A low-power-consumption high-precision high-sensitivity inclination intelligent sensing system and method

Abstract

The present application relates to the technical field of intelligent sensor, in particular to a low-power high-precision high-sensitivity inclination intelligent sensing system and method. The system utilizes multi-element sensing components to synchronously sense MEMS core and shell double-source temperature, environmental acceleration and composite inclination signal; the inclination signal solving and error compensation unit executes physical signal conditioning, and removes dynamic vibration noise through frequency domain stop band suppression; a zero drift compensator is adopted to reconstruct the nonlinear evolution characteristics of long-term creep error according to the temperature historical reference sequence, and real-time calibration of zero drift amount is performed to execute reference correction on the inclination signal; the measurement state management unit monitors amplitude overrun and triggers reporting. Through multi-source physical error decoupling and zero self-calibration mechanism, the present application realizes long-term stable maintenance of inclination measurement reference and accurate acquisition of gravity vector in field environment.

Description

Technical Field

[0001] This invention relates to the field of intelligent sensor technology, specifically to a low-power, high-precision, and high-sensitivity tilt angle intelligent sensing system and method. Background Technology

[0002] Precise tilt measurement is a fundamental task in geodesy, topographic surveying, and inertial navigation orientation. In leveling scenarios such as geospatial orientation perception and geophysical field observation, the core requirement is to accurately obtain the tilt angle and horizontal attitude vector of the carrier relative to the Earth's gravity vertical. These data are the fundamental basis for establishing a high-precision geographic reference system and horizon benchmark.

[0003] While traditional precision optical theodolites or levels can provide reliable geodetic benchmarks, their optomechanical structures are ill-suited for long-term, unattended automated surveying in harsh field environments. In contrast, electronic levels and tilt sensors based on microelectromechanical systems (MEMS) offer advantages such as small size and low power consumption, making them ideal alternative surveying instruments. However, when these smart sensors are deployed as core attitude-determining components in complex outdoor environments for extended periods, they face the metrological defect of unstable zero points, i.e., long-term zero-point drift.

[0004] Specifically, the thermal hysteresis effect caused by environmental temperature differences and the long-term creep caused by the aging of packaging materials can lead to a continuous and nonlinear drift of the zero-point reference point of the sensitive axis inside the instrument, thereby masking the true changes in gravity level. Existing external data post-processing methods are unable to decouple this complex nonlinear response in real time, nor can they eliminate azimuth reference deviations without disassembling the instrument. Therefore, there is an urgent need in this field for an intelligent tilt measurement system that can deeply integrate attitude calculation and instrument error compensation and has self-calibration capabilities.

[0005] To address the technical problem of intelligent sensors struggling to autonomously decouple and eliminate measurement reference failure caused by the superposition of nonlinear thermal hysteresis and long-term creep in complex field environments, this invention proposes a low-power, high-precision, and high-sensitivity tilt angle intelligent sensing system and method. Summary of the Invention

[0006] The purpose of this invention is to provide a low-power, high-precision, and high-sensitivity tilt angle intelligent sensing system and method. Through the deep integration of multi-dimensional signal conditioning and time-series memory zero-position compensation, it can achieve accurate calculation of gravity vectors and autonomous calibration of long-term creep errors in complex environments.

[0007] To achieve the above objectives, the present invention provides the following technical solution: A low-power, high-precision, and high-sensitivity tilt angle intelligent sensing system includes: The multi-sensor module simultaneously collects MEMS core temperature, package shell temperature, environmental acceleration interference signals, and the original composite tilt angle signal generated by superimposing the true static tilt angle, thermal drift error component, creep error component, and dynamic vibration noise. The tilt angle signal calculation and error compensation unit operates the signal conditioning logic based on MEMS core temperature, package shell temperature, and environmental acceleration interference signal to remove dynamic vibration noise in the composite tilt angle signal and output preprocessed tilt angle data. Based on the preprocessed tilt angle data and temperature history reference sequence, a nonlinear mapping correction relationship between creep error component and history sequence is established using a zero-point drift compensator based on time memory, generating a real-time zero-point correction value. The real-time zero-point correction value is then used to perform reference correction with the preprocessed tilt angle data to obtain the final calibrated tilt angle data. The measurement status management unit generates an over-range status flag when the amplitude of the final calibration tilt angle data exceeds the measurement range limit. The wireless communication module responds to the over-range status flag and performs measurement data reporting.

[0008] Preferably, the tilt signal calculation and error compensation unit is specifically configured as follows when running the signal conditioning logic: The transient temperature difference between the MEMS core temperature and the packaging shell temperature is calculated in real time, and the transient temperature difference is input into a preset thermal hysteresis transfer function. The thermal hysteresis transfer function characterizes the thermal drift delay characteristics caused by the temperature gradient based on the principle of inertial links, and outputs the sensor thermal drift response coefficient. The tilt signal solution and error compensation unit also configures frequency domain stopband suppression logic based on the frequency domain characteristics of the environmental acceleration interference signal. The signal conditioning logic is configured to comprehensively superimpose the sensor thermal drift response coefficient and the suppression parameters output by the frequency domain stopband suppression logic to perform physical error elimination on the composite tilt signal.

[0009] Preferably, the tilt signal calculation and error compensation unit is specifically configured as follows when configuring the frequency domain stopband suppression logic: Spectral analysis is performed on the environmental acceleration interference signal to extract the main interference frequency point and the secondary interference frequency point of the interference gravity vector measurement. Using the main interference frequency point and the secondary interference frequency point, a set of adaptive notch filters is configured. The adaptive notch filters dynamically lock the center notch frequency on the main interference frequency point and the secondary interference frequency point, thereby attenuating the signal amplitude of the dynamic vibration noise in the frequency domain.

[0010] Preferably, the tilt signal calculation and error compensation unit is specifically configured as follows when configuring the zero-position drift compensator based on time-series memory: Based on the preprocessed tilt angle data, the long-term historical reference sequence of the MEMS core temperature and the packaging shell temperature, a recursive polynomial fitting network is constructed. The recursive polynomial fitting network uses the long-term historical reference sequence as the input state vector and the zero-point drift value of the preprocessed tilt angle data in the quasi-static calibration interval as the fitting convergence target. The creep error component is approximated by a nonlinear curve through the recursive polynomial fitting network.

[0011] Preferably, the measurement status management unit is further configured to: The rate of change of the internal correction coefficient of the tilt angle signal calculation and error compensation unit is monitored, and the instrument calibration cycle is dynamically set in combination with the current power supply; when the instrument calibration cycle is reached, an online calibration request is generated.

[0012] Preferably, the wireless communication module is further configured to: In response to the online calibration request, the preprocessed tilt angle data, the final calibrated tilt angle data, and a set of instrument operating parameters for remote calibration output by the tilt angle signal calculation and error compensation unit are reported synchronously. The instrument operating parameter set specifically includes: the current coefficients of the thermal hysteresis transfer function, the locking frequency of the adaptive notch filter, and the current weighting coefficient matrix of the recursive polynomial fitting network.

