An unmanned aerial vehicle high-dynamic inertial navigation system based on multi-stage impact-resistant MEMS

Abstract

The application discloses a kind of unmanned plane high dynamic inertial navigation system based on multistage impact resistance MEMS, and specifically relates to unmanned plane navigation technical field.The system includes high dynamic carrier adaptation module, redundant sensor array module, gradient impact resistance protection module, intelligent signal acquisition module, multi-modal fusion processing module, environment self-adaptive compensation module, standardization output interface module and system health management module.The system is realized wide frequency vibration inhibition by three-stage impact resistance structure, uses 120 ° space symmetry layout redundant MEMS sensor array to guarantee fault tolerance capability, combines dynamic weighted fusion algorithm and multi-physical field coupling compensation technology to improve navigation precision under dynamic environment.System integrates intelligent health management function, supports real-time state monitoring and predictive maintenance, and outputs anti-interference navigation data conforming to military standard.The application significantly improves the reliability and measurement accuracy of the inertial navigation system of the unmanned plane in a strong impact and high vibration environment.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) navigation technology, and more specifically to a high-dynamic inertial navigation system for UAVs based on multi-level shock-resistant MEMS. Background Technology

[0002] With the widespread application of drones in military reconnaissance, disaster relief, and other fields, higher demands are placed on the reliability of inertial navigation systems. Traditional MEMS inertial measurement units have significant shortcomings in high-dynamic environments: in terms of mechanical protection, single-layer buffer structures have poor isolation effects against high-frequency impacts, with vibration transmission rates exceeding 60% under a 100g impact; in terms of sensor configuration, single-chip designs have the risk of single-point failure, while redundant solutions face layout optimization challenges; in terms of signal processing, insufficient temperature compensation leads to an accuracy loss of up to 30% under operating conditions ranging from -40℃ to 85℃.

[0003] These problems are particularly prominent in military scenarios: extreme conditions such as 20,000g overload during gun-launched takeoff, 20-2000Hz wideband vibration during high-speed maneuvers, and sudden temperature changes make commercial systems unsuitable. Although fiber optic gyroscope solutions offer superior performance, their size and power consumption are unsuitable for small UAVs. The system architecture suffers from three major limitations: non-modular design increases maintenance costs, environmental adaptability is insufficient, and health monitoring functionality is lacking. At the algorithm level, traditional Kalman filtering performs poorly in multi-sensor data fusion and fault reconstruction, and poor interface compatibility hinders its widespread application.

[0004] Therefore, developing novel MEMS inertial navigation systems with high shock resistance, strong environmental adaptability, and intelligent health management is imperative. Innovation is needed across multiple dimensions, including mechanical protection, sensor configuration, and signal processing, to overcome existing technological bottlenecks. This invention proposes a complete solution to these key technical challenges. Summary of the Invention

[0005] The purpose of this invention is to provide a high-dynamic inertial navigation system for unmanned aerial vehicles (UAVs) based on multi-level shock-resistant MEMS. This invention achieves a performance leap through three core technological breakthroughs: a unique silicon-based gel-titanium alloy honeycomb composite buffer structure reduces the transmission rate of a 20,000g gun-firing overload to below 5%, improving shock resistance by 12 times compared to traditional designs; a triple-redundant MEMS sensor array with a 120° spatial symmetry layout, combined with a dynamic reconstruction algorithm, maintains over 95% measurement accuracy even when a single sensor fails; and an integrated LSTM neural network environmental compensation system controls the zero-drift error within 0.05mg under temperature variations ranging from -40℃ to 85℃. The system also achieves breakthroughs in key technologies such as modular quick-release design, multi-source intelligent fusion, and military-standard interfaces, reducing maintenance time by 70%, increasing data update rate to 1kHz, and covering over 90% of military platforms with interface compatibility, perfectly solving the reliability, accuracy, and adaptability challenges of traditional solutions in extreme environments.

[0006] The objective of this invention can be achieved through the following technical solutions:

[0007] A high-dynamic inertial navigation system for unmanned aerial vehicles (UAVs) based on multi-level shock-resistant MEMS, the system comprising a high-dynamic carrier adaptation module, a redundant sensor array module, a gradient shock-resistant protection module, an intelligent signal acquisition module, a multi-modal fusion processing module, an environmental adaptive compensation module, a standardized output interface module, and a system health management module, wherein:

[0008] The high-dynamic carrier adaptation module is used to perform modal matching and frequency isolation of the body vibration energy through the UAV centroid positioning architecture to obtain a stable measurement benchmark of the natural frequency.

[0009] The redundant sensor array module is used to perform collaborative measurement and cross-validation of three sets of MEMS accelerometers based on a 120° spatially symmetrical topology, and output multi-axis inertial data with fault reconstruction capability.

[0010] The gradient shock protection module is used to implement multi-level frequency band selective attenuation of mechanical shock loads using a silicon-based gel-titanium alloy honeycomb composite structure, thereby achieving significant suppression of vibration energy in a wide frequency range.

[0011] The intelligent signal acquisition module is used to perform anti-aliasing filtering on the original analog signal of the MEMS sensor based on a high-precision ADC conversion circuit to obtain a 16-bit digitized acceleration sampling sequence.

[0012] The multimodal fusion processing module is used to perform singular value decomposition on heterogeneous sensor data streams using a dynamic weighting factor matrix to generate attitude quaternions with error tolerance optimization.

[0013] The environmental adaptive compensation module is used to combine LSTM neural network to complete nonlinear modeling of temperature drift characteristics and establish an environmental error prediction model.

[0014] The standardized output interface module is used to encapsulate navigation information into frame structures based on the MIL-STD-1553B protocol stack, forming communication messages that conform to military standards.

[0015] The system health management module is used to perform fatigue analysis on historical impact data based on the vibration cumulative damage model and output the remaining life prediction curve.

[0016] Optionally, the step of performing modal matching and frequency isolation on the body vibration energy through the UAV centroid positioning architecture to obtain a stable measurement reference for the natural frequency includes:

[0017] Dynamic stiffness optimization of the UAV frame structure is performed based on finite element modal analysis to form an anti-resonance mechanical transmission path;

[0018] A frequency-selective attenuation channel is constructed by using a magnesium alloy honeycomb sandwich to dissipate high-frequency vibration energy through waveguides.

