Automatic calibration method and system suitable for mass production of laser radars

Abstract

The invention provides an automatic calibration method and system suitable for mass production of laser radars, especially quantum laser radars. Comprising the steps of collecting an optical axis deviation angle and a mechanical positioning error of a to-be-calibrated radar, and establishing an initial correlation characteristic set by using a support vector machine; extracting peak delay and amplitude fluctuation characteristics of the electronic response signal, and calculating a photoelectric coupling coefficient through clustering analysis; and generating a radar difference vector in combination with the offset vector, calculating a calibration parameter based on the mode length, and driving an execution mechanism to finely adjust the optical module or write firmware compensation to complete closed-loop calibration. According to the invention, quantitative modeling and soft-hard cooperative compensation of multi-physics field coupling errors are realized, the calibration precision is improved to a millimeter level, the distance measurement deviation of the quantum laser radar caused by temperature drift is effectively solved, and the production efficiency and consistency are both considered.

Description

Technical Field

[0001] This invention relates to the field of radio direction finding and navigation technology, and more particularly to an automated calibration method and system suitable for mass-produced lidar. This invention is especially applicable to production line calibration of quantum lidar, solid-state lidar, and mechanical lidar, which require extremely high assembly precision and signal-to-noise ratio. Background Technology

[0002] With the rapid development of autonomous driving and intelligent sensing technologies, the consistency and precision control of LiDAR, as a core sensor, during mass production has become a key bottleneck restricting the industry's development. In particular, for quantum LiDAR that uses single-photon detection technology (such as single-photon avalanche diodes (SPADs) or silicon photomultiplier tubes (SiPMs), its high sensitivity at the single-photon level makes the requirements for the system's minute mechanical deformation of optical components and the stability of circuit response unprecedentedly high.

[0003] However, in current mass production processes, the multi-physics coupling effect between the optical system and electronic circuits makes it difficult for traditional calibration methods to meet high-precision requirements. Existing technologies typically employ a step-by-step calibration strategy of "adjusting the light first, then the electricity," independently adjusting the mechanical position of optical components and the signal parameters of the electronic system, without considering the dynamic impact of optomechanical structural deviations on the photoelectric detection signal. Minor optical axis offsets and mechanical positioning errors during assembly can cause the echo spot to shift on the photodetector. For quantum lidar, this shift distorts the statistical distribution of photon arrival times, altering the signal rise time characteristics and causing distance delay (walk error) and amplitude fluctuations. Such cross-domain coupling effects make system-level optimization impossible through compensation in a single domain; minute physical deviations are often amplified into significant false alarm noise or measurement drift.

[0004] Furthermore, calibration results under static standard environments cannot reflect performance drift caused by temperature changes or mechanical vibrations in actual applications, leading to measurement inaccuracies in quantum lidar under complex operating conditions. Currently, the industry lacks a unified model that can quantify the correlation between optomechanical errors and electronic responses, and there is no effective mechanism to achieve closed-loop calibration through hardware and software collaboration in mass production environments.

[0005] Therefore, there is an urgent need to provide a calibration method that can be automated, takes into account the effects of multi-physics interactions, and supports real-time feedback adjustment, in order to improve the measurement accuracy and batch consistency of lidar, especially quantum lidar products.

[0006] It should be noted that the information disclosed in the background section above is only used to enhance the understanding of the background of the present invention, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0007] In view of this, the present invention provides an automated calibration method and system suitable for mass-produced lidar, aiming to solve the problems of insufficient calibration accuracy, unpredictable temperature drift, and poor performance consistency between batches caused by the neglect of multi-physics coupling effects between optical systems and electronic circuits in the prior art. By constructing a fully automated calibration process of "data acquisition-feature coupling analysis-closed-loop execution", combined with support vector machine classification, cluster regression modeling and software-hardware collaborative compensation mechanism, the present invention achieves quantitative analysis and closed-loop correction of optical-mechanical-electronic coupling error, significantly improving the ranging accuracy and long-term stability of lidar under complex working conditions, and meeting the stringent requirements of efficiency and consistency for large-scale production.

[0008] This invention provides an automated calibration method suitable for mass-produced lidar, the method comprising: The optical axis deviation angle data and mechanical positioning error data of the lidar to be calibrated on the production line are collected by high-precision sensors; The support vector machine classification algorithm is used to perform multidimensional feature mapping and classification on optical axis deviation angle data and mechanical positioning error data, identify calibration instances with potential coupling deviations, and establish an initial set of correlation characteristics including assembly gap size and offset vector. The electronic circuit response signal output by the example to be calibrated was acquired under a standard test environment, and the signal peak delay characteristics and response amplitude fluctuation characteristics were extracted from the electronic circuit response signal. Cluster analysis is performed on the peak delay characteristics and response amplitude fluctuation characteristics of the signal to calculate the coupling coefficient of the interaction between the quantified optical system and the electronic circuit, and the radar difference vector characterizing the radar integrated error state is generated by combining the offset vector. Determine whether the magnitude of the radar difference vector exceeds the preset tolerance threshold. If so, calculate the calibration parameters based on the coupling coefficient. The calibration parameters include the assembly gap adjustment for the optical module or the signal compensation coefficient for the electronic circuit. The actuator is driven to physically adjust the optical module according to the calibration parameters, or the signal compensation coefficient is written into the firmware storage area of ​​the lidar to be calibrated to complete the closed-loop calibration.

[0009] In some optional embodiments, a support vector machine classification algorithm is used to perform multi-dimensional feature mapping and classification processing on the optical axis deviation angle data and mechanical positioning error data, including: The optical axis deviation angle data and mechanical positioning error data are used to construct a two-dimensional feature input vector; The radial basis function is used to map the two-dimensional feature input vector to a high-dimensional feature space; The distance between the mapped two-dimensional feature input vector and the preset support vector machine hyperplane is calculated, and the lidar to be calibrated is marked as a normal instance, a linear deviation instance, or a nonlinear coupling deviation instance based on the distance.

[0010] In some optional embodiments, extracting signal peak delay characteristics and response amplitude fluctuation characteristics from the electronic circuit response signal includes: Wavelet transform is used to perform multi-level decomposition on the acquired electronic circuit response signal to filter out background noise; Locate the peak time point of the denoised electronic circuit response signal, calculate the time difference between the peak time point and the transmitted trigger signal, and determine the time difference as the signal peak delay feature; A fast Fourier transform is performed on the denoised electronic circuit response signal to calculate the standard deviation of the amplitude distribution in the frequency domain, and the standard deviation of the amplitude distribution is determined as the response amplitude fluctuation characteristic.

[0011] In some alternative embodiments, calculating the coupling coefficient of the interaction between the quantized optical system and the electronic circuitry includes: Normalize the assembly gap size, signal peak delay characteristics, and response amplitude fluctuation characteristics; The K-means clustering algorithm was used to group the normalized data to obtain feature clusters under different error modes; For each feature cluster, the slope of the influence of the optical axis deviation angle change on the response amplitude fluctuation characteristics is fitted by linear regression analysis, and the influence slope is determined as the coupling coefficient.

[0012] In some optional embodiments, the actuator is driven to perform physical displacement adjustment of the optical module according to calibration parameters, including: The assembly gap adjustment is converted into a drive voltage signal for the piezoelectric ceramic actuator. Drive the piezoelectric ceramic actuator connected to the optical module to perform micron-level stepping movement along the optical axis; During movement, the response signals of the electronic circuits are continuously acquired, and the magnitude of the radar difference vector is updated in real time. When the magnitude of the updated radar difference vector reaches its minimum value, the drive is stopped and the current position of the optical module is locked.

[0013] In some optional embodiments, after locking the current position of the optical module, the method further includes an active environmental disturbance testing step, which includes: The temperature of the test environment is controlled to undergo a step change within a preset time window to apply thermal stress disturbance to the lidar to be calibrated. During temperature changes, point cloud coordinate data output by the lidar to be calibrated is continuously collected; The drift gradient of point cloud coordinate data with temperature is calculated. If the drift gradient exceeds the preset thermal stability threshold, a secondary fine-tuning command is generated according to the direction of the drift gradient, and the piezoelectric ceramic actuator is driven again to correct the assembly gap size.

[0014] In some optional embodiments, the active environmental disturbance testing step further includes: The vibration table controlling the test environment applies mechanical vibration to the lidar to be calibrated at a preset frequency; Monitor the signal-to-noise ratio change of the electronic circuit response signal under mechanical vibration. If the signal-to-noise ratio is lower than the preset standard, the vibration frequency of the test environment is adjusted to match the resonance characteristics of the lidar to be calibrated, and the resonance frequency at this time is recorded as calibration reference data.

[0015] In some optional embodiments, writing the signal compensation coefficients into the firmware storage area of ​​the lidar to be calibrated includes: Based on the residual error data after calibration, a multidimensional electronic compensation lookup table is constructed, which establishes the mapping relationship of distance correction values ​​under different operating temperatures and detection distances. The multidimensional electronic compensation lookup table is burned into the non-volatile memory of the lidar to be calibrated via the communication interface; The main control chip of the lidar to be calibrated reads the real-time temperature and measurement distance during operation, and performs real-time data correction according to the multi-dimensional electronic compensation lookup table.