[0013] Preferably, it also includes a remote precision calibration module, which is configured as follows: The system receives measurement data reported by the wireless communication module; based on the received data and combined with the physical reference simulation model of the structure under test, it calculates the theoretical true value of the creep error component using a global error inversion algorithm; the global error inversion algorithm integrates a numerical solution algorithm for partial differential equations based on the laws of mechanics and thermodynamics of materials; based on the deviation analysis between the theoretical true value and the instrument operating parameter set, it generates a set of calibration correction constants; and the calibration correction constants are transmitted via the wireless communication module through remote configuration.

[0014] Preferably, the tilt signal calculation and error compensation unit is further configured as follows: Based on the calibration correction constant issued by the remote precision calibration module, the coefficient injection update is performed on the zero-point drift compensator based on time-series memory. Specifically, the update involves applying the calibration correction constant to fine-tune the internal weighting coefficients of the recursive polynomial fitting network, thereby eliminating the long-term zero-point drift of the multivariate sensing module.

[0015] A low-power, high-precision, and high-sensitivity tilt angle intelligent sensing method includes: The system simultaneously acquires MEMS core temperature, package shell temperature, environmental acceleration interference signals, and the original composite tilt angle signal generated by superimposing the true static tilt angle, thermal drift error component, creep error component, and dynamic vibration noise. Based on the MEMS core temperature, packaging shell temperature, and environmental acceleration interference signal operation signal conditioning logic, dynamic vibration noise in the composite tilt angle signal is removed, and preprocessed tilt angle data is output. Based on the preprocessed tilt angle data and temperature history reference sequence, a nonlinear mapping correction relationship between creep error components and history sequence is established using a zero-point drift compensator based on time-series memory, generating a real-time zero-point correction value. The real-time zero-point correction value is then used to perform reference correction with the preprocessed tilt angle data to obtain the final calibrated tilt angle data. An over-range status flag is generated when the amplitude of the final calibration tilt angle data exceeds the measurement range limit. In response to the over-range status flag, the measurement data is reported.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. By synchronously acquiring dual-source temperature data from the core and shell, and combining this with a thermal hysteresis transfer function based on the principle of inertial links, along with frequency domain stopband suppression logic, physical-level decoupling and elimination of nonlinear thermal hysteresis errors and dynamic vibration noise in composite tilt angle signals are achieved. This technique effectively solves the problem that existing single-point temperature measurements cannot cope with the temperature gradient effect caused by drastic day-night temperature changes outdoors, as well as vibration interference caused by strong winds or vehicle traffic. Therefore, it significantly improves the purity and signal-to-noise ratio of preprocessed tilt angle data in harsh outdoor dynamic environments.

[0017] 2. By constructing a zero-point drift compensator based on time-series memory and establishing a nonlinear mapping relationship between long-term creep error and historical reference sequences using a recursive polynomial fitting network, autonomous prediction and correction of zero-point drift caused by sensor aging due to packaging material aging and internal stress release are achieved. This technique overcomes the shortcomings of traditional polynomial fitting, which relies solely on the current state and ignores historical cumulative effects. It ensures that the system can maintain a high-precision zero-point reference in long-term unattended monitoring scenarios, significantly extending the maintenance-free cycle of the equipment.

[0018] 3. A remote precision calibration module employs a global error inversion algorithm based on a physical benchmark simulation model. This algorithm interacts in a closed loop with the instrument operating parameter set reported by the local wireless communication module. This enables the use of high-performance computing in the cloud to solve partial differential equations to obtain theoretical truth values, and the remote injection of calibration correction constants to fine-tune the recursive network coefficients. This architecture combines low-power real-time monitoring at the edge with physical-level precision calibration in the cloud, resolving the contradiction that low-power embedded devices struggle to perform complex physical modeling calculations. This ensures high sensitivity and data reliability of the sensing system throughout its entire lifecycle. Attached Figure Description

[0019] Figure 1 This is a block diagram of a low-power, high-precision, and high-sensitivity tilt angle intelligent sensing system according to the present invention. Figure 2 This is a flowchart illustrating the internal architecture and error compensation process of the recursive multinomial fitting network in an embodiment of the present invention. Figure 3 This is a flowchart of a low-power, high-precision, and high-sensitivity tilt angle intelligent sensing method according to the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Other embodiments obtained by those skilled in the art based on the ideas in this specification without creative effort all fall within the protection scope of this invention.

[0021] Reference Figures 1 to 3 This invention provides a low-power, high-precision, and high-sensitivity tilt angle intelligent sensing system, the specific technical solution of which is as follows: The multi-sensor module simultaneously collects MEMS core temperature, package shell temperature, environmental acceleration interference signals, and the original composite tilt angle signal generated by superimposing the true static tilt angle, thermal drift error component, creep error component, and dynamic vibration noise. The tilt angle signal calculation and error compensation unit operates the signal conditioning logic based on MEMS core temperature, package shell temperature, and environmental acceleration interference signal to remove dynamic vibration noise in the composite tilt angle signal and output preprocessed tilt angle data. Based on the preprocessed tilt angle data and temperature history reference sequence, a nonlinear mapping correction relationship between creep error component and history sequence is established using a zero-point drift compensator based on time memory, generating a real-time zero-point correction value. The real-time zero-point correction value is then used to perform reference correction with the preprocessed tilt angle data to obtain the final calibrated tilt angle data. The measurement status management unit generates an over-range status flag when the amplitude of the final calibration tilt angle data exceeds the measurement range limit. The wireless communication module responds to the over-range status flag and performs measurement data reporting.

[0022] Example 1: This embodiment uses the attitude monitoring of the main tower and bridge deck of a long-span suspension bridge in a mountainous canyon environment as a specific application scenario. In such scenarios, the monitoring equipment not only faces the thermal shock caused by huge diurnal temperature differences, but also needs to maintain high accuracy under continuous dynamic vibrations caused by vehicle traffic and crosswinds in the canyon. At the same time, it must monitor the slight long-term tilt caused by concrete shrinkage and creep or foundation settlement.

[0023] In the tilt angle intelligent sensing system detailed in this embodiment of the invention, the core processing flow after the system is powered on is as follows: The multi-sensor module periodically and synchronously captures the tilt angle state of the bridge monitoring points, the dual-point temperature distribution of the MEMS core and shell, and the vibration acceleration of the bridge body; The tilt angle signal calculation and error compensation unit receives the above multi-dimensional data stream, immediately starts the signal conditioning logic, and removes the high-frequency vibration noise caused by wind load and vehicle traffic from both the frequency domain and time domain. Then, it calls the internally integrated zero-point drift compensator based on time-series memory, calculates the zero-point drift amount of the sensor caused by long-term operation at the current moment based on the historical temperature and tilt angle reference sequence, deducts it, and outputs the true tilt angle value of the bridge structure; The measurement status management unit continuously monitors the true tilt angle value. Once the tilt angle is detected to exceed the preset bridge safety warning threshold (i.e., the range limit), it immediately triggers the over-range status flag, activates the wireless communication module to send alarm data to the remote health monitoring center, thereby achieving high-reliability monitoring under low power consumption.

