System and method for controlling MEMS micromirror array

Abstract

The invention relates to the technical field of MEMS (Micro Electro Mechanical System) control, and discloses a system and a method for controlling an MEMS (Micro Electro Mechanical System) micromirror array. The method comprises the following steps: establishing a space-time characteristic evaluation model containing an energy flow analysis domain and a mechanical response domain, synchronously acquiring feed-in and load energy waveform characteristics of a system, and obtaining stress distribution data of a driving mechanism and real-time attitude data of a reflecting surface of each micro-mirror in parallel; and the data are fused to generate a multi-dimensional feature vector, and the multi-dimensional feature vector is compared with a preset stability boundary condition, so that a targeted regulation and control strategy is output. According to the method, multi-dimensional state sensing from energy transfer efficiency to internal stress of a mechanical structure is realized, the hysteresis quality caused by the fact that a traditional method only depends on motion feedback is overcome, the performance degradation trend and potential mechanical damage can be recognized in advance, a control system is enabled to turn from passive error correction to active early warning and health management, and the reliability of the system is improved. And the working stability and adaptability of the micromirror array are improved, and the service life of the device is prolonged.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of micro-electro-mechanical system control, in particular to a system and method for controlling a MEMS micromirror array. BACKGROUND

[0002] As a key executive component of optical communication, laser radar and other systems, the dynamic performance and long-term reliability of the MEMS micromirror array directly affect the overall system performance. At present, the regulation and control of such devices is mainly based on a simple "electric signal input-mirror angle feedback" control mode. This method combines the pre-set driving voltage or current waveform with the monitoring of the actual deflection angle of the mirror by an optical sensor to form a closed-loop control. The technical core lies in how to quickly eliminate the angle deviation by adjusting the electrical parameters.

[0003] This traditional regulation and control method has some shortcomings. The system regards the micromirror array as a black box, only focusing on the external electrical characteristics and the final motion result, and cannot understand the internal dynamic process. The control system lacks monitoring of energy transfer efficiency. There are multiple loss paths in the process of converting input electrical energy into mirror mechanical energy. When there is a response anomaly or mechanical characteristic drift in the array, the energy loss pattern will change. The existing technology cannot identify these early signs of performance degradation because it cannot simultaneously compare the input energy and the effective load energy, and can only passively compensate after the motion error appears.

[0004] The existing method completely ignores real-time evaluation of the mechanical load of the driving mechanism. When the micromirror moves at high frequency, its cantilever beam, torsion bar and other structures are subjected to periodic stress, which is prone to fatigue accumulation. Traditional control only focuses on whether the mirror is in place, but has no knowledge of the stress distribution inside the driving mechanism. This lack of information on the key mechanical state has prevented the system from implementing protective regulation at the early stage of structural damage, limiting the service life of the array and increasing the risk of sudden failure under high load conditions. The existing regulation and control technology needs to be broken through from a deeper state perception dimension. SUMMARY

[0005] The purpose of the present application is to provide a system and method for controlling a MEMS micromirror array to solve the problems raised in the background art.

[0006] To achieve the above purpose, the present application provides a method for controlling a MEMS micromirror array, comprising: establishing a space-time characteristic evaluation model for the MEMS micromirror array, which includes an energy flow analysis domain and a mechanical response domain; in the energy flow analysis domain, synchronously collecting the input energy waveform features and load energy waveform features of the MEMS micromirror array; In the mechanical response domain, stress distribution data of each micro-mirror driving mechanism in the MEMS micro-mirror array and real-time attitude data of each micro-mirror reflecting surface are acquired in parallel; The input of the feed-in energy waveform feature, the load energy waveform feature, the stress distribution data and the real-time attitude data into the space-time characteristic evaluation model generates a multi-dimensional feature vector; Based on the comparison of the multi-dimensional feature vector with the preset stability boundary condition, the regulation and control strategy of the MEMS micro-mirror array is output.

[0007] Preferably, the step of establishing a space-time characteristic evaluation model specifically includes: The data sampling period of the energy flow analysis domain and the data sampling period of the mechanical response domain are defined and kept synchronized; In the energy flow analysis domain, an input sensor network for monitoring the feed-in energy waveform feature and an output sensor network for monitoring the load energy waveform feature are set up; In the mechanical response domain, a micro-strain gauge array for measuring the stress distribution data and an optical distance measuring unit for capturing the real-time attitude data are deployed; The data fusion core of the space-time characteristic evaluation model is configured to receive and time-align asynchronous data streams from the input sensor network, the output sensor network, the micro-strain gauge array and the optical distance measuring unit.

[0008] Preferably, the step of synchronously collecting the feed-in energy waveform feature and the load energy waveform feature includes: The voltage transient value and the current transient value in the power supply circuit of the MEMS micro-mirror array are continuously recorded through the input sensor network; The voltage transient value and the current transient value in the load circuit of the MEMS micro-mirror array are continuously recorded through the output sensor network; The recorded power supply circuit voltage transient value and current transient value are subjected to convolution operation to extract the feed-in energy waveform feature; The recorded load circuit voltage transient value and current transient value are subjected to convolution operation to extract the load energy waveform feature.

