An image measurement method for precision dimensional measurement

Abstract

This invention discloses an image measurement method for precision dimensional measurement, comprising the following steps: Step 1: Selecting a laser scanner and an optical image measuring instrument to form a measurement array based on the workpiece's geometric characteristics; Step 2: Calibrating and verifying the measurement array, obtaining the equipment's basic error curve using standard gauge blocks, and acquiring real-time environmental parameters after successful verification; Step 3: Calculating the environmental interference coefficient K. env The process involves several steps: Step 4: Real-time monitoring of the equipment's acceleration sensor data to generate the equipment stability index Si; Step 5: Executing the measurement and outputting the result when Si > Ssafe; Step 6: Performing a consistency analysis on the results from multiple devices. If the results are consistent, the final value is output; otherwise, dynamic recalibration is triggered. This invention enables multi-dimensional collaborative control, breaking through the accuracy bottleneck of traditional image measurement in dynamic environments, and is suitable for precision manufacturing scenarios with high requirements for measurement stability and reliability.

Description

Technical Field

[0001] This invention relates to the field of precision measurement, and more specifically to an image measurement method for precision dimensional measurement. Background Technology

[0002] Precision dimensional measurement is a core component of high-end manufacturing, with accuracy requirements typically reaching the micrometer or even nanometer level. For example, the width tolerance of semiconductor chip circuits needs to be controlled within ±10nm, and the profile error of aero-engine blades needs to be ≤5μm. Using image measurement for close-range dimensional measurement is a technique employed in precision dimensional measurement.

[0003] Image measurement methods acquire workpiece images through optical imaging systems, such as laser scanners and optical imagers, and combine them with digital image processing technology to achieve non-contact dimensional analysis, offering advantages such as high efficiency and non-destructiveness. However, existing technologies suffer from problems such as ineffective quantification of environmental interference, lack of equipment stability assessment, insufficient accuracy of multi-device collaboration, and limited image blur repair capabilities. Therefore, this paper proposes an image measurement method for precision dimensional measurement. Summary of the Invention

[0004] This invention solves the problems of the prior art through the following technical solution, which includes the following steps:

[0005] Step 1: Select a laser scanner and an optical image measuring instrument to form a measuring array based on the workpiece's geometric characteristics;

[0006] Step 2: Calibrate and verify the measurement array. Obtain the basic error curve of the equipment using standard gauge blocks. After verification, collect real-time environmental parameters.

[0007] Step 3: Calculate the environmental interference coefficient ,when Start measurement at the specified time;

[0008] Step 4: Monitor the equipment's acceleration sensor data in real time and generate the equipment stability index Si;

[0009] Step 5: When Si > Ssafe, perform the measurement and output the result;

[0010] Step 6: Perform consistency analysis on the results from multiple devices. If the results are consistent, output the final value; otherwise, trigger dynamic recalibration.

[0011] Furthermore, the calculation process for the environmental interference coefficient includes:

[0012] Temperature change ∆T and humidity change ∆H are obtained by temperature and humidity sensors, and spectrum data A(f) is collected by vibration sensor.

[0013] Recall the device's pre-stored frequency response sensitivity curve W(f);

[0014] Calculate the overall interference coefficient using the formula:

[0015] ;

[0016] Where α, β and γ are weighting coefficients, T0 is the reference temperature, H0 is the reference humidity, and fmin and fmax are the effective vibration frequency ranges.

[0017] Furthermore, the process for obtaining the device stability index in step four is as follows:

[0018] Triaxial acceleration data is collected in real time by an accelerometer integrated inside the measuring device. The accelerometer continuously monitors the vibration state of the device during operation at a preset sampling rate (e.g., 200Hz), and the sampling time interval is defined as the time differential element dt.

[0019] Secondly, the collected triaxial acceleration data is filtered in the frequency domain using a bandpass filter to extract the effective vibration components within a preset frequency range (such as 0.5-50Hz) and eliminate low-frequency noise and high-frequency electromagnetic interference from the environment.

[0020] Then, based on the time-domain vibration acceleration function a(t) (the time-domain signal output in real time by the accelerometer), the displacement deviation is calculated by quadratic integration:

[0021] ;

[0022] Statistical analysis was performed on the displacement deviation sequence in the time domain, and its variance was calculated. (Characterizing the degree of fluctuation in displacement deviation), combined with the equipment's inherent sensitivity coefficient w (determined by the vibration sensitivity parameter calibrated at the equipment's factory), a stability index is generated, specifically:

[0023] .

[0024] Furthermore, the dynamic recalibration process in step six includes:

[0025] Based on the environmental interference coefficient K of each device env Calculate environmental adaptation factors Selecting environmental adaptation factors The largest device was used as the benchmark, as it exhibits the best measurement stability under the current conditions.

[0026] The reference device scans the standard gauge block from multiple angles (such as 0°, 90°, and 180°) to generate a reference point cloud dataset P containing three-dimensional coordinate information. ref The device to be calibrated is synchronously controlled to scan the same standard gauge block from the same perspective, generating a calibration dataset P. test ;

[0027] The spatial transformation matrix is ​​calculated using the Iterative Closest Point (ICP) algorithm. The specific process is as follows:

[0028] ;

[0029] Where T(・) is a rigid transformation function, used to transform the dataset Ptest to be calibrated to the reference coordinate system through the rotation matrix R and the translation vector t, so as to minimize the sum of squared Euclidean distances between the transformed point cloud and Pref;

[0030] Finally, based on the calculated transformation matrix T calib Update the internal calibration parameter matrix of the equipment to be calibrated to complete the one-time calibration of the equipment's spatial coordinate system.