[0024] Furthermore, when running the signal conditioning logic, the tilt signal calculation and error compensation unit is specifically configured to: calculate the transient temperature difference between the MEMS core temperature and the packaging shell temperature in real time, and input the transient temperature difference into a preset thermal hysteresis transfer function; the thermal hysteresis transfer function characterizes the thermal drift delay characteristics caused by the temperature gradient based on the principle of inertial links, and outputs the sensor thermal drift response coefficient; the tilt signal calculation and error compensation unit also configures frequency domain stopband suppression logic based on the frequency domain characteristics of the environmental acceleration interference signal, and the signal conditioning logic is configured to comprehensively superimpose the sensor thermal drift response coefficient and the suppression parameters output by the frequency domain stopband suppression logic to perform physical error elimination on the composite tilt signal.

[0025] Specifically, in the long-span suspension bridge monitoring scenario of this embodiment, to demonstrate the edge autonomous perception capability of the intelligent sensor in complex environments, the system abandons the passive mode of traditional sensors that only output raw analog quantities, and adopts a hardware-software co-operation dual-source temperature sensing mechanism. The intelligent sensor's calculation core directly obtains the MEMS core temperature generated by the PTAT (Absolute Temperature Proportional) semiconductor temperature measurement circuit embedded in the chip by reading the internal registers of the MEMS chip; simultaneously, it acquires the high-precision NTC thermistor signal attached to the inner wall of the sensor's metal protective shell to obtain the packaging shell temperature. The system uses a high-precision synchronous clock to lock these two signals and calculates the transient temperature difference between the core and the shell in real time. This step is the basis for the sensor's self-state diagnosis.

[0026] Subsequently, the embedded processing unit inside the sensor invokes a pre-defined digital thermal hysteresis transfer function. This function constructs a first-order recursive filtering model simulating the time lag effect of heat conduction using an edge-side algorithm. Specifically, this model employs a discretized inertial filtering algorithm: the thermal drift response coefficient at the current moment is equal to the transient temperature difference at the current moment multiplied by the thermal inertia factor, plus the thermal drift response coefficient at the previous moment multiplied by (1 minus the thermal inertia factor). Here, the thermal inertia factor is a constant between 0 and 1, a physical parameter pre-calibrated based on the specific heat capacity and thermal conductivity of the sensor's packaging material (such as 316L stainless steel). Through this edge computing based on time-series memory, the sensor can output an accurate thermal drift response coefficient, signifying that the sensor possesses the intelligent characteristic of "understanding" its own internal thermal stress release process, rather than merely mechanically responding to external temperature.

[0027] Meanwhile, to address the broadband vibrations of the suspension bridge under vehicle traffic or strong winds, the frequency-domain stopband suppression logic within the unit initiates an environmental adaptive sensing mode. It performs real-time time-frequency domain analysis on the synchronously acquired environmental acceleration signals, intelligently identifying energy-concentrated interference frequency bands (such as specific high-frequency vibration intervals caused by heavy vehicles) and dynamically generating corresponding digital bandstop filter parameters. Finally, the signal conditioning logic performs physical error removal at the sensor edge: subtracting the thermal drift component corrected by the thermal hysteresis model from the original composite signal, and filtering out vibration noise through an adaptive filter. This process is entirely completed within the sensor, outputting physically cleaned pre-processed tilt angle data.

[0028] In a preferred embodiment, the tilt signal calculation and error compensation unit is also configured to perform vibration rectification error compensation when running the signal conditioning logic: based on the amplitude of the AC component of the environmental acceleration interference signal, the estimated value of the DC bias error induced by the nonlinear effect is calculated using a pre-calibrated second-order rectification coefficient, and the estimated value of the DC bias error is subtracted from the DC component of the composite tilt signal to eliminate the cross-coupling interference of high-frequency vibration on the static tilt measurement.

[0029] Specifically, when a suspension bridge faces strong crosswinds or heavy truck convoys, the sensors are subjected to continuous high-frequency broadband vibrations. Although notch filters can filter out the fluctuating portion of the signal, the asymmetry of the MEMS microstructure "rectifies" some of the high-frequency vibration energy into a spurious DC voltage offset (i.e., vibration rectification error, VRE), causing the filtered average value to still deviate from the true tilt angle. To address this, the system activates the VRE compensation module, first extracting the root mean square (RMS) value of the environmental acceleration signal to quantify the vibration energy intensity at the current moment. Subsequently, the system calls upon the sensor's "vibration-drift sensitivity coefficient" (usually proportional to the square of the vibration amplitude) measured during the factory calibration phase, and calculates the expected DC bias under the current vibration intensity using a second-order polynomial (e.g., a spurious offset of 0.02 degrees under 0.5g vibration). Finally, the system directly subtracts this estimated value from the tilt angle reading.

[0030] This embodiment improves upon the physical problem that traditional filtering algorithms cannot remove "vibration-induced DC drift," significantly enhancing the static measurement accuracy of the sensor in dynamic and harsh environments. It ensures that even when the bridge is violently shaking, the system can still accurately output the minute changes in its static equilibrium position.

[0031] This embodiment achieves a leap from "passive measurement" to "intelligent sensing" by integrating dual-source temperature monitoring and a digital thermal hysteresis model based on inertial principles at the sensor edge. Combined with environmentally adaptive frequency-domain stopband suppression, the system can autonomously cope with thermal gradient errors caused by rapid temperature changes in the field and dynamic traffic flow noise, significantly reducing the computing burden on cloud servers and demonstrating the efficiency and reliability of intelligent sensors in edge computing and data cleaning.

[0032] Furthermore, when configuring the frequency domain stopband suppression logic, the tilt signal calculation and error compensation unit is specifically configured to: perform spectral analysis on the environmental acceleration interference signal to extract the main interference frequency point and the secondary interference frequency point that interfere with the gravity vector measurement; and configure a set of adaptive notch filters using the main interference frequency point and the secondary interference frequency point, wherein the adaptive notch filters dynamically lock the center notch frequency on the main interference frequency point and the secondary interference frequency point, thereby attenuating the signal amplitude of the dynamic vibration noise in the frequency domain.

