A lighting regulation decision-making method and system fusing multi-source perception and edge computing

Abstract

The application relates to the technical field of industrial intelligent control, and particularly discloses a lighting regulation and control decision method and system fusing multi-source sensing and edge computing, which comprises the following steps: constructing a lighting-biology-heat-vision-polarization-color temperature coupling constraint model in an edge node by deploying biological rhythm, polarized light field and suspended dust sensing nodes; based on multi-target optimization decoupling of six types of constraints, including frequency flash-shutter synchronization, thermal deformation inhibition, spectrum modulation, polarization maintenance, color temperature consistency and biological rhythm adaptation, generating a phase-locked pulse width modulation sequence, spectrum light mixing ratio and polarization-direction regulation parameters; collecting feedback in real time to build an efficiency evaluation matrix, and adaptively correcting the weight and online fine-tuning. The application solves the problems of polarization degradation, color temperature space phase confusion and frequency flash-shutter frequency mixing in the dust environment in the field of precision manufacturing, and realizes multi-target collaborative optimization of personnel health, equipment safety and visual detection quality.

Description

Technical Field

[0001] This invention relates to the field of industrial intelligent control technology, and in particular to a lighting control decision-making method and system that integrates multi-source sensing and edge computing. Background Technology

[0002] Current industrial lighting control technologies mainly rely on simple closed-loop control based on illuminance sensor feedback or time-based timing strategies, failing to deeply integrate multi-source heterogeneous data to achieve refined control. In submicron-level manufacturing scenarios such as semiconductor wafer inspection, precision optical assembly, and micro / nano fabrication, lighting systems are not only the basic guarantee for visual operations but also a key engineering element affecting measurement accuracy, product yield, and the long-term occupational health of workers.

[0003] However, existing technologies suffer from the following deep-seated technical shortcomings: First, they neglect the beat frequency effect between the rolling shutter of high-speed industrial cameras and the strobe drive of LED lighting. In high frame rate visual inspection, the asynchronous coupling between the lighting strobe frequency and the camera's line scanning frequency introduces periodic artifacts in the image, leading to misjudgments by defect detection algorithms. Second, they fail to consider the impact of the thermal radiation from high-power LED lighting arrays on the thermoelastic micro-deformation of the precision measurement base. Micrometer-level thermal deformation can cause submicrometer-level measurement inaccuracies, but existing controls lack thermo-mechanical coupling constraints. Third, they neglect the depolarization scattering effect of suspended dust in industrial environments on polarized lighting systems. Different dust materials have different attenuation mechanisms for polarization, and existing controls lack real-time perception of dust particle size, material, and spatial distribution, making them unable to dynamically adjust. Fourth, no active maintenance mechanism for color temperature spatial consistency within the same work surface has been established. Existing technologies only monitor the color temperature at a single point, ignoring the spatial differences in color temperature drift under multiple luminaire layouts, leading to biorhythmic phase disorder and visual cognitive load when personnel move their gaze across areas. Fifth, the real-time coupling between eye movement characteristics and lighting control has not been fully considered. Existing biorhythm monitoring relies solely on static pupil data and fails to utilize dynamic features such as fixation trajectory and saccade speed to predict visual fatigue accumulation. Sixth, the traditional centralized control architecture uploads all data to the cloud, resulting in large control delays. Furthermore, existing edge lighting control nodes operate independently, lacking the systematic decoupling capability for conflicts in multi-physical field constraints, and are unable to generate collaborative lighting strategies that balance personnel health, equipment safety, and work quality.

[0004] Therefore, there is an urgent need for a lighting control decision-making method that integrates multi-source sensing and edge computing to solve the long-neglected problem of synergistic optimization of optical-mechanical-biological multi-physics fields in the field of precision manufacturing. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by proposing a lighting control decision-making method and system that integrates multi-source sensing and edge computing.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: A lighting control decision-making method integrating multi-source sensing and edge computing includes: S1. Acquire illumination direction polarization degree and suspended dust data through polarization field sensing nodes, acquire gaze point trajectory through eye tracking device, acquire mid-temperature data through multispectral sensor network, construct illumination constraint model at edge computing nodes, and output illumination regulation constraint feature tensor; S2. Receive the illumination control constraint feature tensor, decouple the polarization degree and color temperature consistency constraint, and generate polarization-direction control parameters and spectral mixing ratio strategy; S3. The polarization-direction control parameters and spectral mixing ratio strategy are sent out for execution, and the actual polarization degree, actual color temperature and eye-tracking data are collected in real time to construct the lighting performance evaluation matrix and adaptively correct the feature fusion weights of the lighting constraint model.

[0007] As a further technical solution of the present invention, S1 specifically includes: S11. Collect melatonin secretion phase data and core body temperature rhythm data of workers through biorhythm sensing nodes; collect the rolling shutter scanning frequency and current frame exposure start time data of industrial cameras through high-speed visual sensing nodes; collect the thermoelastic micro-deformation and material thermal expansion coefficient data of key measurement bases through precision mechanical deformation sensing nodes; generate melatonin inhibition sensitivity curve, stroboscopic-shutter frequency avoidance phase constraint matrix, and thermal deformation compensation demand vector respectively. S12. Collect the multispectral reflectance fingerprint of the workpiece, collect the polarization attenuation in each direction and the material distribution of dust particle size, and construct the spectral identification weight vector and the three-dimensional optical field polarization degradation tensor of space-direction-material. S13. Collect eye-tracking feature data sets, collect multispectral color temperature difference sequences and color temperature drift rates, and construct spatial spectral degradation tensors; S14. Input all collected data into the lighting-biology-thermal-visual coupling constraint model, perform multi-physics spatiotemporal alignment and correlation fusion processing, and output the lighting regulation constraint feature tensor representing the coupling relationship between flicker-shutter frequency avoidance requirements, thermal deformation suppression requirements, spectral fingerprint modulation requirements, polarization degree preservation requirements, color temperature spatial consistency requirements, and biological rhythm adaptation requirements.

[0008] As a further technical solution of the present invention, S12 specifically includes: S121. Through the material surface sensing node, the surface of the workpiece under test is sequentially illuminated under multiple preset multispectral illumination conditions, and the surface reflection intensity value corresponding to each spectral band is collected to form the original reflection spectrum curve; the original reflection spectrum curve is normalized and the peak wavelength position and half width at half maximum feature are extracted to construct the spectral identification weight vector of the target material, wherein the weight value of each spectral band is positively correlated with the reflection intensity distinction between different materials under that band; S122. Through the polarization field sensing node, collect the real-time polarization degree values ​​of each lighting actuator in multiple light output directions. In the formula: For linear polarization degree, The polarization degree component is represented by a circular polarization degree component, and its difference value is calculated with the initial polarization degree value of the equipment in a clean state to generate a polarization degree attenuation sequence for each light output direction, i.e.: In the formula: This is the degree of polarization attenuation. For the first Initial polarization values ​​in each direction, For the first Real-time polarization degree values ​​in each direction; S123. Using a suspended dust concentration sensor, the dust particle size distribution and dust material classification signals at multiple spatial locations are simultaneously acquired using the multi-angle polarization scattering principle. The dust material classification signals can distinguish at least three types of dust: iron filings, aluminum powder, and copper dust. S124. Using the spatial grid coordinates of the industrial lighting area as the first dimension, the light output direction of the lighting device as the second dimension, and the dust material type as the third dimension, the polarization attenuation sequence generated in S122 is correlated and mapped with the dust particle size distribution and dust material classification signals generated in S123. The polarization attenuation rate of each spatial grid position facing different dust materials in each light output direction and the polarization scattering cross section coefficient corresponding to each dust material are stored to complete the construction of the light field polarization degradation tensor.

[0009] As a further technical solution of the present invention, S13 specifically includes: S131. Continuously collect eye movement feature data sets of the current worker using an eye-tracking device. These data sets include pupil diameter change sequences, fixation point trajectory, blink frequency, and saccade speed. Perform sliding window filtering on the pupil diameter change sequences to extract the pupil's response delay time to light reflection and steady-state contraction amplitude, which are used as initial feature parameters for visual fatigue assessment. In the formula: In response to the delay time, This is the moment when the pupil diameter begins to continuously contract. This refers to the moment of sudden change in light intensity; In the formula: For steady-state contraction amplitude, To determine the stable pupil diameter before constriction, This represents the stable pupil diameter after constriction. S132. Through a multispectral sensor network, the actual color temperature values ​​at multiple spatial grid locations within the industrial lighting area are simultaneously acquired. The instantaneous deviation between the color temperature value of each spatial grid and the average color temperature of all grids within the same working surface is calculated to generate a color temperature difference sequence. At the same time, the rate of change of the color temperature value of each lighting device in different light output directions over time is recorded to generate a color temperature drift rate sequence. S133. Perform spatiotemporal correlation analysis on the color temperature difference sequence and the color temperature drift rate sequence of each lighting device to identify the main contributing sources of color temperature spatial inconsistency. When the color temperature drift rate of a certain lighting device in a specific light output direction exceeds the threshold of the adjacent device, it is marked as the dominant source of color temperature deviation. S134. Using the spatial coordinates of the industrial lighting area as the first dimension and the light output direction of the lighting device as the second dimension, the color temperature difference sequence, color temperature drift rate sequence, and color temperature deviation dominant source marker are associated and stored to construct a spatial spectral degradation tensor. The color temperature attenuation coefficient of each element in the spatial spectral degradation tensor is positively correlated with the aging degree of the corresponding device and the amount of dust deposition in the light output direction.