[0009] Preferably, the step of acquiring the stress distribution data and the real-time attitude data in parallel includes: The surface strain of each micro-mirror driving mechanism at different driving phases is measured by the micro-strain gauge array in a spatial interpolation manner; The surface strain data are converted into the stress distribution data; The measuring light beam is emitted to each micro-mirror reflecting surface by the optical distance measuring unit, and the phase shift of the reflected light beam is received; According to the phase offset, a tilt angle and an offset of each micro-mirror reflecting surface relative to a reference position are calculated to form the real-time attitude data.

[0010] Preferably, the step of generating a multi-dimensional feature vector comprises: Performing frequency domain decomposition on the input energy waveform feature to obtain energy input spectrum components; Performing frequency domain decomposition on the load energy waveform feature to obtain energy output spectrum components; Performing principal component analysis on the stress distribution data to extract key stress modes; Performing cluster analysis on the real-time attitude data to identify a coordinated motion mode of a micro-mirror group; Combining the energy input spectrum components, the energy output spectrum components, the key stress modes and the coordinated motion mode into the multi-dimensional feature vector.

[0011] Preferably, the step of comparing the multi-dimensional feature vector with stability boundary conditions comprises: Mapping the multi-dimensional feature vector into a pre-labeled stability atlas; Calculating Mahalanobis distances between the multi-dimensional feature vector and center feature vectors of each stable region in the stability atlas; Determining a target stable region with the minimum Mahalanobis distance to the multi-dimensional feature vector; Querying a historical control parameter record associated with the target stable region; Generating the current control strategy according to the historical control parameter record; The step of mapping the multi-dimensional feature vector into a pre-labeled stability atlas comprises: Loading the pre-labeled stability atlas, which is constructed from historical experimental data and contains multiple stable regions, each defined by a center feature vector and a covariance matrix; Performing standardization processing on the multi-dimensional feature vector to make it have the same scale as the feature vectors in the stability atlas; Calculating Mahalanobis distances between the standardized multi-dimensional feature vector and center feature vectors of each stable region, using the covariance matrix of the corresponding stable region as the weight; Comparing the values of all Mahalanobis distances to identify the stable region corresponding to the smallest Mahalanobis distance as the target stable region.

[0012] Preferably, the step of outputting a control strategy is followed by: Executing the control strategy to adjust the driving signal parameters applied to the MEMS micro-mirror array; After the adjustment is completed, a new round of the feed-in energy waveform features, the load energy waveform features, the stress distribution data and the real-time attitude data are re-acquired; The re-acquired data is input into the space-time characteristic evaluation model to generate a verification multi-dimensional feature vector; Whether the verification multi-dimensional feature vector falls into the target stable region is verified; If not, the adjustment and control strategy is iteratively corrected until the verification multi-dimensional feature vector falls into the target stable region.

[0013] Preferably, the step of generating the current adjustment and control strategy according to the historical adjustment and control parameter records comprises: All historical adjustment and control parameter records associated with the target stable region are retrieved from the storage system, each record containing a historical adjustment and control parameter and its corresponding system performance indicator; The performance indicator weighted score of each historical adjustment and control parameter record is calculated, the weighted score being determined based on the importance weight of the performance indicator and the historical use frequency; The historical adjustment and control parameter record with the highest weighted score is selected as the preferred record; The historical adjustment and control parameter in the preferred record is extracted, and parameter fine-tuning is performed in combination with the current running state of the MEMS micromirror array to generate the current adjustment and control strategy, the adjustment and control strategy including amplitude, frequency and phase adjustment instructions of the driving signal.

[0014] Preferably, the adjustment and control strategy specifically includes a set of instructions for cooperatively correcting the driving waveform shape, driving frequency spectrum and driving phase difference between micromirrors of the MEMS micromirror array.

[0015] Preferably, the present application further comprises a system for controlling a MEMS micromirror array, the system comprising a memory, a processor and a computer program stored in the memory and running on the processor, the processor implementing the steps of a method for adjusting and controlling a MEMS micromirror array when executing the computer program.

[0016] Compared with the prior art, the present application has the following beneficial effects: Synchronous acquisition of feed-in and load energy waveform features realizes dynamic monitoring of system energy transmission efficiency. By accurately comparing the dynamic characteristics of input and output energy, abnormal energy consumption caused by micro-performance changes can be sensitively captured. This technology can identify early signals indicating unit failure or efficiency decay before visible deviations in mirror movement occur, enabling adjustment and control behavior to change from lagging correction to proactive intervention based on energy efficiency evaluation, thereby improving system stability.

[0017] The stress distribution of the parallel acquisition driving mechanism and the real-time attitude data of the mirror surface are acquired, and the direct correlation between the mechanical internal state and the motion performance is established. The technology realizes real-time quantification of the microscopic strain of the driving structure, and can accurately evaluate the mechanical load state by analyzing the stress response under a specific motion instruction. When abnormal drift of the stress required to maintain the attitude is detected, the change in structural stiffness or potential damage can be judged in time, so that protective adjustment is implemented based on the mechanical health state, thereby prolonging the service life of the device.

