertical landing velocity updates." The trajectory tracker, based on its robust internal control algorithm, smoothly tracks this new target, thus achieving non-intrusive, fault-tolerant control takeover.
[0023] S5, Graceful Degradation Survival Support Steps. (For example...) Figure 6As shown, the graceful degradation survivability support unit collects real-time temperature and health status data from the flight control computer, sensor interfaces, and actuator drivers at a high sampling rate. When the temperature of a component exceeds a preset threshold, the thermal control path planning subunit is triggered, employing a GPU-based parallel shortest path algorithm to replan the computational task path. The local thermal control subunit adjusts heat dissipation power through a MEMS variable thermal resistance structure. The pre-reconfiguration subunit predicts high-failure-risk areas based on a hybrid neural network model and pre-plans alternative computational paths. The performance evaluation subunit calculates temperature uniformity and heat dissipation efficiency indicators to determine whether a secondary reconfiguration should be triggered.
[0024] Example 2: Thermal Protection and Pre-Reconfiguration of Propulsion System Sensors
[0025] This embodiment demonstrates the temperature monitoring and pre-reconstruction mechanism of a turbopump sensor.
[0026] During the launch vehicle landing phase, the engine turbopump vibration sensor experienced a temperature rise due to thermal radiation. The graceful degradation unit's condition monitoring subunit detected that the temperature exceeded a preset threshold, triggering a pre-reconfiguration. The pre-reconfiguration subunit inputs historical temperature data to predict the failure probability. When the predicted probability exceeds the preset threshold, it pre-plans a backup sensor path, switches the vibration monitoring task to the redundant sensor, and pre-adjusts the MEMS thermal resistance to enhance heat dissipation. After reconfiguration, the sensor temperature stabilizes, vibration data continuity is maintained, and thrust performance assessment remains unaffected.
[0027] Example 3: Temperature Compensation and Thermal Control of Structural Strain Sensors
[0028] This embodiment demonstrates the synergy between temperature compensation and thermal control in structural strain sensors.
[0029] The fiber Bragg grating sensor at the root of the stainless steel rocket flap experiences a temperature rise due to reentry heat flux. The fusion prediction unit, based on the sensor's real-time temperature, calls upon a pre-stored thermal-strain coupling compensation model to correct the strain measurement. The graceful degradation unit triggers thermal control reconfiguration, enhancing heat dissipation in the sensor area. Strain measurement accuracy is maintained, structural stiffness assessment is accurate, and attitude angular rate boundary calculations are reliable.
[0030] Example 4: Flight Control Computer Thermal-Computational Co-optimization
[0031] This embodiment demonstrates the closed-loop thermal-computational collaborative optimization of the flight control computer.
[0032] like Figure 8As shown, the collaborative optimization closed-loop process is as follows: the downgraded digital twin model of the fusion prediction unit runs on the GPU; the graceful degradation unit detects an upward temperature trend and sends an early warning to the fusion prediction unit; the fusion prediction unit responds to the early warning and transfers some model inference tasks to a backup computing unit with a lower temperature; when the temperature continues to rise to a dangerous threshold, the graceful degradation unit triggers thermal control reconstruction; after reconstruction, the thermal capacity margin recovery information is fed back to the fusion prediction unit; the fusion prediction unit gradually restores task allocation based on the updated thermal capacity margin.
[0033] Example 5: Thermal Protection of the Trajectory Reconstruction Solver
[0034] This embodiment demonstrates the thermal protection during the operation of the trajectory reconstruction solver.
[0035] Envelope-trajectory conflict triggers online trajectory reconstruction, and the solver runs a sequential quadratic programming algorithm on the computational unit. High computational load leads to temperature rise, and the graceful degradation unit's predicted failure probability exceeds a preset threshold, triggering pre-reconstruction. Part of the solver's computational tasks are migrated to backup computational units, reducing the load on the main computational unit. Thermal control enhances heat dissipation, stabilizing the temperature. Trajectory reconstruction is completed within twice the control cycle time, meeting real-time requirements.