[0031] Furthermore, image compensation is performed when Si ≤ Ssafe, and the specific compensation process is as follows:

[0032] Based on the x-direction displacement deviation δu calculated in step four x The displacement deviation δuᵧ in the y-direction, combined with the lens optical characteristic parameter λ (unit: 1 / m) 2 (Determined by the optical degradation coefficient specified by the lens manufacturer), constructing a point spread function model:

[0033] ;

[0034] This model is used to quantify the optical imaging diffusion effect caused by equipment vibration and serves as the core input parameter for subsequent deconvolution operations;

[0035] Secondly, an improved deconvolution algorithm is used to restore the blurred image. The specific iterative calculation process is as follows:

[0036] ;

[0037] in Let Iobs be the image estimate for the k-th iteration, ⊗ represent the convolution operation, ⊛ represent the correlation operation (defined as: for matrices A and B, the result of the correlation operation is the convolution of the transposes of A and B), and PSF. T The matrix transpose of the point spread function PSF is used, and the PSF model constructed in the first step is directly called for calculation during the iteration process;

[0038] Then, an adaptive iteration stopping condition is introduced to avoid overfitting or underfitting:

[0039] ;

[0040] Where ε is the convergence threshold (set by the image accuracy requirements, such as ε = 1 × 10⁻).4 When this condition is met, the iteration terminates and outputs a clear image compensated based on the PSF model.

[0041] Furthermore, the specific process for performing consistency analysis on results from multiple devices is as follows:

[0042] First, calculate the environmental adaptability factor of each device. The specific process is as follows:

[0043] ,in The attenuation coefficient controls the sensitivity to environmental influences. Let be the environmental interference coefficient of device i. This factor converts environmental interference into a 0-1 weight parameter through an exponential decay model. The smaller the interference, the larger the factor value.

[0044] The confidence weights are generated by combining environmental adaptation factors and historical equipment reliability. The specific process is as follows:

[0045] ;

[0046] Where Ri represents the historical reliability of device i, giving higher weight to devices that are highly adaptable to the environment and have a stable historical performance;

[0047] Finally, the consistency of the measurement results is evaluated using a weighted consistency criterion formula. The specific process is as follows:

[0048] ;

[0049] Where Mi is the measurement result of device i; For the weighted average result, ;

[0050] D th This is the consistency threshold, in the same unit as Mi.

[0051] Furthermore, The following process will be executed at that time:

[0052] choose The device corresponding to the maximum value is used as the reference device, which is least affected by interference and has the best measurement reliability in the current environment.

[0053] The reference equipment performs a high-density scan on the key geometric feature points of the workpiece (such as the tip circle of the gear tooth and the boundary point of the groove width), with a scanning density of no less than 3 times that of the conventional scan, to generate a feature point cloud dataset containing sub-pixel level coordinate information.

[0054] Subpixel-level feature templates are generated based on encrypted scanned point cloud data. Subpixel interpolation algorithms are used to improve the accuracy of feature points to within 0.1 pixels, forming a standard geometric model for local calibration.

[0055] Non-reference devices optimize local parameters based on pixel-level feature templates. The deviation between the template and the measured feature points is fitted by least squares method. Local parameters such as lens distortion coefficient and coordinate system offset are iteratively corrected. After optimization, the consistency analysis process is re-executed.

[0056] Furthermore, the output result flow includes the following process:

[0057] A time decay factor is applied to the confidence weights verified by the consistency analysis. The specific calculation process is as follows:

[0058] ,in Here, t is the time decay coefficient, and t is the test duration.

[0059] The weighted fusion result is calculated based on the weights after time decay. The specific process is as follows:

[0060] ;

[0061] Where Mi represents the measurement result of device i, and weighted fusion is used to suppress the impact of abnormal data on the final result;

[0062] Finally, output the results with confidence level annotations. The specific process is as follows:

[0063] ;

[0064] Where S min Let D be the minimum stability index and D be the consistency distance. This is the consistency attenuation coefficient.

[0065] Furthermore, if the convergence condition is not met even after the number of iterations in the image compensation process exceeds a preset threshold N (N≥20, set by the image accuracy requirements). When this happens, execute the following failure handling procedure:

[0066] The multi-frame fusion algorithm is activated to repair the blurred image. The specific process is as follows:

[0067] Acquire K frames (K≥3, set according to vibration frequency) of the same scene {I1, I2, ..., I K The restored image is output through Fourier transform domain fusion calculation:

[0068] If the iteration count exceeds N without convergence, multi-frame fusion is initiated to output the restored image. The specific process is as follows:

[0069] ;

[0070] in For Fourier transform operators, OTF is the optical transfer function (i.e., the Fourier transform of the point spread function PSF). express The conjugate of the complex number, μ is the regularization parameter (value 1×10). −6 ~1×10 −4 (to prevent the denominator from returning to zero).

[0071] Then, the validity of the multi-frame fusion result is verified by calculating the edge sharpness index of the fused image.

[0072] If Sedge≥Sth (Sth is the sharpness threshold, set by the workpiece feature accuracy requirements), then the fused image is output.

[0073] Finally, if multi-frame fusion still does not satisfy Sedge≥Sth, a device self-test command is triggered, automatically executing the following troubleshooting process:

[0074] Lens focal length calibration: The Z-axis focal length offset of the lens is calibrated using a standard gauge block, and automatic compensation is performed when the deviation exceeds ±0.05mm;

[0075] Diagnostic accelerometer: Detects the consistency of the response of 3-axis vibration signals in the 0.5-50Hz frequency band. A deviation exceeding 15% indicates a hardware fault.

[0076] The measurement process is paused and a fault code is output. The measurement will be restarted after manual inspection.