[0019] The low-frequency sway is passively suppressed by the mass block-spring damping system, resulting in a stable reference measurement environment;

[0020] A laser alignment instrument is used to perform micron-level leveling of the sensor mounting plane, establishing the hardware foundation for geometric error compensation;

[0021] The structural resonance characteristics were experimentally calibrated using a modal testing system, resulting in closed-loop control parameters for vibration suppression.

[0022] Optionally, the method for performing collaborative measurement and cross-validation of three sets of MEMS accelerometers based on a 120° spatially symmetrical topology, and outputting multi-axis inertial data with fault reconstruction capability, includes:

[0023] Based on the spatial vector synthesis algorithm, the triaxial redundant data is orthogonalized to construct a full-dimensional inertial reference frame;

[0024] A fault early warning mechanism is established by dynamically assessing the sensor's operating status through real-time health monitoring.

[0025] Least squares fitting is used to reconstruct outlier measurements, maintaining continuous navigation parameter output.

[0026] A temperature-stress coupling model is used to compensate for installation errors online, thereby improving the consistency of array measurements.

[0027] By relying on covariance analysis, sensor weights are adaptively allocated to optimize the accuracy of multi-source data fusion.

[0028] Optionally, the use of a silicon-based gel-titanium alloy honeycomb composite structure to selectively attenuate mechanical impact loads across multiple frequency bands, thereby significantly suppressing vibration energy over a wide frequency range, includes:

[0029] Impedance matching of shock stress waves is achieved based on a viscoelastic-metal gradient structure to form a broadband energy dissipation network.

[0030] Selective filtering of vibrations in specific frequency bands is achieved through porous cellular topology optimization, thus establishing a frequency domain protection barrier.

[0031] Nonlinear damping materials are used to diffuse transient impact energy in the time domain, reducing the peak load transfer rate.

[0032] A multi-layer composite interface design is used to perform waveguide control on the vibration propagation path to achieve directional energy attenuation;

[0033] The protective performance was quantitatively verified by relying on the impact test bench, forming an optimization basis for impact-resistant design.

[0034] Optionally, the step of performing anti-aliasing filtering on the original analog signal of the MEMS sensor based on a high-precision ADC conversion circuit to obtain a 16-bit digitized acceleration sampling sequence includes:

[0035] A 16-bit Σ-Δ modulator is used to perform high-precision sampling of the sensor's analog signal to obtain a digital measurement sequence;

[0036] Anti-aliasing filters are used to cut off high-frequency noise and improve the fidelity of the effective signal.

[0037] Time domain consistency is ensured by performing time alignment on multi-channel data through synchronous sampling clocks;

[0038] A digital gain calibration circuit is used to automatically adjust the signal amplitude and maintain dynamic range stability.

[0039] The self-testing circuit is used to monitor the integrity of the data acquisition link in real time, ensuring data reliability.

[0040] Optionally, the step of performing singular value decomposition on the heterogeneous sensor data stream using a dynamic weighting factor matrix to generate an error-tolerant optimized attitude quaternion includes:

[0041] Based on the Lie group filtering framework, spatiotemporal alignment of heterogeneous sensor data is performed to eliminate coordinate system inconsistency errors.

[0042] Dynamic suppression of measurement noise is achieved through adaptive Kalman gain, thereby improving the signal-to-noise ratio.

[0043] We utilize robust estimation theory to perform weighted removal of outlier data points, thereby enhancing the algorithm's ability to resist interference.

[0044] A federated filtering architecture is used to perform distributed fusion of multi-source information, balancing computational efficiency and accuracy.

[0045] By relying on fuzzy logic control to perform online optimization of weight factors, intelligent data fusion is achieved.

[0046] Optionally, the step of combining LSTM neural networks to perform nonlinear modeling of temperature drift characteristics and establishing an environmental error prediction model includes:

[0047] Feature extraction of the temperature-drift nonlinear relationship is performed based on deep neural networks to establish an error prediction model;

[0048] Multi-resolution analysis of the sensor noise spectrum is performed using wavelet packet transform to achieve frequency domain adaptive filtering.

[0049] The particle filter algorithm is used to dynamically estimate the system state error, thereby improving environmental adaptability.

[0050] Multi-sensor cross-validation is used to perform online calibration of compensation parameters to maintain long-term stability;

[0051] By leveraging digital twin technology, the compensation effect is virtually verified, forming a closed-loop optimization system.

[0052] Optionally, the step of encapsulating navigation information into a frame structure based on the MIL-STD-1553B protocol stack to form a communication message conforming to military standards includes:

[0053] A time-triggered architecture is used to perform deterministic scheduling of the data communication process, thereby building a low-latency transmission channel;

[0054] Differential Manchester encoding is used to encapsulate navigation commands to prevent interference and enhance the robustness of signal transmission.

[0055] Cyclic redundancy check is used to perform multiple verifications on data packet integrity to ensure the reliability of information transmission;

[0056] A double-buffered architecture is used to perform asynchronous management of the communication process, balancing real-time performance and throughput.

[0057] Electromagnetic hardening of the interface physical layer is achieved by relying on military connector standards to improve environmental adaptability.

[0058] Optionally, the fatigue analysis based on the vibration cumulative damage model on historical impact data, and the output of the remaining life prediction curve, includes:

[0059] Based on vibration power spectral density analysis, the characteristics of impact energy distribution are extracted, and a damage accumulation model is established.

[0060] Online assessment of sensor degradation trends is performed using deep reinforcement learning to predict remaining lifespan;

[0061] Fault tree analysis is used to comprehensively identify system risk factors and construct a health assessment system.

[0062] Blockchain technology is used to perform distributed storage of maintenance records, ensuring that the data is tamper-proof;

[0063] By leveraging a cloud platform, remote monitoring of system status data can be achieved, enabling predictive maintenance.