[0036] Example 6: Redundancy switching of instruction injection channels
[0037] This embodiment demonstrates the handling of temperature anomalies in the command injection channel.
[0038] When the strategy injection unit sends high-level guidance parameters to the main system, the main communication bus interface temperature becomes abnormal. The graceful degradation unit predicts that the failure probability exceeds a preset threshold, and the command injection switches to the backup communication bus, where the interface temperature returns to normal. The high-level guidance parameters are successfully injected into the main system via the backup bus. The command transmission latency increases but remains within acceptable limits, and the fault-tolerant trajectory executes normally.
[0039] Example 7: Summary of Performance Indicators Across Multiple Scenarios
[0040] This embodiment summarizes the performance indicators of the system of the present invention in different scenarios.
[0041]
[0042] Example 8: Comparison and verification with existing technologies
[0043] This embodiment compares and verifies the present invention with existing technologies.
[0044]
[0045] Example 9: Preferred Example
[0046] Taking a stainless steel liquid oxygen-methane carrier integrating the system of this invention as an example, a complete scenario is illustrated: During the landing descent phase, the dedicated channel reported that the engine turbopump vibration energy increased at the second harmonic of the shaft frequency; at the same time, the temperature strain sensor on the dorsal wall at the root of the left leading flap showed that the thermal load in this area was higher than expected.
[0047] The fusion prediction unit integrates this information and predicts that the upper limit of future thrust will decrease, and the upper limit of the left roll angular velocity needs to be reduced. The envelope-trajectory conflict prediction unit determines that the original high-precision landing trajectory cannot be completed within this contracting envelope.
[0048] The online trajectory reconstruction unit plans a new trajectory: extending the descent distance, reducing the descent rate, and laterally shifting the landing point to the backup platform. All maneuver requirements of this trajectory fall within the decayed performance envelope. During the solution process, the computational unit temperature rises, triggering a thermal control reconfiguration by the graceful degradation unit. The task is partially migrated to the backup computational unit, ensuring real-time completion of the solution.
[0049] The strategy injection management unit sends the new target point coordinates and velocity constraints to the main system. The main system controls the rocket to fly smoothly along the new trajectory and finally land safely on the backup platform. Figure 5 As shown. The entire process was completed within seconds, and the mission commander only perceived a smooth change in course.
[0050] Industrial applicability
[0051] The intelligent fault-tolerant landing decision-making method and system for launch vehicles based on multi-dimensional physical field fusion prediction provided by this invention has broad industrial applicability. The method is based on mature sensing technology, digital twins, convex optimization, and artificial intelligence technologies, and can be implemented on existing flight control computer platforms. The status monitoring unit can be implemented using a high-speed ADC and a temperature sensor array in collaboration; the thermal control path planning unit can adopt a GPU parallel computing architecture; and the pre-reconfiguration unit can deploy a hybrid neural network model using an inference engine. Hardware-in-the-loop simulation has verified that the system is fully functional and can be directly applied to liquid oxygen-methane reusable launch vehicles, providing them with a complete technical solution from health monitoring to fault-tolerant control, and ensuring the survivability of flight control hardware in extreme environments.