[0077] Compared with existing technologies, this invention has the following advantages: This image measurement method for precision dimensional measurement uses an array of laser scanners and optical image measuring instruments. It allows for flexible selection of equipment based on workpiece geometry, adapting to measurement scenarios with varying dimensional and precision requirements. It calculates environmental interference coefficients and sets trigger thresholds, dynamically determining measurement timing based on stability indices. This effectively eliminates interference from environmental factors such as temperature, humidity, and vibration. When measurement results from multiple devices are inconsistent, recalibration is triggered. Using the device with the best environmental adaptability as a benchmark, calibration parameters are updated through a point cloud matching algorithm to ensure consistency across multiple devices. Based on a point spread function model and an improved deconvolution algorithm, real-time compensation is provided for image blurring caused by device vibration, improving image clarity and dimensional measurement accuracy. Environmental adaptability factors, historical reliability, and time decay factors are introduced for weighted calculation, combined with consistency threshold judgment, to generate measurement results with credibility annotations, enhancing data reliability. An image compensation failure handling process is set up to avoid measurement interruptions due to single-stage failures, improving fault tolerance. Attached Figure Description

[0078] Figure 1 This is the overall flowchart of the present invention. Detailed Implementation

[0079] The embodiments of the present invention are described in detail below. These embodiments are implemented based on the technical solution of the present invention, and provide detailed implementation methods and specific operation processes. However, the scope of protection of the present invention is not limited to the following embodiments.

[0080] like Figure 1 As shown, this embodiment provides a technical solution: an image measurement method for precision dimensional measurement, comprising the following steps:

[0081] Step 1: Select a laser scanner and an optical image measuring instrument to form a measuring array based on the workpiece's geometric characteristics;

[0082] Step 2: Calibrate and verify the measurement array. Obtain the basic error curve of the equipment using standard gauge blocks. After verification, collect real-time environmental parameters.

[0083] Step 3: Calculate the environmental interference coefficient ,when Start measurement at the specified time;

[0084] Step 4: Monitor the equipment's acceleration sensor data in real time and generate the equipment stability index Si;

[0085] Step 5: When Si > Ssafe, perform the measurement and output the result;

[0086] Step 6: Perform consistency analysis on the results from multiple devices. If the results are consistent, output the final value; otherwise, trigger dynamic recalibration.

[0087] The environmental interference coefficient calculation process includes:

[0088] Temperature change ∆T and humidity change ∆H are obtained by temperature and humidity sensors, and spectrum data A(f) is collected by vibration sensor.

[0089] Recall the device's pre-stored frequency response sensitivity curve W(f);

[0090] Calculate the overall interference coefficient using the formula:

[0091] ;

[0092] Where α, β and γ are weighting coefficients, T0 is the reference temperature, H0 is the reference humidity, and fmin and fmax are the effective vibration frequency ranges;

[0093] By simultaneously collecting temperature and humidity changes ΔT and ΔH, and vibration spectrum data A(f), environmental factors are transformed into calculable quantitative indicators, avoiding measurement errors caused by environmental fluctuations.

[0094] For example, temperature changes can cause equipment to expand and contract, humidity changes can cause lens fogging, and vibration can cause image blurring. These factors are all included in the interference coefficient calculation to form a comprehensive evaluation.

[0095] The pre-stored frequency response sensitivity curve W(f) of the device is invoked to match the interference calculation with the actual performance of the device. Different devices have different sensitivities to vibration frequencies. For example, laser scanners are more sensitive to high-frequency vibrations. W(f) can specifically quantify the interference weight of each frequency band to improve the accuracy of the assessment.

[0096] By calculating the environmental interference coefficient Kenv and comparing it with a threshold η, an automatic criterion for measurement initiation is achieved. When environmental interference exceeds the threshold, the system pauses the measurement to avoid generating erroneous data in harsh environments and ensure the reliability of the measurement results.

[0097] For example, a precision parts factory uses a laser scanner to measure the dimensions of micro gears. There are temperature fluctuations (ΔT=5℃) caused by the start and stop of the air conditioning and vibrations (main frequency 10-30Hz) generated by the operation of machine tools in the workshop.

[0098] Data acquisition: Temperature and humidity sensors measured ΔT=5℃, reference temperature T0=20℃, ΔH=10%, and reference humidity H0=50%RH; vibration sensors acquired spectrum data A(f) in the 10-30Hz frequency band.

[0099] Coefficient calculation: Temperature disturbance term: α・|ΔT / T0|=0.4・|5 / 20|=0.1;

[0100] Humidity interference term: β・|ΔH / H0|=0.3・|10 / 50|=0.06;

[0101] Vibration disturbance term: γ・∫[10,30]A(f)・W(f)df, assuming the calculation result is 0.15;

[0102] The overall Kenv = 0.1 + 0.06 + 0.15 = 0.31;

[0103] Threshold judgment: If the preset η=0.2, then Kenv>η, the system will not start the measurement until the air conditioner runs stably and the vibration is reduced. Then Kenv drops below η before the measurement is performed to avoid scanning size deviation caused by temperature and vibration.

[0104] The process for obtaining the equipment stability index in step four is as follows:

[0105] Triaxial acceleration data is collected in real time by an accelerometer integrated inside the measuring device. The accelerometer continuously monitors the vibration state of the device during operation at a preset sampling rate (e.g., 200Hz), and the sampling time interval is defined as the time differential element dt.

[0106] Secondly, the collected triaxial acceleration data is filtered in the frequency domain using a bandpass filter to extract the effective vibration components within a preset frequency range (such as 0.5-50Hz) and eliminate low-frequency noise and high-frequency electromagnetic interference from the environment.

[0107] Then, based on the time-domain vibration acceleration function a(t) (the time-domain signal output in real time by the accelerometer), the displacement deviation is calculated by quadratic integration:

[0108] ;

[0109] Statistical analysis was performed on the displacement deviation sequence in the time domain, and its variance was calculated. (Characterizing the degree of fluctuation in displacement deviation), combined with the equipment's inherent sensitivity coefficient w (determined by the vibration sensitivity parameter calibrated at the equipment's factory), a stability index is generated, specifically:

[0110] ;

[0111] By acquiring triaxial acceleration data through 200Hz high-frequency sampling, transient vibrations during machine tool operation can be captured, such as impact vibrations during tool changes. This avoids the loss of vibration characteristics due to insufficient sampling frequency, ensuring real-time stability assessment.