[0010] As a further technical solution of the present invention, S2 specifically includes: S21. Extract the stroboscopic-shutter shooting frequency avoidance requirements and shutter scanning frequency, constrained by integer multiples of the scanning frequency, and phase-locked with the exposure start time to generate a stroboscopic-shutter phase-locked pulse width modulation coding sequence. S22. Extract the requirements for thermal deformation suppression and spectral fingerprint modulation. With the goal of maximizing the thermal deformation threshold and spectral discriminability, and combined with minimizing the melatonin suppression rate, use multi-objective particle swarm optimization to solve the spectral power distribution curve and the mixing ratio strategy. S23. Extract the polarization degree retention requirements and color temperature spatial consistency requirements. Calculate the predicted polarization degree of each spatial grid under the current dust concentration based on the light field polarization degradation tensor. With the predicted polarization degree of each grid not lower than the preset lower limit and the maximum color temperature deviation within the same working surface not exceeding the preset consistency threshold as hard constraints, solve the target polarization output angle and target light output direction deflection angle of each lighting execution device to generate polarization-direction control parameters.

[0011] As a further technical solution of the present invention, S21 specifically includes: S211. Extract the flicker-shutter shooting frequency avoidance requirement from the lighting control constraint feature tensor, and simultaneously obtain the rolling shutter scanning frequency and the current frame exposure start time acquired by the high-speed visual perception node. S212. Use positive integer multiples of the rolling shutter scanning frequency as the candidate stroboscopic fundamental frequency set, and calculate the exposure time difference between two adjacent rows of pixels at each candidate fundamental frequency. With lighting fluctuation cycle ratio The fundamental frequency whose ratio is closest to an integer is selected as the optimal stroboscopic fundamental frequency; S213. Align the rising edge of the pulse width modulation coding sequence to the exposure start time of the current frame, and determine the period length and initial duty cycle of the coding sequence according to the optimal stroboscopic base frequency to generate a stroboscopic-shutter phase-locked pulse width modulation coding sequence. S214. Synchronize and verify the generated encoded sequence with the shutter scanning frequency, and calculate the frequency deviation rate between the fundamental harmonic component of the encoded sequence and the shutter scanning frequency. In the formula: The actual output fundamental frequency of the encoded sequence. For optimal harmonic coefficients, The shutter scanning frequency is used; when the frequency deviation rate exceeds the preset synchronization tolerance threshold, the phase offset of the encoded sequence is automatically adjusted until the deviation rate is lower than the threshold.

[0012] As a further technical solution of the present invention, S22 specifically includes: S221. Extract the key measurement base thermal deformation threshold in the thermal deformation suppression requirement and the spectral identification weight vector of the target material in the spectral fingerprint modulation requirement from the lighting control constraint feature tensor, and extract the upper limit of melatonin inhibition rate in the photobiological safety constraint. S222. Initialize the spectral power distribution particle swarm of the multi-channel LED. The position vector of each particle is composed of the peak wavelength, half width at half maximum and driving current amplitude of each channel LED. The velocity vector of the particle controls the iteration step size of each parameter. S223. A multi-objective fitness function is constructed with three parallel optimization objectives: minimizing the predicted value of base thermal deformation, maximizing the weighted spectral fingerprint recognition, and minimizing the predicted value of melatonin inhibition rate. The Pareto dominance relation is used to update the historical optimal position and the global optimal position of the particle swarm, and the non-dominated solution set is solved iteratively. S224. Select the solution closest to the ideal point from the Pareto front, output the corresponding multi-channel LED spectral power distribution curve and the mixing ratio strategy of each channel, and record the thermal deformation prediction value, spectral fingerprint identification and melatonin suppression rate under the solution for deviation comparison in step S3.

[0013] As a further technical solution of the present invention, S3 specifically includes: S31. The strobe-shutter phase-locked pulse width modulation coding sequence, multi-channel LED spectral mixing ratio strategy, polarization-direction control parameters and thermal radiation power upper limit constraint parameters generated in step S2 are sent to the lighting drive execution unit to trigger each lighting execution device to perform dimming, color adjustment, polarization adjustment and light direction adjustment operations. S32. Real-time acquisition of the flicker artifact rate of industrial camera output images, real-time micro-variables of precision measurement base, confidence level of material surface spectral fingerprint recognition algorithm, actual polarization feedback value and actual color temperature feedback value of each spatial grid, as well as actual eye movement feature data set and actual melatonin secretion phase change data of operators; S33. Compare the various feedback data collected in S32 with the corresponding target values ​​in the lighting control constraint feature tensor output in step S1 to calculate the flicker artifact rate deviation, micro-variable deviation, spectral fingerprint confidence deviation, polarization deviation, color temperature deviation and eye-tracking feature deviation, and construct the lighting multi-physics coupling effectiveness evaluation matrix. S34. Based on the evaluation matrix bias, gradient descent is used to correct the multi-physics feature fusion weight coefficients, while fine-tuning the phase shift of the coding sequence, the duty cycle of the light mixing ratio, and the polarization output angle.

[0014] As a further technical solution of the present invention, S34 specifically includes: S341. Calculate the cumulative error integral of each deviation value in the lighting multiphysics coupling effectiveness evaluation matrix within a preset time window. In the formula: Indicates the preset time window length. Indicates time No. The deviation values ​​of each dimension are updated using the gradient descent method to update the multiphysics feature fusion weight coefficients of the lighting-biology-thermal-visual coupling constraint model in step S1. The update step size of the weight coefficients is proportional to the cumulative error integral of the corresponding deviation value, so that the perception dimension that produces a large deviation receives a higher weight correction magnitude. S342. Based on the sign and amplitude of the stroboscopic artifact rate deviation, adjust the phase offset of the stroboscopic-shutter phase-locked pulse width modulation coding sequence in the reverse direction. If the sign is positive, it indicates that the artifact rate is too high. Then, finely adjust the phase offset by one step unit in the positive direction until the deviation converges. S343. Based on the weighted sum of the spectral fingerprint confidence deviation and the color temperature deviation, adjust the duty cycle of each channel in the multi-channel LED spectral mixing strategy. When the confidence deviation is dominant, prioritize optimizing the spectral shape; when the color temperature deviation is dominant, prioritize optimizing the white balance point. At the same time, based on the sign of the polarization deviation, fine-tune the polarization output angle in the polarization-direction control parameters to form a closed-loop correction.

[0015] A lighting control decision system integrating multi-source sensing and edge computing is provided to implement a lighting control decision method that integrates multi-source sensing and edge computing. The system includes a multi-source heterogeneous sensing module, edge computing nodes, a lighting driver execution unit, and a cloud management platform, wherein: The multi-source heterogeneous sensing module includes a biological rhythm sensing node, a high-speed visual sensing node, a precision mechanical deformation sensing node, a material surface sensing node, a polarized light field sensing node, an eye-tracking device, and a multispectral sensor network. These are used to collect data on the melatonin secretion phase and core body temperature rhythm of workers, the rolling shutter scanning frequency and exposure start time data of industrial cameras, the thermoelastic micro-deformation and thermal expansion coefficient data of key measurement bases, the surface reflectance spectral fingerprint data of the workpiece under test, the multi-directional polarization attenuation sequence and dust particle size distribution and material classification signals of various lighting actuators, the eye movement feature data set of workers, and the color temperature difference sequence and color temperature drift rate of various spatial locations. An edge computing node, connected to the multi-source heterogeneous sensing module, includes: The coupling constraint model construction module is used to perform spatiotemporal alignment and multiphysics field fusion on the above heterogeneous data, and output lighting control constraint feature tensors that characterize the requirements for stroboscopic-shutter frequency avoidance, thermal deformation suppression, spectral fingerprint modulation, polarization degree preservation, color temperature consistency and biological rhythm adaptation. The multiphysics optimization solution module is used to decouple multiple constraints and solve the spectral power distribution and driving waveform that satisfy shutter synchronization conditions, thermal deformation threshold, maximization of spectral discriminability, lower limit of polarization degree, upper limit of color temperature difference and minimum melatonin suppression rate. It generates strobe-shutter phase-locked PWM encoding sequence, spectral mixing ratio strategy, polarization-direction control parameters and thermal radiation upper limit constraint parameters. The performance verification and weight evolution module is used to send control parameters to the lighting drive execution unit, collect flicker artifact rate, micro-variables, spectral recognition confidence, actual polarization degree, actual color temperature, eye movement characteristics and melatonin phase change data in real time, construct a multi-physics field coupling performance evaluation matrix, adaptively correct the fusion weight coefficient of the coupling constraint model based on deviation analysis, and fine-tune the PWM encoding sequence phase shift, light mixing ratio duty cycle and polarization output angle online. The lighting drive execution unit is connected to the edge computing node and is used to receive and execute the control parameters to drive the lighting execution device to perform dimming, color adjustment, polarization adjustment and light direction adjustment operations. The cloud management platform communicates with the edge computing nodes to receive uploaded data and perform incremental training on the multiphysics optimization algorithm, and then sends the updated parameters to the edge computing nodes. The edge computing nodes and the cloud management platform use a mechanism to resume transmission after network interruption.

[0016] The beneficial effects of this invention are as follows: 1. By integrating multi-source data such as biological rhythms, polarized light fields, suspended dust, and eye tracking, a six-dimensional coupled constraint model of lighting-biology-thermal-vision-polarization-color temperature is constructed at the edge. This model breaks through the limitation of traditional lighting that only focuses on illuminance and solves the rare technical problems such as polarization degradation in dust environments, color temperature spatial phase disorder, and stroboscopic-shutter frequency, enabling real-time multi-physics collaborative perception for submicron precision manufacturing.

[0017] 2. Based on multi-objective particle swarm optimization and sequential quadratic programming algorithms, the stroboscopic-shutter phase locking, thermal deformation suppression, spectral fingerprint modulation, polarization degree preservation, and color temperature consistency hard constraints are simultaneously decoupled to generate differentiated spectral mixing and polarization-direction control strategies. This breaks through the traditional paradigm of single energy-saving objectives and achieves multi-objective Pareto optimality of personnel biological rhythms, equipment thermal stability, and visual inspection quality.