[0064] This invention provides a high-dynamic inertial navigation system for unmanned aerial vehicles (UAVs) based on multi-level shock-resistant MEMS. Through innovative modular design and advanced algorithms, this system significantly improves the navigation performance and reliability of UAVs in extreme environments. The system comprises eight core functional modules: a high-dynamic carrier adaptation module employing a magnesium alloy honeycomb structure and a mass-spring damping system to effectively isolate vibration energy across a wide frequency range of 20-2000Hz; a redundant sensor array module using three sets of MEMS accelerometers arranged in a 120° spatial symmetry and a real-time health monitoring algorithm to ensure data reconstruction capability in the event of a single sensor failure; and a gradient shock-resistant protection module combining a silicon-based gel layer and a titanium alloy honeycomb structure to form a frequency-selective multi-level buffer system capable of withstanding instantaneous impact loads up to 20,000g.

[0065] In signal processing, the system employs several innovative technologies: the intelligent signal acquisition module, based on a 16-bit Σ-Δ modulator and anti-aliasing filter, achieves high-fidelity signal digitization; the multi-modal fusion processing module utilizes a Lie group filtering framework and federated filtering architecture to effectively solve the spatiotemporal registration problem of multi-source heterogeneous sensors; and the environmental adaptive compensation module achieves intelligent compensation for temperature drift and vibration noise through LSTM neural networks and wavelet packet transform. Particularly noteworthy is the introduction of digital twin technology to virtually verify the compensation effect, forming a complete closed-loop optimization mechanism. The standardized output interface module adopts the MIL-STD-1553B protocol and differential Manchester encoding to ensure reliable data transmission in complex electromagnetic environments.

[0066] This system also innovates in intelligent health management: through vibration power spectral density analysis and deep reinforcement learning algorithms, the system can accurately predict the remaining service life of each component; distributed maintenance record storage using blockchain technology ensures the authenticity and immutability of the data; and the remote monitoring function of the cloud platform greatly improves the maintainability of the system. Experimental data shows that the system has an attitude measurement error of less than 0.1° within a temperature range of -40℃ to 85℃, a survival rate of 99.9% under a 20,000g impact load, and a fault reconstruction accuracy of over 95%, with all performance indicators significantly outperforming existing commercial solutions.

[0067] Compared with the prior art, the present invention has the following beneficial effects:

[0068] 1. This invention significantly improves the reliability of the system in extreme environments through an innovative three-level impact-resistant protection structure and intelligent health management module. The composite design of silicon-based gel layer and titanium alloy honeycomb achieves selective attenuation of vibration energy across a wide frequency band, improving vibration suppression by more than 40% compared to traditional solutions; combined with a real-time damage accumulation analysis model, it can provide early warning of potential faults, extending the system's mean time between failures (MTBF) by 3-5 times.

[0069] 2. This invention employs a redundant sensor array with a 120° spatial symmetry layout and an advanced data fusion algorithm, significantly improving measurement accuracy and fault tolerance. The dynamic weighted fusion algorithm, combined with an LSTM temperature compensation model, controls the navigation error to within 0.1° within a temperature range of -40°C to 85°C. When a single sensor fails, the system can still reconstruct complete inertial parameters from the remaining sensor data, achieving a fault reconstruction accuracy of over 95%.

[0070] 3. The modular design and standardized interfaces of this invention significantly improve the system's applicability and maintenance convenience. Each functional module adopts a quick-release connection design, reducing maintenance time by 70%; it supports multiple protocol outputs such as MIL-STD-1553B and CAN, adapting to more than 90% of UAV flight control systems; through the remote monitoring function of the cloud platform, predictive maintenance can be achieved, reducing operation and maintenance costs by more than 30%. Attached Figure Description

[0071] Figure 1 This is a schematic diagram of the structure of a high-dynamic inertial navigation system for unmanned aerial vehicles based on multi-level shock-resistant MEMS according to the present invention. Detailed Implementation

[0072] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0073] Example: Figure 1 A schematic diagram of a high-dynamic inertial navigation system for unmanned aerial vehicles (UAVs) based on multi-level shock-resistant MEMS is provided. This high-dynamic inertial navigation system for UAVs based on multi-level shock-resistant MEMS includes the following modules:

[0074] High dynamic carrier adaptation module: By using the UAV centroid positioning architecture to perform modal matching and frequency isolation on the vibration energy of the airframe, a stable measurement benchmark for the natural frequency is obtained.

[0075] Redundant sensor array module: Based on a 120° spatially symmetrical topology, three sets of MEMS accelerometers perform collaborative measurement and cross-validation, and output multi-axis inertial data with fault reconstruction capability.

[0076] Gradient shock-resistant protection module: It adopts a silicon-based gel-titanium alloy honeycomb composite structure to selectively attenuate mechanical impact loads in multiple frequency bands, thereby significantly suppressing vibration energy in a wide frequency range.

[0077] Intelligent signal acquisition module: Based on a high-precision ADC conversion circuit, the module performs anti-aliasing filtering on the original analog signal of the MEMS sensor to obtain a 16-bit digitized acceleration sampling sequence.

[0078] Multimodal fusion processing module: It uses a dynamic weighting factor matrix to perform singular value decomposition on heterogeneous sensor data streams and generates attitude quaternions with error tolerance optimization.

[0079] Environmental adaptive compensation module: Combines LSTM neural network to complete nonlinear modeling of temperature drift characteristics and establishes environmental error prediction model.

[0080] Standardized output interface module: Based on the MIL-STD-1553B protocol stack, navigation information is encapsulated in a frame structure to form communication messages that conform to military standards.

[0081] System health management module: Based on the vibration cumulative damage model, fatigue analysis is performed on historical impact data, and the remaining life prediction curve is output.

[0082] By employing a UAV centroid positioning architecture to perform modal matching and frequency isolation on the airframe vibration energy, a stable measurement benchmark for the natural frequency is obtained. Specifically, the implementation is as follows:

[0083] The main frame of the module is cast from aerospace-grade magnesium alloy ZL114A, which possesses excellent specific strength and damping characteristics. Through finite element topology optimization technology, a honeycomb-shaped vibration damping structure with gradient porosity was designed, gradually increasing from 30% in the core region to 60% in the edge region. This innovative structural design successfully achieved a 40% weight reduction while maintaining a tensile strength of 300MPa, significantly improving the payload capacity of the UAV. The frame surface undergoes micro-arc oxidation treatment to form a 20-30μm thick ceramic layer, improving corrosion resistance by more than 5 times.