[0112] A 0.5-50Hz bandpass filter is used to eliminate environmental noise, such as low-frequency vibrations from ventilation systems and high-frequency noise from electromagnetic interference. Only the mechanical vibration frequency band that has the greatest impact on the measuring equipment is retained, thereby improving the accuracy of stability assessment.

[0113] Acceleration data is converted into displacement deviation variance and quantified into a 0-1 index using the stability index Si, which facilitates the system's automatic determination of measurement timing.

[0114] The closer Si is to 1, the higher the stability. It can be directly compared with the safety threshold (Ssafe) to achieve digital start-stop control.

[0115] For example, a precision parts factory uses a laser scanner and an optical image measuring instrument array to measure a micro gear with a module of 0.5mm. The machine tools in the workshop generate vibrations of 20-30Hz when running, and the air conditioner periodically starts and stops, causing temperature fluctuations of ΔT=3℃.

[0116] Triaxial acceleration data were acquired at a sampling rate of 200Hz. The vibration components of 20-30Hz were extracted by passing the data through a 0.5-50Hz bandpass filter to obtain the acceleration function a(t).

[0117] By performing a second integral over a(t), the displacement deviation in the x-direction δux = 1.8 μm and in the y-direction δuy = 1.2 μm are calculated, and the variance σ is obtained. 2 δu=(1.82 +1.2 2 ) / 2 = 2.34μm 2 .

[0118] Then, the stability index was calculated: the inherent sensitivity coefficient of the device was w = 0.6 / μm. 2 (Based on the default settings in the laser scanner manual), the stability index is: ;

[0119] Dynamic management and control applications:

[0120] If the preset safety threshold S safe =0.5, at this time Si=0.41 safe The system triggers the image compensation process: constructing the point spread function. , where λ = 10^6 m⁻ 2 Lens optical parameters;

[0121] The Richardson-Lucy deconvolution algorithm was used to repair image blur caused by vibration, avoiding a deviation of ±0.3μm in gear tooth profile measurement due to vibration. The actual tooth width should be 500μm.

[0122] If stability monitoring is not enabled, direct measurement may result in the tooth width being misjudged as 500.3-500.6μm due to vibration;

[0123] After activation, through Si evaluation and image compensation, the measured value returned to 500±0.1μm, meeting the inspection requirements of precision gears.

[0124] The dynamic recalibration process in step six includes:

[0125] Based on the environmental interference coefficient K of each device env Calculate environmental adaptation factors Selecting environmental adaptation factors The largest device was used as the benchmark, as it exhibits the best measurement stability under the current conditions.

[0126] The reference device scans the standard gauge block from multiple angles (such as 0°, 90°, and 180°) to generate a reference point cloud dataset P containing three-dimensional coordinate information. ref The device to be calibrated is synchronously controlled to scan the same standard gauge block from the same perspective, generating a calibration dataset P. test ;

[0127] The spatial transformation matrix is ​​calculated using the Iterative Closest Point (ICP) algorithm. The specific process is as follows:

[0128] ;

[0129] ​Where T(・) is a rigid transformation function used to transform the dataset P to be calibrated. test Transform the point cloud to the reference coordinate system using rotation matrix R and translation vector t, so that the transformed point cloud is consistent with P. ref The sum of squared Euclidean distances is minimized;

[0130] Finally, based on the calculated transformation matrix T calib Update the internal calibration parameter matrix of the equipment to be calibrated and complete the one-time calibration of the equipment's spatial coordinate system;

[0131] When measurement results from multiple devices are inconsistent, dynamic recalibration with a reference device can eliminate systematic errors between devices.

[0132] For example, when laser scanners and optical imaging instruments experience reference shifts due to environmental vibrations, they can be corrected synchronously through recalibration, thus avoiding measurement errors caused by the accumulation of equipment deviations.

[0133] The Iterative Closest Point (ICP) algorithm is used to calculate the transformation matrix, aligning the point cloud data of the device to be calibrated with the reference point cloud to achieve spatial transformation calibration with sub-millimeter accuracy. This algorithm demonstrates significant matching effectiveness for complex geometric features such as gear tooth profiles and can effectively correct coordinate offsets caused by equipment installation deviations or temperature drift.

[0134] By using the equipment with the best environmental adaptability as a benchmark, the calibration process is coupled with the real-time environment, avoiding environmental mismatch problems caused by using a fixed benchmark.

[0135] For example, when the workshop temperature changes, the system automatically selects the equipment least affected by the temperature as the benchmark, thereby improving calibration reliability.

[0136] For example, if a precision parts factory uses a laser scanner (device A) and an optical image measuring instrument (device B) to measure a miniature gear (10mm tip circle diameter), the measurement results from equipment A and equipment B will deviate due to the continuous vibration of the machine tool:

[0137] Measurement value from device A: 10.02 mm;

[0138] Measurement value from device B: 9.98 mm;

[0139] Preset consistency threshold D th =0.03mm, at this point the deviation is 0.04mm > D th This triggers the recalibration process.

[0140] The specific process and effects of recalibration are as follows:

[0141] Calculate the environmental adaptability factors of the two devices Where κ = 0.1 / ℃, and K of device A env =0.25, K of device B env=0.18, Device B is less affected by environmental interference, so Device B is selected as the benchmark.

[0142] Baseline point cloud acquisition:

[0143] Device B performs multi-angle (0°, 90°, 180°) scanning on a standard gauge block (10mm standard cylinder), generating a reference point cloud dataset P containing 10,000 points. ref .

[0144] Scanning of equipment to be calibrated:

[0145] Device A synchronously scans the same standard block to generate a point cloud dataset P. tes The point cloud shifted by approximately 0.03 mm due to vibration.