Claims
1. A method for intelligent fault-tolerant landing decision-making of launch vehicles based on multi-dimensional physical field fusion prediction, characterized in that, While the launch vehicle's main flight control system performs landing phase control tasks, an independent fault-tolerant decision-making system executes a closed-loop prediction-decision process in parallel. This process includes: S1. Multidimensional physical field fusion prediction step: Through a dedicated monitoring channel independent of the main flight control sensor network, direct sensing data from the physical fields of the launch vehicle propulsion system and the structural system are collected and fused in real time. Based on the direct sensing data of the physical fields of the launch vehicle propulsion system and the structural system, the time-varying dynamic performance envelope characterizing the overall control capability of the launch vehicle in the future landing segment is predicted in a rolling manner. S2, Envelope-Trajectory Conflict Prediction Step: The continuously updated time-varying dynamic performance envelope is compared with the real-time performance requirements of the nominal landing trajectory currently tracked by the main flight control system. When it is predicted that the time-varying dynamic performance envelope cannot meet the real-time performance requirements at a future preset key decision point, a fault-tolerant replanning trigger command is generated. S3. Online adaptive trajectory reconstruction step: In response to the trigger command, using the latest time-varying dynamic performance envelope as the fundamental constraint, and with the goal of safely touching the ground within the preset landing area, a set of parameterized adaptive landing strategies is generated online in real time. S4. Strategy Verification and Seamless Injection Steps: The adaptive landing strategy is verified for safety, and the main system is guided to smoothly transition to a new safe landing target by injecting modified high-level guidance parameters into the trajectory tracking module of the main flight control system. S5. Graceful Degradation Survival Support Steps: Real-time reference data from the flight control computer, sensor interfaces, and actuator drivers are collected via a dedicated monitoring channel independent of the main flight control system. Based on the real-time reference data, overheated components are identified. A GPU-based parallel shortest path algorithm is used to recalculate the computational task path for the overheated components. The heat dissipation power of components surrounding the damaged area is adjusted through a MEMS variable thermal resistance structure. The real-time reference data includes temperature, power consumption, and health status data.
2. The method according to claim 1, characterized in that, The direct sensing data of the propulsion system physical field collected by the dedicated monitoring channel includes at least one of the following: high-frequency vibration spectrum, combustion chamber pressure oscillation waveform, and turbopump speed micro-deviation, used for online evaluation of the instantaneous performance margin of the thrust generation device; the performance margin is directly related to the decay model of the maximum available thrust boundary in the time-varying dynamic performance envelope. Among them, the temperature status of the turbopump sensor is monitored in real time, and pre-reconstruction is triggered when the sensor temperature exceeds a preset threshold.
3. The method according to claim 2, characterized in that, The direct sensing data of the propulsion system's physical field comes from a full-flow staged combustion cycle engine using liquid oxygen and liquid methane as propellants, and the analysis of the high-frequency vibration spectrum focuses on characteristic frequency bands associated with the turbine pump rotor dynamics and combustion instability.
4. The method according to claim 1, characterized in that, The direct sensing data of the physical field of the structural system collected by the dedicated monitoring channel includes at least one of the following: distributed strain field distribution, temperature field gradient, and acoustic emission event signals for composite material structures, used for online evaluation of the stiffness and damping characteristics of the structural system; the stiffness and damping characteristics are directly related to the contraction model of the maximum safe attitude angular rate boundary in the time-varying dynamic performance envelope. Among these measures, the temperature status of the structural strain sensor is monitored in real time to compensate for the impact of temperature on the accuracy of strain measurement.
5. The method according to claim 4, characterized in that, The direct sensing data of the physical field of the structure system comes from the arrow body, which is mainly supported by austenitic stainless steel or carbon fiber composite material. For the stainless steel arrow body, the distributed sensor network is a fiber optic temperature strain sensing network embedded or attached to the back wall of the thermal protection system. For the composite material arrow body, the sensor network is an acoustic emission sensor array that monitors matrix cracking and fiber breakage.
6. The method according to claim 1, characterized in that, The time-varying dynamic performance envelope is quantitatively characterized at least by the real-time evolving maximum available thrust boundary and maximum safe attitude angular rate boundary. The contraction of the time-varying dynamic performance envelope is driven by the attenuation of the thrust margin of the propulsion system and / or the degradation of the stiffness and damping characteristics of the structural system. Among them, the temperature of the flight control computer is monitored in real time, and the computing power allocation of the envelope prediction model is dynamically adjusted when the temperature exceeds the preset threshold.
7. The method according to claim 1, characterized in that, The online real-time solution generates a set of parameterized adaptive landing strategies, specifically including: A convex optimization solver based on sequential quadratic programming is used to generate a set of parameterized adaptive landing strategies online in real time within a time window that is less than twice the typical control period of the launch vehicle's landing phase. Among them, the temperature of the solver is monitored in real time, and when the temperature exceeds a preset threshold, the task is triggered to migrate to a backup computing unit.