[0033] Specifically, in this embodiment, the core of the intelligent sensor's calculation no longer relies on a fixed low-pass filter, but instead enables an adaptive spectrum sensing mechanism at the edge. The tilt signal calculation and error compensation unit first performs time windowing on the acquired environmental acceleration interference signal (e.g., using a Hanning window to reduce spectral leakage), and then performs a Fast Fourier Transform (FFT) at the edge. By calculating the power spectral density of the signal, the unit can intelligently scan and identify the two peak signals with the highest energy amplitude in the frequency domain. The system marks the frequency component with the highest amplitude as the primary interference frequency point (e.g., the first-order natural frequency of the bridge), and the frequency component with the second highest amplitude as the secondary interference frequency point (e.g., specific harmonic frequencies caused by heavy vehicles).

[0034] Subsequently, using the two identified specific interference frequency values, the sensor's internal algorithm dynamically reconstructs the digital signal processing link and configures a set of cascaded adaptive notch filters. This filter is constructed based on a second-order infinite impulse response (IRR) digital filter structure. The system calculates the required feedback and feedforward coefficients in real time using table lookup or polynomial approximation, based on the currently identified primary and secondary interference frequencies and a preset narrow bandwidth factor. Specifically, the system uses the transfer function design formula of the second-order IRR notch filter, based on the current interference frequency and the preset sampling rate, to calculate specific coefficients that reduce the gain to zero at the interference frequency. This calculation ensures an extremely narrow notch bandwidth, attenuating only the energy at the interference frequency while maintaining a constant gain for the DC component. This process is equivalent to precisely "etching" two extremely narrow stopband depressions within the sensor's signal passband.

[0035] Finally, the system injects these parameters into the filtering logic in real time, ensuring that the center frequency of the notch filter is dynamically locked to the frequency of the interference source, which varies with the environment. When the composite tilt signal flows through the filter, dynamic vibration noise at specific frequencies (i.e., the signal components corresponding to the primary and secondary interference points) is precisely suppressed, with its amplitude attenuation depth set to greater than 20 dB. This reduces the interference energy within the notch bandwidth to less than 10% of its original value, while DC or extremely low-frequency components containing real gravity information pass through almost without loss, and the passband gain remains near 0 dB. This processing method acts like a smart "surgical knife," removing specific lesion noise rather than crudely cutting off all high-frequency signals.

[0036] This embodiment achieves proactive sensing and precise suppression of dynamic environmental noise by integrating real-time spectrum analysis and adaptive notch filtering technology at the edge of the smart sensor. This technique effectively overcomes the shortcomings of traditional fixed filters in dealing with frequency drift (such as the frequency of a bridge structure changing with its mass), and maximizes the signal-to-noise ratio and fidelity of tilt angle measurement while preserving the high-frequency dynamic response bandwidth.

[0037] Furthermore, when configuring the zero-point drift compensator based on time-series memory, the tilt signal calculation and error compensation unit is specifically configured as follows: based on the long-term historical reference sequence of the preprocessed tilt data, the MEMS core temperature, and the packaging shell temperature, a recursive polynomial fitting network is constructed; the recursive polynomial fitting network uses the long-term historical reference sequence as the input state vector and the zero-point drift value of the preprocessed tilt data in the quasi-static calibration interval as the fitting convergence target; the creep error component is approximated nonlinearly through the recursive polynomial fitting network.

[0038] Specifically, refer to Figure 2 In this embodiment, the smart sensor incorporates a recursive multinomial fitting network with edge online learning capabilities. Instead of being based on a high-energy-consuming deep neural network, this network is constructed as a lightweight adaptive filter architecture based on the recursive least squares principle, specifically designed for operation in low-power embedded controllers.

[0039] The system first maintains a circular buffer in the sensor's internal storage unit to store a long-term historical reference sequence. This sequence includes the MEMS core temperature, package shell temperature, and operating timestamps over a past period. When the sensor is in a quasi-static calibration interval (the specific determination logic is: calculate the standard deviation of the environmental acceleration signal within a preset time window; if this standard deviation is consistently less than a preset static threshold (e.g., 0.005g) and lasts for more than 10 seconds, then the bridge deck is determined to be in a static state with no vehicle traffic and minimal wind load), the system locks the deviation between the current preprocessed tilt angle data and the theoretical zero point (or initial installation reference) as the fitting convergence target, which is considered the current true creep value.

[0040] In a preferred implementation, the tilt signal calculation and error compensation unit is configured to run a sparsification update logic based on curvature saliency when constructing the temperature history reference sequence: it calculates the trajectory curvature formed by the currently acquired data and the most recently stored historical reference point in real time, and only writes the temperature and tilt data of the current moment as keyframes into the historical reference sequence when the trajectory curvature exceeds a preset nonlinear threshold; otherwise, it performs a data discard operation, thereby minimizing the sequence length while maintaining the integrity of the long-term creep trend characteristics.

[0041] Specifically, in the long-term monitoring of long-span suspension bridges, the zero-point drift of sensors often exhibits non-uniform characteristics of "fast in the initial stage and slow in the later stage" and "fluctuating with the seasons." To record drift trajectories spanning several years within limited Flash storage space, the system no longer employs a timed storage mode but instead introduces a geometric curvature judgment algorithm. The system constructs a three-point sliding window in memory, consisting of the "previous keyframe," the "current storage frame," and the "real-time sampling point," and calculates the geometric curvature of these three points on the "temperature-drift" plane. When the sensor is in a stable aging period, the drift changes approximately linearly with temperature, and the curvature is extremely small; the system automatically discards intermediate redundant data. However, when a drastic temperature change or a sudden change in material stress release causes an inflection point in the drift curve (i.e., the curvature exceeds a set threshold of 0.05), the system determines that the current data contains important non-linear characteristics, immediately locks it as a "keyframe," and permanently writes it into the historical sequence.

[0042] This embodiment successfully compresses a year's worth of high-frequency monitoring data into a feature sequence containing only a few hundred keyframes through this "on-demand memory" sparsity strategy. While reducing the data storage volume by more than 90%, it perfectly preserves all the nonlinear details of the zero-point drift, greatly reducing the computational overhead of the subsequent fitting network.