[0146] ICP algorithm point cloud matching:

[0147] By iteratively calculating the transformation matrix Tcalib, Minimize this to obtain the rotation matrix R and the translation vector t: rotation angle: 0.5° around the z-axis;

[0148] Translation amount: x direction +0.025mm, y direction -0.01mm;

[0149] After the transformation matrix is ​​applied, the overlap between the point cloud of device A and the reference point cloud is improved.

[0150] Calibration parameter update:

[0151] Update the internal calibration matrix of device A to include rotation and translation parameters so that subsequent measurements can automatically compensate for spatial deviations.

[0152] Comparison of effects before and after recalibration:

[0153] Before recalibration: Equipment A measured the gear tip circle as 10.02mm, the actual value was 10.00mm, and the deviation was +0.02mm; Equipment B measured it as 9.98mm, and the deviation was -0.02mm.

[0154] After recalibration: the remeasured value of equipment A is 10.005mm, and that of equipment B is 9.998mm, with deviations of ≤0.005mm, meeting the gear tolerance requirement of ±0.01mm, and the results of the two equipment are consistent (0.007mm). <D th =0.03mm.

[0155] By setting up dynamic recalibration, measurement deviations can be automatically corrected when there is environmental interference or equipment drift, avoiding the time-consuming cost of manual calibration and improving the continuity and reliability of precision measurements.

[0156] Image compensation is performed when Si ≤ Ssafe. The specific compensation process is as follows:

[0157] Based on the x-direction displacement deviation δu calculated in step four x The displacement deviation δuᵧ in the y-direction, combined with the lens optical characteristic parameter λ (unit: 1 / m) 2 (Determined by the optical degradation coefficient specified by the lens manufacturer), constructing a point spread function model:

[0158] ;

[0159] This model is used to quantify the optical imaging diffusion effect caused by equipment vibration and serves as the core input parameter for subsequent deconvolution operations;

[0160] Secondly, an improved deconvolution algorithm is used to restore the blurred image. The specific iterative calculation process is as follows:

[0161] ;

[0162] in Let Iobs be the image estimate for the k-th iteration, ⊗ represent the convolution operation, ⊛ represent the correlation operation (defined as: for matrices A and B, the result of the correlation operation is the convolution of the transposes of A and B), and PSF. T The matrix transpose of the point spread function PSF is used, and the PSF model constructed in the first step is directly called for calculation during the iteration process;

[0163] Then, an adaptive iteration stopping condition is introduced to avoid overfitting or underfitting:

[0164] ;

[0165] Where ε is the convergence threshold (set by the image accuracy requirements, such as ε = 1 × 10⁻). 4 When this condition is met, the iteration terminates and a clear image based on PSF model compensation is output.

[0166] The mechanism of image blurring caused by vibration is quantified using the point spread function (PSF) model, and the displacement deviation δu is... x The transformation of δuᵧ into the optical degradation parameter λ provides a physical basis for the restoration of blurred images.

[0167] For example, device vibration causes pixel shift during lens imaging. The PSF model can describe the diffusion effect of this shift on the image, providing a mathematical basis for deconvolution inpainting.

[0168] The Richardson-Lucy deconvolution algorithm is used to iteratively optimize blurred images, combined with an adaptive stopping condition, i.e., a convergence threshold ε, to avoid overfitting or underfitting. This algorithm has a significant effect on restoring edge features such as gear tooth profiles, and can repair pixel-level deviations in blurred edges to the sub-pixel level.

[0169] By dynamically controlling the number of iterations through the convergence threshold ε, infinite computation is avoided. At the same time, multi-frame fusion is triggered when compensation fails, improving the system's adaptability to complex vibration scenarios.

[0170] For example, when a laser scanner in a precision parts factory is measuring micro-gears, machine tool vibration can cause the stability index Si to drop to 0.41. safe =0.5, triggering the image compensation process:

[0171] The specific process and effects of image compensation are as follows:

[0172] PSF model construction:

[0173] Based on the calculated displacement deviation δu x =1.8μm, δuᵧ=1.2μm, optical degradation coefficient λ=10 6 m⁻ 2 (Lens parameters), construct the point spread function:

[0174] PSF(x,y)=exp(−10 6 ×(1.8 2 ×10 −12 +1.2 2 ×10− 12 ))≈exp(−0.00468)≈0.995;

[0175] This PSF describes the diffusion range of a point light source caused by vibration, used to simulate the blurring process.

[0176] Richardson-Lucy deconvolution iteration:

[0177] The initial blurred image I0 is a grayscale image with blurred edges of the gear tooth profile. The actual tooth width edges should be sharp, and the blurred edges spread by about 2 pixels.

[0178] Iteration formula: ;

[0179] Adaptive stopping condition: when The iteration stopped at the 15th iteration and converged.

[0180] Image comparison before and after compensation:

[0181] ​Before compensation: The width of the grayscale transition band at the edge of the tooth tip circle is 4 pixels, corresponding to an actual size of 8μm. The measured diameter of the tooth tip circle is 10.02mm, with a deviation of +20μm.

[0182] After compensation: the edge transition band is compressed to 1 pixel (2μm), and the measured diameter is 10.003mm (deviation +3μm), which meets the gear tolerance requirement of ±10μm;

[0183] By combining physical modeling with intelligent algorithms, image blurring caused by vibration can be repaired in real time, enabling precision measuring equipment to maintain sub-micron level accuracy even in dynamic interference environments and avoiding misjudgment of dimensions due to image quality issues.

[0184] The specific process for performing consistency analysis on results from multiple devices is as follows:

[0185] First, calculate the environmental adaptability factor of each device. The specific process is as follows:

[0186] ,in The attenuation coefficient controls the sensitivity to environmental influences. Let be the environmental interference coefficient of device i. This factor converts environmental interference into a 0-1 weight parameter through an exponential decay model. The smaller the interference, the larger the factor value.