[0018] The energy flow characteristics and mechanical response data are fused to generate a multi-dimensional feature vector, and a unified state evaluation framework is constructed. The technology makes the control strategy a multi-objective comprehensive decision of energy efficiency optimization and mechanical protection. The system can intelligently diagnose whether the problem source is energy efficiency anomaly or mechanical fatigue according to the specific mode of the feature vector deviating from the stability boundary, and output targeted instructions to realize the coordinated improvement of system performance and reliability under complex working conditions. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 The working principle diagram of the method for controlling the MEMS micromirror array is described. Figure 2 The flowchart for establishing the space-time characteristic evaluation model is described. Figure 3 The flowchart for synchronously collecting the waveform characteristics of the input energy and the load energy is described. Figure 4 The three-dimensional visualization diagram of the stress distribution is described. Figure 5 The micromirror attitude change diagram is described. DETAILED DESCRIPTION

[0020] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0021] Please refer to Figure 1The application provides a method for controlling a MEMS micromirror array, which comprises: an energy flow analysis domain focusing on monitoring the energy input and output dynamics of the MEMS micromirror array, capturing the features of the input energy waveform and the features of the load energy waveform in real time through a high-precision sensor network, which reflect the energy transmission efficiency of the driving signal and the load matching state; a mechanical response domain aiming at the mechanical structure behavior of the micromirror, collecting the stress distribution data of each micromirror driving mechanism and the real-time attitude data of each micromirror reflecting surface in parallel through a distributed sensing unit, so as to comprehensively evaluate the mechanical stability and motion consistency of the micromirror array; a space-time characteristic evaluation model integrating the energy flow and mechanical response data through a data fusion technology, generating a multi-dimensional feature vector which comprehensively reflects the key parameters of the energy dynamics and the mechanical state; and a comparison between the multi-dimensional feature vector and preset stability boundary conditions, the stability boundary conditions being previously calibrated based on historical experimental data and defining the parameter range of the safe and stable operation of the micromirror array. The comparison process involves calculating the similarity between the feature vector and the stability region, and finally outputting a control strategy, which includes adjusting the waveform, frequency or phase of the driving signal and the like to optimize the performance of the micromirror array.

[0022] Embodiment 1: refer to Figure 2 In specific implementation, defining the data sampling period of the energy flow analysis domain and the data sampling period of the mechanical response domain is an initial step, the data sampling period of the energy flow analysis domain is set to be in the order of microseconds, for example, 1 microsecond, to adapt to high-frequency energy fluctuations, and the data sampling period of the mechanical response domain is set to be in the order of milliseconds, for example, 1 millisecond, to match the mechanical inertia, the two sampling periods are synchronized through a global clock signal and a phase-locked loop, and the phase-locked loop generates a synchronization pulse to ensure that the sampling events are aligned in time. In some embodiments, the data sampling period of the energy flow analysis domain can be dynamically adjusted according to the array operating frequency, for example, automatically shortening the sampling period when the driving signal frequency is increased, and the data sampling period of the mechanical response domain is fixed to be in the order of milliseconds according to the micromirror response delay, the synchronization mechanism uses a digital phase-locked loop circuit, the phase-locked loop reference clock source is a crystal oscillator, and the output synchronization signal is distributed to each sampling module. Optionally, the sampling period definition process includes a calibration stage, which verifies the sampling synchronization accuracy by injecting test signals to ensure the time consistency of the energy flow and mechanical response data.

[0023] In a specific implementation, input sensor network and output sensor network are arranged in the energy flow analysis domain, the input sensor network is deployed in the power supply loop of the MEMS micromirror array, the input sensor network includes high-bandwidth differential voltage probes and Hall effect current sensors, the voltage probes are connected to the positive and negative terminals of the power supply loop, and the current sensors are installed around the power supply wire to continuously monitor voltage transients and current transients in a parallel manner. The output sensor network is similarly installed in the load loop, which is located at the output end of the micromirror driving circuit, and the output sensor network adds an opto-isolator to prevent load feedback interference, and the sensor data is transmitted to the processing unit through a high-speed serial interface such as SPI or I2C. In some embodiments, the layout of the input sensor network takes into account circuit impedance matching, the voltage probes use a high input impedance design to reduce load effects, and the current sensors use a Rogowski coil to improve frequency response characteristics, and all sensor nodes are connected through a bus topology to achieve centralized data collection.

[0024] In a specific implementation, a micro-strain gauge array and an optical distance measuring unit are deployed in the mechanical response domain, the micro-strain gauge array is attached to the surface of each micromirror driving mechanism in a grid form, the micro-strain gauge array uses a foil-type resistance strain gauge, and the spatial arrangement of the strain gauge is based on the geometric shape of the driving mechanism, for example, a uniform grid is used for a square mechanism, and the grid spacing is in microns to achieve spatial interpolation measurement. The optical distance measuring unit is integrated around the array, and the optical distance measuring unit includes a laser emitter and an interferometer detector, the laser emitter generates a coherent light beam, the light beam is guided to the micromirror reflecting surface through a beam splitter, and the reflected light beam is received by the detector and analyzed for phase shift. Optionally, the installation of the micro-strain gauge array uses an epoxy adhesive with temperature stability to reduce thermal drift, and the optical path calibration of the optical distance measuring unit is completed through an autofocus mechanism to ensure that the light beam is perpendicular to the reflecting surface.