[0043] At this point, the recursive multinomial fitting network is activated. It constructs a multidimensional input state vector containing a time power term (representing the aging trend over time) and a temperature power term (representing residual thermal stress). The network uses recursive least squares (RLS) as its core update algorithm. The specific operational logic is as follows: First, the system constructs an input state vector, which consists of a constant bias term, a first-order normalized time term, a second-order normalized time term, a first-order temperature term, and a second-order temperature term. To ensure long-term numerical stability, the normalized time is defined as the ratio of the relative duration of the current moment with respect to the previous calibration time point to a preset period. When the system performs a new online calibration or reaches the preset cycle limit, the time term is automatically reset to zero to prevent computational overflow caused by the unbounded growth of the time parameter. Next, the prior prediction error is calculated, which equals the measured drift value at the current moment minus the transpose of the weighted coefficient matrix from the previous moment and the current input state vector. Then, the gain vector is updated, calculated by multiplying the covariance matrix from the previous moment and the current input state vector, dividing by the sum of a forgetting factor and the weighted normalization term of the input state vector. Finally, the weighted coefficient matrix is ​​updated; the new weighted coefficients equal the old weighted coefficients plus the product of the gain vector and the prior prediction error. The forgetting factor is set between 0.98 and 0.995 to strike a balance between smoothing historical noise and tracking recent trends. Mathematically, this process is equivalent to approximating a creep curve that slowly evolves over time in a multidimensional space. Whenever a new quasi-static observation is obtained, the network performs a coefficient iteration, causing its output prediction value to continuously approach the true creep error component. During the non-calibration period (i.e., during normal monitoring), the system directly uses the predicted value output by the network as the current zero-point correction amount, which is subtracted from the measurement results, thereby achieving mathematical-level compensation for long-term physical creep.

[0044] This embodiment endows the sensor with "autonomous evolution" characteristics by constructing a recursive fitting network with temporal memory capabilities at the edge of the smart sensor. It can perform self-training using fragmented quasi-static time slices without relying on external manual calibration, accurately tracking and eliminating zero-point drift caused by device aging and long-term environmental effects, ensuring that the system maintains a high-precision measurement benchmark throughout its entire lifespan of several years.

[0045] Furthermore, the measurement status management unit is also configured to: monitor the rate of change of the internal correction coefficient of the tilt signal calculation and error compensation unit, and dynamically set the instrument calibration cycle in conjunction with the current power supply; and generate an online calibration request when the instrument calibration cycle is reached.

[0046] In a preferred implementation, when generating an over-range status flag, the measurement status management unit is configured to use dynamic thermodynamic threshold judgment logic: read the current MEMS core temperature in real time, look up the structural safety reference tilt angle at the current temperature according to the thermal expansion coefficient of the material of the structure under test, and set the measurement range limit as the superposition value of the structural safety reference tilt angle and the preset safety margin, thereby realizing adaptive alarm that automatically drifts with the ambient temperature.

[0047] Specifically, as a typical flexible structure, suspension bridges are highly sensitive to temperature. During the high temperatures of summer, the main cables elongate due to thermal expansion, causing the bridge deck at mid-span to naturally deflect, resulting in a physical change in the tower's tilt angle; the opposite occurs in winter. If the system uses a fixed alarm threshold (e.g., ±1 degree), false alarms are easily triggered under extreme temperature differences. Therefore, the intelligent sensors incorporate a "temperature-morphology reference table" for the bridge. Before each judgment, the state management unit first reads the current temperature, queries the theoretical "healthy attitude angle" of the bridge at that temperature (e.g., a theoretical tilt angle of 0.5 degrees at 30℃), and then adds a safety margin (e.g., ±0.2 degrees) to dynamically generate the current alarm threshold (i.e., 0.7 degrees). Only when the measured data exceeds this dynamic threshold is a structural anomaly determined.

[0048] This embodiment introduces a dynamic threshold mechanism based on environmental perception, which enables the sensor to intelligently distinguish between "normal thermal expansion and contraction" and "structural damage." This completely solves the problem of false alarms caused by seasonal temperature differences in the monitoring of large outdoor structures, and significantly improves the sensitivity and reliability of the system's alarms.

[0049] Specifically, in this embodiment, the measurement state management unit of the intelligent sensor incorporates an adaptive task scheduling strategy. This unit continuously monitors the internal state of the recursive polynomial fitting network in the tilt signal calculation and error compensation unit, specifically calculating the discrete rate of change of the weighting coefficient matrix over the most recent few time steps. If the rate of change is low, it indicates that the creep trend inside the sensor has stabilized or is in a linear development stage, and the model confidence is high; conversely, if the rate of change fluctuates drastically, it suggests that a sudden change in the environment has occurred or the sensor has entered a new aging stage, and the model accuracy is at risk.

[0050] Meanwhile, the unit reads the remaining power data of the power management chip in real time through the ADC interface. Based on these two sets of status data, the system runs a dynamic scheduling algorithm: when the system power is sufficient (e.g., greater than 80%) and the coefficient change rate is high, the system sets the instrument calibration cycle to a high-frequency mode (e.g., once per hour), sacrificing power in exchange for real-time tracking of rapid drift; when the system power is critical (e.g., below 20%), regardless of how the coefficient changes, the system is forced into a "survival priority mode", significantly extending the calibration cycle (e.g., once per day) to ensure the continuation of basic functions.

[0051] Once the system's internal timer reaches the dynamically set calibration cycle, the management unit immediately generates a high-priority online calibration request interrupt signal at the software layer. This signal wakes up the communication and computing modules, which are in a low-power sleep state, triggering subsequent data interaction processes. This mechanism allows the sensor to no longer operate rigidly at a fixed frequency, but rather, like a living organism, intelligently adjust its working rhythm based on its own "health status" (model stability) and "energy reserves" (power).

[0052] This embodiment breaks away from the rigid mode of fixed-period calibration of traditional sensors by introducing an adaptive scheduling mechanism based on model stability and remaining energy perception. This technology ensures that the system can keenly capture minute error changes in a "power-rich" state, and maximize its battery life in a "power-depleted" state, thereby achieving an intelligent dynamic balance between measurement accuracy and power consumption control at the edge where hardware resources are limited.

[0053] Furthermore, the wireless communication module is also configured to: respond to the online calibration request, synchronously report the preprocessed tilt angle data, the final calibrated tilt angle data output by the tilt angle signal calculation and error compensation unit, and a set of instrument operating parameters for remote calibration; the instrument operating parameter set specifically includes: the current coefficients of the thermal hysteresis transfer function, the locking frequency of the adaptive notch filter, and the current weighting coefficient matrix of the recursive polynomial fitting network.

[0054] Specifically, in this embodiment, when the smart sensor determines that online calibration is required through the aforementioned adaptive state management mechanism, the wireless communication module (e.g., a communication chip using NB-IoT or LoRaWAN low-power wide-area network protocols) is activated and establishes an encrypted data link with the remote cloud server. The sensor then executes a "fully transparent" data packaging strategy to construct a structured data packet containing multi-level information.

[0055] The first level of this data packet contains preprocessed tilt angle data and final calibrated tilt angle data. The system pairs the "semi-finished" data (which has undergone physical error removal but has not yet been zero-point corrected) with the "final" data (which has undergone time-series memory compensation). This paired reporting mechanism allows the remote center to intuitively compare and analyze the correction magnitude of the edge-end algorithm for long-term creep errors, verifying the rationality of the compensation logic.