[0018] 3. Construct a multi-physics coupling performance evaluation matrix for lighting, dynamically correct feature fusion weights through cumulative error integral gradient descent, and fine-tune phase shift, mixing duty cycle and polarization angle online to form a closed-loop adaptive evolution mechanism from perception, decision-making to execution, enabling the lighting system to have the ability to self-optimize based on physiological stress, dust concentration and detection quality feedback. Attached Figure Description

[0019] Figure 1 This is a flowchart of the method proposed in Embodiment 1 of the present invention; Figure 2 This is an architectural diagram of the lighting-biology-thermal-visual coupling constraint model in Embodiment 1 of the present invention; Figure 3 This is a framework diagram of the system proposed in Embodiment 2 of the present invention; Figure 4 This is a comparison diagram of the effects of the prior art and the present invention in Embodiment 3 of the present invention. Detailed Implementation

[0020] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0021] Example 1 Please see the appendix Figure 1 - Appendix Figure 2 A lighting control decision-making method integrating multi-source sensing and edge computing, comprising: S1. Acquire illumination direction polarization degree and suspended dust data through polarization field sensing nodes, acquire gaze point trajectory through eye-tracking devices, acquire mid-temperature data through multispectral sensor networks, construct an illumination constraint model at edge computing nodes, and output illumination regulation constraint feature tensors; specifically including: S11. Multi-source physiological and device state perception and initial constraint parameter generation, specifically: S111. By deploying biorhythm sensing nodes worn by workers at the precision manufacturing site, melatonin secretion phase data and core body temperature rhythm data are continuously collected; the node has a built-in non-invasive salivary melatonin concentration detection sensor and body surface temperature sensor to collect physiological signals and record timestamps at preset time intervals.

[0022] The collected melatonin secretion phase data were smoothed using spline interpolation. A circadian rhythm phase baseline was established by combining this baseline with the trough times of the core body temperature rhythm. The phase shift of the current time relative to this baseline was then calculated. ,Right now: In the formula: For the current moment, The core body temperature rhythm trough time is used as the reference point; combined with a preset melatonin inhibition spectral sensitivity weighting coefficient, a melatonin inhibition sensitivity curve is generated, where: the melatonin inhibition sensitivity weighting coefficient... for: In the formula: As a baseline sensitivity, This is the phase-sensitive adjustment coefficient.

[0023] S112. Acquire the rolling shutter scanning frequency and current frame exposure start time data of the industrial camera through a high-speed visual sensing node; the node establishes microsecond-level clock synchronization with the industrial camera through a hardware synchronization signal line or the IEEE1588 precise time protocol, and obtains the camera's line scanning frequency parameters and the absolute timestamp of the exposure start of each frame in real time.

[0024] The range of sensitive frequency bands for stroboscopic shooting frequency is calculated based on the scanning frequency of the rolling shutter, namely: In the formula: The beat frequency, LED driving frequency, The scanning frequency of the rolling shutter. The frequency multiplication factor is an integer, with classic values ​​of n = 1 (fundamental frequency synchronization mode, suitable for low frame rate detection scenarios) or n = 2 (second harmonic synchronization mode, suitable for high frame rate scenarios where power frequency interference needs to be avoided); a phase constraint matrix for stroboscopic-shutter frequency avoidance is established by combining the exposure start time data. The dimension of this matrix corresponds to the phase difference space between the camera scan line period and the illumination drive period.

[0025] S113. Acquire data on the thermoelastic micro-deformation and material thermal expansion coefficient of the key measurement base through a precision mechanical deformation sensing node; the node is equipped with fiber Bragg grating strain sensors or laser interferometric displacement sensors at key structural points of the measurement base to monitor micron-level structural deformation in real time.

[0026] Simultaneously, the thermal expansion coefficient of the measuring base material is retrieved from the equipment material database. Combined with historical data on current ambient temperature and lighting thermal radiation power, a thermal deformation compensation demand vector is calculated using a thermoelastic mechanical model. The components of this vector represent the expected deformation of each structural point under a specific heat load and the corresponding reduction demand for lighting thermal radiation power. Specifically, the thermoelastic micro-deformation... for: In the formula: The original length of the base For the temperature rise, The coefficient of thermal expansion of the material is 1.0 × 10⁻⁶. −6 ℃ −1 Up to 25×10 −6 ℃ −1 ; Demand for reduction of thermal radiation power for: In the formula: This represents the deformation-to-power conversion factor, ranging from 0.5 W / μm to 5.0 W / μm, with the classic value being γ = 2.0 W / μm (for a power density of 100 W / m). 2 (A typical LED array lighting configuration with a distance of 0.5m, with a conservative compensation coefficient corresponding to submicron level measurement accuracy requirements).

[0027] S12. Multispectral and polarization light field sensing and construction of light field degradation tensor, specifically: S121. Material surface spectral fingerprint acquisition and recognition weight vector construction, specifically: A tunable multispectral illumination array is controlled by a material surface sensing node. This array consists of multiple independently controllable narrowband LED light sources with wavelengths covering the visible to near-infrared bands. The node activates each narrowband LED light source sequentially according to a preset wavelength sequence, and the illumination duration of each wavelength ensures that the light response of the workpiece surface reaches a steady state.

[0028] A synchronously triggered spectral sensor acquires the reflection intensity values ​​of the workpiece surface under illumination of various wavelengths, forming a discretized original reflection spectrum curve. The original reflection spectrum curve is then normalized to eliminate amplitude differences caused by fluctuations in light source intensity and changes in measurement distance. Specifically, the normalized reflection intensity is calculated as follows: In the formula: For the first Normalized reflection intensity of each wavelength, The intensity of light reflected from the collected sample. The light intensity is for reference to a standard whiteboard. This represents the background light intensity of the dark current. Indicates the first Each wavelength sampling point.

[0029] The peak wavelength position and full width at half maximum (FWHM) feature of the normalized curve are extracted. The reflectance intensity discrimination between the target material and the background interference material in each spectral band is calculated. Based on the discrimination value, a spectral discrimination weight vector is constructed. The weight value of each band is positively correlated with the material discrimination ability, that is: ; In the formula: For the first The weight value of each band, For the target material in the first Normalized reflection intensity of each band, The background material is in the first Normalized reflection intensity of each band, This represents the total number of bands.

[0030] S122. Multi-directional sensing of polarized light field and generation of polarization degree attenuation sequence, specifically: A rotatable polarization detection unit is arranged in the light output path of each lighting device through a polarization field sensing node. This unit consists of a precision rotating bracket combined with a linear polarizer and a quarter-wave plate. The node controls the detection unit to sequentially measure the polarization state of the transmitted light at multiple preset light output angle positions. By rotating the linear polarizer, it collects the extreme values ​​of light intensity at different angles, calculates the linear polarization degree and circular polarization degree components in each light output direction, and synthesizes the real-time polarization degree value in that direction. Specifically, the linear polarization degree... for: In the formula: This represents the maximum light intensity value measured by the rotating polarizer. Minimum light intensity value; circular polarization component for: In the formula: The light intensity of the right-hand circularly polarized component is the light intensity value measured when the fast axis of the quarter-wave plate is aligned with the horizontal direction. The intensity of the left-hand circularly polarized component is the light intensity measured when the fast axis of the quarter-wave plate is aligned vertically; real-time degree of polarization value. for: .

[0031] The system retrieves the initial polarization values ​​for each light emission direction under clean conditions stored during the device's factory calibration. It then performs a difference calculation between the real-time polarization value and the initial polarization value for the corresponding direction to obtain the polarization attenuation in each direction. ,Right now: In the formula: For the first Initial polarization values ​​in each direction, For the first Real-time polarization degree values ​​in each direction; label each output light direction with its corresponding polarization degree attenuation. Arrange them in an orderly manner to generate a polarization attenuation sequence.

[0032] S123. Multi-parameter sensing and material classification of suspended dust, specifically: Sampling probes are deployed at multiple spatial locations within an industrial lighting area using a suspended dust concentration sensor. Each probe has a built-in multi-angle polarized light emitter and receiver. The sensor emits a detection beam with a specific polarization state into the detection area. Dust particles scatter the beam, and the receiver captures the polarization intensity distribution signal at different scattering angles.

[0033] Based on the analysis of the polarization state change and angular distribution characteristics of scattered light using Mie scattering theory, the particle size distribution of dust at each spatial location is calculated by inversion, yielding the volume concentration percentage of different particle size ranges. Specifically, the particle size distribution inversion calculation is as follows: In the formula: For particle size The number density of dust particles, To at the scattering angle The intensity of scattered light measured at that location, The scattering coefficient function For the first Particle size classification The complex refractive index of the dust.

[0034] Based on the differences in the real and imaginary parts of the complex refractive index of iron filings, aluminum powder, and copper dust, a material classification discriminant function is established to identify and classify the dust types in the detection area, generating a dust material classification signal. Specifically, the material classification discriminant calculation is as follows: ; In the formula: For the classification results, For material type index, The intensity of vertically polarized scattered light. The intensity of horizontally polarized scattered light. For material The theoretical polarization ratio, This represents the number of scattering angles.

[0035] S124. Construction of 3D Optical Field Polarization Degradation Tensor and Data Association Mapping Using the spatial grid coordinates of the industrial lighting area as the first dimension, a discrete spatial coordinate system covering the entire lighting operation area is established; using the light output direction of each lighting device as the second dimension, a discrete direction index set is established; using the dust material type identified in S123 as the third dimension, a discrete material type index set is established. The polarization attenuation sequence generated in S122 is three-dimensionally correlated and mapped with the dust particle size distribution and dust material classification signals generated in S123. The polarization attenuation rate of each spatial grid position facing a specific dust material in each light output direction is calculated, i.e.: In the formula: For the first The spatial grid in the ... Facing the material from each light-emitting direction The rate of polarization attenuation at that time For the first The first grid The degree of polarization attenuation in each direction, For the first Material at each grid point Dust concentration.