[0025] In a specific implementation, a data fusion core configured to evaluate the spatio-temporal characteristics model is implemented based on a field programmable gate array hardware, the data fusion core receives asynchronous data streams from the input sensor network, the output sensor network, the micro-strain gauge array, and the optical distance measuring unit, the asynchronous data streams are time-stamped, and the time-stamps are generated by a global clock. The data fusion core performs a time alignment algorithm, the alignment algorithm uses an interpolation method to compensate for sampling delays, for example, for the time difference between energy flow data and mechanical response data, a linear interpolation formula is applied to calculate the intermediate value, and the interpolation formula is expressed as:

[0026] wherein: represents the time after alignment, and are the original values of adjacent sampling points and , The target alignment time point is aligned. The data fusion core has a built-in buffer to store temporary data, and the logic unit performs interpolation operations to output a synchronized data set for subsequent feature generation.

[0027] Embodiment 2: refer to Figure 3 In specific implementations, the voltage transient value and the current transient value in the power supply circuit of the MEMS micromirror array are continuously recorded by the input sensor network, which includes a high-speed differential amplifier connected to the positive and negative terminals of the power supply circuit and an analog-to-digital converter with a sampling rate set to 10 million times per second to capture microsecond-level transients. The voltage transient value is digitized by the analog-to-digital converter after being adjusted by the amplifier gain, and the current transient value is synchronously sampled after being converted to a voltage signal by a Rogowski coil. In some embodiments, the recording process of the input sensor network uses a ping-pong buffer structure, with two storage areas working alternately, one storing data while the other transmitting data to the processor, ensuring continuous recording without interruption. It can be understood that the recording of the voltage transient value and the current transient value in the power supply circuit needs to consider common-mode noise suppression, and the differential amplifier is configured with a common-mode rejection ratio greater than 80 decibels, and the analog-to-digital converter has a resolution of not less than 16 bits to ensure signal accuracy.

[0028] In specific implementations, the voltage transient value and the current transient value in the load circuit of the MEMS micromirror array are continuously recorded by the output sensor network, which is installed on the load circuit between the drive circuit and the micromirror array. The load circuit voltage transient value measurement uses a high-impedance probe to reduce the load effect, and the load circuit current transient value measurement uses a current transformer to isolate the high-voltage side. The sampling clock of the output sensor network is synchronized with the input sensor network, and the sampling clock is distributed by the same clock source. The load circuit data recording is simultaneously subjected to digital filtering processing, and the digital filter uses a finite impulse response filter to eliminate high-frequency noise. Optionally, the load circuit recording increases the oversampling technology, with an oversampling rate set to more than 4 times, followed by downsampling processing to improve the signal-to-noise ratio. It can be understood that the recording of the load circuit voltage transient value and the current transient value needs to be strictly synchronized with the power supply circuit data, with a synchronization error controlled within the nanosecond level.

[0029] In specific implementations, the recorded load loop voltage transient values and current transient values are convoluted to extract load energy waveform features, the load loop convolution uses the same kernel function as the power supply loop but is processed independently, and the convolution results contain the time-frequency characteristics of the load energy waveform. The load energy waveform feature extraction process increases the calibration coefficient compensation, which is determined by the previous calibration experiment, and is used to eliminate the sensor system error. Optionally, the load loop convolution is followed by normalization processing, which scales the energy value to a standard range for subsequent comparison. It can be understood that the load energy waveform features reflect the actual energy consumption state of the micro-mirror array, and together with the feed-in energy waveform features, they form a complete data set for energy flow analysis.

[0030] In specific implementations, the surface strain data of each micro-mirror driving mechanism at different driving phases is measured by a spatial interpolation method using a micro-strain gauge array, the micro-strain gauge array uses a foil resistance strain gauge attached to the surface of the driving mechanism in a grid form, and the grid spacing is 50 microns to achieve spatial interpolation measurement. The spatial interpolation algorithm uses a bilinear interpolation method to calculate the strain value of any point in the region according to the readings of the adjacent four strain gauges, and the surface strain data is converted into a voltage signal through a Wheatstone bridge circuit, which is then amplified and sampled by an analog-to-digital converter. In some embodiments, the surface strain measurement at different driving phases is synchronized with the driving signal, and the driving phase signal is provided by the controller, and the measurement data of the micro-strain gauge array is stored in groups according to the phase angle. It can be understood that temperature compensation needs to be considered in surface strain data measurement, and the micro-strain gauge array integrates a temperature sensor for real-time temperature drift correction of the measurement data.