[0056] The second layer of this data packet is the core set of instrument operating parameters, which is equivalent to the sensor reporting its own "mental snapshot." Specifically, the system serializes the currently used thermal inertia weighting factor in the thermal hysteresis transfer function into a floating-point array to inform the server of the sensor's current sensitivity assessment to environmental thermal shock. Simultaneously, it writes the currently dynamically locked center frequency value of the adaptive notch filter (e.g., the primary and secondary vibration frequencies of a bridge accurate to 0.1 Hz) into the load, enabling the server to verify whether the sensor has accurately captured the true structural resonance point. Most importantly, the system compresses and encodes the massive weighting coefficient matrix within the recursive polynomial fitting network and uploads it. This matrix represents the aging trend model "learned" by the sensor over long-term operation. By reporting this parameter, the remote server can directly read the sensor's "memory" at the edge, thereby determining whether overfitting or model divergence has occurred, achieving interpretable auditing and remote health diagnosis of the edge intelligent algorithm.

[0057] This embodiment constructs a "white-box" intelligent monitoring link by synchronously reporting measurement process data and internal algorithm state parameters. This technology not only enables the remote center to obtain high-precision tilt angle results, but also allows it to grasp the logical basis (i.e., operating parameters) for the sensor to obtain these results. This provides complete data support for subsequent global error inversion and precision calibration based on cloud-based physical models, completely solving the pain points of traditional sensor data being untraceable and fault diagnosis being difficult.

[0058] Furthermore, it also includes a remote precision calibration module, which is configured to: receive measurement data reported by the wireless communication module; calculate the theoretical true value of the creep error component based on the received data and a physical reference simulation model of the structure under test using a global error inversion algorithm; the global error inversion algorithm integrates a numerical solution algorithm for partial differential equations based on the laws of mechanics and thermodynamics of materials; generate a set of calibration correction constants based on the deviation analysis between the theoretical true value and the instrument operating parameter set; and transmit the calibration correction constants via the wireless communication module through a remote configuration method.

[0059] Specifically, in this embodiment, the remote precision calibration module deployed in the monitoring center is hosted on a high-performance cloud computing cluster. Its core is a digital twin system that is highly consistent with the actual monitored object (a long-span suspension bridge). This module first receives packaged data reported by the front-end sensors and parses it to obtain the original state set, which includes ambient temperature, vibration frequency, and pre-processed tilt angle.

[0060] Subsequently, the module invokes the physical benchmark simulation model of the structure under test. This model is a high-precision digital structure built based on the finite element analysis (FEA) theory, which internally defines in detail the geometric dimensions of the bridge towers and main cables, the thermal expansion coefficients, elastic moduli, and nonlinear creep characteristics of the materials. Using the received environmental data as boundary conditions, the module initiates a global error inversion algorithm. The core of this algorithm lies in numerically solving two sets of key physical equations: First, by solving the unsteady-state heat conduction equation, which describes the relationship between the rate of temperature change over time, the thermal diffusivity coefficient, and the spatial distribution of temperature under the action of the Laplace operator, the transient temperature field distribution generated by solar radiation and air temperature changes inside the bridge structure is simulated; Second, based on this temperature field, the static equilibrium equations of thermo-structural coupling are solved simultaneously, that is, the product of the structural stiffness matrix and the displacement vector is equal to the sum of the external load vector and the thermal expansion equivalent load vector, calculating the theoretical deformation of the bridge under the combined action of thermal stress and gravity, and then deriving the theoretical structural attitude angle at the sensor installation location. This system, based on the pre-defined logic that the confidence level of the digital twin model is higher than that of the aging sensor, uses the theoretical structural attitude angle calculated by the cloud-based physical model as the thermoelastic response benchmark, assuming that the current weather and load conditions are within the normal range and no sudden impact events are detected. The system calculates the residual between the final calibration tilt angle data uploaded by the sensor and the theoretical structural attitude angle, and executes a "drift-deformation" binary discrimination logic: First, it calculates the correlation between the residual sequence and the current temperature change sequence. If the two show a high correlation (indicating that the deviation mainly fluctuates with temperature), it is determined to be sensor thermal drift or model thermal parameter deviation, and a calibration command is generated. If the correlation is low and the residual shows a non-regressive monotonically divergent trend (indicating that the deviation is unrelated to temperature and may be due to foundation settlement or structural plastic damage), it is determined to be a real structural anomaly, the system forcibly terminates the calibration process, and triggers the highest-level structural safety alarm. This process utilizes cloud computing power to reconstruct the theoretical response of the physical world, serving as a relative reference benchmark for calibration.

[0061] Next, the module performs deviation analysis. The system compares the "theoretical true value" calculated in the cloud with the "final calibration tilt data" reported by the sensor, calculating the residual sequence. More importantly, the system uses this residual sequence to perform sensitivity analysis on the "instrument operating parameter set" reported by the sensor (especially the weighting coefficients of the recursive polynomial fitting network), identifying specific coefficient terms that cause prediction deviations at the edge. Based on a gradient descent optimization strategy, the algorithm generates a set of calibration correction constants that minimize this residual. This set of constants is essentially a fine-tuning instruction for the weights of the edge fitting network. Finally, the remote module encapsulates this set of instructions into a configuration message and sends it to the edge sensor via a wireless network, completing the remote correction of the data model by the physical model.

[0062] This embodiment solves the problem of "unknown zero-point drift" caused by the lack of external benchmarks after long-term operation of a single sensor by constructing a collaborative architecture of "cloud-based physical simulation + edge data calibration". By using cloud-based solutions to complex material mechanics and thermodynamics partial differential equations, the system can obtain the theoretical truth at the physical level, thereby periodically and precisely correcting the lightweight fitting algorithm at the edge, ensuring the absolute accuracy and physical interpretability of the monitoring data.

[0063] Furthermore, the tilt signal calculation and error compensation unit is also configured to: perform coefficient injection update on the zero-point drift compensator based on time-series memory based on the calibration correction constant issued by the remote precision calibration module. Specifically, the update is to apply the calibration correction constant to fine-tune the internal weighting coefficients of the recursive polynomial fitting network to eliminate the long-term zero-point drift of the multi-element sensing module.

[0064] Specifically, in this embodiment, after the wireless communication module receives the configuration message from the cloud-based remote precision calibration module, the intelligent sensor immediately initiates the online parameter reconstruction mechanism. The tilt signal calculation and error compensation unit first performs integrity verification on the message and extracts the calibration correction constant obtained through inversion calculation using the cloud-based physical model. Mathematically, this constant represents the deviation gradient value of the weight matrix in the current recursive polynomial fitting network.

[0065] To complete the update without interrupting the monitoring task, the unit performs an atomic-level coefficient injection update operation. The system pauses the next-time prediction operation of the fitted network and performs a weighted superposition or replacement operation on the extracted calibration correction constants and the current weighted coefficient matrix stored in non-volatile memory (such as Flash or EEPROM). This process is equivalent to using the "omniscient perspective" of the cloud to correct the "local memory" of the edge algorithm, forcibly pulling the convergence trajectory of the fitted network back to the real physical baseline.