[0187] The confidence weights are generated by combining environmental adaptation factors and historical equipment reliability. The specific process is as follows:

[0188] ;

[0189] Where Ri represents the historical reliability of device i, giving higher weight to devices that are highly adaptable to the environment and have a stable historical performance;

[0190] Finally, the consistency of the measurement results is evaluated using a weighted consistency criterion formula. The specific process is as follows:

[0191] ;

[0192] Where Mi is the measurement result of device i; For the weighted average result, ;

[0193] D th This is the consistency threshold, in the same unit as Mi;

[0194] Through environmental adaptation factors The confidence weight wi is generated from the historical reliability Ri, so that the measurement results of devices that are less affected by environmental interference and have stable historical performance are given higher weights, thus avoiding the averaging error of bad money driving out good.

[0195] For example, if a device measures more stably under the current temperature and humidity, its result will account for a larger proportion in the weighted average, thus improving the overall accuracy.

[0196] The environmental adaptability factor adopts an exponential decay model, converting the environmental interference coefficient into a weighting factor of 0-1. When the temperature and humidity in the workshop fluctuate, the weight of the equipment that is more affected is automatically reduced, ensuring that the fusion result is biased towards equipment with strong environmental adaptability.

[0197] By introducing historical reliability (Ri) of the equipment and combining it with long-term measurement error statistics (such as the standard deviation of the past 100 measurements), the weight of new measurement results depends not only on the real-time environment, but also on the historical performance of the equipment.

[0198] For example, if a device has a small long-term measurement error, it may still receive a higher weight due to its higher Ri, even if the current environmental adaptation factor is slightly lower.

[0199] For example, if a precision parts manufacturer uses a laser scanner (device A) and an optical image measuring instrument (device B) to measure a miniature gear (10mm tip circle diameter), the current environmental interference coefficient and historical equipment data are as follows:

[0200] Device A: Kenv_A=0.25, Ri_A=0.9 (historical error standard deviation 0.01mm);

[0201] Device B: Kenv_B=0.18, Ri_B=0.8 (historical error standard deviation 0.02mm);

[0202] The attenuation coefficient κ = 0.1 / unit Kenv, and the consistency threshold Dth = 0.03mm;

[0203] The specific process and results of consistency analysis are as follows:

[0204] Calculation of environmental adaptation factors:

[0205] Equipment A: C env A=e (-0.1×0.25) ≈0.975;

[0206] Equipment B: C env B=e (-0.1×0.18) ≈0.982;

[0207] Confidence weight generation:

[0208] Device A: wA=C env A×RiA=0.975×0.9=0.8775;

[0209] Device B: wB=C env B×RiB=0.982×0.8=0.7856;

[0210] Weight normalization: w'A=0.8775 / (0.8775+0.7856)≈0.53, w'B≈0.47;

[0211] Weighted average and consistency judgment:

[0212] Equipment measurements: MA = 10.02 mm, MB = 9.98 mm;

[0213] Weighted average result: =0.53+0.470.53×10.02+0.47×9.98≈10.0016mm;

[0214] Consistency distance calculation:

[0215] ;

[0216] Since D=0.015mm <D th =0.03mm, the judgment result is consistent, and the final output value is 10.00mm;

[0217] After weighted fusion, device A has a higher environmental adaptability factor and historical reliability, with a weighting of 53%, making the result closer to its measured value. At the same time, consistency judgment ensures that the deviation is within the threshold.

[0218] If the current measurement value of device B deviates to 9.95mm due to temporary vibration, then D=0.035mm>Dth, triggering dynamic recalibration to avoid erroneous data output.

[0219] When consistency is not satisfied The following process will be executed at that time:

[0220] choose The device corresponding to the maximum value is used as the reference device, which is least affected by interference and has the best measurement reliability in the current environment.

[0221] The reference equipment performs a high-density scan on the key geometric feature points of the workpiece (such as the tip circle of the gear tooth and the boundary point of the groove width), with a scanning density of no less than 3 times that of the conventional scan, to generate a feature point cloud dataset containing sub-pixel level coordinate information.

[0222] Subpixel-level feature templates are generated based on encrypted scanned point cloud data. Subpixel interpolation algorithms are used to improve the accuracy of feature points to within 0.1 pixels, forming a standard geometric model for local calibration.

[0223] Non-reference devices optimize local parameters based on pixel-level feature templates. The deviation between the template and the measured feature points is fitted by least squares method. Local parameters such as lens distortion coefficient and coordinate system offset are iteratively corrected. After optimization, the consistency analysis process is re-executed.

[0224] For example, if a device is least affected by interference under the current temperature and humidity, its measurement results will automatically become the calibration benchmark, improving calibration efficiency and reliability.

[0225] By performing intensive scanning of key feature points on the workpiece, such as the vertices and root transition curves of the gear tooth profile, sub-pixel-level feature templates are generated, breaking through the limitations of pixel-level resolution and achieving geometric feature matching with micron-level precision. This template can accurately describe the microscopic morphology of the gear tooth profile, providing a high-precision reference for subsequent calibration.

[0226] Non-reference devices optimize local parameters, such as lens focal length and coordinate system offset, based on sub-pixel templates, avoiding the time-consuming operation of global recalibration. For local areas susceptible to vibration in gear measurements, such as the tooth tip edge, deviations can be specifically corrected, improving calibration efficiency.

[0227] The output result process includes the following steps:

[0228] A time decay factor is applied to the confidence weights verified by the consistency analysis. The specific calculation process is as follows:

[0229] ,in Here, t is the time decay coefficient, and t is the test duration.