[0016] In some optional embodiments, the method further includes a batch-consistency-based dynamic path planning step: Statistically analyze the mechanical positioning error data of a continuously preset number of lidars to be calibrated, and calculate the consistency variance between batches; If the consistency variance is lower than the first threshold, the current batch is determined to be a high consistency batch, and a simplified calibration path is automatically selected. The simplified calibration path only performs electronic circuit parameter compensation. If the consistency variance is higher than the second threshold, the current batch is determined to be an abnormal fluctuation batch, and a high-intensity calibration path is automatically selected. The high-intensity calibration path includes the physical displacement adjustment and active environmental disturbance test steps of the entire process.

[0017] In some optional embodiments, the dynamic path planning step further includes: When a batch is identified as having abnormal fluctuations, the feature vector with the largest deviation in that batch is extracted. The feature vector is fed back to the next higher level of production to generate tolerance adjustment instructions for the front-end machining equipment.

[0018] In some optional embodiments, the formula for calculating the magnitude of the radar difference vector is: Vector synthesis is performed on the optical axis deviation angular component and the circuit response error component after weighting by the coupling coefficient; The Euclidean norm of the composite vector is calculated as the magnitude of the radar difference vector.

[0019] In some optional embodiments, a step of establishing a closed-loop feedback model is included before calculating the calibration parameters based on the calibration parameters: Collect historical calibration data, which includes the initial set of correlation characteristics before calibration and the radar performance improvement indicators after calibration; A mapping model from an initial set of associated features to optimal calibration parameters is trained using a neural network; The optimal assembly gap adjustment amount is predicted using the trained mapping model, and the optimal assembly gap adjustment amount is used as the initial calibration parameter.

[0020] In some optional embodiments, locking the current position of the optical module specifically includes: After confirming that the magnitude of the radar difference vector has converged, UV-curable adhesive is injected into the joint between the optical module and the housing using a dispensing device. Turn on the ultraviolet light source to irradiate the joint, and complete the adhesive curing while keeping the position of the piezoelectric ceramic actuator unchanged.

[0021] In some alternative embodiments, the electronic circuit response signal includes timestamp data output by a time-to-digital converter and echo waveform digitization data output by an analog-to-digital converter.

[0022] In some alternative embodiments, the method is applied to automated production lines, where high-precision sensors, actuators, and test environment controllers are connected to a central processing unit via an industrial bus to achieve fully automated assembly line operations.

[0023] This invention provides an automated calibration system suitable for mass-produced lidar, the system comprising: A multi-dimensional data acquisition unit is set up at the production line station and is configured to acquire the optical axis deviation angle data and mechanical positioning error data of the lidar to be calibrated, as well as the electronic circuit response signal output by the lidar to be calibrated under a standard test environment. The central processing unit (CPU) is connected in communication with the multidimensional data acquisition unit and is programmed to run the following logical modules: The classification mapping module is used to run the support vector machine algorithm to map the optical axis deviation angle data and mechanical positioning error data to the preset deviation classification space, and generate an initial set of associated features containing assembly gap dimensions and offset vectors. The coupling analysis module is used to extract the peak delay characteristics and response amplitude fluctuation characteristics from the electronic circuit response signal, and calculate the coupling coefficient of the interaction between the optical system and the electronic circuit based on cluster analysis, thereby constructing the radar difference vector. The decision generation module is used to calculate calibration parameters when the magnitude of the radar difference vector exceeds the tolerance threshold. The calibration parameters include physical assembly gap adjustment instructions or firmware signal compensation coefficients. The execution subsystem, controlled by the central processing center, includes: A precision mechanical adjustment mechanism is used to perform displacement operations on the optical module of the lidar to be calibrated in response to physical assembly gap adjustment commands. The firmware programming interface is used to perform data writing operations on the internal memory of the lidar to be calibrated in response to the firmware signal compensation coefficient.

[0024] In some optional embodiments, the precision mechanical adjustment mechanism is an integrated micro-alignment device, which includes: A flexible six-axis gripper, designed for non-destructive clamping of optical modules; A piezoelectric ceramic drive array, mechanically connected to a flexible six-axis gripper, is configured to provide linear displacement drive force with nanometer-scale resolution. The in-situ curing component, integrated into the side of the flexible six-axis holder, includes a directional ultraviolet light source or laser welding head, and is configured to instantly lock the relative position of the optical module and the radar housing after the calibration is confirmed at the central processing computing center.

[0025] In some optional embodiments, the system also integrates an environmental stress screening device, which includes: The rapid temperature change test chamber has a test cavity to accommodate the lidar to be calibrated, and is equipped with a high-power thermocouple and an air circulation system to generate a controlled temperature step environment. The dynamic response monitoring circuit, independent of the internal circuitry of the lidar to be calibrated, is configured to capture the echo intensity drift physical quantity of the lidar to be calibrated in real time during temperature steps and feed the echo intensity drift physical quantity back to the central processing computing center to correct the calibration parameters.

[0026] In some optional embodiments, the multidimensional data acquisition unit includes a physical synchronization trigger: The physical synchronization trigger is electrically connected to the high-precision vision sensor, oscilloscope, and the transmit trigger pin of the lidar to be calibrated in the multi-dimensional data acquisition unit, and is configured to generate a unified time base signal to ensure that the acquired optical morphology data and electronic signal waveform data are strictly aligned on the time axis.

[0027] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit the invention.

[0028] The automated calibration method and system for mass-produced lidar of the present invention has the following beneficial effects: By synchronously acquiring optical axis deviation angle and mechanical positioning error data, and combining feature extraction and coupling analysis of electronic circuit response signals, quantitative modeling of opto-mechanical-electronic multi-physics coupling errors was achieved. The radar difference vector was used to comprehensively characterize the system-level error state, guiding the coordinated calibration of hardware fine-tuning and firmware compensation. This significantly improved the ranging accuracy and point cloud stability of the lidar under mass production conditions, effectively solving the performance dispersion problem caused by accumulated assembly tolerances and ensuring the long-term reliability of the product across the entire temperature range. Attached Figure Description

[0029] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings.

[0030] Figure 1 This is a flowchart of an automated calibration method for mass-produced lidar according to an embodiment of the present invention; Figure 2 This is a schematic diagram of an automated calibration system for mass-produced lidar according to an embodiment of the present invention. Detailed Implementation

[0031] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the invention will be more comprehensive and complete, and will fully convey the concept of the exemplary embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0032] Furthermore, the accompanying drawings are merely illustrative of the invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0033] The flowchart shown in the attached diagram is merely an illustrative example and does not necessarily include all steps. For example, some steps may be broken down, while others may be combined or partially combined. Therefore, the actual execution order may change depending on the specific circumstances.

[0034] In the manufacturing process of lidar, the performance of the optical and electronic systems is not independent but interconnected through their physical assembly state. When there is a slight axial offset or angular deviation in the optical lens, the landing position of the echo spot on the photodetector changes, leading to a change in the rising slope of the photocurrent response, which in turn affects the timing and amplitude stability of the signal peak. This cross-physical domain influence manifests as distance deviation and reflection intensity fluctuations in the ranging results, essentially a nonlinear mapping of mechanical tolerances in the photoelectric conversion process. By performing multi-dimensional spatial modeling of the collected optical error parameters and corresponding electronic response characteristics, the quantitative relationship between the two can be revealed. Using statistical learning methods to cluster the response characteristics under different error modes, sample groups with similar coupling behaviors can be identified, and regression models can be established within each group to extract the slope parameter characterizing the system sensitivity. This parameter reflects the amount of electronic signal change caused by a unit mechanical deviation, providing a basis for assessing the overall error contribution. Weighting the mechanical deviation with this slope and combining it with the circuit response error to form components in the vector space, the magnitude obtained through vector synthesis can quantify the overall deviation of a single lidar instance from the ideal state. When the deviation exceeds the allowable range, it indicates that subsequent signal compensation alone cannot completely correct the impact of the original hardware deviation, and a physical adjustment mechanism needs to be introduced. Conversely, efficient correction can be achieved through firmware-level parameter writing. This approach realizes the transformation from single-dimensional adjustment to system-level collaborative optimization, enabling the calibration process not only to cope with static assembly errors but also to provide data support for performance stability under subsequent dynamic environments.

[0035] like Figure 1 As shown, this embodiment of the invention provides an automated calibration method suitable for mass-produced lidar, which includes the following steps: Step S100: Collect optical and mechanical status data of the lidar to be calibrated.

[0036] In one embodiment, the lidar to be calibrated can be a conventional mechanically rotating lidar or a quantum lidar based on single-photon detection technology. For quantum lidar, since its receiver typically employs a single-photon avalanche diode (SPAD) array that is extremely sensitive to the photon incident angle, even a small mismatch in the optical system will directly lead to a drastic decrease in photon detection efficiency (PDE). Therefore, when acquiring the physical parameters of the lidar to be calibrated in its assembled state using a high-precision sensor system deployed at the production line station, the system must possess sufficient spatial resolution to capture sub-micron-level assembly deviations, thereby accurately identifying subtle geometric errors that may affect the quantum lidar's quantum efficiency and signal-to-noise ratio. Specific parameters include the deviation angle of its optical axis relative to the ideal optical axis and the mechanical positioning errors generated during the installation of the optical components. In other alternative embodiments, multidimensional geometric errors can be simultaneously captured using non-contact optical measurement equipment or multimodal sensing fusion technology.