[0036] Simultaneously, the polarization scattering cross-sectional coefficients corresponding to each dust material are stored to construct a three-dimensional optical field polarization degradation tensor of space-direction-material. This tensor comprehensively characterizes the degradation effect of the dust environment on the polarized illumination light field, namely: In the formula: Tensor 1 The element value at position, Indicates material The polarization scattering cross-section coefficient ranges from 5 to 50 m. 2 / kg.

[0037] S13. Eye tracking, color temperature monitoring, and construction of the spatial spectral degradation tensor, specifically: S131. Continuous acquisition of eye-tracking features and extraction of initial features for visual fatigue, specifically: The eye-tracking device continuously illuminates the eye area with a built-in near-infrared light source and captures a sequence of eye images at a fixed frame rate using a high-speed image sensor. For each frame, pupil contour detection and corneal reflective spot localization are performed, and the pupil center coordinates and pupil diameter are calculated to generate a sequence of pupil diameter changes. By tracking the displacement of the corneal reflective spot relative to the pupil center, the direction of gaze is calculated to generate the fixation point trajectory. The eyelid opening and closing state is detected, and the number of times the eyelids completely close per unit time is counted to generate the blink frequency. The ratio of the displacement to the time interval between consecutive fixation points is calculated to generate the saccade speed.

[0038] A sliding window filter is applied to the pupil diameter variation sequence, with the window length covering multiple frame periods. The arithmetic mean of the pupil diameter within the window is calculated as the baseline value, i.e.: In the formula: For the first The average pupil diameter of each window. The number of sampling points contained in the sliding window. For the first The original pupil diameter value at each sampling time is used; the standard deviation of the pupil diameter within the window is calculated as a fluctuation index to eliminate transient spikes caused by blinking and sensor noise interference.

[0039] Extract the pupil response delay time to light reflection from the filtered pupil diameter change curve; that is, the time interval from the moment of sudden change in light intensity to the start of sustained pupil diameter contraction. In the formula: In response to the delay time, This is the moment when the pupil diameter begins to continuously contract. The moment of sudden change in light intensity; extract the steady-state contraction amplitude, that is, the absolute change in pupil diameter from the initial stable value to the new steady state, i.e.: In the formula: For steady-state contraction amplitude, To determine the stable pupil diameter before constriction, The stable pupil diameter after constriction; the response delay time With steady-state contraction amplitude As initial characteristic parameters for visual fatigue assessment.

[0040] S132. Multi-space color temperature synchronous acquisition and differential drift sequence generation, specifically: Multiple color temperature sensing nodes are arranged in a uniform spatial grid within an industrial lighting area using a multispectral sensor network. Each node has a built-in spectral decomposition element and photoelectric detection array. The network triggers all nodes to collect the actual color temperature value at their respective spatial grid positions at the same time through a time synchronization protocol, and records the spatial coordinate position of each node.

[0041] Calculate the arithmetic mean of the color temperature values ​​of all spatial grids within the same working surface as the spatial reference color temperature. Then, calculate the difference between the actual color temperature value of each spatial grid and the spatial reference color temperature, i.e.: In the formula: For the first Color temperature difference values ​​of each spatial grid. For the first The actual color temperature value of each grid cell. The total number of spatial grids, For the first The actual color temperature values ​​of each grid are calculated; the differences between the grids are arranged in spatial coordinate order to generate a color temperature difference sequence.

[0042] For each lighting device, color temperature monitoring points are deployed in multiple light-emitting directions. The color temperature value of each monitoring point at the current moment is recorded with respect to the color temperature value at the previous moment at a fixed sampling period. The ratio of the change in color temperature value between adjacent moments to the time interval is calculated to generate a color temperature drift rate sequence, i.e.: In the formula: Color temperature drift rate, The color temperature value at the current moment. The color temperature value at the previous moment. This represents the sampling time interval.

[0043] S133. Spatiotemporal correlation analysis and identification of the dominant source of color temperature deviation, specifically: A spatiotemporal correlation analysis is performed between the color temperature difference sequence and the color temperature drift rate sequence of each lighting actuator. A sliding analysis window of fixed time length is established. Within each time window, the color temperature difference value of each spatial grid is compared with the color temperature drift rate of the lighting actuators near that grid in each light emission direction. The dependence weight of color temperature spatial inconsistency on each lighting actuator is calculated. When the absolute value of the color temperature drift rate of a certain lighting actuator in a specific light emission direction exceeds a preset proportional threshold of the average drift rate of adjacent devices in that direction, the device is determined to have inferior color temperature stability in that direction compared to surrounding devices. It is marked as the dominant source of color temperature deviation, and its device identifier, light emission direction identifier, and the time interval exceeding the threshold are recorded. Specifically, the determination of the dominant source of deviation is calculated as follows: In the formula: For the first The lighting actuator in the first The drift deviation index in the direction of light emission. This refers to the color temperature drift rate of the device in this direction. The preset proportional threshold coefficient ranges from 1.2 to 2.5, suitable for typical monitoring scenarios. Taking 1.5, for precision manufacturing scenarios with high stability requirements. Take 2.0, This represents the number of adjacent devices. For the first The adjacent devices in the first Color temperature drift rate in each direction; when It was determined to be the dominant source at that time.

[0044] S134. Two-dimensional construction and data association storage of spatial spectral degradation tensor, specifically: Using the spatial grid coordinates of the industrial lighting area as the first dimension index, a discrete spatial coordinate system covering the entire lighting operation area is established. Using the light emission direction of each lighting device as the second dimension index, a discrete directional index set is established. The color temperature difference sequence and color temperature drift rate sequence generated in S132, along with the dominant color temperature deviation source identifier marked in S133, are stored in a two-dimensional associative mapping according to the correspondence between spatial coordinates and light emission direction. A spatial spectral degradation tensor is constructed, where each element represents the color temperature attenuation coefficient at a specific spatial location in a specific light emission direction. This coefficient is determined by a weighted average of the absolute value of the color temperature difference at the corresponding location, the absolute value of the color temperature drift rate in that direction, and the dominant color temperature deviation source identifier. It is also positively correlated with the aging degree of the corresponding lighting device and the amount of dust deposition in the light emission direction. Specifically, the color temperature attenuation coefficient... for: In the formula: For the first The spatial grid in the ... Color temperature attenuation coefficient in each light-emitting direction, The color temperature difference weighting coefficient is used. This is the color temperature drift weighting coefficient. As the primary source, the weighting coefficient is used during routine quality inspections. =0.3, =0.3, =0.4, Precision Vision Inspection Station =0.6, =0.2, =0.2, long-term continuous production scenario =0.2, =0.6, =0.2, when equipment maintenance warning is issued =0.2, =0.2, =0.6; For the first The color temperature difference values ​​of each grid. For the first The grid corresponding device in the first Color temperature shift rate in each direction, This is the dominant source label value. It is 1 when the position and direction are labeled as the dominant source, and 0 otherwise.

[0045] S14. Multiphysics data fusion and constraint feature tensor output, specifically: S141. The melatonin suppression sensitivity curve, stroboscopic-shutter frequency avoidance phase constraint matrix, thermal deformation compensation requirement vector generated in S11, the spectral discrimination weight vector and spatial-direction-material three-dimensional light field polarization degradation tensor constructed in S12, and the spatial spectral degradation tensor and initial feature parameters for visual fatigue assessment constructed in S13 are jointly input into the lighting-biology-thermal-visual coupling constraint model deployed on the edge computing node. This model consists of a spatiotemporal alignment engine, a multiphysics coupling calculator, and a constraint feature encoder, which are cascaded together.

[0046] S142. The spatiotemporal alignment engine uses the absolute timestamps recorded when each sensing node collects data. It employs a timestamp synchronization algorithm to align heterogeneous data with different sampling frequencies to a unified time base. It also employs a spatial coordinate transformation algorithm to map personnel positions, equipment coordinates, and lighting area grid coordinates to a unified industrial site digital twin coordinate system.

[0047] S143. Multiphysics Coupling Calculator performs multiphysics correlation and fusion processing: By correlating circadian rhythm data with color temperature spatial distribution data, a mapping relationship between melatonin inhibition sensitivity and local color temperature values ​​is established. Specifically, this relates to the intensity of circadian rhythm adaptation requirements. for: In the formula: This represents the weighting coefficient for melatonin inhibition sensitivity, ranging from 0.3 to 1.0, applicable to 24-hour shift work scenarios. =0.8, high-precision visual inspection task in short time period =0.4; This represents the visual fatigue response weighting coefficient, ranging from 0.2 to 0.8, for continuous work lasting 8 hours or more. =0.6, intermittent operation or automated inspection position =0.3; This indicates melatonin inhibition sensitivity. This indicates the delay time in the pupil's response to light reflection. This represents the baseline value for response delay under normal physiological conditions.

[0048] By associating the stroboscopic-shutter phase constraint with the illumination drive parameters, a mapping relationship between the phase-locked constraint and the drive frequency is established. Specifically, the intensity of the stroboscopic-shutter synchronization requirement... for: In the formula: This represents the allowable width of the phase deviation range for the phase constraint matrix.

[0049] By linking the thermal deformation compensation requirement with the lighting thermal radiation power, a thermo-mechanical transfer function is established. Specifically, the intensity of the thermal deformation suppression requirement is... for: In the formula: The maximum component value of the thermal deformation compensation demand vector. The allowable deformation threshold for precise measurement is set in the range of 0.01μm–10μm.