[0230] The weighted fusion result is calculated based on the weights after time decay. The specific process is as follows:

[0231] ;

[0232] Where Mi represents the measurement result of device i, and weighted fusion is used to suppress the impact of abnormal data on the final result;

[0233] Finally, output the results with confidence level annotations. The specific process is as follows:

[0234] ;

[0235] Where S min Let D be the minimum stability index and D be the consistency distance. The consistency attenuation coefficient;

[0236] By applying a time decay factor to the measurement data, more recent measurement results gain higher confidence, avoiding the accumulation of biases in earlier data caused by gradual environmental changes (such as equipment temperature drift). For example, during long-term continuous measurements, the weight of later data is automatically higher than that of earlier data, adapting to dynamic environmental changes.

[0237] The final value is calculated by incorporating time decay weights to suppress the influence of outlier data. When a device experiences a sudden change in value due to temporary vibration, the weight of its long-term test data decreases due to time decay, preventing single-point errors from dominating the result.

[0238] Output results with confidence level labels, allowing operators to intuitively assess measurement reliability. For example, a confidence level Q < 0.7 prompts for retesting, reducing the risk of misjudgment.

[0239] For example, a precision parts manufacturer performs three measurements on the same miniature gear, with the time intervals and equipment data as follows:

[0240] First measurement (t=0min): Equipment A measured value M1=10.02mm, stability index S1=0.8, environmental interference coefficient K env =0.2;

[0241] Second measurement (t=10min): Equipment B measured values ​​M2=9.98mm, S2=0.75, K env =0.22;

[0242] Third measurement (t=20min): Equipment A measured values ​​M3=10.01mm, S3=0.85, K env =0.18;

[0243] The time decay coefficient τ = 0.05 / min, the consistency decay coefficient ξ = 1, and the consistency distance D = 0.01mm;

[0244] Final result generation process and effects;

[0245] Time decay weight calculation:

[0246] First measurement: w1′=w1×e −0.05×0 =w1×1=0.53;

[0247] Second measurement: w2′=0.47×e −0.05×10 ≈0.47×0.6065≈0.285;

[0248] Third measurement: w3′=0.53× e−0.05 ×20≈0.53×0.3679≈0.195;

[0249] The final value of the weighted fusion calculation is:

[0250] ;

[0251] Credibility labeling calculation:

[0252] Minimum stability index S min =0.75 (second measurement);

[0253] Substitute into the formula, ;

[0254] Result annotation: Mfinal=10.01mm, confidence level after rounding Q=0.70;

[0255] Without time decay enabled: The simple weighted average is (10.02+9.98+10.01) / 3≈10.003mm, but it does not take into account the possibility that the second measurement (10 minutes ago) may be affected by environmental changes, resulting in an artificially high weight.

[0256] The third measurement in this case was more recent, and although its weight was reduced due to decay, it was still more reliable than the earlier data, and the final value was closer to the current true size (10.01 mm).

[0257] A confidence level of Q=0.70 indicates moderate reliability. Operators can decide whether to conduct additional measurements based on their needs. If Q<0.7, a rapid retest should be initiated.

[0258] Furthermore, if the convergence condition is not met even after the number of iterations in the image compensation process exceeds a preset threshold N (N≥20, set by the image accuracy requirements). When this happens, execute the following failure handling procedure:

[0259] The multi-frame fusion algorithm is activated to repair the blurred image. The specific process is as follows:

[0260] Acquire K frames (K≥3, set according to vibration frequency) of the same scene {I1, I2, ..., I K The restored image is output through Fourier transform domain fusion calculation:

[0261] If the iteration count exceeds N without convergence, multi-frame fusion is initiated to output the restored image. The specific process is as follows:

[0262] ;

[0263] in For Fourier transform operators, OTF is the optical transfer function (i.e., the Fourier transform of the point spread function PSF). express The conjugate of the complex number, μ is the regularization parameter (value 1×10). −6 ~1×10 −4 (to prevent the denominator from returning to zero).

[0264] Then, the validity of the multi-frame fusion result is verified by calculating the edge sharpness index of the fused image.

[0265] If Sedge≥Sth (Sth is the sharpness threshold, set by the workpiece feature accuracy requirements), then the fused image is output.

[0266] Finally, if multi-frame fusion still does not satisfy Sedge≥Sth, a device self-test command is triggered, automatically executing the following troubleshooting process:

[0267] Lens focal length calibration: The Z-axis focal length offset of the lens is calibrated using a standard gauge block, and automatic compensation is performed when the deviation exceeds ±0.05mm;

[0268] Diagnostic accelerometer: Detects the consistency of the response of the three-axis vibration signal in the 0.5-50Hz frequency band. A deviation exceeding 15% indicates a hardware fault.

[0269] The measurement process is paused and a fault code is output. The measurement will be restarted after manual inspection.

[0270] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0271] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0272] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. An image measurement method for precision dimensional measurement, characterized in that, Includes the following steps: Step 1: Select a laser scanner and an optical image measuring instrument to form a measuring array based on the workpiece's geometric characteristics; Step 2: Calibrate and verify the measurement array. Obtain the basic error curve of the equipment using standard gauge blocks. After verification, collect real-time environmental parameters. Step 3: Calculate the environmental interference coefficient K env When K env Measurement is initiated when the value is less than or equal to η, where η is the environmental interference coefficient threshold. The environmental interference coefficient calculation process includes: Temperature change ∆T and humidity change ∆H are obtained by temperature and humidity sensors, and spectrum data A(f) is collected by vibration sensor. Recall the device's pre-stored frequency response sensitivity curve W(f); Calculate the overall interference coefficient using the formula: ； Where α, β and γ are weighting coefficients, T0 is the reference temperature, H0 is the reference humidity, and fmin and fmax are the effective vibration frequency ranges; Step 4: Monitor the equipment's acceleration sensor data in real time and generate the equipment stability index Si; The process for obtaining the equipment stability index in step four is as follows: Triaxial acceleration data is collected in real time by an accelerometer integrated inside the measuring device. The accelerometer continuously monitors the vibration state of the device during operation at a preset sampling rate, and the sampling time interval is defined as the time differential element dt. Secondly, a bandpass filter is used to filter the collected triaxial acceleration data in the frequency domain, extract the effective vibration components within the preset frequency range, and eliminate low-frequency noise and high-frequency electromagnetic interference from the environment. Then, based on the time-domain vibration acceleration function a(t), the displacement deviation is calculated by quadratic integration: ； Statistical analysis was performed on the displacement deviation sequence in the time domain, and its variance was calculated. Combining the inherent sensitivity coefficient w of the equipment, a stability index is generated, specifically: ； Step 5: When Si > Ssafe, perform the measurement and output the result, where Ssafe is the device stability safety threshold; Step 6: Perform consistency analysis on the results from multiple devices. If the results are consistent, output the final value; otherwise, trigger dynamic recalibration.