[0066] After fine-tuning, the system immediately restarts the recursive computation logic. At this point, the real-time zero-point correction value output by the recursive polynomial fitting network contains the latest physical calibration information, which can accurately offset the cumulative deviation caused by sensor packaging stress release, material aging, or mounting base deformation. Through this closed-loop control of "cloud inversion-edge injection," the system successfully eliminates the nonlinear zero-point drift generated by the multi-sensor module during long-term operation, enabling the sensor to "rejuvenate" without the need for manual on-site disassembly and calibration, maintaining factory-grade measurement accuracy.

[0067] By constructing a closed-loop calibration mechanism of "cloud-based physical inversion - edge coefficient injection," this invention solves the problem of persistent drift caused by the inability of traditional edge sensors to run complex physical models due to limited computing power. The system utilizes remotely distributed calibration correction constants to perform atomic-level internal weighting coefficient fine-tuning on the recursive polynomial fitting network at the edge. In effect, it "distills" the "cognitive ability" of high-precision cloud-based physical simulation (based on the theoretical truth value of partial differential equations) through parameterization and injects it into a lightweight edge algorithm. This mechanism enables the sensor to correct its nonlinear zero-point drift in real time using external physical benchmarks without requiring manual on-site disassembly and calibration. It achieves deep integration of the data-driven model and the physical mechanism model, thereby ensuring that the sensing system maintains factory-grade absolute measurement accuracy throughout its entire lifecycle.

[0068] This invention achieves precise decoupling and autonomous compensation for dynamic interference and static drift in complex environments under low-power hardware conditions through a multi-dimensional sensing and temporal memory deep integration architecture. The system utilizes multiple sensing modules to simultaneously capture dual-source temperature and environmental acceleration, and, in conjunction with the signal conditioning logic of the tilt angle signal calculation and error compensation unit, physically isolates dynamic vibration noise. Simultaneously, it innovatively introduces a zero-point drift compensator based on temporal memory, using historical benchmark sequences to establish a nonlinear mapping, accurately predicting and eliminating long-term creep errors caused by sensor material aging. Combined with the over-range triggering mechanism of the measurement state management unit, the system significantly reduces the power consumption of transmitting invalid data while ensuring high sensitivity in capturing key structural deformations. This provides a truly intelligent monitoring solution with "environmental self-adaptation, error self-calibration, and energy efficiency self-balancing" characteristics, making it particularly suitable for long-term health monitoring of large infrastructure in harsh outdoor environments.

[0069] Example 2: This embodiment provides a low-power, high-precision, and high-sensitivity tilt angle intelligent sensing method, referring to... Figure 3 This method operates within intelligent sensors deployed at critical stress nodes (such as tower tops or main cable clamps) of long-span suspension bridges. The method is executed by an embedded microprocessor according to a predetermined program logic; the specific processing flow is as follows: The system first initiates a hardware-level synchronous acquisition task. Driven by a microsecond-level synchronization clock, the sensor simultaneously reads data from the PTAT semiconductor temperature measurement circuit integrated within the MEMS chip (i.e., the MEMS core temperature) via the internal bus, and acquires data from the NTC thermistor attached to the inner wall of the casing (i.e., the package temperature) via the ADC interface. Simultaneously, the accelerometer continuously captures environmental acceleration interference signals at a high sampling rate (e.g., 100Hz), as well as the raw composite tilt signal, which is a mixture of the true static tilt angle, thermal drift, long-term creep, and vibration noise. This synchronization mechanism ensures that all physical quantities within the same time slice have strict temporal correlation, providing an aligned data foundation for subsequent calculations.

[0070] After acquiring the raw data, the method enters the signal conditioning phase. The processor executes the signal conditioning logic, which contains two branches that process in parallel: The first branch handles thermal hysteresis error: the method calculates the transient temperature difference between the core temperature and the shell temperature, and inputs it into a digitized thermal hysteresis transfer function. This function uses a recursive algorithm based on material thermal inertia calibration to calculate the thermal drift response coefficient of the sensor at the current moment due to the release of internal thermal stress.

[0071] The second branch handles dynamic vibration noise: the method performs frequency domain analysis on the environmental acceleration signal, identifies the high-frequency interference energy band caused by bridge vibration or vehicle traffic, and configures the corresponding frequency domain stopband suppression logic.

[0072] Subsequently, the method superimposes the calculated thermal drift response coefficient and frequency domain suppression parameter onto the original composite tilt angle signal, and removes the dynamic vibration component and part of the thermal transient error through mathematical operations, outputting preprocessed tilt angle data with high signal-to-noise ratio.

[0073] After obtaining clean preprocessed data, the method initiates the core static compensation process. The processor calls a zero-drift compensator based on time-series memory. This step uses the temperature and tilt angle historical reference sequence stored locally by the sensor to construct a state vector. The method runs a recursive polynomial fitting network algorithm to predict the current creep error component (i.e., zero-drift value) of the sensor in real time based on the aging trend over a period of time and the current temperature state. Then, the method performs reference correction, subtracting the real-time generated zero-drift correction value from the preprocessed tilt angle data, thereby eliminating the systematic deviation caused by the aging of the encapsulation material or the creep of the mounting base, and obtaining the final calibrated tilt angle data that is infinitely close to the physical truth.

[0074] After obtaining the final data, the method executes low-power decision logic. The measurement state management unit compares the amplitude of the final calibrated tilt angle data with a preset measurement range limit (e.g., the safety warning angle for bridges). If the data amplitude does not exceed the limit, the system determines that the current structure is in a safe state and only stores the data in a locally cyclically overlaid low-power storage area without triggering remote transmission; if the data amplitude exceeds the limit, the system immediately generates an over-range status flag, which is marked as a high-priority event.

[0075] Once an over-range status indicator is detected, the method immediately wakes up the dormant wireless communication module. In response to the indicator, the system establishes a communication link with the remote monitoring center and uploads the current measurement data. This data packet contains not only the final calibrated tilt angle data that triggers the alarm, but also pre-processed tilt angle data for fault tracing. After data transmission is complete, the system automatically shuts down the communication RF circuit and re-enters the low-power monitoring cycle.

[0076] This invention establishes a standardized intelligent workflow from physical environment perception to deep data cleaning through a multi-dimensional physical quantity synchronous acquisition and hierarchical solution process. The method first runs signal conditioning logic, effectively eliminating dynamic interference at the front end of signal processing using thermal hysteresis and frequency domain characteristics, providing high signal-to-noise ratio clean samples for subsequent processing. Then, it executes a zero-point drift compensation step based on time-series memory, using historical data streams to reconstruct the sensor's aging trend online, achieving nonlinear approximation and real-time correction of long-term creep errors. Combined with an event-driven reporting strategy based on over-range identification, this method constructs a highly efficient closed loop of "perception-cleaning-decision-transmission" at the software algorithm level. This achieves both extremely high reliability of monitoring data and extremely low system power consumption within limited embedded computing resources, making it particularly suitable for automated monitoring scenarios with stringent requirements for data timeliness and accuracy.