[0031] In specific implementations, the surface strain data is converted into stress distribution data, the conversion process is based on the Hooke's Law formula, and the stress distribution data calculation uses the material Young's modulus parameter, and for silicon-based MEMS material, the Young's modulus value is 169 Giga Pascals. The mapping relationship between the surface strain data and the stress distribution data is a linear transformation, and the stress value of each measurement point is obtained by multiplying the strain value by the Young's modulus. Optionally, the stress distribution data post-processing uses a smoothing filter algorithm, and the smoothing filter uses a Gaussian kernel convolution to reduce noise effects. The optical distance measuring unit emits a measurement beam to each micro-mirror reflecting surface and receives the phase shift of the reflected beam, the optical distance measuring unit uses a laser interferometer system, the measurement beam is a helium-neon laser with a wavelength of 632.8 nanometers, and the beam diameter is focused to 5 microns. The phase shift of the reflected beam is received by a photodetector, and the phase demodulation uses a heterodyne interference technology, and the phase shift amount and the micro-mirror displacement relationship are calculated by the optical path difference. In some embodiments, the measurement frequency of the optical distance measuring unit is 10 kilohertz, each micro-mirror reflecting surface is configured with an independent measurement light path, and the phase shift data is transmitted to the processing unit in real time. It can be understood that the real-time attitude data contains tilt angle and offset information, and the tilt angle is calculated by differentiating the phase shift of different points on the reflecting surface.

[0032] In a specific implementation, the real-time attitude data is formed by calculating the tilt angle and offset of each micro-mirror reflecting surface relative to a reference position according to the phase shift, the reference position is determined by a calibration process which records the initial phase value in a zero driving state. The tilt angle calculation uses geometric optics formula, the offset is derived from the relationship between the phase shift and the wavelength, the real-time attitude data is stored in an array form containing the pitch angle and yaw angle of each micro-mirror. Optionally, the real-time attitude data verification uses a redundant measurement mechanism, the comparison results of two measurement light paths arranged on the same micro-mirror reflecting surface. It can be understood that the real-time attitude data accurately reflects the motion state of the micro-mirror and is the basis for cooperative motion pattern recognition.

[0033] In a specific implementation, the energy input spectral component is obtained by frequency domain decomposition on the input energy waveform feature, the frequency domain decomposition uses fast Fourier transform algorithm, the fast Fourier transform point number is 1024 points, and the sampling frequency matches the sampling rate of the energy waveform feature. The energy input spectral component contains amplitude spectrum and phase spectrum information, and the amplitude and phase of the main frequency component are extracted as a feature vector. In some embodiments, the input energy waveform feature is windowed before frequency domain decomposition, and the Hanning window is used to reduce spectral leakage. It can be understood that the energy input spectral component represents the frequency distribution characteristics of the driving energy. The energy output spectral component is obtained by frequency domain decomposition on the load energy waveform feature, the extraction process of the energy output spectral component is consistent with that of the energy input spectral component but is processed independently, and the energy output spectral component focuses on the difference frequency component from the energy input spectral component. The frequency domain decomposition result is logarithmically scaled to enhance the dynamic range, and the amplitude ratio of the main harmonic component is taken as a feature quantity. Optionally, the energy output spectral component analysis increases the impedance matching index calculation, and the impedance matching index reflects the energy transmission efficiency. The key stress modal is extracted by principal component analysis on the stress distribution data, the principal component analysis algorithm calculates the covariance matrix of the stress distribution data, and solves the eigenvalue and eigenvector of the covariance matrix. The key stress modal corresponds to the eigenvector of the maximum eigenvalue, and the principal components with a cumulative contribution rate of more than 85% are retained. The stress distribution data is preprocessed to eliminate the influence of dimension by standardization, and the standardization makes each stress value subtract the mean value and then divide by the standard deviation.

[0034] In a specific implementation, the real-time attitude data is preprocessed to identify the cooperative motion pattern of the micro-mirror group by cluster analysis, the cluster analysis uses the K-means algorithm, and the cluster number is set to 3-5 groups according to the size of the micro-mirror array. The cooperative motion pattern is described by the cluster center, and the center vector of each cluster represents a typical motion pattern. The real-time attitude data is preprocessed to be normalized, and the normalization maps the angle value to the [0, 1] interval. It can be understood that the cooperative motion pattern reflects the motion correlation between the micro-mirrors.

[0035] In specific implementations, the energy input spectral component, the energy output spectral component, the critical stress modal and the concerted motion pattern are combined into a multi-dimensional feature vector, the combination process adopts a feature splicing method, and the dimension of the multi-dimensional feature vector is the sum of the dimensions of each feature. The feature vector is normalized to make the magnitudes of each feature consistent, and the normalization uses the min-max scaling method. The mathematical representation of the multi-dimensional feature vector is:

[0036] wherein: is the multi-dimensional feature vector, is the energy input spectral component vector, is the energy output spectral component vector, is the critical stress modal vector, is the concerted motion pattern vector.

[0037] Referring to Figure 4 , the stress distribution on the surface of the MEMS micromirror driving mechanism is shown. In the figure, the X-axis and Y-axis represent the spatial coordinate position of the driving mechanism surface, and the Z-axis corresponds to the stress value calculated according to Hooke's law. The entire surface uses the thermal map color gradient from dark blue to bright yellow, clearly showing the distribution difference and change trend of stress intensity in space. It can be seen from the figure that the stress forms an obvious concentrated distribution pattern in some areas, which reflects the complex mechanical response characteristics of the micromirror under different driving phases. This three-dimensional visualization method can intuitively reveal the detailed features and overall laws of stress distribution, providing an important visual analysis tool for evaluating the mechanical stability of the micromirror mechanism and predicting potential failure risks.