2. The image measurement method for precision dimensional measurement according to claim 1, characterized in that: Image compensation is performed when Si ≤ Ssafe. The specific compensation process is as follows: Based on the displacement deviation δu calculated in step four, the displacement deviation δu in the x-direction is extracted from it. x Using the displacement deviation δuᵧ in the y-direction and the lens optical characteristic parameter λ, a point spread function model is constructed: ； This model is used to quantify the optical imaging diffusion effect caused by equipment vibration and serves as the core input parameter for subsequent deconvolution operations; Secondly, an improved deconvolution algorithm is used to restore the blurred image. The specific iterative calculation process is as follows: ； in Let Iobs be the image estimate for the k-th iteration, and Iobs be the observed blurred image. Then, an adaptive iteration stopping condition is introduced to avoid overfitting or underfitting: ； Where ε is the convergence threshold, the iteration terminates when this condition is met, and the clear image after compensation based on the PSF model is output.

3. The image measurement method for precision dimensional measurement according to claim 2, characterized in that: The dynamic recalibration process in step six includes: Based on the environmental interference coefficient K of each device env Calculate environmental adaptation factors Selecting environmental adaptation factors The largest device is used as the benchmark device; The reference device performs multi-angle scanning of the standard gauge block to generate a reference point cloud dataset P containing three-dimensional coordinate information. ref The device to be calibrated is synchronously controlled to scan the same standard gauge block from the same perspective, generating a calibration dataset P. test ; The spatial transformation matrix is ​​calculated using the iterative nearest point algorithm. The specific process is as follows: ； Where T(・) is a rigid transformation function used to transform the dataset P to be calibrated. test Transform the point cloud to the reference coordinate system using rotation matrix R and translation vector t, so that the transformed point cloud is consistent with P. ref The sum of squared Euclidean distances is minimized; Finally, based on the calculated transformation matrix T calib Update the internal calibration parameter matrix of the equipment to be calibrated to complete the one-time calibration of the equipment's spatial coordinate system.

4. The image measurement method for precision dimensional measurement according to claim 3, characterized in that: The specific process for performing consistency analysis on results from multiple devices is as follows: First, calculate the environmental adaptability factor of each device. The specific process is as follows: ,in The attenuation coefficient controls the sensitivity to environmental influences. The confidence weights are generated by combining environmental adaptation factors and historical equipment reliability. The specific process is as follows: ； Where Ri represents the historical reliability of device i, giving higher weight to devices that are highly adaptable to the environment and have a stable historical performance; Finally, the consistency of the measurement results is evaluated using a weighted consistency criterion formula. The specific process is as follows: ； Where Mi is the measurement result of device i; For the weighted average result, ; D th This is the consistency threshold, in the same unit as Mi.

5. The image measurement method for precision dimensional measurement according to claim 4, characterized in that: When consistency is not satisfied The following process will be executed at that time: choose The device corresponding to the maximum value is used as the reference device; The reference equipment performs a high-density scan of the key geometric feature points of the workpiece to generate a feature point cloud dataset containing sub-pixel level coordinate information. Generating subpixel-level feature templates based on encrypted scanned point cloud data; Non-reference devices optimize local parameters based on pixel-level feature templates. The deviation between the template and the measured feature points is fitted by least squares method, and the local parameters are iteratively corrected. The local parameters include lens distortion coefficients and coordinate system offset. After optimization, the consistency analysis process is re-executed.

6. The image measurement method for precision dimensional measurement according to claim 5, characterized in that: The output result process includes the following steps: A time decay factor is applied to the confidence weights verified by the consistency analysis. The specific calculation process is as follows: ,in Here, t is the time decay coefficient, and t is the test duration. The weighted fusion result is calculated based on the weights after time decay. The specific process is as follows: ； Where Mi represents the measurement result of device i, and weighted fusion is used to suppress the impact of abnormal data on the final result; Finally, output the results with confidence level annotations. The specific process is as follows: ； Where S min Let D be the minimum stability index and D be the consistency distance. This is the consistency attenuation coefficient.

7. The image measurement method for precision dimension measurement according to claim 6, characterized in that: If the convergence condition is not met after the preset threshold N iterations in the image compensation process. When this happens, execute the following failure handling procedure: The multi-frame fusion algorithm is activated to repair the blurred image. The specific process is as follows: Acquire K frames of images of the same scene {I1, I2, ..., I K The restored image is output through Fourier transform domain fusion calculation: If the iteration count exceeds N without convergence, multi-frame fusion is initiated to output the restored image. The specific process is as follows: ； in Here, represents the Fourier transform operator, and OTF represents the optical transfer function. express The conjugate of complex numbers, For regularization parameters; Then, the validity of the multi-frame fusion result is verified by calculating the edge sharpness index of the fused image: If Sedge≥Sth, then output the fused image, where Sth is the sharpness threshold; Finally, if the multi-frame fusion still does not satisfy Sedge≥Sth, a device self-test command is triggered.

Images