8. The method according to claim 1, characterized in that, The modified high-level guidance parameters include at least: the position offset vector of the target landing point in the launch coordinate system, the allowable threshold of each component of the terminal velocity vector, and the relaxation of the trajectory flight path angle constraint; the modified high-level guidance parameters are encapsulated into a command message that can be parsed by the trajectory tracking module of the main flight control system, and injected without interrupting the execution of its underlying control law cycle. Among them, the interface temperature of the command injection channel is monitored in real time, and the system switches to the backup communication bus when the temperature is abnormal.
9. The method according to claim 1, characterized in that, The trigger command is generated within a preset warning time window after it is predicted that the performance envelope cannot meet the demand, but before the actual physical exceedance occurs; the warning time window is dynamically calculated based on the time-varying dynamic performance envelope contraction rate and trajectory maneuvering demand. The calculation of the warning time window takes into account the real-time temperature status of the flight control computer. When the temperature is high and the computing performance degrades, the warning window is dynamically extended.
10. The method according to claim 1, characterized in that, The time-varying dynamic performance envelope is generated by a real-time running reduced-order digital twin model. The reduced-order digital twin model takes the direct sensing data of the physical fields of the propulsion system and the structural system of the launch vehicle as input, and dynamically maps and outputs the continuous evolution trajectory of the maximum available thrust boundary and the maximum safe attitude angular rate boundary based on the launch vehicle dynamic equation. The reasoning calculation of the reduced-order digital twin model is accelerated by the flight control computer, which monitors the temperature of the computing unit in real time. When the temperature exceeds a preset threshold, a pre-reconstruction is triggered, and the model reasoning task is migrated to the backup computing unit.
11. The method according to claim 1, characterized in that, After step S4 is executed, the system immediately returns to step S1 and continues to make rolling predictions and evaluations based on the new state after injection, forming a closed-loop monitoring and re-decision capability for the effect of adaptive strategy execution. During the closed-loop monitoring process, the power consumption and temperature of each unit are continuously monitored, and the task allocation is dynamically adjusted to maintain the thermal balance of the system.
12. A launch vehicle intelligent fault-tolerant landing decision system based on multi-dimensional physics field fusion prediction for implementing the method of any one of claims 1-11, characterized in that, As an independent module physically or logically isolated from the main flight control system, the system includes: A multidimensional physics field fusion prediction unit is used to perform step S1; An envelope-trajectory conflict prediction unit is used to execute step S2. An online trajectory reconstruction unit is used to perform step S3; The strategy injection management unit is used to execute the S4 step; A graceful degradation survival support unit is used to perform the S5 step, including: The status monitoring subunit is used to collect real-time temperature, power consumption, and health status data of the flight control computer, sensor interfaces, and actuator drivers. The thermal control path planning subunit determines the overheated component based on the real-time reference data and replans the computation task path for the overheated component using a GPU-based parallel shortest path algorithm. The local thermal control subunit adjusts the heat dissipation power of the components surrounding the damaged area through a MEMS variable thermal resistance structure. The pre-reconfigurable sub-unit is used to predict high-failure-risk areas and plan alternative computation paths in advance. The performance evaluation sub-unit is used to evaluate the temperature uniformity and heat dissipation efficiency after reconstruction and to trigger a secondary reconstruction.
13. The system according to claim 12, characterized in that, The graceful degradation survival guarantee unit and the multi-dimensional physics field fusion prediction unit form a thermal-computation collaborative optimization closed loop: the fusion prediction unit adjusts the allocation of model inference tasks according to the real-time temperature of each coprocessor, giving priority to allocating high-load tasks to coprocessors with lower temperatures; the graceful degradation unit triggers thermal control reconstruction when the temperature is abnormal, and feeds back the reconstructed thermal capacity margin to the fusion prediction unit as the basis for the next round of task scheduling.
14. A reusable vehicle, characterized in that, It integrates the intelligent fault-tolerant landing decision system for launch vehicles based on multi-dimensional physical field fusion prediction as described in claim 12 or 13.