[0037] Step S200: Identify instances with coupling bias risk based on classification algorithms and establish an initial set of association characteristics.

[0038] In one embodiment, the collected optical axis deviation angle and mechanical positioning error are used as input features. A classification model in machine learning is employed for multi-dimensional feature mapping and pattern recognition to determine whether the current radar is in an assembly state region that may lead to abnormal photoelectric response. For calibration instances identified as having potential nonlinear coupling effects, an initial set of correlation characteristics containing information on their original assembly gap dimensions and spatial offset direction is generated as the basic data structure for subsequent closed-loop analysis. In other optional embodiments, different types of supervised or semi-supervised classifiers can be used to replace this model to adapt to the data distribution characteristics of different production lines.

[0039] Step S300: Acquire the electronic circuit response signal and extract dynamic features.

[0040] In one embodiment, the instance to be calibrated is triggered to emit a probe pulse under a standard test environment, and its output analog echo signal is simultaneously acquired. Indicators reflecting the system's dynamic performance are extracted from this signal, specifically including the time delay for the signal to reach its peak value and the degree of fluctuation in signal amplitude in the time or frequency domain. These characteristics can sensitively reflect back-end electrical signal distortion caused by front-end optical mismatch. In other optional implementations, different combinations of signal acquisition channels can be selected according to the actual hardware configuration, such as parallel acquisition of timestamp sequences and digitized waveforms, to enhance feature characterization capabilities.

[0041] Step S400: Perform cluster analysis to quantify the effects of optical-mechanical-electrical interactions and construct a comprehensive error vector.

[0042] In one embodiment, unsupervised clustering is used to group the signal features of multiple samples, identifying typical pattern clusters with similar "assembly-response" behaviors. For each pattern, a quantitative relationship between optical deviation parameters and electronic response changes is further established to obtain a coupling coefficient describing the strength of their interaction. Combining the previously obtained offset vector information, a radar difference vector is constructed that can uniformly express the degree of deviation between mechanical deviation and circuit response; its mathematical form is a directivity quantity in a multidimensional error space. In other optional embodiments, a weighted fusion strategy or nonlinear modeling method can be introduced to optimize the construction process of this vector, improving its accuracy in representing complex coupling behaviors.

[0043] Step S500: Based on the overall error status, decide whether to initiate the calibration action and calculate the corresponding parameters.

[0044] In one embodiment, the magnitude of the radar difference vector is calculated and compared with a preset tolerance threshold. If the magnitude exceeds the allowable range, it indicates that the current system error cannot be fully compensated by post-processing alone, requiring physical structure adjustment or deep firmware correction. In this case, the required calibration intervention parameters are derived in reverse based on the aforementioned coupling coefficient and error model. This parameter set covers mechanical adjustment commands for adjusting the relative position of the optical modules, or electronic compensation coefficients for correcting ranging results. In other optional embodiments, a predictive model can be trained using historical data to estimate the optimal adjustment path in advance, reducing the number of iterations.

[0045] Step S600: Perform a closed-loop calibration operation to eliminate system bias.

[0046] In one embodiment, based on the calculated calibration parameters, an external actuator is driven to apply precise displacement control to the optical module of the lidar to be calibrated, thereby reducing assembly gaps or correcting optical axis misalignment; alternatively, the generated compensation coefficients are written into the lidar's internal non-volatile memory for the main control chip to recall during runtime. This entire process constitutes a feedback control loop, ensuring that the calibration action directly addresses the root causes of performance degradation. In other alternative embodiments, the actuator may employ precision drive devices based on different principles, such as electromagnetic actuation, voice coil motors, or thermal expansion drive elements, to adapt to diverse product packaging structures.

[0047] The aforementioned technical features form a closely coordinated working mechanism: First, high-precision sensing methods are used to acquire the actual physical state of the lidar during the manufacturing process. Then, an intelligent classification model is used to screen out individuals with strong coupling risks. Subsequently, deep feature mining of electronic response signals reveals how optical assembly errors are transformed into observable circuit performance degradation. Based on this, a cross-domain mapping relationship is established using data-driven methods to quantify the impact of "optical-mechanical" deviations on "electrical" performance, and a unified error evaluation index—the lidar difference vector—is constructed. Finally, differentiated decisions are made based on the severity of this vector, triggering a hardware-software dual-channel calibration mechanism to achieve a complete closed loop from problem identification to root cause management. This multi-level, cross-domain collaborative design effectively overcomes the cumulative error problem caused by the fragmented processing of optical, mechanical, and electrical parameters in traditional calibration methods, and solves technical challenges such as decreased point cloud accuracy, significant temperature drift, and poor batch consistency caused by multi-physics coupling.

[0048] Compared with existing technologies, this embodiment, through the above-described solution, enables fully automated, high-precision system-level calibration of lidar under mass production conditions, significantly improving product accuracy and long-term stability while also taking into account production cycle time and quality control requirements. It is particularly suitable for applications with stringent reliability requirements, such as automotive and industrial applications.

[0049] In one specific implementation, a support vector machine (SVM) classification algorithm is used to perform multi-dimensional feature mapping and classification processing on optical axis deviation angle data and mechanical positioning error data, including: The acquired optical axis deviation angle data and mechanical positioning error data are constructed into a two-dimensional feature input vector. First, the central processing and computing center obtains the optical axis deviation angle of the current lidar to be calibrated from the multi-dimensional data acquisition unit. (Unit: degrees) and mechanical positioning error (Unit: mm), and these two parameters are combined as feature dimensions to form the input vector. Specifically, the vector is fed into a pre-trained Support Vector Machine (SVM) classification model for processing. The SVM model uses a radial basis function (RBF) as its kernel function, and its expression is: ,in The kernel parameters, pre-determined through optimization using historical calibration datasets, are used to control the spatial curvature of the high-dimensional mapping.

[0050] The kernel function is used to nonlinearly map the original two-dimensional feature vector to a high-dimensional feature space, and an optimal classification hyperplane is constructed in this space to distinguish different assembly states.

[0051] The geometric distance from the test sample to the hyperplane in high-dimensional space is calculated, and its category is determined based on the sign and magnitude of the distance: if the distance is within a preset normal range, it is marked as a "normal instance"; if it exceeds the linearly correctable range but does not show a strong nonlinear response trend, it is classified as a "linear deviation instance"; if it falls within a known nonlinear coupling sensitive region, it is determined as a "nonlinear coupling deviation instance," triggering the subsequent deep coupling analysis process. In some other optional implementations, the kernel function can also be replaced with a polynomial kernel or a sigmoid kernel to adapt to the changes in assembly error distribution patterns in different types of production lines; in addition, the classification decision boundary can be dynamically updated through an online learning mechanism, continuously optimizing the classifier parameters using recent calibration results within a sliding window.

[0052] Through the above scheme, this embodiment can achieve refined classification and identification of the initial assembly state of lidar, not only distinguishing between qualified and unqualified products, but also further classifying specific deviation types with potential strong optical-mechanical-electrical coupling risks, thereby providing reliable criteria for the selection of subsequent differentiated calibration strategies and improving the pertinence and resource utilization efficiency of the overall calibration process.

[0053] In one specific implementation, extracting signal peak delay characteristics and response amplitude fluctuation characteristics from the electronic circuit response signal includes: When denoising the acquired electronic circuit response signal, wavelet transform is first used to perform multi-level decomposition on the signal. Specifically, the db4 wavelet basis function is selected to perform four-level discrete wavelet decomposition on the original echo waveform data, decomposing the signal into detail coefficients and approximation coefficients of different frequency bands. Then, based on the signal energy distribution characteristics, soft thresholding is applied to the detail coefficients of each level to suppress high-frequency noise components and retain the effective components that characterize the signal's abrupt changes. Finally, the denoised clean signal waveform is recovered through wavelet reconstruction algorithm.

[0054] The time coordinates corresponding to the point of maximum amplitude on the denoised signal waveform are located. Specifically, the peak position is fitted between sampling points using an interpolation algorithm to improve time resolution. Then, the precise start time of the radar transmission trigger signal is read, the time difference between the two is calculated, and this difference is output as the signal peak delay feature for subsequent error analysis.

[0055] Then, a fast Fourier transform is performed on the same denoised signal waveform to convert it from the time domain to the frequency domain; the amplitude sequence at each frequency point after the transformation is extracted, and the standard deviation of the amplitude sequence is calculated, that is, the square root of the variance is obtained after normalizing its mean. The obtained value is defined as the response amplitude fluctuation characteristic, which reflects the degree of energy dispersion of the signal in the frequency domain.