[0050] By associating spectral discrimination weights with multispectral illumination parameters, a mapping relationship between material identification requirements and spectral power distribution is established. Specifically, the intensity of spectral fingerprint modulation requirements is... for: In the formula: For the first Spectral discrimination weights for each spectral band For the first Normalized reflection intensity of each band, This represents the total number of spectral bands.

[0051] By correlating the optical field polarization degradation tensor with polarized illumination parameters, a mapping relationship is established between the polarization preservation requirements in dusty environments and the polarization output angle. Specifically, the polarization degree preservation requirement intensity... for: In the formula: For the first The spatial grid number The first direction of light output The three-dimensional light field polarization degradation tensor element values ​​of the material. For material The polarization scattering cross-sectional coefficient, For the first Material at each spatial grid Dust concentration.

[0052] By correlating the spatial spectral degradation tensor with color temperature consistency control parameters, a mapping relationship is established between the dominant sources of color temperature deviation and color temperature compensation strategies. Specifically, the intensity of color temperature spatial consistency requirements is determined. for: In the formula: For the first The spatial grid number Color temperature attenuation coefficient in each light-emitting direction, The dominant source penalty coefficient for color temperature deviation ranges from 1.0 to 5.0, with a mild penalty mode of γ=1.5. Standard penalty mode γ=3.0, severe penalty mode γ=5.0; This is the dominant source marker value for color temperature deviation. It is 1 when this position and direction are marked as the dominant source, and 0 otherwise.

[0053] S144. The constraint feature encoder normalizes and encodes the six-dimensional demand intensity into an illumination regulation constraint feature tensor, namely: In the formula: The first characteristic tensor representing the lighting control constraint The element values ​​of each dimension, where It covers six dimensions: biological rhythm, strobe synchronization, thermal deformation, spectral fingerprint, polarization preservation, and color temperature consistency. Indicates the first Demand intensity in each dimension; Indicates the first Demand intensity in each dimension, among which Iterate through all six dimensions from 1 to 6; This represents the overall constraint budget strength, used to normalize and scale the overall amplitude of the lighting control constraint feature tensor. The dimensions of this tensor correspond to flicker-shutter frequency avoidance requirements, thermal deformation suppression requirements, spectral fingerprint modulation requirements, polarization degree preservation requirements, color temperature space consistency requirements, and biological rhythm adaptation requirements, respectively. The numerical values ​​in each dimension represent the strength level and priority weight of the corresponding constraint conditions, forming the input constraint conditions for subsequent lighting control decisions.

[0054] S2. Receive the illumination control constraint feature tensor, decouple the polarization degree and color temperature consistency constraints, and generate polarization-direction control parameters and spectral mixing ratio strategies; specifically including: S21. Generation of stroboscopic-shutter-frequency avoidance and phase-locked coding sequence, specifically: S211. Extraction of flicker avoidance requirements and acquisition of high-speed visual parameters, specifically: From the lighting control constraint feature tensor storage area deployed on the edge computing node, dimensional data values ​​characterizing the intensity of flicker-shutter frequency avoidance requirements are extracted. Simultaneously, the rolling shutter scanning frequency parameter values ​​stored in the industrial camera's internal registers are read in real-time via a hardware synchronization signal line or precision time protocol link between the high-speed vision sensing node and the industrial camera. The absolute timestamp value of the current frame's exposure start moment is obtained through a timestamp capture module. This timestamp is transmitted from the industrial camera to the high-speed vision sensing node via a hardware trigger signal at the moment of exposure start, ensuring microsecond-level time synchronization accuracy. The extracted flicker avoidance requirement intensity value, rolling shutter scanning frequency value, and the current frame's exposure start moment timestamp are temporarily stored in a high-speed cache, providing input data sources for subsequent optimal fundamental frequency calculation and phase alignment.

[0055] S212. Candidate fundamental frequency construction and optimal stroboscopic fundamental frequency selection, specifically: Using the obtained shutter scanning frequency value as the reference frequency, a candidate stroboscopic fundamental frequency set is generated sequentially according to the positive integer multiplication factor relationship. The multiplication factor starts from 1 and increases until the preset maximum multiplication factor limit is reached, that is: In the formula: For the first One candidate stroboscopic base frequency The frequency multipliers are positive integers. This refers to the scanning frequency of the rolling shutter.

[0056] For each candidate stroboscopic fundamental frequency, calculate its corresponding illumination fluctuation period duration. Based on the line scanning mechanism of the rolling shutter, calculate the time difference between the start times of exposure for adjacent rows of pixels. Ratio the exposure time difference between adjacent rows to the illumination fluctuation period duration corresponding to the current candidate fundamental frequency to obtain a sequence of ratios between line exposure difference and illumination period, i.e.: In the formula: For the first The ratio of line exposure time difference to illumination fluctuation period at each candidate fundamental frequency. This represents the time difference between the start times of exposure for two adjacent rows of pixels. For the first The lighting fluctuation period corresponding to each candidate fundamental frequency.

[0057] Iterate through all candidate fundamental frequencies and select the candidate stroboscopic fundamental frequency whose ratio is closest to an integer as the optimal stroboscopic fundamental frequency, that is: In the formula: To achieve the optimal frequency multiplication factor, the exposure times of adjacent row pixels correspond to the same phase points in the illumination drive cycle, fundamentally eliminating beat frequency stripe artifacts caused by non-integer multiples of frequency.

[0058] S213. Phase alignment and pulse width modulation coded sequence generation, specifically: By precisely aligning the rising edge trigger time of the pulse width modulation (PWM) coded sequence to the exposure start time of the current frame, strict phase locking between illumination drive and camera exposure is achieved. Based on the determined optimal stroboscopic fundamental frequency value, the period length parameter of the PWM coded sequence is calculated. That is, the optimal stroboscopic fundamental frequency The reciprocal of, that is: .

[0059] Based on the target lighting brightness requirements and the LED driving characteristic curve, the initial duty cycle parameter of the pulse width modulation encoding sequence is set. By combining the period length parameter, duty cycle parameter and phase alignment time, a complete strobe-shutter phase-locked pulse width modulation encoding sequence is generated. The waveform data of this encoding sequence is written into the waveform buffer of the lighting drive execution unit. The hardware timer is configured to output a drive signal to the LED constant current drive circuit at a specified phase time, thereby realizing precise timing control of the lighting source.

[0060] S214. Synchronous verification and adaptive phase shift adjustment, specifically: The actual output fundamental frequency of the generated pulse width modulation (PWM) coded sequence is compared and verified in real time with the rolling shutter scanning frequency. The actual output frequency value of the coded sequence is measured using a frequency counter module, and the frequency deviation ratio between it and an integer multiple of the rolling shutter scanning frequency is calculated. ,Right now: In the formula: This is the actual output fundamental frequency of the encoded sequence.

[0061] When the detected frequency deviation rate exceeds the preset synchronization tolerance threshold, a phase adaptive adjustment mechanism is triggered: by fine-tuning the phase offset parameter of the pulse width modulation coding sequence, that is, changing the time delay of the rising edge of the coding sequence relative to the camera exposure start time, the frequency deviation rate is gradually reduced. Specifically, In the formula: This is the adjusted phase offset. This is the current phase offset. This is the phase adjustment gain coefficient.

[0062] The measurement and adjustment process is iteratively executed until the frequency deviation rate converges to below the synchronization tolerance threshold, ensuring that the lighting drive and camera scanning maintain a stable phase-locked relationship during long-term operation. The synchronization tolerance threshold ranges from 0.05% to 5%, with 0.1% for nanometer-level precision vision inspection scenarios, 0.5% for high-precision optical character recognition and precision assembly guidance, 1% for conventional industrial vision inspection, and 3% for quality sampling scenarios with lower dynamic range requirements.

[0063] S22. Optimization of spectral power distribution and mixing ratio under multi-objective constraints, specifically: S221. Multi-objective constraint parameter extraction and particle swarm initialization preparation, specifically: From the lighting control constraint feature tensor storage area of ​​the edge computing node, the key measurement base thermal deformation threshold parameter in the thermal deformation suppression requirement dimension is extracted. This parameter characterizes the maximum thermoelastic deformation allowed for precise measurement. Simultaneously, the spectral discrimination weight vector in the spectral fingerprint modulation requirement dimension is extracted. Each component of this vector reflects the contribution of different spectral bands to distinguishing the target material from the background material. Furthermore, the upper limit parameter of melatonin suppression rate in the circadian rhythm adaptation requirement dimension is extracted. This parameter limits the degree of interference of the blue light component in the lighting spectrum on the melatonin secretion of the workers. Based on the extracted three constraint parameters, the search space boundary conditions for multi-objective optimization are determined, providing a constraint framework for subsequent particle swarm initialization.

[0064] S222. Multi-channel LED particle swarm initialization and parameter encoding, specifically: Initialize a particle swarm for the multi-channel LED spectral power distribution, setting the swarm size and maximum number of iterations. The position vector of each particle uses real-number encoding, sequentially concatenating the peak wavelength, full width at half maximum (FWHM), and driving current amplitude parameters of each LED channel. The vector dimension is equal to the number of channels multiplied by three. The particle velocity vector has the same dimension as the position vector, controlling the step size and direction of parameter changes during iteration. The velocity value range is initially limited based on the physical feasible region of each parameter. Randomly generate an initial particle swarm distribution within the constraint boundaries to ensure the initial solution covers a reasonable parameter space for the multi-channel LED spectral power distribution.

[0065] S223. Multi-objective fitness evaluation and Pareto optimal iteration, specifically: A multi-objective fitness function is constructed with three parallel optimization objectives: minimizing the predicted base thermal deformation, maximizing the weighted spectral fingerprint recognition, and minimizing the predicted melatonin suppression rate. Three objective function values ​​are calculated for each particle: the predicted thermal deformation value... calculate: In the formula: The coefficient of thermal expansion of the material. The original length of the base For the first Thermal radiation conversion coefficient of each channel For the number of LED channels, For the first The driving current amplitude of each LED channel; negative spectral resolution. calculate: In the formula: For the first Weight of each band, For the target material in the first Reflectivity of each band The background material is in the first Reflectivity of each band The total number of discrete spectral bands divided in the spectral identification calculation; melatonin suppression rate. calculate: In the formula: wavelength Melatonin inhibition sensitivity at the site For the first The spectral distribution function of each channel.