[0084] The installation platform integrates a three-stage active-passive composite vibration damping system, forming a complete vibration suppression system. The first stage uses a high-damping rubber-metal composite vibration isolator, whose unique layered structure design keeps the stiffness coefficient stable within the range of 50N / mm±5%, achieving an isolation efficiency of 85% for mid-frequency vibrations of 20-100Hz. The second stage is an intelligent air pressure regulation system, containing four independently controlled air springs. Through a PID algorithm, the air spring pressure is adjusted in real time, allowing precise control of the natural frequency within the 5-20Hz range. The third stage is equipped with a voice coil motor active damper, employing a neodymium iron boron magnetic circuit with a magnetic field strength of 1.2T, combined with a 2000-line photoelectric encoder, achieving an ultra-fast response of <1ms, which can cancel more than 98% of residual high-frequency vibrations.

[0085] A Leica AT960 laser tracker was used for precise centroid positioning, achieving a measurement accuracy of ±0.01 mm / m. Fine adjustments were made using a six-degree-of-freedom micro-adjustment platform to ensure the spatial deviation between the mounting platform and the UAV's theoretical centroid was controlled within 0.5 mm. The mounting surface was machined using diamond turning, achieving a surface roughness Ra < 0.4 μm and flatness < 0.01 mm / m. All fasteners were 12.9 grade high-strength bolts, with preload controlled within ±5% of the design value. After positioning, epoxy structural adhesive was used for auxiliary fixation, achieving a shear strength ≥ 20 MPa after curing.

[0086] The system's dynamic characteristics were verified through modal impact testing. Test results showed that the first-order resonant frequency reached 450Hz±5%, and the second-order resonant frequency reached 680Hz±5%. Within the full frequency range of 20-2000Hz, the vibration transmissibility remained below 10%, more than five times higher than that of traditional rigid mounting methods. The modularly integrated overload monitoring system includes 12 strain gauges with a sampling rate of 1kHz, capable of real-time measurement of impact loads in the XYZ directions with a measurement error of <1%. Data is transmitted via a CAN bus with an update rate of 100Hz, providing real-time vibration status feedback to the flight control system.

[0087] Utilizing a 120° spatially symmetrical topology, three sets of MEMS accelerometers perform collaborative measurements and cross-validation, outputting multi-axis inertial data with fault reconstruction capabilities. The specific implementation is as follows:

[0088] The core sensing unit of this system is the newly released ADXL206 triaxial MEMS accelerometer from Analog Devices (ADI). This device features an ultra-wide range of ±200g and a noise density as low as 25μg / L. It maintains excellent stability within an operating temperature range of -40℃ to 125℃. The three sensor sets are arranged symmetrically in a 120° circle and mounted on a precision-machined TC4 titanium alloy base. The base surface is mirror-polished, with a flatness controlled within 0.005mm. To ensure installation accuracy, a Renishaw XL-80 laser interferometer is used for angle positioning, ensuring that the installation angle tolerance between the three sensor sets is strictly controlled within ±0.05°. Each sensor unit adopts a completely independent modular packaging design, including multiple protective structures: the outer layer is a TC4 titanium alloy protective shell with a wall thickness of 1mm±0.05mm, which undergoes vacuum annealing to eliminate internal stress; the middle layer is a custom silicone buffer pad with a hardness of 40±2Shore A, employing a special groove structure design to enhance damping effect; the inner layer is a double-layer electromagnetic shielding cover made of copper-plated nickel material, with a shielding effectiveness >60dB@1GHz. All connectors are selected from the MIL-DTL-38999 series, with a contact resistance <5mΩ, ensuring reliable signal transmission. The protective structure, optimized through finite element analysis, effectively protects the internal sensors under a 20,000g impact. The signal conditioning circuit utilizes the Analog Devices AD8421 ultra-low noise instrumentation amplifier, achieving a gain accuracy of 0.01%, a temperature drift coefficient of <0.5ppm / ℃, and an input voltage noise of only 1.2nV / ℃. Each sensor channel is equipped with an independent 24-bit Σ-Δ ADC using a TI ADS1256 chip, with a sampling rate programmable from 1Hz to 10kHz via software. The system incorporates a digital anti-aliasing filter with a cutoff frequency of 40% of the sampling rate and stopband attenuation >120dB. The power management system utilizes an LT3045 ultra-low noise LDO with an output noise of 0.8μVRMS, providing clean power to the analog circuitry. Sensor calibration is performed on a CS-Z series six-axis rotary table with an accuracy of 0.001°, employing least squares fitting of the calibration parameter matrix to ensure 99.5% sensitivity consistency across axes and a cross-coupling error <0.3%. The system's real-time health monitoring module includes: an anomaly detection algorithm based on Mahalanobis distance with a detection threshold of 3σ; a fault prediction model based on time series analysis with a prediction window of 1s; and a sensor performance degradation assessment system with an assessment cycle of 24 hours. These functions are processed in real-time by a dedicated DSP, ensuring a response time of <10ms from fault occurrence to warning output. All monitoring data is recorded in FRAM, supporting fault backtracking analysis.

[0089] A silicon-based gel-titanium alloy honeycomb composite structure is used to selectively attenuate mechanical impact loads across multiple frequency bands, achieving significant suppression of vibration energy over a wide frequency range. Specifically, the implementation is as follows:

[0090] The outer cushioning layer uses Dow Corning SE1700 special silicone-based gel material, which possesses excellent damping characteristics and temperature stability. Through precision injection molding, the material is processed to a precise thickness of 2±0.1 mm, with a surface roughness controlled within Ra0.8 μm. The material has a loss factor >0.5, maintaining stable elastic properties within an extreme temperature range of -55℃ to 150℃, with performance fluctuations within this temperature range not exceeding 10%. The material also possesses self-healing properties, recovering more than 95% of its original shape within 24 hours after being subjected to extrusion deformation.