[0056] In some other alternative implementations, the wavelet basis used in the wavelet transform can be replaced with the sym5 or coif3 wavelet basis, which is suitable for signal denoising scenarios under different signal-to-noise ratio conditions; the number of layers in the multi-layer decomposition can be dynamically adjusted to 3 to 6 layers according to the signal bandwidth; in the frequency domain feature extraction stage, the fluctuation energy ratio in the power spectral density integral interval can also be used instead of the standard deviation calculation to enhance the sensitivity to periodic interference.

[0057] Through the above scheme, this embodiment can effectively improve the feature extraction accuracy in the response signal of electronic circuits. In particular, it can still stably obtain the signal peak delay and amplitude fluctuation characteristics in a strong background noise environment, providing high-fidelity input data for subsequent modeling of opto-mechanical-electrical coupling relationship, thereby enhancing the accuracy and robustness of calibration decisions.

[0058] In one specific implementation, calculating the coupling coefficient of the interaction between the quantized optical system and the electronic circuitry includes: The acquired assembly gap dimensions, signal peak delay characteristics, and response amplitude fluctuation characteristics are first subjected to Z-score normalization to eliminate the interference of different physical dimensions on subsequent analysis. Specifically, the mean of each feature variable in the current batch of samples is subtracted and divided by its standard deviation, so that the processed data follows a distribution characteristic of zero mean and unit variance. Then, the K-means clustering algorithm is used to perform unsupervised grouping on the normalized multidimensional feature dataset. The preset number of clusters K is set to 3 to 5, and the optimal number of clusters is determined by the elbow method, thereby dividing different types of error mode clusters. Each cluster corresponds to a typical "mechanical deviation-circuit response" coupling behavior, such as lens tilt-dominated type, assembly stress release type, or detector sensitive area offset type.

[0059] For each generated feature cluster, an independent linear regression analysis is performed: using the optical axis deviation angle of the samples within that cluster as the independent variable and the response amplitude fluctuation characteristics as the dependent variable, a least-squares linear function curve is fitted. The slope of the regression model is defined as the coupling coefficient under that error mode, used to characterize the intensity of circuit amplitude fluctuations caused by a unit angular deviation. This coupling coefficient, as one of the input parameters for subsequent difference vector construction and calibration decisions, is stored in the local database of the central processing center and associated with the identifier of the corresponding cluster.

[0060] In other alternative implementations, normalization can be replaced by min-max scaling or robust scaling (based on median and interquartile range) to accommodate data distributions with outliers; clustering algorithms can use Gaussian mixture models (GMM) or hierarchical clustering instead of K-means to improve adaptability to non-spherical data distributions; and in the linear regression stage, ridge regression or Lasso regression can be introduced to enhance model stability and suppress overfitting.

[0061] Through the above-described scheme, this embodiment enables refined modeling of the opto-mechanical-electrical coupling effect, effectively identifying and separating different types of assembly defect modes, and providing customized quantitative evaluation indicators for each mode. Compared with existing technologies, this embodiment not only improves the accuracy and robustness of coupling coefficient calculation, but also enhances the calibration system's ability to identify complex nonlinear deviations, thereby supporting more accurate hardware-software co-compensation strategies and significantly reducing the point cloud distortion rate and range drift error of mass-produced lidar.

[0062] In one specific implementation, the actuator is driven to adjust the physical displacement of the optical module according to calibration parameters, including: The calculated assembly gap adjustment is converted into a drive voltage signal for the piezoelectric ceramic actuator. This conversion is based on a pre-calibrated displacement-voltage response function relationship and is corrected by a hysteresis compensation model to ensure accurate and controllable output displacement. Specifically, the central processing computing center calculates the required drive voltage amplitude based on the static sensitivity coefficient of the piezoelectric ceramic material and the historical loading state, and generates the corresponding analog voltage output command. The execution subsystem receives the instruction and drives the piezoelectric ceramic actuator connected to the optical module to perform micron-level stepping movement along the optical axis, with each step increment being less than 0.5μm. During the movement, the multidimensional data acquisition unit continuously acquires the electronic circuit response signal output by the photodetector in real time through an oscilloscope and transmits it synchronously to the central processing computing center. The coupling analysis module re-extracts the peak delay feature and response amplitude fluctuation feature of the newly acquired signal, and updates the mode length of the radar difference vector in combination with the current optomechanical state parameters; the system judges the trend of the change of the mode length, and when it detects that it has reached a local minimum and the rate of change is lower than the preset convergence threshold, it immediately terminates the drive signal output and stops the actuator action; The actuator initiates a position locking mechanism, using a curing device integrated in the clamping structure to physically fix the relative position between the optical module and the housing, thereby completing the mechanical adjustment closed loop.

[0063] In some other alternative implementations, the position locking mechanism can be implemented using UV-curable adhesive, hot melt adhesive, or laser welding; the generation of the drive voltage signal can also introduce feedback closed-loop control, which provides actual displacement feedback through strain sensors or capacitive displacement sensors integrated inside the actuator, further improving the adjustment accuracy.

[0064] Through the above scheme, this embodiment can achieve high-precision dynamic fine-tuning of the optical module. During the adjustment process, the calibration progress is quantified by monitoring changes in the electronic response signal in real time, and the adjustment process is automatically terminated with the goal of minimizing the radar difference vector magnitude. This avoids the uncertainty caused by manual intervention and significantly improves the consistency and reliability of calibration. At the same time, the use of piezoelectric ceramic actuators as driving elements provides nanometer-level resolution and fast response capability, which is suitable for the high-speed and high-precision calibration requirements of automated production lines. In addition, by continuously feeding back electronic signal characteristics in a closed loop during the adjustment process, the mechanical adjustment and electrical performance optimization are directly linked, effectively solving the systematic error problem caused by the coupling of multiple physical fields of opto-mechanical-electricity.

[0065] In one specific implementation, based on the above embodiments, after locking the current position of the optical module, an active environmental disturbance test is further performed to evaluate the stability of the calibration state under non-standard operating conditions.

[0066] First, the control system drives the environmental simulation device to apply a controlled temperature step excitation to the calibrated lidar. Specifically, a high-power thermocouple rapidly raises the ambient temperature of the test chamber from room temperature to a high temperature within a preset time window, introducing thermal stress into the lidar's internal structure. Then, during the dynamic temperature change, the lidar continuously samples the distance of a standard reflective target at a fixed distance, while the central processing center simultaneously receives and records the output point cloud coordinate data sequence. Next, the system aligns the distance measurements in the point cloud data based on timestamps and fits a function curve of the distance reading versus temperature, calculating the average rate of change of this curve over the temperature rise interval as the drift gradient. If this drift gradient exceeds a preset thermal stability threshold, it is determined that the current assembly state exhibits unacceptable geometric deformation or material stress release behavior under thermal load. At this point, the system generates a secondary fine-tuning command based on the positive or negative direction of the drift gradient. This command is interpreted as the displacement direction and amplitude of the piezoelectric ceramic actuator, and the precision mechanical adjustment mechanism is reactivated to apply a small compensating displacement to the optical module along the optical axis to counteract the optical path offset caused by thermal expansion.

[0067] In other alternative implementations, the external excitations applied in the active environmental disturbance test may include periodic vibrations, humidity changes, or electromagnetic interference, and the induced response data may include echo signal-to-noise ratio, time jitter, or intensity fluctuations, to characterize the system robustness under different physical field couplings.

[0068] Through the above scheme, this embodiment can actively excite and quantify the performance degradation trend of lidar under thermal stress conditions after calibration, breaking through the technical limitation that traditional static calibration is only applicable to room temperature conditions; by introducing the drift gradient as a feedback quantity into the closed-loop control process, the dynamic identification and re-correction of potential thermal errors are realized, which significantly improves the long-term stability and measurement consistency of the product in the entire operating temperature range.

[0069] In one specific implementation, the active environmental disturbance testing step further includes: A vibration table controlling the test environment applies mechanical vibration to the lidar to be calibrated, which is in a calibrated state, at a preset frequency. Specifically, the vibration table is driven by an electromagnetic exciter, and its vibration frequency is initially set to a linear sweep frequency signal in the range of 10 Hz to 200 Hz, covering the frequency bands of mechanical interference commonly encountered in vehicle environments, such as engine, road surface excitation, and wind vibration. Simultaneously with the vibration application, the central processing center acquires the echo signal output from the lidar's photodetector in real time using a digital oscilloscope and performs joint time-domain and frequency-domain analysis on the signal.

[0070] The signal-to-noise ratio (SNR) of the echo signal under the current vibration state is calculated, defined as the ratio of the peak power of the effective echo pulse to the root mean square value of the background noise. If the SNR is detected to be lower than a preset standard (e.g., 15 dB), significant vibration sensitivity is determined. Then, the system pauses the current frequency sweep process and enters resonance tracking mode: the control system gradually adjusts the vibration frequency step (0.5 Hz step size) and maintains steady-state excitation for 3 seconds at each frequency point, simultaneously recording the corresponding SNR value. When the SNR reaches a local minimum, it is determined that the frequency point is close to the mechanical resonant frequency of the lidar optical module or internal structure.

[0071] This resonant frequency is recorded as a characteristic parameter in the equipment log and uploaded to the central database, forming a unique "mechanical robustness fingerprint" for each radar. This data can be used to generate subsequent factory test reports or fed back to the product design team to optimize structural stiffness and damping configuration.