[0066] The Pareto dominance relation is used to determine the superiority of particles. If particle a is not inferior to particle b on all targets and is strictly superior on at least one target, then a dominates b. The historical best position of each particle and the global best position of the particle swarm are updated. The particle swarm is iteratively evolved through velocity update formulas and position update formulas to obtain the Pareto front composed of the non-dominated solution set. Specifically, the velocity update is as follows: ; In the formula: For the first The generation Dimensional speed, For inertial weights, and As a learning factor, and For a range of random numbers, The best position in history dimensional components, The globally optimal position is the Dimensional components; Position update: In the formula: For the first The generation Dimensional position.

[0067] S224. Pareto solution selection and spectral strategy output, specifically: Calculate the normalized Euclidean distance between each non-dominated solution and the ideal point in the Pareto front. The ideal point is composed of the optimal values ​​of the three objectives, i.e.: In the formula: For the first Normalized Euclidean distance of each non-dominated solution For the first The solution of the first... One target value, For the first The ideal optimal value of each objective and The Pareto frontier is the first The maximum and minimum values ​​of each target.

[0068] The non-dominated solution closest to the ideal point is selected as the optimal compromise solution, which achieves the best balance among multiple objectives. The position vector of this solution is decoded, and the peak wavelength, full width at half maximum (FWHM), and driving current amplitude parameters of each LED channel are extracted to reconstruct the spectral power distribution curves of the multi-channel LEDs. Based on the proportion of the driving current amplitude of each channel to the total current, the mixing ratio strategy for each channel is calculated, i.e.: In the formula: For the first The mixing ratio of each channel is determined; the predicted thermal deformation value, spectral fingerprint identification value, and melatonin inhibition rate corresponding to the solution are recorded simultaneously for deviation comparison and feedback correction in subsequent steps.

[0069] S23. Solving for polarization-direction control parameters and satisfying hard constraints, specifically: The polarization degree requirement value for maintaining consistency with the color temperature space is extracted from the lighting control constraint feature tensor stored in the edge computing nodes. Based on the spatial-direction-material three-dimensional light field polarization degradation tensor, the predicted polarization degree of each spatial grid within the industrial lighting area under the current dust concentration distribution is calculated, i.e.: In the formula: Indicates the first Predicted polarization degree of each spatial grid, This indicates the initial degree of polarization of the lighting actuator in a clean environment. Indicates the total number of light directions, This indicates the total number of dust material types. Indicates the first The spatial grid number The first direction of light output The light field polarization degradation tensor element values ​​of this material Indicates the first At the first spatial grid position The dust concentration of the material Indicates the first The polarization scattering cross-sectional coefficient of the material.

[0070] An optimization problem model with hard constraints was established, using the polarization output angle and light output direction deflection angle of each lighting execution device as optimization variables. The first type of hard constraint requires that the predicted polarization degree of each spatial grid not be lower than a preset lower limit of polarization degree, ensuring sufficient polarization contrast for visual inspection even in dusty environments. The second type of hard constraint requires that the maximum color temperature deviation between any two spatial grids within the same work surface not exceed a preset color temperature consistency threshold, avoiding visual adaptation load and circadian rhythm phase disruption when personnel move across areas. Specifically: the preset lower limit of polarization degree ranges from 0.60 to 0.95, with 0.90 for submicron-level surface defect detection scenarios, 0.85 for conventional polarization imaging detection scenarios, and 0.75 for high-dust-concentration industrial inspection scenarios. The preset color temperature consistency threshold ranges from 50K to 500K, with 50K for high-end display panel inspection, 100K for precision color sorting, 200K for conventional assembly inspection, and 300K for long-term inspection scenarios.

[0071] A sequential quadratic programming algorithm is employed to solve the problem. This algorithm consists of four cascaded functional units: a Lagrange function builder, a quadratic programming subproblem solver, a line search module, and a convergence arbiter. The Lagrange function builder integrates the objective function and hard constraints to form an augmented form. The quadratic programming subproblem solver constructs a quadratic approximation and linear constraints for the current iteration point, and solves for the optimal search direction using the effective set method or interior point method. The line search module performs a backtracking search along the direction to determine the optimal step size. The convergence arbiter calculates the constraint violation degree and multiplier update amount. If the result is lower than the preset accuracy, it outputs the target polarization output angle and the target light emission direction deflection angle of each lighting actuator; otherwise, it continues iterating. The solved target polarization output angle and target light emission direction deflection angle are encoded as polarization-direction control parameters, which are used to control the polarizer rotation mechanism and mechanical steering actuator of each lighting actuator.

[0072] S3. The polarization-direction control parameters and spectral mixing ratio strategy are sent out for execution, and actual polarization degree, actual color temperature, and eye-tracking data are collected in real time to construct a lighting performance evaluation matrix. The feature fusion weights of the lighting constraint model are adaptively adjusted accordingly. Specifically, this includes: S31. Lighting control parameter distribution and execution unit driving, specifically: The strobe-shutter phase-locked pulse width modulation (PWM) coded sequence generated in step S2 is written into the hardware waveform buffer of the lighting drive execution unit. A timer interrupt triggering mechanism is configured to ensure precise synchronization between the sequence and the industrial camera exposure. The duty cycle parameters of each channel in the multi-channel LED spectral mixing and matching strategy are converted into drive current amplitude commands and output to the constant current drive circuit of each channel LED through a digital-to-analog converter. The target polarization output angle in the polarization-direction adjustment parameters is converted into a rotation angle command for a stepper motor or servo motor to drive the polarizer rotation mechanism to the specified azimuth angle. The target light output direction deflection angle is converted into an angle control command for a mechanical steering actuator to adjust the light output axis direction of the lamp. The thermal radiation power upper limit constraint parameter is converted into the maximum allowable drive current limit value for each channel LED and the overcurrent protection logic is embedded in the drive circuit to ensure that the total lighting power does not exceed the thermal deformation suppression requirements. The above parameters are sent to each lighting execution device through an industrial Ethernet or controller area network bus to trigger synchronous dimming, color adjustment, polarization adjustment, and light output direction adjustment operations.

[0073] S32. Real-time feedback and acquisition of multi-dimensional control effectiveness, specifically: The current frame image data is acquired in real time through the image output interface of an industrial camera. A stroboscopic artifact detection algorithm is used to extract the grayscale distribution curve along the image row direction. A Fast Fourier Transform analysis is performed on this curve to extract the frequency components and amplitudes of the periodic stripes, and the stroboscopic artifact rate is calculated. ,Right now: In the formula: This represents the standard deviation of pixel grayscale values ​​along the row direction of the image. It is the arithmetic mean of the pixel grayscale values ​​along the row direction of the image.

[0074] The strain values ​​at key measurement points on the base surface are read in real time by a fiber Bragg grating strain sensor built into the precision mechanical deformation sensing node. The real-time micro-deformation is then calculated by combining the sensor gauge length. ,Right now: In the formula: The real-time strain value measured by the fiber Bragg grating sensor; This is the gauge length of the fiber Bragg grating sensor.

[0075] By running a spectral fingerprinting algorithm through a material surface sensing node, the reflectance spectrum of the workpiece surface collected under the current multispectral illumination is compared with the standard material spectral database. The algorithm outputs the material category with the highest matching degree and its corresponding recognition confidence level. In the formula: Spectral identification confidence level; For the first The identification weight of each spectral band; For the first Actual observed reflectance of each band; For the first The standard material in the first Standard reflectance of each band; This represents the total number of spectral bands involved in the calculation.

[0076] By measuring the Stokes vector parameters in real time at each spatial grid location using polarization field sensing nodes, the combined value of linear polarization and circular polarization is calculated to obtain the actual polarization feedback value. ,Right now: In the formula: The Stokes first parameter, i.e., total light intensity, is measured in real time. These are the second, third, and fourth Stokes parameters measured in real time.

[0077] The spectral power distribution data of each spatial grid location is collected in real time through a multispectral sensor network, and the correlated color temperature value is calculated to form a sequence of actual color temperature feedback values, namely: ,in: This is the actual color temperature feedback value. and These are the CIE chromaticity coordinates calculated based on real-time spectral data.

[0078] An eye-tracking device was used to collect real-time sequences of pupil diameter changes, fixation point trajectory coordinates, blink frequency, and saccade speed of the operator, constructing a dataset of actual eye movement characteristics, and calculating the coefficient of variation of saccade speed. As a quantitative indicator of visual fatigue, namely: In the formula: The standard deviation of real-time scanning speed. This is the arithmetic mean of the real-time scanning speed.

[0079] By collecting real-time data on salivary melatonin concentration or core body temperature through biorhythm sensing nodes and comparing it with a baseline circadian rhythm curve, the actual phase shift of melatonin secretion at the current moment is calculated. ,Right now: In the formula: For the real-time melatonin secretion phase, The baseline circadian rhythm phase.

[0080] S33. Multiphysics performance deviation comparison and evaluation matrix construction, specifically: S331. Extract the target values ​​for flicker-shutter frequency avoidance, thermal deformation suppression, spectral fingerprint modulation, polarization preservation, color temperature space consistency, and circadian rhythm adaptation generated in step S1 from the lighting control constraint feature tensor storage area, and use them as the expected benchmark values ​​for each dimension; map the flicker artifact rate, real-time micro-variables, spectral recognition confidence, actual polarization feedback value, actual color temperature feedback value, eye movement feature variation coefficient, and melatonin secretion phase shift obtained in real time in step S32 to the corresponding six dimensions.