[0091] The middle layer utilizes a TC4 titanium alloy honeycomb structure, integrally formed using laser selective melting additive manufacturing technology. The honeycomb cell size is precisely controlled to 5mm, with a wall thickness of 0.15mm±0.01mm. Optimized cell structure design achieves a high porosity of 85%±2%. The material undergoes vacuum annealing at 850℃ for 2 hours to eliminate residual stress, resulting in a compressive strength exceeding 300MPa. The honeycomb structure surface is treated with micro-arc oxidation to form a 5-10μm thick ceramic layer, significantly improving wear resistance and corrosion resistance. The structure has undergone finite element topology optimization, achieving a 40% weight reduction while maintaining mechanical properties.

[0092] An advanced vacuum hot-pressing composite process is used to combine the buffer layer with the honeycomb structure. The process parameters are strictly controlled as follows: molding pressure 2MPa±0.1, temperature 180℃±5℃, and holding time 2 hours. Impact testing verifies that under a 300g / 1ms half-sine wave impact condition, the peak acceleration transmitted to the internal MEMS chip is <30g, and the energy attenuation rate is >90%. Vibration testing shows that the vibration transmissibility of this structure is <15% in the 20-2000Hz frequency range, and the first-order natural frequency reaches above 450Hz. After 1 million fatigue cycles, the structural performance degradation is <3%, demonstrating excellent durability.

[0093] The module employs a patented quick-release locking mechanism with an innovative four-way symmetrical locking structure, allowing for precise adjustment of the preload force within the range of 5-20 N·m. The locking mechanism is made of high-strength stainless steel, surface-nitrided to achieve a hardness of HRC60 or higher. Module replacement is simple, taking less than 3 minutes in battlefield environments. Rigorous environmental adaptability testing has verified: no visible corrosion after 96 hours of salt spray testing; insulation resistance >100 MΩ under 95% RH humid heat; and no signs of surface growth after 28 days of mold cultivation. All tests indicate performance degradation of <5%, fully meeting the stringent requirements of military equipment.

[0094] Anti-aliasing filtering is performed on the original analog signal of the MEMS sensor based on a high-precision ADC conversion circuit to obtain a 16-bit digitized acceleration sampling sequence. The specific implementation is as follows:

[0095] The analog-to-digital converter (ADC) utilizes the high-performance ADS8588S chip from TI, featuring 18-bit high resolution and integral nonlinearity error controlled within ±2 LSB. The chip supports a programmable sampling rate of 500 kSPS, with eight different sampling rate settings available via the SPI interface. The analog input range is ±10V, and the built-in overvoltage protection circuitry can withstand instantaneous overvoltages of ±15V. The chip's typical power consumption is only 120mW@500kSPS, and its performance remains undegraded within an operating temperature range of -40℃ to 125℃. The conversion result is output via a parallel interface with a settling time of only 50ns, meeting the requirements for high-speed data acquisition.

[0096] The analog front-end employs a three-stage precision signal conditioning circuit: the first stage uses an LTC1562 chip as a 5th-order elliptic anti-aliasing filter, with a cutoff frequency precisely set at 2.5kHz, in-band ripple <0.1dB, and stopband attenuation >90dB. The second stage uses an Analog Devices PGA280 programmable gain amplifier, providing eight programmable gain levels from 1 to 128 times (in the sequence 1-2-4-8-16-32-64-128), with a gain error of <0.05% for each level and a temperature drift coefficient of <2ppm / ℃. The third stage is an automatic zero-reset compensation circuit based on an LTC2057 zero-drift operational amplifier design, with a compensation range of ±100mV, a resolution of 10μV, and a time drift of <1μV / hour. The three-stage circuit is cascaded, with a total noise contribution of <5μVrms.

[0097] The clock system uses the Analog Devices AD9548 network synchronization chip, paired with an OCXO temperature-controlled crystal oscillator, and outputs an ultra-low jitter clock through phase-locked loop technology. The digital processing section integrates a Xilinx Artix-7 FPGA to achieve real-time online diagnostics: gain drift monitoring uses a 24-bit Σ-Δ ADC with an accuracy of 0.1%; zero-point deviation detection uses a self-calibrating comparator with an accuracy of 0.5mV; channel consistency verification is implemented through a cross-correlation algorithm. All key parameters are stored in Fujitsu MB85RS256A FRAM, with a read / write endurance of 10^12 cycles and a data retention period exceeding 100 years.

[0098] Data output utilizes an LVDS differential interface, compliant with the IEEE 1596.3 standard, with a transmission rate up to 1Gbps. The interface circuit employs a DS90LV047A driver with a slew rate of 400ps and jitter <10ps. The transmission link uses twisted-pair shielded cable with impedance matching controlled at 100Ω±5%. The bit error rate (BER) testing system, based on PRBS31 code verification, achieves a BER <10^-12 after 72 hours of continuous testing at 1Gbps. The module supports hot-swapping, and the power sequence control circuit ensures that the insertion and removal process does not lead to data errors or device damage. All interface connectors are from the ERNI series, with a contact resistance <30mΩ and a mating life >500 cycles.

[0099] Singular value decomposition is performed on the heterogeneous sensor data stream using a dynamic weighting factor matrix to generate attitude quaternions with optimized error tolerance. The specific implementation is as follows:

[0100] The main processor utilizes the Xilinx Zynq UltraScale+ MPSoC heterogeneous computing platform, integrating a dual-core ARM Cortex-A72@1.5GHz application processor and a quad-core Cortex-R5@600MHz real-time processor. The processing platform is equipped with 4GB of LPDDR4 memory and 256GB of eMMC storage, achieving on-chip data exchange speeds of up to 32Gbps via the AXI bus. The chip is manufactured using a 16nm FinFET process, with power consumption controlled below 8W, and supports an industrial-grade temperature range of -40℃ to 100℃. The processor integrates an improved federated filtering algorithm accelerator, reducing filtering computation latency to below 50μs, meeting real-time requirements in highly dynamic environments.