[0072] In some other alternative implementations, the vibration table can be replaced by a multi-axis electric vibration platform to apply vibration excitation in composite directions; signal-to-noise ratio monitoring can be achieved in real time at the hardware level through dedicated signal conditioning circuits; the resonant frequency identification algorithm can also employ fast Fourier transform combined with peak search, or a spectrum estimation method based on an autoregressive model to improve frequency resolution.

[0073] Through the above scheme, this embodiment can actively identify the performance degradation critical point of lidar under mechanical vibration and obtain its structural resonance characteristics as calibration reference data. This not only enhances the reliability verification capability of the product in complex dynamic environments, but also provides quantifiable physical basis for subsequent design iterations and quality traceability.

[0074] In one specific implementation, writing the signal compensation coefficients into the firmware storage area of ​​the lidar to be calibrated includes: A multidimensional electronic compensation lookup table is constructed based on the residual error data after calibration. Specifically, the central processing center first collects raw ranging deviation data at standard temperature points and different test distances, forming a two-dimensional or three-dimensional error matrix, where the dimension includes the operating temperature. Original distance measurement value and the corresponding correction amount The lookup table's data structure is organized as a mapping array indexed by temperature and distance. Its correction logic combines the thermal drift coefficient and the nonlinear response residual, expressed as: in It is obtained by fitting multiple sets of static measurement data under standard conditions at the end of the calibration period, and modeling is performed using piecewise linear interpolation or spline functions.

[0075] Then, the generated lookup table is burned into a preset address segment in the Flash memory of the LiDAR main control chip via a UART or JTAG communication interface in binary file format. The burning process includes a verification step: after writing, the system reads back a portion of the data block and compares it with the source file to ensure transmission integrity.

[0076] Configure the data processing module in the LiDAR firmware to call this lookup table in real time during runtime. Whenever a ranging result is received, the main control chip first reads the current ambient temperature value output by the on-chip temperature sensor, and combines it with the original ranging result. Perform bilinear or multidimensional interpolation in the lookup table to obtain the corresponding corrected offset. It outputs the final corrected distance value. .

[0077] In other alternative implementations, the multidimensional electronic compensation lookup table can be extended to include a third-dimensional input parameter, such as the transmit pulse width or target reflectivity, to further improve ranging accuracy in complex scenarios; or, the storage medium of the lookup table can be replaced with EEPROM or other non-volatile memory that supports multiple erase and rewrite operations; or, the compensation logic can be dynamically loaded by an embedded script interpreter rather than being fixed in the firmware code.

[0078] Through the above solution, this embodiment can achieve high-precision ranging output of lidar in the entire operating temperature range and full range, effectively suppressing systematic deviations caused by temperature changes and residual assembly errors, and significantly improving the stability and reliability of the product in practical applications by taking into account both real-time performance and flexibility through a programmable lookup table mechanism.

[0079] In one specific implementation, based on the above embodiments, the central processing center further executes dynamic path planning logic based on batch consistency. First, it continuously collects and caches the mechanical positioning error data of the most recently preset number of lidars to be calibrated in their initial state, forming a statistical sample set within a sliding time window. Specifically, the system performs variance analysis on this sample set to calculate the batch-to-batch consistency variance of the mechanical positioning error data. Then, it compares this consistency variance with two preset judgment thresholds: if the consistency variance is lower than the first threshold, the current production batch is determined to be in a stable process state and belongs to a high-consistency batch; at this time, the system automatically switches to a simplified calibration path, which skips the physical displacement adjustment of the optical module and the active environmental disturbance test steps, retaining only the extraction of electronic circuit parameters and firmware compensation writing process to shorten the calibration cycle per unit. Next, if the consistency variance is higher than the second threshold, the current batch is determined to have significant assembly fluctuations and belongs to an abnormal fluctuation batch; the system automatically activates a high-intensity calibration path, which fully executes the entire calibration process, including precision mechanical adjustment, closed-loop feedback locking, and active environmental disturbance testing, ensuring that each lidar meets stringent performance standards.

[0080] In other alternative implementations, the system may also employ a multi-level adaptive strategy based on the changing trend of the consistency variance, such as setting an intermediate threshold range to enable some enhanced calibration steps, or combining a moving average filtering algorithm to improve the robustness of variance calculation.

[0081] Through the above solution, this embodiment enables intelligent dynamic adjustment of the calibration process, significantly improving production line cycle efficiency while ensuring product quality. For production batches with stable processes, unnecessary mechanical fine-tuning and environmental stress testing expenses are avoided, reducing equipment wear and energy consumption; while for batches with abnormal fluctuations, potential defective products are intercepted in a timely manner by strengthening the calibration path, enhancing the quality control capability of the manufacturing process.

[0082] In one specific implementation, the dynamic path planning step further includes: When the system determines that the current batch is an abnormally fluctuating batch, it first extracts the feature vector with the largest deviation from the initial correlation characteristic set of all lidars to be calibrated in the batch. Specifically, this feature vector includes the maximum optical axis deviation angle, the maximum mechanical positioning error, and the corresponding offset direction information. Then, the central processing computing center transmits this feature vector to the Manufacturing Execution System (MES) via the industrial communication bus, triggering the front-end process feedback module. This module converts the feature vector into tolerance adjustment instructions for upstream machining equipment based on preset mapping rules. For example, if a batch lens installation misalignment problem is detected, it generates a compensation displacement for the injection mold fixture center positioning mechanism, or corrects the offset of the coordinate system origin in the CNC machining program. Then, the tolerance adjustment instructions are automatically sent to the corresponding CNC equipment controller to achieve closed-loop update of machining parameters. In some other optional implementations, the feedback of the feature vector can be pushed to a remote operation and maintenance platform in real time via wireless industrial IoT protocols (such as 5G-U or Wi-Fi 6), where process engineers can manually confirm and execute the adjustment; or, the tolerance adjustment instructions can be encapsulated in the OPC UA standard data format, compatible with different brands of PLC control systems.

[0083] Through the above scheme, this embodiment can reverse the calibration data analysis results at the end of the production line to the front-end manufacturing process, forming a closed loop of dynamic optimization of process parameters across processes. This not only improves the response capability to abnormal batches, but also realizes the quality control upgrade from passive screening to proactive prevention, effectively reducing the risk of large-scale assembly deviations caused by equipment wear or process drift.

[0084] In one specific implementation, the magnitude of the radar difference vector is quantized through the following mathematical operations. First, the optical axis deviation angle is... Coupling coefficient calculated in the above steps Multiply to obtain the weighted optical error components. This component reflects the actual impact of mechanical assembly deviations on the amplitude fluctuations of the electronic signal. Then, the error index in the circuit response is selected as the second component; specifically, the normalized signal peak delay characteristic is used. And introduce weighting factors Scale matching is performed to form a time-domain error term. These two components constitute a two-dimensional error vector: Next, perform Euclidean norm operations on the vector to calculate its magnitude: This modulus value serves as a comprehensive error metric to determine whether the current radar instance deviates from the preset performance boundary. In other optional implementations, the vector synthesis method can be extended to higher dimensions, such as by introducing drift gradients under temperature perturbations. As a third component, the normalization strategy and weight allocation of each component are adjusted accordingly; or, the Manhattan norm or Chebyshev norm is used instead of the Euclidean norm for distance measurement to adapt to the different priority requirements of specific production lines for error-sensitive directions.

[0085] Through the above scheme, this embodiment can uniformly map multi-source heterogeneous physical and electrical errors onto a comparable and decisionable scalar index, achieving an objective quantitative assessment of lidar system-level deviations. Compared with existing technologies, this embodiment avoids the risk of misjudgment caused by single parameter threshold judgments, improving the robustness of calibration decisions. Simultaneously, this vector synthesis mechanism supports flexible expansion, facilitating the integration of more error dimensions under various environments or dynamic operating conditions, enhancing the adaptability and foresight of the calibration system.

[0086] In one specific implementation, before calculating the calibration parameters based on the calibration parameters, the step of establishing a closed-loop feedback model is also included: Calibration data from multiple batches of lidars accumulated during historical calibration processes were collected. Specifically, this included the initial set of associated characteristics measured before calibration for each lidar (including optical axis deviation angle, mechanical positioning error, assembly gap size, and offset vector direction), and the corresponding performance improvement indicators measured after calibration. The latter were quantitatively characterized by the improvement rate of the standard deviation of point cloud distance accuracy, the signal-to-noise ratio improvement, and the decrease in the coefficient of variation of full-field uniformity. Specifically, the initial set of associated characteristics was used as the input feature vector, and the corresponding optimal assembly gap adjustment and electronic compensation coefficient were used as the target output labels to construct a supervised learning dataset for training the neural network.

[0087] Then, a nonlinear mapping model from input features to calibration parameters is established using a feedforward fully connected neural network architecture. This neural network consists of an input layer (with a dimension matching the number of features in the initial associated feature set), two hidden layers (configured with 64 and 32 ReLU activation units respectively), and a linear output layer. During training, a mean squared error loss function and the Adam optimizer are used, and the network weights are iteratively updated using the backpropagation algorithm until the prediction error on the validation set converges. The trained mapping model is then deployed in the real-time inference engine of the central processing unit.