[0038] In specific implementations, the multi-dimensional feature vector is mapped into a pre-labeled stability atlas, the pre-labeled stability atlas is constructed through historical experimental data, the historical experimental data covers performance records of the MEMS micromirror array under multiple working conditions, the stability atlas contains multiple stable regions, each stable region is defined by a center feature vector and a covariance matrix, the center feature vector represents a set of feature values of an ideal stable state, and the covariance matrix describes the distribution characteristics of data points in the region. The pre-labeled stability atlas is loaded from the non-volatile memory, the atlas data is stored in the form of a matrix, including a region identifier, a center feature vector array and a covariance matrix array, and the loading process uses a high-speed data interface to ensure real-time access. In some embodiments, the construction of the stability atlas is based on a clustering algorithm such as K-means clustering, which groups historical feature vectors to form stable regions, the center feature vector of each region is the mean value within the group, and the covariance matrix calculates the dispersion degree of data within the group. It can be understood that the mapping process is to match the current multi-dimensional feature vector with the reference state in the atlas.

[0039] In a specific implementation, the multi-dimensional feature vector is normalized using the Z-score method, each element of the multi-dimensional feature vector is subtracted by its historical mean and divided by the historical standard deviation, the historical mean and standard deviation are derived from the global data of the stability map. The normalization ensures that the multi-dimensional feature vector has the same scale as the feature vectors in the stability map, avoiding the influence of dimension difference on distance calculation. The normalized multi-dimensional feature vector element value range is adjusted to zero mean and unit variance, and the processing process is completed in real time on the embedded processor. Optionally, the normalization process adds an outlier detection step, which truncates or scales if a feature value exceeds the preset range. It can be understood that the normalization process is a prerequisite for distance calculation.

[0040] In a specific implementation, the Mahalanobis distance between the normalized multi-dimensional feature vector and the center feature vector of each stable region is calculated, and the calculation formula of the Mahalanobis distance is:

[0041] Wherein: represents the Mahalanobis distance, is the normalized multi-dimensional feature vector, is the center feature vector of a certain stable region, is the transpose operation, is the covariance matrix of the stable region, is the inverse matrix of the covariance matrix. The Mahalanobis distance calculation uses matrix operation, first calculates the vector difference , then multiplies by the inverse covariance matrix, and finally takes the square root to get the distance value. The calculation process optimizes numerical stability and avoids matrix singularity problems by using regularization methods to process the covariance matrix.

[0042] In a specific implementation, the target stable region with the smallest Mahalanobis distance from the multi-dimensional feature vector is determined, the calculated Mahalanobis distance values are compared, and the smallest value is found using a sorting algorithm. The stable region identifier corresponding to the smallest value is the target stable region. The selection of the target stable region is based on the minimum distance principle, ensuring that the current state is closest to a certain stable state. The historical control parameter record associated with the target stable region is queried, and the historical control parameter record is stored in the database, each record containing historical control parameters such as drive signal amplitude, frequency and phase, as well as corresponding system performance indicators. Referring to Table 1, the structure of the stability map is shown, including region identifier, center feature vector dimension and covariance matrix size.

[0043] Table 1: Stability map region table

[0044] In specific implementations, the current control strategy is generated from the historical control parameter records, all historical records of the target stable region are extracted from the database, and each record contains parameter values and performance indicators. When generating the strategy, the performance indicator weighted score of the historical record is calculated, the weighted score is based on the indicator importance weight and the historical usage frequency, the record with the highest score is selected as the preferred record, and the parameter fine-tuning is extracted and applied. In some embodiments, the control strategy includes adjustment instructions for the amplitude, frequency and phase of the drive signal, and the instructions are issued to the drive circuit through a digital interface. Optionally, the strategy is verified for its effectiveness after generation, and the performance improvement is confirmed through simulation or real-time testing.

[0045] Referring to Figure 5 , the real-time change trend of the tilt angle of a plurality of micro-mirrors is shown. The horizontal axis represents the time process, and the vertical axis represents the tilt angle value of the micro-mirror. A plurality of curves of different colors respectively represent the attitude change trajectories of different micro-mirrors. These curves show certain periodicity and regularity of change, reflecting the dynamic response process of the micro-mirror under the action of the drive signal. By comparing the change patterns of different curves, the motion coordination and phase relationship between the micro-mirror groups can be observed. This time sequence analysis chart can directly present the motion state and dynamic characteristics of the micro-mirror system, and provides time dimension data support for identifying the coordinated motion mode and evaluating the system stability.

[0046] In specific implementations, all historical control parameter records associated with the target stable region are retrieved from the storage system, the storage system uses a relational database management system, and the database table structure includes a region identifier, a historical control parameter field, and a system performance indicator field. The retrieval operation is performed using structured query language, the query condition is based on the unique identifier of the target stable region, and the result set returned contains multiple historical records. Each historical control parameter record specifically records the drive signal amplitude value, the drive signal frequency value, the drive signal phase value, and the corresponding system performance indicators such as the positioning error value and the power consumption value. In some embodiments, the database query uses a precompiled statement to optimize query efficiency, and the query result is sorted by timestamp to return the latest record. It can be understood that the retrieval of historical control parameter records needs to ensure data integrity and consistency.