[0101] The system employs a three-stage pipelined data processing architecture: In the preprocessing stage, a 5-level decomposition is performed using the db6 wavelet basis function, followed by an improved SURE soft thresholding algorithm for denoising, resulting in a signal-to-noise ratio improvement of >20dB. In the registration stage, a six-DOF motion model is established based on the Lie group SE, and quaternion interpolation is used to achieve spatiotemporal alignment, achieving an angle alignment accuracy of 0.01° and a position accuracy of 0.1mm. In the fusion stage, constrained weighted singular value decomposition is performed, with QR decomposition accelerating computation; the weight matrix is ​​updated every 10ms. The processing pipeline uses a double-buffering mechanism to ensure a stable data throughput of over 1000 frames per second.

[0102] The dynamic weighting system employs a multi-dimensional evaluation strategy: real-time Mahalanobis distance detection sets a 3σ dynamic threshold, and outlier removal is achieved based on chi-square distribution. Historical data consistency analysis uses a 1-second sliding window, and sensor reliability is evaluated using the Pearson correlation coefficient. The environmental adaptability module integrates data from multiple sensors, including temperature, vibration, and electromagnetic sensors, and uses fuzzy logic for comprehensive scoring. Weight calculation uses a normalized exponential function to ensure that the sum of the weights of all sensors is always 1. The system supports a weight smoothing transition algorithm to avoid output jumps caused by abrupt changes, and the transition time constant is adjustable.

[0103] The system maintains excellent performance even under single sensor failure conditions: static testing (turntable accuracy 0.001°) shows an attitude error of <0.15°, and dynamic testing (angular rate 300° / s) shows an error of <0.3°. The online upgrade system uses AES-256 encrypted transmission (CBC mode, 256-bit key length) coupled with SHA-3 digital signature verification. The upgrade package uses differential compression technology, keeping the typical upgrade package size below 20MB. The upgrade process employs a dual-bank design to ensure automatic rollback in case of upgrade failure, with the entire upgrade process taking <3 minutes (under 100Mbps network conditions). The system log records a complete operation history, supporting fault diagnosis and performance analysis.

[0104] By combining LSTM neural networks to perform nonlinear modeling of temperature drift characteristics, an environmental error prediction model is established, specifically implemented as follows:

[0105] The LSTM neural network employs a 6-128-64-3 hierarchical structure. The input layer contains data from six PT1000 temperature sensors and three vibration spectrum features. The hidden layers utilize a gated recurrent unit structure, with the forget gate bias initialized to 1.0 to enhance long-term memory. The output layer outputs triaxial compensation values ​​via a sigmoid activation function, with a dynamic range of ±50mg. The network has approximately 25,000 parameters, initialized using the Xavier method to ensure balanced gradient distribution across layers. Model training uses mixed-precision computation, with a single training session taking approximately 2 hours on an NVIDIA Tesla V100.

[0106] The training dataset covers the full operating temperature range of -40℃ to 125℃, containing 20,000 sets of laboratory calibration data and 5,000 sets of field-measured data. Data augmentation techniques were used to expand the dataset by 3 times by adding Gaussian noise and random jitter. Training was performed using the Adam optimizer with an initial learning rate of 0.001, decreasing by 0.9 every 100 epochs, and a batch size of 128. The Huber loss function was used, with the δ parameter set to 0.1, to enhance model robustness while maintaining accuracy. The final model achieved a mean absolute error of 0.02 mg on the validation set.

[0107] The online compensation system hardware includes: a PT1000 temperature acquisition module, a 1024-point FFT analysis unit based on Cortex-M7, and an Intel Movidius Myriad X AI accelerator. The system workflow is as follows: environmental data is acquired every 10ms, preprocessed, and then input into the neural network; the AI ​​accelerator ensures that the compensation calculation latency is <1ms. The temperature acquisition channel uses a four-wire connection to eliminate the influence of wire resistance; vibration analysis uses a sliding DFT algorithm with a 100Hz update rate. The compensation output is smoothed by a PID controller to avoid abrupt changes.

[0108] The automatic calibration system starts every 5 minutes, employing a modified RANSAC algorithm: K-means clustering is used to initially select the interior point set, followed by least-squares fitting with 100 iterations and an interior point threshold of 2σ. The calibration process runs in a background thread, with CPU usage <5% and memory overhead of 15MB. Actual performance tests show that under a temperature change of 20℃ / min, the zero-bias stability of the X / Y / Z axes reaches 0.048mg, 0.052mg, and 0.046mg, respectively; the angular random walk is better than 0.001 / The compensation parameters are stored in the Fujitsu MB85RS128A FRAM, equipped with ECC verification, and have a data retention period of >10 years at a high temperature of 85°C.

[0109] The navigation information is encapsulated in a frame structure based on the MIL-STD-1553B protocol stack to form a communication message that conforms to military standards. The specific implementation is as follows:

[0110] The communication protocol stack is developed based on the VxWorks 7.0 real-time operating system and adopts a dual-protocol stack parallel architecture design. The MIL-STD-1553B protocol stack implements three working modes: bus controller, remote terminal, and bus monitor, with a data transmission rate of 1Mbps and a message response time of <12μs. The CAN FD protocol stack supports high-speed transmission of 5Mbps, is compatible with the traditional CAN2.0B protocol, and extends the payload to 64 bytes. The dual protocol stacks achieve data interaction through shared memory and employ a priority scheduling algorithm to ensure that the transmission latency of critical messages is <1ms. The protocol stack has a built-in flow control mechanism, supporting a stable transmission of 200 messages / second.

[0111] The data frame format strictly adheres to the STANAG 3838 military standard, with the frame header containing a 16-bit message ID and an 8-bit priority flag. A 32-bit CRC checksum is used, with a generator polynomial of 0x04C11DB7, capable of detecting all burst errors ≤32 bits. The automatic retransmission mechanism is implemented based on a sliding window protocol, with a configurable window size and dynamically adjusted retransmission timeout. The system ensures transmission reliability >99.999% through dual checksum verification and sequence number verification. The encryption module supports the AES-256 algorithm, with encryption / decryption latency <50μs / frame, meeting real-time requirements.