[0088] Next, before performing formal calibration on the instance to be calibrated, the system first calls the pre-trained neural network model, inputs the currently acquired initial set of correlation characteristics, and predicts a set of initial calibration parameters, including the recommended optimal assembly gap adjustment. This prediction result serves as the initial setpoint for subsequent closed-loop adjustment, driving the piezoelectric ceramic actuator for rapid coarse adjustment, thereby significantly shortening the search time required to approach the optimal position.

[0089] In some other alternative implementations, the neural network model can be replaced with a gradient boosting decision tree (GBDT) or a Gaussian process regression (GPR) model to adapt to the data distribution characteristics of different production lines; or, historical calibration data can be subjected to principal component analysis (PCA) dimensionality reduction before being input into the model to reduce the impact of redundant features on the model's generalization ability.

[0090] Through the above scheme, this embodiment can use historical production data to build a closed-loop feedback model with predictive capabilities, providing high-confidence initial calibration parameter suggestions before formal calibration begins, thereby effectively shortening the calibration convergence path, reducing reliance on repeated fine-tuning of the actuator, improving overall calibration efficiency, and enhancing adaptability to new models or process fluctuations.

[0091] In one specific implementation, locking the current position of the optical module specifically includes: After confirming that the magnitude of the radar difference vector has reached the convergence condition, it is determined that the optical module has been adjusted to the optimal position. At this point, the piezoelectric ceramic actuator maintains its current output force and displacement state unchanged. Specifically, using a dispensing device integrated into the execution subsystem, a quantitative amount of UV-curable adhesive is injected into the annular joint area between the optical module and the radar housing. The amount of adhesive is controlled within the range of 0.02 to 0.05 mL to ensure sufficient filling without overflowing onto the optical surface. Next, the directional UV light source installed on the side of the flexible six-axis holder is activated, emitting UV light with a wavelength of 365 nm and a light intensity of not less than 100 mW / cm². The light is irradiated at the joint for more than 3 seconds to allow the adhesive to complete the cross-linking and curing reaction, thereby physically locking the relative position of the optical module and the housing. In some other alternative implementations, in-situ curing can also use laser welding instead of UV curing, in which a laser welding head is aimed at the joint of the metal shell and a pulsed laser is applied to melt the local material to achieve metallurgical bonding; or a thermosetting adhesive is used in conjunction with an infrared heating device to complete the curing, with the heating temperature controlled between 80°C and 120°C to avoid thermal damage to the internal electronic components.

[0092] Through the above solution, this embodiment can achieve long-term stable maintenance of the mechanical position after nanometer-level precision adjustment, effectively preventing micro-displacement rebound caused by vibration during subsequent handling or use; at the same time, the use of ultraviolet curing adhesive combined with point irradiation can complete high-strength bonding within seconds, taking into account both process efficiency and connection reliability, and is suitable for the cycle time requirements of automated production lines.

[0093] In one specific implementation, the acquisition of the electronic circuit response signal is achieved through an integrated time-to-digital converter (TDC) and analog-to-digital converter (ADC). Specifically, firstly, the lidar to be calibrated emits a laser pulse, triggering the TDC to start timing. The TDC records the time point when the echo photons arrive at the photodetector with picosecond resolution and outputs high-precision timestamp data for accurate reconstruction of time of flight (TOF). Then, the analog echo signal is conditioned by a preamplifier circuit and sent to a high-speed ADC for sampling at a sampling rate of no less than 1 GS / s. The ADC converts the continuous voltage signal into discrete digital waveform data, preserving the signal amplitude, rise slope, and noise distribution characteristics. The central processing unit simultaneously receives the timestamp sequence from the TDC and the waveform sampling points from the ADC, constructing a composite signal data structure containing both timing and amplitude information for use by the subsequent feature extraction module.

[0094] In other alternative implementations, the TDC can achieve parallel acquisition of multiple echo events through a multi-channel time measurement architecture to support multi-echo identification capabilities; the ADC can employ alternating sampling or a multi-stage pipeline structure to increase the effective number of bits (ENOB), thereby enhancing small-signal resolution. Furthermore, the sampling clocks of the TDC and ADC are driven by the same low-phase-noise crystal oscillator, ensuring global time base consistency between the timestamp and waveform sampling.

[0095] Through the above scheme, this embodiment can completely obtain the time dimension and amplitude dimension information in the electronic circuit response signal, providing original data support for the subsequent accurate extraction of signal peak delay characteristics and response amplitude fluctuation characteristics, thereby improving the accuracy and robustness of opto-mechanical-electrical coupling error analysis.

[0096] In one specific implementation, this automated calibration method is integrated into a continuously operating assembly line. The control units of high-precision vision sensors, digital oscilloscopes, piezoelectric ceramic drive controllers, and environmental stress screening devices are all bidirectionally connected to a central processing unit (CSU) via an industrial bus. Specifically, a real-time industrial Ethernet based on the EtherCAT protocol is first used as the backbone communication architecture. Data acquisition devices and actuators distributed across multiple workstations are uniformly connected to the CSU's scheduling system, ensuring strict synchronization of the operation sequence of all devices. Then, the CSU periodically issues control commands to coordinate the subsystems to automatically complete data acquisition, analysis, decision-making, and closed-loop execution actions according to a preset rhythm, without manual intervention. Finally, at the end of each calibration cycle, the system automatically generates a log file containing raw error data, coupling coefficients, difference vector magnitudes, and compensation parameters, and uploads it to the factory-level Manufacturing Execution System (MES) via the OPC UA interface, achieving full traceability of the calibration process.

[0097] In other alternative implementations, the industrial bus can be replaced with PROFINET or Modbus TCP protocols, suitable for production line scenarios with relatively low real-time requirements; command interaction between the central processing unit and the actuator can also adopt an extended scheme based on Time Sensitive Network (TSN) to support concurrent control of larger-scale equipment; in addition, temperature step commands in the test environment controller can be transmitted independently via CAN bus to enhance electromagnetic interference immunity.

[0098] Through the above solution, this embodiment can achieve seamless integration of the LiDAR calibration process into a high-speed automated production line, significantly improving the calibration throughput per unit time, while ensuring the timing consistency and operational reliability of multiple devices working together, thereby meeting the dual requirements of efficiency and stability for large-scale production of automotive-grade products.

[0099] This invention provides an automated calibration system suitable for mass-produced lidar. For example... Figure 2 As shown, the system includes: The multi-dimensional data acquisition unit M100 is installed at the production line station and is configured to acquire the optical axis deviation angle data and mechanical positioning error data of the lidar to be calibrated, as well as the electronic circuit response signal output by the lidar to be calibrated under standard test conditions. The central processing unit M200 is communicatively connected to the multidimensional data acquisition unit M100. The central processing unit M200 is programmed to run the following logic modules: The classification mapping module M300 is used to run the support vector machine algorithm to map the optical axis deviation angle data and mechanical positioning error data to the preset deviation classification space, and generate an initial set of associated features containing assembly gap dimensions and offset vectors. The coupling analysis module M400 is used to extract signal peak delay features and response amplitude fluctuation features from the electronic circuit response signal, and calculate the coupling coefficient of the interaction between the optical system and the electronic circuit based on cluster analysis, thereby constructing the radar difference vector. The decision generation module M500 is used to calculate calibration parameters when the magnitude of the radar difference vector exceeds the tolerance threshold. The calibration parameters include physical assembly gap adjustment instructions or firmware signal compensation coefficients. In some embodiments, the central processing computing center M200 also includes a front-end process feedback module. This module is configured with statistical process control (SPC) logic for continuously monitoring consistency variance between batches. When abnormally fluctuating batches are detected, the module extracts key deviation feature vectors and sends reverse correction instructions to the factory-level manufacturing execution system (MES) or the front-end injection molding / CNC equipment controller via an industrial internet interface (such as OPC UA) to achieve a quality closed loop across processes.

[0100] The execution subsystem M600, controlled by the central processing computing center M200, includes: A precision mechanical adjustment mechanism is used to perform displacement operations on the optical module of the lidar to be calibrated in response to physical assembly gap adjustment commands. The firmware programming interface is used to perform data writing operations on the internal memory of the lidar to be calibrated in response to the firmware signal compensation coefficient.