[0047] In specific implementations, the performance indicator weighted score of each historical control parameter record is calculated, and the performance indicator weighted score calculation formula is:

[0048] Wherein: represents the weighted score of the jth historical record, is the importance weight of the i-th performance indicator, is the normalized value of the i-th performance indicator in the j-th record, is the total number of performance indicators, is the historical usage frequency of the jth record, is the adjustment coefficient of the usage frequency. The performance indicator importance weight is determined in advance by the analytic hierarchy process, and the historical usage frequency is obtained from the database access log statistics. The calculation process traverses all the retrieved records and generates a quantitative score for each record. Optionally, the performance indicator normalization processing uses the minimum-maximum normalization method to map different dimension indicators to the [0, 1] interval. It can be understood that the weighted score calculation comprehensively considers the performance and usage history.

[0049] In specific implementations, the historical control parameter record with the highest weighted score is selected as the preferred record, and the weighted score values of all records are compared to determine the record corresponding to the maximum value using the bubble sort or quicksort algorithm. The historical control parameters in the preferred record are extracted, including the specific drive signal amplitude value, drive signal frequency value, and drive signal phase value. The extracted parameters are transmitted to the parameter fine-tuning module, which adjusts the parameters in combination with the current MEMS micromirror array operating state such as the ambient temperature value and the device aging coefficient. In some embodiments, the parameter fine-tuning is implemented using a fuzzy logic controller, with the operating state parameters as inputs and the parameter adjustment amount as output. It can be understood that the selection of the preferred record based on quantitative evaluation ensures reliability. The control strategy specifically includes a set of drive signal amplitude adjustment instructions, drive signal frequency adjustment instructions, and drive signal phase adjustment instructions. The modification of the drive signal waveform shape is achieved by changing the lookup table data of the digital waveform synthesizer, the adjustment of the drive frequency spectrum is achieved using a digital filter to modify the output frequency spectrum, and the cooperative correction of the inter-micro-mirror drive phase difference is achieved by adjusting the delay parameters of each channel. The control strategy is encapsulated as a control command frame and issued to the drive circuit through the communication interface. Optionally, a safety check is added during the generation of the control strategy to ensure that the parameter values are within the allowed range. It can be understood that the generation of the control strategy is the core link to achieve closed-loop control.

[0050] In specific implementations, the adjustment of the driving signal parameters applied to the MEMS micromirror array is performed according to the regulation strategy, which changes the output voltage amplitude through a digital-to-analog converter, modifies the output frequency through a frequency synthesizer, and adjusts the phase difference of each channel through a phase delay circuit. The adjustment process adopts a gradual change mode to avoid sudden changes, and the driving signal parameters are gradually transitioned to the target values in specified steps. Immediately after the adjustment is completed, data reacquisition is started, which reacquires the energy waveform features, the load energy waveform features, the stress distribution data, and the real-time attitude data. In some embodiments, the adjustment of the driving signal parameters is synchronized with the data acquisition, and the acquisition time window covers the entire adjustment process. The reacquired data is input into the spatiotemporal characteristic evaluation model to generate a verification multidimensional feature vector, and the generation process of the verification multidimensional feature vector is completely consistent with that of the original multidimensional feature vector, including the same feature extraction and combination steps. The generated verification multidimensional feature vector is compared with the target stable region, and the verification process calculates the Mahalanobis distance between the verification multidimensional feature vector and the center feature vector of the target stable region. If the Mahalanobis distance is less than the region radius threshold, it is determined that the verification multidimensional feature vector falls within the target stable region; otherwise, the iterative correction process is started. Optionally, the verification process sets a maximum number of iterations to prevent infinite loops. It can be understood that the verification link ensures the effectiveness of the regulation strategy.

[0051] In specific implementations, the iterative correction of the regulation strategy is performed until the verification multidimensional feature vector falls within the target stable region. The iterative correction is based on the gradient descent method, which adjusts the regulation parameters according to the deviation direction of the verification result and the target region. After each iteration, the verification multidimensional feature vector is regenerated and reverified until the convergence condition is met or the maximum number of iterations is reached. The adjustment amount of the regulation strategy in the correction process is gradually reduced to achieve fine adjustment. The iterative correction records the parameters and results of each adjustment, which are used for optimization of subsequent regulation.

[0052] Although embodiments of the present application have been shown and described, it is to be understood that various modifications, substitutions, replacements and changes can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A method of regulating a MEMS micromirror array, characterized by, The method comprises: establishing a space-time characteristic evaluation model for the MEMS micromirror array, which contains an energy flow analysis domain and a mechanical response domain; in the energy flow analysis domain, synchronously collecting the feed-in energy waveform features and the load energy waveform features of the MEMS micromirror array; in the mechanical response domain, acquiring the stress distribution data of each micromirror driving mechanism and the real-time attitude data of each micromirror reflecting surface of the MEMS micromirror array in parallel; inputting the feed-in energy waveform features, the load energy waveform features, the stress distribution data and the real-time attitude data into the space-time characteristic evaluation model to generate a multi-dimensional feature vector; comparing the multi-dimensional feature vector with a preset stability boundary condition to output a regulation strategy of the MEMS micromirror array.