[0112] The physical layer interface uses MIL-DTL-38999 series Type III connectors, employing a triple-locking mechanism to ensure connection reliability. The gold plating thickness of the contacts is ≥1.27μm, with a contact resistance <10mΩ and a mating life >500 cycles. The housing is made of aluminum alloy with a hard anodized surface, achieving an IP67 protection rating. The interface module's structural strength has been optimized through FEA simulation, capable of withstanding 50g mechanical shock and random vibration from 10-2000Hz. Anti-mismating design includes key coding and color markings, supporting blind mating. All cables feature double-shielded design, with a transfer impedance <50mΩ / m@100MHz.

[0113] The electromagnetic compatibility design meets all test items of GJB151B-2013: conducted emissions are below the limit by 6dB in the 10kHz-10MHz frequency band; radiated emissions have a compliance margin of >10dB in the 2MHz-18GHz range; conducted susceptibility testing passes with a 200mA injection current. Radiated immunity reaches 200V / m, employing multi-layer shielding and filtering design. Regarding environmental adaptability, the module passes high and low temperature cycling tests from -55℃ to 85℃, and insulation resistance is >100MΩ after damp heat testing. Contact resistance changes are <5% after salt spray and mold tests, fully meeting the stringent environmental requirements of military equipment.

[0114] Based on the vibration cumulative damage model, fatigue analysis is performed on historical impact data to output a remaining life prediction curve. The specific implementation is as follows:

[0115] The vibration monitoring core utilizes the PCB company's 356A01 IEPE accelerometer, which features a high sensitivity of 100mV / g and a measurement range of ±50g. The signal conditioning circuitry includes a 4th-order Butterworth anti-aliasing filter and a 24-bit Σ-Δ ADC, achieving a 10kHz sampling rate. The system's analysis bandwidth covers DC-5kHz, with a dynamic range of 90dB and a noise floor of <10μg / m³. The sensor is mounted using an M4 thread and secured with a special coupling agent to ensure a transmission loss of <1dB in the 20Hz-5kHz frequency band. Each monitoring node includes a triaxial accelerometer, and synchronous sampling technology ensures phase consistency of <0.1° between channels.

[0116] The damage model is based on an improved Miner cumulative damage theory, innovatively introducing a stress interaction effect coefficient and a load sequence correction factor. The model extracts the stress spectrum using rainflow counting and employs a bilinear damage accumulation rule, keeping the prediction error within 15%. Life prediction utilizes a proximal policy optimized reinforcement learning algorithm; the network structure includes a 128-node hidden layer and a ReLU activation function. The training dataset contains 2000 sets of accelerated life test data, covering multi-stress coupled conditions of temperature, vibration, and shock. The algorithm automatically updates model parameters every 24 hours to continuously optimize prediction accuracy.

[0117] The blockchain-based evidence storage system is built on Hyperledger Fabric 2.3 and uses the Kafka consensus mechanism, achieving a throughput of 500 TPS and a transaction confirmation latency of <50ms. Each data block contains device lifecycle data, and a Merkle tree structure ensures immutability. The remote diagnostic interface supports 4G / 5G dual-mode connectivity, MQTT protocol enables 1Hz real-time data streaming, and HTTPS is used for batch data download. Local storage uses industrial-grade 128GB SSDs, supporting operating temperatures from -40℃ to 85℃, and a wear leveling algorithm extends write life to 30,000 PE cycles. Data management software automatically categorizes and stores data, retaining critical data for 3 years and general data for 1 year.

[0118] The three-tiered early warning system employs a hierarchical response strategy: Level 1 warnings (remaining lifespan <100 hours) trigger a slow flashing yellow LED; Level 2 warnings (remaining lifespan <50 hours) activate a fast flashing red LED and intermittent buzzer; Level 3 warnings (remaining lifespan <10 hours) automatically send an emergency notification via satellite link. The early warning algorithm integrates three dimensions: equipment health index, failure probability, and risk level, achieving an accuracy rate >95%. The false alarm prevention system utilizes multi-sensor cross-validation and trend analysis to suppress the false alarm rate to below 1%. All early warning records automatically generate electronic work orders, with traceability ensured through blockchain storage.

[0119] The results of comparing the embodiments with conventional technical solutions are shown in Table 1:

[0120] Table 1

[0121]

[0122] As shown in Table 1, this invention completely solves the technical bottlenecks of traditional MEMS inertial navigation systems in high-dynamic environments through multi-dimensional innovation. The unique three-level shock-resistant structure improves shock isolation efficiency by 12 times; the intelligent redundancy design ensures that a single point of failure does not affect system operation; the LSTM temperature compensation algorithm reduces temperature drift error to 1 / 600 of traditional solutions; and the federated filtering architecture increases data processing speed by 10 times. These groundbreaking innovations are integrated into a single system, enabling it to maintain zero-bias stability of 0.05mg and a stability of 0.001°C / g even under extreme conditions such as 20,000g shock and temperature variations from -40℃ to 85℃. With its random walk capability, the drone's performance indicators surpass those of existing commercial solutions by 1-2 orders of magnitude, providing reliable technical support for the application of drones in demanding scenarios such as military and emergency response.

[0123] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0124] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A high dynamic inertial navigation system for unmanned aerial vehicles based on a multi-stage impact-resistant MEMS, characterized by, The system includes a high-dynamic carrier adaptation module, a redundant sensor array module, a gradient shock protection module, an intelligent signal acquisition module, a multi-modal fusion processing module, an environmental adaptive compensation module, a standardized output interface module, and a system health management module, wherein: The high-dynamic carrier adaptation module is used to perform modal matching and frequency isolation of the body vibration energy through the UAV centroid positioning architecture to obtain a stable measurement benchmark of the natural frequency. The redundant sensor array module is used to perform collaborative measurement and cross-validation of three sets of MEMS accelerometers based on a 120° spatially symmetrical topology, and output multi-axis inertial data with fault reconstruction capability. The gradient shock protection module is used to implement multi-level frequency band selective attenuation of mechanical shock loads using a silicon-based gel-titanium alloy honeycomb composite structure, thereby achieving significant suppression of vibration energy in a wide frequency range. The intelligent signal acquisition module is used to perform anti-aliasing filtering on the original analog signal of the MEMS sensor based on a high-precision ADC conversion circuit to obtain a 16-bit digitized acceleration sampling sequence. The multimodal fusion processing module is used to perform singular value decomposition on heterogeneous sensor data streams using a dynamic weighting factor matrix to generate attitude quaternions with error tolerance optimization. The environmental adaptive compensation module is used to combine an LSTM neural network to perform nonlinear modeling of temperature drift characteristics and establish an environmental error prediction model. The standardized output interface module is used to encapsulate navigation information into a frame structure based on the MIL-STD-1553B protocol stack to form a communication message that conforms to military standards. The system health management module is used to perform fatigue analysis on historical impact data based on the vibration cumulative damage model and output the remaining life prediction curve.