[0101] In one embodiment, the multi-dimensional data acquisition unit M100 includes a high-precision vision sensor and a high-speed signal acquisition device. The high-precision vision sensor employs a CMOS imaging device with a resolution of at least 5 megapixels, coupled with a telecentric lens, to capture images of the optical module's installation posture and to resolve the lens's positional offset relative to the ideal optical axis using a sub-pixel-level edge detection algorithm. The high-speed signal acquisition device is a digital oscilloscope or a time-to-digital converter (TDC) with a sampling rate of at least 1 GS / s, used to capture the analog echo waveform or digital timestamp information output by the photodetector. The central processing unit M200 integrates a GPU acceleration unit to perform classification, clustering, and regression analysis tasks in parallel, improving overall calibration throughput efficiency. The support vector machine model run by the classification mapping module M300 uses radial basis functions as kernel functions and performs offline learning using historical calibration samples during the training phase to form stable classification boundaries, effectively distinguishing between normal assembly states and states with significant nonlinear coupling risks. The coupling analysis module M400 first performs wavelet denoising on the acquired electronic circuit response signal. Then, it extracts the signal peak delay feature (the time difference between the maximum amplitude of the echo signal and the transmitted trigger signal) and the response amplitude fluctuation feature, obtained by performing a Fast Fourier Transform on the denoised signal and calculating the standard deviation of the frequency domain amplitude. Based on this, the K-Means clustering algorithm is used to divide the current batch of data into several typical error pattern clusters. For each cluster, a linear relationship between the optical axis deviation angle and the response amplitude fluctuation is independently fitted, and the resulting slope serves as the coupling coefficient for that cluster, reflecting the sensitivity of the photoelectric response under a specific assembly condition. The decision generation module M500 constructs a radar difference vector based on the coupling coefficient and the initial correlation characteristic set. Its components represent the weighted mechanical deviation influence and the circuit response distortion degree, respectively, and the modulus represents the overall error level. When the modulus exceeds a preset threshold, the system determines that a closed-loop calibration process needs to be initiated, and uses the gradient descent principle to deduce the optimal assembly gap adjustment amount or generate corresponding electronic compensation parameters. The precision mechanical adjustment mechanism in the M600 execution subsystem possesses micron-level adjustment capabilities, enabling it to drive the optical module to perform precise displacement along the optical axis upon receiving a physical assembly gap adjustment command. The firmware programming interface establishes a connection with the main control chip of the lidar to be calibrated via UART or JTAG protocols, allowing for the secure writing of compensation parameters. In other optional implementations, the high-speed signal acquisition device can be replaced with an integrated data acquisition card, and the vision sensor can employ a CCD device to improve dynamic range. The central processing computing center M200 can be deployed on a local server or edge computing node, communicating with each acquisition and execution unit via industrial Ethernet.

[0102] The aforementioned components work collaboratively to form a closed-loop control architecture of "perception-analysis-decision-execution". The multi-dimensional data acquisition unit M100 simultaneously acquires optical structural parameters and electronic response characteristics, providing realistic input for subsequent modeling. The classification and mapping module M300 within the central processing computing center M200 initially screens out target instances with severe coupling risks, avoiding redundant and complex operations on all products. The coupling analysis module M400 delves into the quantitative relationship between mechanical errors and circuit performance degradation, overcoming the technical bottleneck of the separation of optical, mechanical, and electronic aspects in traditional calibration. The decision generation module M500 determines whether hardware intervention is needed based on this quantitative model and generates precise control commands. The execution subsystem M600 then implements physical structure adjustments or electronic parameter solidification according to the commands, completing system-level performance optimization. Through this hardware-software collaborative, data-driven closed-loop mechanism, the system solves the problem of strong multi-physics coupling caused by minute assembly deviations in mass-produced lidar, achieving automated calibration from individual error identification to global performance convergence.

[0103] Compared with existing technologies, this embodiment introduces a machine learning-based multidimensional feature classification and coupled modeling method to achieve quantitative characterization of the cross-domain error propagation path of "optical-mechanical-electronic", so that the calibration process no longer relies on manual experience or isolated parameter adjustment. Combined with a dual-channel execution architecture, it can implement mechanical fine-tuning with nanometer-level precision and dynamically inject electronic compensation strategies, which significantly improves calibration accuracy and adaptability. The entire system is integrated into an automated production line environment, supports continuous online operation, greatly improves production cycle and product consistency, and meets the urgent need for high-precision and high-efficiency calibration in the large-scale manufacturing of automotive-grade LiDAR.

[0104] In one specific embodiment, the precision mechanical adjustment mechanism employs an integrated micro-motion alignment device, which integrates a flexible six-axis gripper, a piezoelectric ceramic drive array, and an in-situ curing component. The flexible six-axis gripper is constructed of a high-elasticity nickel-titanium alloy sheet, with its ends designed as arc-shaped contact surfaces and covered with a 0.1mm thick polytetrafluoroethylene (PTFE) buffer layer. This layer is used for non-destructive clamping of the outer wall of the lidar's optical module, ensuring clamping stability while avoiding the introduction of additional stress deformation. The piezoelectric ceramic drive array consists of six linear piezoelectric actuation units, arranged in a ring around the gripper base. Each actuation unit has a displacement resolution of 0.05μm and a maximum stroke of ±15μm. By coordinating and controlling the extension and retraction of each unit, six-dimensional nanometer-level precision adjustment of the optical module is achieved in the X, Y, and Z translational degrees of freedom and rotation around the three axes. The in-situ curing component is integrated into the side bracket of the holder and includes a directional ultraviolet light source with an adjustable intensity and a wavelength of 365nm. Its irradiation angle is precisely limited to ±15° via an optical guide, ensuring that it only acts on the pre-coated photosensitive adhesive area at the joint between the optical module and the housing. After the central processing unit M200 confirms the convergence of calibration parameters, the component automatically starts, continuously irradiating for 3 seconds at an intensity of 3W / cm² to complete the rapid cross-linking and curing of the adhesive, thereby locking the optical module in the optimal assembly position. In some alternative embodiments, the in-situ curing component can be replaced with a micro-laser welding head, equipped with a pulsed fiber laser with an output power of 5W and a wavelength of 1064nm, for spot welding and fixing the metal lens barrel to the housing.

[0105] Through the above scheme, this embodiment can achieve non-destructive clamping and six-degree-of-freedom submicron dynamic alignment of the optical module, and combined with the physical locking mechanism of instant curing, ensure the long-term stability of the mechanical state after calibration; at the same time, the clamping, driving and curing functions are highly integrated into an integrated structure, which reduces the cumulative error and action delay caused by multi-device collaboration, and improves the reliability and cycle efficiency of the automated calibration process.

[0106] In one specific implementation, the automated calibration system further integrates an environmental stress screening device, which includes a rapid temperature-changing test chamber and a dynamic response monitoring circuit independent of the internal circuitry of the lidar to be calibrated. The rapid temperature-changing test chamber has a sealed test cavity to house the lidar during the calibration process; the inner wall of the chamber integrates high-power thermocouple assemblies, enabling a temperature step change from 25°C to 65°C within 30 seconds, while a built-in forced air circulation system ensures that the temperature field uniformity within the cavity is controlled within ±0.5°C. Furthermore, the environmental stress screening device includes a precision mechanical vibration table, physically integrated into the bottom of the test cavity or extending into the cavity via a rigid linkage, for supporting the lidar to be calibrated. This vibration table is driven by a broadband electromagnetic exciter or a hydraulic servo system, equipped with a closed-loop vibration controller, capable of responding to commands from the central processing unit M200, generating sinusoidal scanning vibrations or random vibration spectra within a frequency range of 10Hz to 2000Hz, with an adjustable acceleration amplitude range of not less than 5g. The dynamic response monitoring circuit is connected to the output port of the lidar photodetector via an external probe to collect the intensity drift of the echo signal in real time during temperature disturbances. Especially for quantum lidar, which uses single-photon detection devices, this monitoring circuit is specifically configured to capture the drift of microscopic physical quantities in the signal, focusing on monitoring thermally induced noise increases in dark count rate (DCR), attenuation of single-photon detection efficiency (PDE), and pulse broadening. These analog electrical signals are converted into digital sequences via a high-precision ADC and transmitted to the central processing unit M200. Based on the received drift data, the central processing unit M200 reassesses the current thermo-mechanical coupling stability of the lidar. If it detects a non-linear decay trend in echo intensity or signal-to-noise ratio with temperature and the gradient exceeds a preset threshold, it triggers a dynamic correction mechanism for calibration parameters, such as adjusting the target displacement of the piezoelectric ceramic actuator or updating the temperature-related entries in the multidimensional compensation lookup table.

[0107] In some other alternative implementations, the rapid temperature change test chamber can be replaced with a liquid-cooled temperature control fixture, which utilizes a circulating heat transfer medium to achieve faster heat exchange; the dynamic response monitoring circuit can also adopt an optically isolated acquisition architecture, which converts the front-end sensing signals into optical signals for long-distance transmission via fiber optic links to eliminate the influence of electromagnetic interference on the measurement of weak echo signals.

[0108] Through the above scheme, this embodiment can actively excite and quantify the performance drift behavior of lidar under transient temperature conditions, obtain the real physical response by means of a monitoring channel independent of the device under test, thereby supporting closed-loop feedback correction of calibration parameters and improving the long-term stability and reliability of calibration results under complex working conditions.