[0077] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be determined by the scope of protection defined in the claims.

Claims

1. A low-power, high-precision, and high-sensitivity tilt angle intelligent sensing system, characterized in that, include: The multi-sensor module simultaneously collects MEMS core temperature, package shell temperature, environmental acceleration interference signals, and the original composite tilt angle signal generated by superimposing the true static tilt angle, thermal drift error component, creep error component, and dynamic vibration noise. The tilt angle signal calculation and error compensation unit operates signal conditioning logic based on MEMS core temperature, package shell temperature, and environmental acceleration interference signal to remove dynamic vibration noise in the composite tilt angle signal and output preprocessed tilt angle data. Based on the preprocessed tilt angle data and the historical temperature reference sequence, a nonlinear mapping correction relationship between the creep error component and the historical temperature reference sequence is established using a zero-point drift compensator based on time-series memory. Real-time zero-point correction values ​​are generated, and the real-time zero-point correction values ​​are used to perform reference correction with the preprocessed tilt angle data to obtain the final calibrated tilt angle data. The measurement status management unit generates an over-range status flag when the amplitude of the final calibration tilt angle data exceeds the measurement range limit. The wireless communication module responds to the over-range status flag and performs measurement data reporting.

2. The low-power, high-precision, and high-sensitivity tilt angle intelligent sensing system according to claim 1, characterized in that, The tilt signal calculation and error compensation unit is specifically configured as follows when running the signal conditioning logic: The transient temperature difference between the MEMS core temperature and the packaging shell temperature is calculated in real time, and the transient temperature difference is input into a preset thermal hysteresis transfer function. The thermal hysteresis transfer function characterizes the thermal drift delay characteristics caused by the temperature gradient based on the principle of inertial links, and outputs the sensor thermal drift response coefficient. The tilt signal solution and error compensation unit also configures frequency domain stopband suppression logic based on the frequency domain characteristics of the environmental acceleration interference signal. The signal conditioning logic is configured to comprehensively superimpose the sensor thermal drift response coefficient and the suppression parameters output by the frequency domain stopband suppression logic to perform physical error elimination on the composite tilt signal.

3. The low-power, high-precision, and high-sensitivity tilt angle intelligent sensing system according to claim 2, characterized in that, When configuring the frequency domain stopband suppression logic, the tilt signal calculation and error compensation unit is specifically configured as follows: Spectral analysis is performed on the environmental acceleration interference signal to extract the main interference frequency point and the secondary interference frequency point of the interference gravity vector measurement. Using the main interference frequency point and the secondary interference frequency point, a set of adaptive notch filters is configured. The adaptive notch filters dynamically lock the center notch frequency on the main interference frequency point and the secondary interference frequency point, thereby attenuating the signal amplitude of the dynamic vibration noise in the frequency domain.

4. The low-power, high-precision, and high-sensitivity tilt angle intelligent sensing system according to claim 1, characterized in that, When configuring the zero-position drift compensator based on time-series memory, the tilt signal calculation and error compensation unit is specifically configured as follows: Based on the preprocessed tilt angle data, the temperature history reference sequence of the MEMS core temperature and the packaging shell temperature, a recursive polynomial fitting network is constructed; the recursive polynomial fitting network uses the temperature history reference sequence as the input state vector and the zero-point drift value of the preprocessed tilt angle data in the quasi-static calibration interval as the fitting convergence target. The creep error components are approximated using a nonlinear curve through the recursive polynomial fitting network.

5. The low-power, high-precision, and high-sensitivity tilt angle intelligent sensing system according to claim 1, characterized in that, The measurement status management unit is also configured to: The rate of change of the internal correction coefficient of the tilt angle signal calculation and error compensation unit is monitored, and the instrument calibration cycle is dynamically set in combination with the current power supply; when the instrument calibration cycle is reached, an online calibration request is generated.

6. The low-power, high-precision, and high-sensitivity tilt angle intelligent sensing system according to claim 5, characterized in that, The wireless communication module is also configured to: In response to the online calibration request, the preprocessed tilt angle data, the final calibrated tilt angle data, and a set of instrument operating parameters for remote calibration output by the tilt angle signal calculation and error compensation unit are reported simultaneously. The instrument operating parameter set specifically includes: the current coefficients of the thermal hysteresis transfer function, the locking frequency of the adaptive notch filter, and the current weighting coefficient matrix of the recursive polynomial fitting network.

7. The low-power, high-precision, and high-sensitivity tilt angle intelligent sensing system according to claim 6, characterized in that, It also includes a remote precision calibration module, which is configured as follows: The system receives measurement data reported by the wireless communication module; based on the received data and combined with the physical reference simulation model of the structure under test, it calculates the theoretical true value of the creep error component using a global error inversion algorithm. The global error inversion algorithm integrates a numerical solution algorithm for partial differential equations based on the laws of materials mechanics and thermodynamics; based on the deviation analysis between the theoretical true value and the instrument operating parameter set, a set of calibration correction constants is generated; The calibration correction constant is transmitted via the wireless communication module through remote configuration.

8. The low-power, high-precision, and high-sensitivity tilt angle intelligent sensing system according to claim 7, characterized in that, The tilt angle signal calculation and error compensation unit is further configured to: Based on the calibration correction constant issued by the remote precision calibration module, the coefficient injection update is performed on the zero-point drift compensator based on time-series memory. Specifically, the update involves applying the calibration correction constant to fine-tune the internal weighting coefficients of the recursive polynomial fitting network to eliminate the long-term zero-point drift of the multivariate sensing module.

9. A low-power, high-precision, and high-sensitivity tilt angle intelligent sensing method, characterized in that, include: The system simultaneously acquires MEMS core temperature, package shell temperature, environmental acceleration interference signals, and the original composite tilt angle signal generated by superimposing the true static tilt angle, thermal drift error component, creep error component, and dynamic vibration noise. Based on the MEMS core temperature, package shell temperature, and environmental acceleration interference signal operation signal conditioning logic, dynamic vibration noise in the composite tilt angle signal is removed, and pre-processed tilt angle data is output. Based on the preprocessed tilt angle data and the historical temperature reference sequence, a nonlinear mapping correction relationship between the creep error component and the historical temperature reference sequence is established using a zero-point drift compensator based on time-series memory. Real-time zero-point correction values ​​are generated, and the real-time zero-point correction values ​​are used to perform reference correction with the preprocessed tilt angle data to obtain the final calibrated tilt angle data. An over-range status flag is generated when the amplitude of the final calibration tilt angle data exceeds the measurement range limit. In response to the over-range status flag, the measurement data is reported.

Images