2. The method of claim 1, wherein, The step of establishing the space-time characteristic evaluation model specifically comprises: defining the data sampling period of the energy flow analysis domain and the data sampling period of the mechanical response domain, and keeping the two sampling periods synchronized; in the energy flow analysis domain, setting an input sensor network for monitoring the feed-in energy waveform features and an output sensor network for monitoring the load energy waveform features; in the mechanical response domain, deploying a micro-strain gauge array for measuring the stress distribution data and an optical distance measuring unit for capturing the real-time attitude data; configuring a data fusion core of the space-time characteristic evaluation model for receiving and time-aligning asynchronous data streams from the input sensor network, the output sensor network, the micro-strain gauge array and the optical distance measuring unit.

3. The method of claim 2, wherein, The step of synchronously collecting the feed-in energy waveform features and the load energy waveform features comprises: continuously recording the voltage transient value and the current transient value in the power supply circuit of the MEMS micromirror array through the input sensor network; continuously recording the voltage transient value and the current transient value in the load circuit of the MEMS micromirror array through the output sensor network; performing convolution operation on the recorded voltage transient value and current transient value of the power supply circuit to extract the feed-in energy waveform features; performing convolution operation on the recorded voltage transient value and current transient value of the load circuit to extract the load energy waveform features.

4. The method of claim 1, wherein, The step of acquiring the stress distribution data and the real-time attitude data in parallel comprises: measuring the surface strain of each micromirror driving mechanism at different driving phases in a spatial interpolation manner through the micro-strain gauge array; converting the surface strain data into the stress distribution data; emitting a measuring light beam to each micromirror reflecting surface through the optical distance measuring unit, and receiving the phase shift of the reflected light beam; calculating the tilt angle and offset of each micromirror reflecting surface relative to the reference position according to the phase shift to form the real-time attitude data.

5. The method of claim 1, wherein, The step of generating a multi-dimensional feature vector comprises: performing frequency domain decomposition on the feed-in energy waveform features to obtain energy input frequency spectrum components; performing frequency domain decomposition on the load energy waveform features to obtain energy output frequency spectrum components; performing principal component analysis on the stress distribution data to extract key stress modes; performing cluster analysis on the real-time attitude data to identify the cooperative motion mode of the micromirror group; combining the energy input spectrum component, the energy output spectrum component, the critical stress mode and the cooperative motion mode into the multi-dimensional feature vector.

6. The method of claim 5, wherein, The step of comparing the multi-dimensional feature vector with the stability boundary condition comprises: mapping the multi-dimensional feature vector into a pre-labeled stability atlas; calculating Mahalanobis distance between the multi-dimensional feature vector and each stability region center in the stability atlas; determining a target stability region with the minimum Mahalanobis distance to the multi-dimensional feature vector; querying a historical control parameter record associated with the target stability region; generating the current control strategy according to the historical control parameter record; The step of mapping the multi-dimensional feature vector into a pre-labeled stability atlas comprises: loading the pre-labeled stability atlas, which is constructed from historical experimental data and contains multiple stability regions, each defined by a center feature vector and a covariance matrix; normalizing the multi-dimensional feature vector to have the same scale as the feature vectors in the stability atlas; calculating Mahalanobis distance between the normalized multi-dimensional feature vector and each stability region center feature vector, using the covariance matrix of the corresponding stability region as the weight; comparing all Mahalanobis distance values to identify the stability region corresponding to the smallest Mahalanobis distance as the target stability region.

7. The method of claim 6, wherein the MEMS micromirror array is a 2D MEMS micromirror array. The step of outputting the control strategy further comprises: executing the control strategy to adjust the driving signal parameters applied to the MEMS micromirror array; after adjustment, re-collecting a new round of the feed-in energy waveform features, the load energy waveform features, the stress distribution data and the real-time attitude data; inputting the re-collected data into the spatiotemporal characteristic evaluation model to generate a verification multi-dimensional feature vector; verifying whether the verification multi-dimensional feature vector falls within the target stability region; if not, iteratively modifying the control strategy until the verification multi-dimensional feature vector falls within the target stability region.

8. The method of claim 6, wherein the MEMS micromirror array is controlled by a plurality of driving electrodes, and the method further comprises: applying a voltage to the plurality of driving electrodes to control the MEMS micromirror array. The step of generating the current control strategy according to the historical control parameter record comprises: retrieving all historical control parameter records associated with the target stability region from the storage system, each record containing a historical control parameter and its corresponding system performance indicator; calculating a performance indicator weighted score for each historical control parameter record, the weighted score being determined based on the importance weight of the performance indicator and the historical usage frequency; selecting the historical control parameter record with the highest weighted score as the preferred record; extracting the historical control parameter in the preferred record and making parameter fine-tuning based on the current MEMS micromirror array operating state to generate the current control strategy, which includes amplitude, frequency and phase adjustment instructions for the driving signal.

9. The method of claim 1, wherein, The control strategy specifically includes a set of instructions for cooperative correction of the driving waveform shape, driving frequency spectrum and driving phase difference between micromirrors of the MEMS micromirror array.

10. A system for controlling an array of MEMS micromirrors, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, The processor, when executing the computer program, implements the steps of the method for regulating a MEMS micromirror array according to any one of claims 1 to 9.

Images