[0109] In one specific implementation, the multidimensional data acquisition unit M100 further includes a physical synchronization trigger, which is a hardware logic circuit module. Its input is electrically connected to the transmit trigger pin (TX_TRIG) of the lidar to be calibrated, and its output is connected to the frame synchronization signal input of a high-precision vision sensor and the acquisition start signal input of an oscilloscope or a time-to-digital converter (TDC). This physical synchronization trigger is configured to immediately generate a unified time-base synchronization pulse upon detecting the rising edge signal of the transmit trigger pin, and simultaneously distribute this pulse signal to both the vision sensor and the oscilloscope via a coaxial cable, thereby forcing both to start data acquisition at the same absolute time starting point. The high-precision vision sensor initiates image exposure and frame capture based on this pulse, ensuring that the captured optical module image corresponds to the precise moment the lidar emits the test pulse. Simultaneously, the oscilloscope timestamps and samples the analog echo waveform based on this synchronization pulse, ensuring that the zero point of each frame of waveform data is strictly bound to the laser emission event. In some other alternative implementations, the physical synchronization trigger may also employ an optically isolated coupling design to enhance immunity to electromagnetic interference; or, alternatively, the IEEE 1588 Precision Time Protocol (PTP) may be used to achieve hardware and software co-synchronization of clocks between the acquisition devices, provided that the time jitter is controlled within ±10 ns.

[0110] Through the above scheme, this embodiment can achieve strict alignment of optical morphology data and electronic signal waveform data in the time dimension, eliminate the data misalignment problem caused by multi-source asynchronous acquisition, and provide reliable time reference support for the subsequent establishment of "optic-mechanical-electronic" multi-physics coupling model.

[0111] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such modifications and substitutions should be considered within the scope of protection of the present invention.

Claims

1. An automated calibration method suitable for mass-produced lidar, characterized in that, The method includes: The optical axis deviation angle data and mechanical positioning error data of the lidar to be calibrated on the production line are collected by high-precision sensors; The optical axis deviation angle data and mechanical positioning error data are processed by multidimensional feature mapping and classification using the support vector machine classification algorithm to identify calibration instances with potential coupling deviations and establish an initial set of associated characteristics including assembly gap dimensions and offset vectors. The electronic circuit response signal output by the example to be calibrated was acquired under a standard test environment, and the signal peak delay feature and response amplitude fluctuation feature were extracted from the electronic circuit response signal. Cluster analysis is performed on the signal peak delay characteristics and response amplitude fluctuation characteristics to calculate the coupling coefficient of the interaction between the quantified optical system and the electronic circuit, and the radar difference vector characterizing the radar integrated error state is generated by combining the offset vector. Determine whether the magnitude of the radar difference vector exceeds a preset tolerance threshold. If so, calculate the calibration parameters based on the coupling coefficient. The calibration parameters include the assembly gap adjustment for the optical module or the signal compensation coefficient for the electronic circuit. The actuator is driven to physically adjust the optical module according to the calibration parameters, or the signal compensation coefficient is written into the firmware storage area of ​​the lidar to be calibrated to complete the closed-loop calibration.

2. The method according to claim 1, characterized in that, The process of using a support vector machine classification algorithm to perform multidimensional feature mapping and classification on the optical axis deviation angle data and mechanical positioning error data includes: The optical axis deviation angle data and mechanical positioning error data are used to construct a two-dimensional feature input vector; The two-dimensional feature input vector is mapped to a high-dimensional feature space using a radial basis kernel function; The distance between the mapped two-dimensional feature input vector and the preset support vector machine hyperplane is calculated, and the lidar to be calibrated is marked as a normal instance, a linear deviation instance, or a nonlinear coupling deviation instance based on the distance.

3. The method according to claim 1, characterized in that, Extracting signal peak delay features and response amplitude fluctuation features from the electronic circuit response signal includes: Wavelet transform is used to perform multi-level decomposition on the acquired electronic circuit response signal to filter out background noise; Locate the peak time point of the denoised electronic circuit response signal, calculate the time difference between the peak time point and the transmitted trigger signal, and determine the time difference as the peak delay feature of the signal; A fast Fourier transform is performed on the denoised electronic circuit response signal to calculate the standard deviation of the amplitude distribution in the frequency domain, and the standard deviation of the amplitude distribution is determined as the response amplitude fluctuation characteristic.

4. The method according to claim 1, characterized in that, The calculation of the coupling coefficient between the optical system and the electronic circuitry includes: The assembly gap size, the signal peak delay characteristic, and the response amplitude fluctuation characteristic are normalized. The K-means clustering algorithm was used to group the normalized data to obtain feature clusters under different error modes; For each of the feature clusters, the slope of the influence of the optical axis deviation angle change on the response amplitude fluctuation feature is fitted by linear regression analysis, and the slope of the influence is determined as the coupling coefficient.

5. The method according to claim 1, characterized in that, The step of driving the actuator to perform physical displacement adjustment of the optical module according to the calibration parameters includes: The assembly gap adjustment amount is converted into a drive voltage signal for the piezoelectric ceramic actuator; The piezoelectric ceramic actuator connected to the optical module is driven to perform micron-level stepping movements along the optical axis. The electronic circuit response signal is continuously acquired during the movement, and the magnitude of the radar difference vector is updated in real time. When the magnitude of the updated radar difference vector reaches a minimum, the drive is stopped and the current position of the optical module is locked.

6. The method according to claim 5, characterized in that, After locking the current position of the optical module, the method further includes an active environmental disturbance test step, which includes: The temperature of the test environment is controlled to undergo a step change within a preset time window to apply thermal stress disturbance to the lidar to be calibrated. During the temperature change process, the point cloud coordinate data output by the lidar to be calibrated is continuously collected; The drift gradient of the point cloud coordinate data with temperature is calculated. If the drift gradient exceeds a preset thermal stability threshold, a secondary fine-tuning command is generated according to the direction of the drift gradient, and the piezoelectric ceramic actuator is driven again to correct the assembly gap size. A vibration table in the controlled test environment applies mechanical vibration to the lidar to be calibrated at a preset frequency; Monitor the signal-to-noise ratio change of the electronic circuit response signal under mechanical vibration conditions; If the signal-to-noise ratio is lower than the preset standard, the vibration frequency of the test environment is adjusted to match the resonance characteristics of the lidar to be calibrated, and the resonance frequency at this time is recorded as calibration reference data.

7. The method according to claim 6, characterized in that, The method also includes a dynamic path planning step based on batch consistency: The mechanical positioning error data of a continuously preset number of lidars to be calibrated are statistically analyzed, and the consistency variance between batches is calculated. If the consistency variance is lower than the first threshold, the current batch is determined to be a high consistency batch, and a simplified calibration path is automatically selected. The simplified calibration path only performs electronic circuit parameter compensation. If the consistency variance is higher than the second threshold, the current batch is determined to be an abnormal fluctuation batch, and a high-intensity calibration path is automatically selected. The high-intensity calibration path includes the physical displacement adjustment and the active environmental disturbance test steps throughout the entire process. When a batch is identified as having abnormal fluctuations, the feature vector with the largest deviation in that batch is extracted. The feature vector is fed back to the next higher level of production to generate tolerance adjustment instructions for the front-end machining equipment.

8. The method according to claim 1, characterized in that, The step of writing the signal compensation coefficient into the firmware storage area of ​​the lidar to be calibrated includes: Based on the residual error data after calibration, a multidimensional electronic compensation lookup table is constructed, which establishes a mapping relationship between distance correction values ​​under different operating temperatures and detection distances. The multidimensional electronic compensation lookup table is burned into the non-volatile memory of the lidar to be calibrated via a communication interface; The main control chip of the lidar to be calibrated is configured to read the real-time temperature and measurement distance during operation, and perform real-time data correction according to the multi-dimensional electronic compensation lookup table.

9. The method according to claim 1, characterized in that, Before calculating the calibration parameters based on the calibration parameters, the process also includes the step of establishing a closed-loop feedback model: Collect historical calibration data, which includes the initial set of correlation characteristics before calibration and the radar performance improvement index after calibration; A mapping model from the initial set of associated features to the optimal calibration parameters is trained using a neural network; The optimal assembly gap adjustment amount is predicted using the trained mapping model, and the optimal assembly gap adjustment amount is used as the initial calibration parameter.

10. An automated calibration system suitable for mass-produced lidar, characterized in that, The system includes: A multi-dimensional data acquisition unit is set at the workstation of the production line and is configured to acquire the optical axis deviation angle data and mechanical positioning error data of the lidar to be calibrated, as well as acquire the electronic circuit response signal output by the lidar to be calibrated under a standard test environment. A central processing and computing center, communicatively connected to the multidimensional data acquisition unit, is programmed to run the following logical modules: The classification mapping module is used to run the support vector machine algorithm to map the optical axis deviation angle data and the mechanical positioning error data to a preset deviation classification space, and generate an initial set of associated features containing assembly gap size and offset vector. The coupling analysis module is used to extract the signal peak delay characteristics and response amplitude fluctuation characteristics from the electronic circuit response signal, and calculate the coupling coefficient of the interaction between the optical system and the electronic circuit based on cluster analysis, thereby constructing the radar difference vector. The decision generation module is used to calculate calibration parameters when the magnitude of the radar difference vector exceeds the tolerance threshold. The calibration parameters include physical assembly gap adjustment instructions or firmware signal compensation coefficients. The execution subsystem, controlled by the central processing center, includes: A precision mechanical adjustment mechanism is used to perform a displacement operation on the optical module of the lidar to be calibrated in response to the physical assembly gap adjustment command. The firmware burning interface is used to perform data writing operations on the internal memory of the lidar to be calibrated in response to the firmware signal compensation coefficient.

Images