[0081] S332. Calculate the normalized deviation between the actual values ​​and the target values ​​for each dimension, where: Flicker artifact rate deviation This is the difference between the actual value and the target upper limit, i.e.: In the formula: For the real-time acquisition of flicker artifact rate, To avoid the required target value for stroboscopic shutter speed; Microvariable bias This is the difference between the actual value and the thermal deformation threshold, i.e.: In the formula: For real-time micro-variables, The target value for suppressing thermal deformation is the thermal deformation threshold. Spectral fingerprint confidence bias The difference between the target confidence level and the actual value, i.e.: In the formula: To achieve the target value required for spectral fingerprint modulation, To determine the confidence level of the real-time acquired spectra; polarization deviation This is the difference between the actual value and the preset lower limit of polarization, i.e.: In the formula: This represents the arithmetic mean of the actual polarization degrees of each spatial grid. To maintain the required target value for polarization degree, i.e., to preset the lower limit of polarization degree; Color temperature deviation This is the difference between the maximum actual color temperature difference and the consistency threshold, i.e.: In the formula: The maximum deviation of color temperature collected in real time. The target value for color temperature space consistency requirements; Eye movement feature deviation The difference between the weighted sum of the saccade velocity variation coefficient and the melatonin phase shift and the target's physiological state is: In the formula: This is the baseline value for the coefficient of variation of saccadic velocity under normal physiological conditions. The target value for adapting to biological rhythms is the melatonin phase shift target.

[0082] S333. Arrange the six deviation values ​​according to the five-dimensional effects of biological, thermal, visual, polarization, and color temperature to construct a multi-physics coupling performance evaluation matrix for lighting. Each element of this matrix represents the degree and direction of the regulation deviation in the corresponding physical field dimension.

[0083] S34. Adaptive weight correction and closed-loop parameter fine-tuning, specifically: S341. Cumulative error integral calculation and feature fusion weight gradient descent update, specifically: The deviation values ​​for each dimension are extracted from the multiphysics coupling effectiveness evaluation matrix of lighting, including flicker artifact rate deviation, microvariable deviation, spectral fingerprint confidence deviation, polarization deviation, color temperature deviation, and eye-tracking feature deviation. A preset time window of fixed length is set, and the deviation values ​​for each dimension are integrated over time within this window to calculate the cumulative error integral value, i.e.: In the formula: Indicates the first The cumulative error integral value of each dimension, Indicates the preset time window length. Indicates time No. Deviation values ​​in each dimension.

[0084] The gradient descent method is used to update the weight coefficients of the multiphysics feature fusion layer in the lighting-biology-thermal-vision coupling constraint model. The update step size of the weight coefficients is proportional to the cumulative error integral value of the corresponding dimension. This ensures that perceptual dimensions with large cumulative biases receive higher weight correction magnitudes in subsequent fusion calculations, achieving adaptive reassessment of the importance of each physics field. In the formula: Indicates the first The updated weight coefficients for each dimension Indicates the first The weight coefficients of each dimension before the update This represents the gradient descent learning rate; the updated weight coefficients are then normalized. In the formula: This represents the normalized weight coefficients.

[0085] S342. Phase offset fine-tuning of flicker artifact rate deviation, specifically: The flicker artifact rate deviation value is extracted from the multiphysics coupling effectiveness evaluation matrix of illumination, and the sign characteristics and absolute amplitude of the deviation value are analyzed. When the deviation value is positive, it indicates that the actual flicker artifact rate is higher than the target value and there is a deviation in phase locking. At this time, the phase offset of the flicker-shutter phase locking pulse width modulation coding sequence is finely adjusted in the positive direction, that is, the delay time of the rising edge of illumination drive relative to the camera exposure start time is increased. When the deviation value is negative, the phase offset is decreased in the negative direction.

[0086] Each fine-tuning increment is a preset step unit. Deviation detection and phase fine-tuning are performed iteratively until the absolute value of the flicker artifact rate deviation converges to within a preset tolerance threshold. The preset tolerance threshold ranges from 0.1% to 5%. For ultra-high precision vision scenarios such as submicron level defect detection, the threshold is 0.2%. For precision optical measurement and automated optical inspection standard production lines, the threshold is 0.5%. For conventional industrial vision guidance and quality sampling scenarios, the threshold is 1.0%. For coarse inspection scenarios with low dynamic range requirements or where stripes can be filtered out by later algorithms, the threshold is 3.0%.

[0087] S343. Adaptive adjustment of light mixing ratio and duty cycle, and closed-loop correction of polarization angle. Specifically: The spectral fingerprint confidence bias and color temperature bias are extracted from the multiphysics coupling effectiveness evaluation matrix of lighting. These are then multiplied by their respective weighting coefficients and summed to calculate the weighted overall bias. ,Right now: In the formula: This is the weighting coefficient for the confidence bias of spectral fingerprints. This is the weighting coefficient for color temperature deviation. The dominant deviation.

[0088] By comparing the relative amplitudes of the weighted spectral fingerprint confidence bias and color temperature bias, when the confidence bias dominates, the duty cycle of each channel in the multi-channel LED spectral mixing strategy is adjusted first to optimize the spectral shape matching and enhance the distinguishing features between the target material and the background; when the color temperature bias dominates, the duty cycle is adjusted first to optimize the white balance point position, so that the color temperature of the mixed spectrum tends to the target value; specifically, In the formula: For the first The updated duty cycle for each channel For the first The current duty cycle of each channel Adjust the step size coefficient for the duty cycle.

[0089] Simultaneously, the polarization degree deviation value is extracted, and the polarization output angle in the polarization-direction control parameter is finely adjusted according to the sign of the deviation. When the polarization degree is lower than the target, the rotation angle of the polarizer is adjusted to enhance the polarization purity of the emitted light, forming a closed-loop control from detection to correction.

[0090] Example 2 Please see the appendix Figure 3 A lighting control decision system integrating multi-source sensing and edge computing is disclosed to implement a lighting control decision method integrating multi-source sensing and edge computing. The system includes a multi-source heterogeneous sensing module, edge computing nodes, a lighting drive execution unit, and a cloud management platform, wherein: A. A multi-source heterogeneous sensing module, including a biological rhythm sensing node, a high-speed visual sensing node, a precision mechanical deformation sensing node, a material surface sensing node, a polarized light field sensing node, an eye-tracking device, and a multispectral sensor network; used to collect data on the melatonin secretion phase and core body temperature rhythm of workers, the rolling shutter scanning frequency and exposure start time data of industrial cameras, the thermoelastic micro-deformation and thermal expansion coefficient data of key measurement bases, the surface reflectance spectral fingerprint data of the workpiece under test, the multi-directional polarization attenuation sequence and dust particle size distribution and material classification signals of various lighting actuators, the eye movement feature data set of workers, and the color temperature difference sequence and color temperature drift rate of each spatial location.

[0091] B. Edge computing nodes, connected to multi-source heterogeneous sensing modules, including: The coupling constraint model construction module is used to perform spatiotemporal alignment and multiphysics field fusion on the above heterogeneous data, and output lighting control constraint feature tensors that characterize the requirements for stroboscopic-shutter frequency avoidance, thermal deformation suppression, spectral fingerprint modulation, polarization degree preservation, color temperature consistency and biological rhythm adaptation. The multiphysics optimization solution module is used to decouple multiple constraints and solve the spectral power distribution and driving waveform that satisfy shutter synchronization conditions, thermal deformation threshold, maximization of spectral discriminability, lower limit of polarization degree, upper limit of color temperature difference and minimum melatonin suppression rate. It generates strobe-shutter phase-locked PWM encoding sequence, spectral mixing ratio strategy, polarization-direction control parameters and thermal radiation upper limit constraint parameters. The performance verification and weight evolution module is used to send control parameters to the lighting drive execution unit, collect flicker artifact rate, micro-variables, spectral recognition confidence, actual polarization degree, actual color temperature, eye movement characteristics and melatonin phase change data in real time, construct a multi-physics coupling performance evaluation matrix, adaptively correct the fusion weight coefficients of the coupling constraint model based on deviation analysis, and fine-tune the PWM encoding sequence phase offset, light mixing ratio duty cycle and polarization output angle online.

[0092] C. Lighting drive execution unit, connected to the edge computing node, is used to receive and execute control parameters to drive the lighting execution device to perform dimming, color adjustment, polarization adjustment and light direction adjustment operations.

[0093] D. The cloud management platform communicates with the edge computing nodes to receive uploaded data and perform incremental training on the multiphysics optimization algorithm, and then sends the updated parameters to the edge computing nodes; the edge computing nodes and the cloud management platform adopt a mechanism for resuming data transmission after network interruption.

[0094] Example 3 To verify the effectiveness of the proposed lighting control decision-making method integrating multi-source sensing and edge computing, this embodiment established an industrial lighting intelligent control experimental platform in a precision electronic assembly workshop, selecting production data from 30 consecutive working days as the verification sample. The experimental area was the precision inspection station of the SMT assembly line, covering an area of ​​approximately 120 square meters, and 12 LED lighting execution devices were deployed. The present invention was compared with three existing technical solutions: Existing technology 1: Traditional timer switch + fixed illuminance lighting solution, without any intelligent control; Existing technology 2: Energy-saving control scheme of single light sensor + personnel infrared detection; Existing technology 3: Intelligent lighting solution based on multi-sensor weighted fusion and cloud-based decision-making; The present invention proposes an edge computing scheme that integrates multi-source heterogeneous sensing, polarization-spectral degradation tensor, stroboscopic-shutter phase-locked PWM, and visual fatigue closed-loop control.