2. The multi-stage impact-resistant MEMS-based UAV high dynamic inertial navigation system of claim 1, wherein, The high-dynamic carrier adaptation module includes the following specific components: Dynamic stiffness optimization of the UAV frame structure is performed based on finite element modal analysis to form an anti-resonance mechanical transmission path; By utilizing magnesium alloy honeycomb sandwich layers to dissipate high-frequency vibration energy through waveguides, a frequency-selective structure can be constructed. Attenuation channel; The low-frequency sway is passively suppressed by the mass block-spring damping system, resulting in a stable reference measurement environment; A laser alignment instrument is used to perform micron-level leveling of the sensor mounting plane, establishing the hardware foundation for geometric error compensation; The structural resonance characteristics were experimentally calibrated using a modal testing system, resulting in closed-loop control parameters for vibration suppression.

3. The multi-stage impact-resistant MEMS-based UAV high dynamic inertial navigation system of claim 1, wherein, The redundant sensor array module includes the following specific components: Based on the spatial vector synthesis algorithm, the triaxial redundant data is orthogonalized to construct a full-dimensional inertial reference frame; A fault early warning mechanism is established by dynamically assessing the sensor's operating status through real-time health monitoring. Least squares fitting is used to reconstruct outlier measurements, maintaining continuous navigation parameter output. A temperature-stress coupling model is used to compensate for installation errors online, thereby improving the consistency of array measurements. By relying on covariance analysis, sensor weights are adaptively allocated to optimize the accuracy of multi-source data fusion.

4. The multi-stage impact-resistant MEMS-based UAV high dynamic inertial navigation system of claim 1, wherein, The gradient impact protection module includes the following specific components: Impedance matching of shock stress waves is achieved based on a viscoelastic-metal gradient structure to form a broadband energy dissipation network. Selective filtering of vibrations in specific frequency bands is achieved through porous cellular topology optimization, thus establishing a frequency domain protection barrier. Nonlinear damping materials are used to diffuse transient impact energy in the time domain, reducing the peak load transfer rate. A multi-layer composite interface design is used to perform waveguide control on the vibration propagation path, achieving directional energy attenuation. reduce; The protective performance was quantitatively verified by relying on the impact test bench, forming an optimization basis for impact-resistant design.

5. The multi-stage impact-resistant MEMS-based UAV high dynamic inertial navigation system of claim 1, wherein, The intelligent signal acquisition module includes the following specific components: A 16-bit Σ-Δ modulator is used to perform high-precision sampling of the sensor's analog signal to obtain a digital measurement sequence; Anti-aliasing filters are used to cut off high-frequency noise and improve the fidelity of the effective signal. Time domain consistency is ensured by performing time alignment on multi-channel data through synchronous sampling clocks; A digital gain calibration circuit is used to automatically adjust the signal amplitude and maintain dynamic range stability. The self-testing circuit is used to monitor the integrity of the data acquisition link in real time, ensuring data reliability.

6. The multi-stage impact-resistant MEMS-based UAV high dynamic inertial navigation system of claim 1, wherein, The multimodal fusion processing module includes the following specific components: Spatiotemporal alignment of heterogeneous sensor data is performed based on the Lie group filtering framework to eliminate coordinate system inconsistency errors. Dynamic suppression of measurement noise is achieved through adaptive Kalman gain, thereby improving the signal-to-noise ratio. We utilize robust estimation theory to perform weighted removal of outlier data points, thereby enhancing the algorithm's ability to resist interference. A federated filtering architecture is used to perform distributed fusion of multi-source information, balancing computational efficiency and accuracy. By relying on fuzzy logic control to perform online optimization of weight factors, intelligent data fusion is achieved.

7. The multi-stage impact-resistant MEMS-based UAV high dynamic inertial navigation system of claim 1, wherein, The environmental adaptive compensation module includes the following specific components: Feature extraction of the temperature-drift nonlinear relationship is performed based on deep neural networks to establish an error prediction model; Multi-resolution analysis of the sensor noise spectrum is performed using wavelet packet transform to achieve frequency domain adaptive filtering. The particle filter algorithm is used to dynamically estimate the system state error, thereby improving environmental adaptability. Multi-sensor cross-validation is used to perform online calibration of compensation parameters to maintain long-term stability; By leveraging digital twin technology, the compensation effect is virtually verified, forming a closed-loop optimization system.

8. The multi-stage impact-resistant MEMS-based UAV high dynamic inertial navigation system of claim 1, wherein, The standardized output interface module includes the following specific components: A time-triggered architecture is used to perform deterministic scheduling of the data communication process, thereby building a low-latency transmission channel; Differential Manchester encoding is used to encapsulate navigation commands to prevent interference and enhance the robustness of signal transmission. Cyclic redundancy check is used to perform multiple verifications on data packet integrity to ensure the reliability of information transmission; A double-buffered architecture is used to perform asynchronous management of the communication process, balancing real-time performance and throughput. Electromagnetic hardening of the interface physical layer is achieved by relying on military connector standards to improve environmental adaptability.

9. The multi-stage impact-resistant MEMS-based UAV high dynamic inertial navigation system of claim 1, wherein, The system health management module includes the following specific components: Based on vibration power spectral density analysis, the characteristics of impact energy distribution are extracted, and a damage accumulation model is established. Online assessment of sensor degradation trends is performed using deep reinforcement learning to predict remaining lifespan; Fault tree analysis is used to comprehensively identify system risk factors and construct a health assessment system. Blockchain technology is used to perform distributed storage of maintenance records, ensuring that the data is tamper-proof; The cloud platform is used for remote monitoring of system state data to realize predictive maintenance function.

Images