[0095] I. Experimental Conditions The ambient temperature is 25±3℃, the suspended dust concentration is 0.5-1.2 mg / m³, mainly iron filings, the production cycle is 1200 pieces / hour, the camera shutter frequency is 60Hz, the number of operators is 6 people working in shifts, and the testing cycle is 30 working days × 8 hours.

[0096] II. Definition of Performance Indicators Comprehensive energy consumption: unit kWh / day, representing the average daily power consumption of the workstation lighting system; Flicker artifact rate: percentage of frames in images captured by industrial cameras showing bright and dark stripes; Color temperature spatial inconsistency: unit K, representing the maximum deviation between the color temperature of each measuring point and the average value within the same work surface; Polarization retention rate: the ratio of the measured polarization degree after 30 days of operation to the initial polarization degree; Visual fatigue index: a comprehensive score based on eye movement characteristics, i.e., the increase rate of blinking frequency + the delay of pupil light reflection, ranging from 0 to 1, with higher values ​​indicating greater fatigue; Spectral fingerprint recognition confidence level: the average confidence level of material surface defect detection; Thermal deformation compensation error: unit μm, representing the deviation between the actual thermal deformation of the precision measurement base and the compensation target value.

[0097] III. Effect Comparison The effects of the present invention and the prior art are compared in Table 1 below. Figure 4 As shown: Table 1: Comparison of Results IV. Analysis of Experimental Results 1. In terms of overall energy consumption: This invention reduces energy consumption by 34.5% compared to the best comparison group of existing technology 3. This is mainly attributed to the polarization-spectral degradation compensation that avoids ineffective overexposure, and the stroboscopic-shutter synchronization optimization that ensures that illumination only provides effective light output within the camera exposure window, thereby reducing unnecessary stroboscopic energy consumption.

[0098] 2. Regarding flicker artifact rate: This invention reduces the flicker artifact rate to 0.9%, far lower than the 8.5% of the prior art 3. The reason is that this invention, for the first time, uses the rolling shutter scanning frequency and the exposure start time as constraints to generate a phase-locked PWM encoding sequence, thus eliminating the beat frequency effect at its source.

[0099] 3. Regarding color temperature spatial inconsistency: This invention controls the color temperature deviation to within 145K, which is far superior to the 390K of the prior art. This is due to the construction of the spatial spectral degradation tensor and the multi-objective optimization with the visual fatigue index as a dynamic constraint.

[0100] 4. Polarization retention rate: After 30 days of operation, the polarization retention rate of this invention is as high as 94%, while the existing technology is generally below 80%; the core is that this invention constructs the optical field polarization degradation tensor by using dust particle size and material classification signal, thereby realizing dynamic compensation for the depolarization effect of metal dust.

[0101] 5. Regarding the visual fatigue index: This invention reduces the visual fatigue index to 0.28, a 47.2% reduction compared to the prior art 3; the eye-tracking driven proactive intervention strategy ensures that lighting parameters always match the physiological state of the person, significantly delaying the accumulation of fatigue.

[0102] 6. Regarding the confidence level of spectral fingerprint recognition: This invention improves the confidence level of defect detection to 97.4%, which is attributed to the fact that the spectral power distribution obtained by multi-objective optimization solution maximizes material discrimination and suppresses ambient light interference.

[0103] 7. Regarding thermal deformation compensation error: This invention controls the thermal deformation compensation error to 1.1μm, which is 74.4% better than the prior art, and meets the micron-level stability requirements of precision measurement base.

[0104] V. Conclusion This embodiment demonstrates that, compared to three existing typical industrial lighting control schemes, the present invention achieves significant improvements in seven dimensions, including energy consumption, imaging quality, light color uniformity, polarization preservation, personnel health, detection accuracy, and thermal stability, thus verifying the advanced nature and engineering applicability of the proposed method.

[0105] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of the invention is limited to these examples; within the framework of the invention, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of the different aspects of the invention as described above, which are not provided in detail for the sake of brevity.

[0106] This invention is intended to cover all such substitutions, modifications, and variations that fall within the broad scope of this specification. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A lighting regulation decision-making method fusing multi-source perception and edge computing, characterized in that, include: The illumination direction polarization degree and suspended dust data are collected by polarization field sensing nodes, the gaze point trajectory is collected by eye tracking device, and mid-temperature data is collected by multispectral sensor network. An illumination constraint model is constructed at edge computing nodes, and the illumination regulation constraint feature tensor is output. Receive the illumination control constraint feature tensor, decouple the polarization degree and color temperature consistency constraint, and generate polarization-direction control parameters and spectral mixing ratio strategy; The polarization-direction control parameters and spectral mixing ratio strategy are sent out for execution, and the actual polarization degree, actual color temperature and eye-tracking data are collected in real time to construct a lighting performance evaluation matrix and adaptively correct the feature fusion weights of the lighting constraint model. 2.The lighting regulation decision method of fusing multi-source perception and edge computing according to claim 1, wherein, The process involves acquiring illumination direction polarization degree and suspended dust data through polarization field sensing nodes, acquiring gaze point trajectory through an eye-tracking device, acquiring mid-temperature data through a multispectral sensor network, constructing an illumination constraint model at edge computing nodes, and outputting an illumination regulation constraint feature tensor, specifically including: The system collects melatonin secretion phase data and core body temperature rhythm data of workers through biorhythm sensing nodes, collects the shutter scanning frequency and current frame exposure start time data of industrial cameras through high-speed vision sensing nodes, and collects thermoelastic micro-deformation data and material thermal expansion coefficient data of key measurement bases through precision mechanical deformation sensing nodes; and generates melatonin inhibition sensitivity curve, stroboscopic-shutter frequency avoidance phase constraint matrix and thermal deformation compensation demand vector respectively. Collect the multispectral reflectance fingerprint of the workpiece, collect the polarization attenuation in each direction and the material distribution of dust particle size, and construct the spectral identification weight vector and the three-dimensional optical field polarization degradation tensor of space-direction-material. Collect eye-tracking feature data sets, collect multispectral color temperature difference sequences and color temperature drift rates, and construct a spatial spectral degradation tensor; All collected data are input into the lighting-biology-thermal-visual coupling constraint model, and multi-physics spatiotemporal alignment and correlation fusion processing are performed to output the lighting regulation constraint feature tensor.

3. The lighting control decision-making method integrating multi-source sensing and edge computing according to claim 2, characterized in that, The process involves collecting the multispectral reflectance fingerprint of the workpiece, acquiring the polarization attenuation in each direction and the material distribution of dust particle size, and constructing a spectral identification weight vector and a three-dimensional optical field polarization degradation tensor of space-direction-material. Specifically, this includes: By using the material surface sensing node, the surface of the workpiece under test is sequentially illuminated under multispectral illumination conditions, and the surface reflection intensity value corresponding to each spectral band is collected to form the original reflection spectrum curve. The peak wavelength position and half width at half maximum feature are extracted to construct the spectral identification weight vector of the target material. By using polarization field sensing nodes, real-time polarization values ​​of each lighting actuator in multiple light-emitting directions are collected. In the formula: For linear polarization degree, The polarization degree component is represented by a circular polarization degree component, and its difference value is calculated with the initial polarization degree value of the equipment in a clean state to generate a polarization degree attenuation sequence for each light output direction, i.e.: In the formula: This is the degree of polarization attenuation. For the first Initial polarization values ​​in each direction, For the first Real-time polarization degree values ​​in each direction; By using a suspended dust concentration sensor, the principle of multi-angle polarization scattering is employed to simultaneously acquire dust particle size distribution and dust material classification signals at multiple spatial locations. Using the spatial grid coordinates of the industrial lighting area as the first dimension, the light emission direction of the lighting device as the second dimension, and the dust material type as the third dimension, the polarization attenuation sequence is correlated and mapped with the dust particle size distribution and dust material classification signals. The polarization attenuation rate of each spatial grid position and the polarization scattering cross section coefficient corresponding to each dust material are stored to complete the construction of the light field polarization degradation tensor.

4. The method of claim 3, wherein, The acquisition of eye-tracking feature data sets includes acquiring multispectral color temperature difference sequences and color temperature drift rates, and constructing a spatial spectral degradation tensor, specifically including: Eye-tracking devices continuously collect eye-movement feature data of the current worker. The pupil diameter change sequence is then processed using a sliding window filter to extract the pupil's response delay time to light reflection and the steady-state contraction amplitude. In the formula: In response to the delay time, This is the moment when the pupil diameter begins to continuously contract. This refers to the moment of sudden change in light intensity; In the formula: For steady-state contraction amplitude, This is the stable value of the pupil diameter before constriction. This represents the stable pupil diameter after constriction. By using a multispectral sensor network, the actual color temperature values ​​at multiple spatial grid locations within the industrial lighting area are simultaneously acquired. The instantaneous deviation between the color temperature value of each spatial grid and the average color temperature of all grids within the same working surface is calculated to generate a color temperature difference sequence. At the same time, the rate of change of the color temperature value of each lighting device over time in different light emission directions is recorded to generate a color temperature drift rate sequence. Spatiotemporal correlation analysis was performed between the color temperature difference sequence and the color temperature drift rate sequence of each lighting device to identify the main contributing sources of color temperature spatial inconsistency and mark them as the dominant sources of color temperature deviation. Using the spatial coordinates of the industrial lighting area as the first dimension and the light output direction of the lighting device as the second dimension, the color temperature difference sequence, color temperature drift rate sequence, and color temperature deviation dominant source marker are associated and stored to construct a spatial spectral degradation tensor.

5. The method of claim 4, wherein, Receive the illumination control constraint feature tensor, decouple the polarization degree and color temperature consistency constraints, and generate polarization-direction control parameters and spectral mixing ratio strategies, specifically including: Extract the stroboscopic-shutter shooting frequency avoidance requirements and shutter scanning frequency, constrained by integer multiples of the scanning frequency, and phase-locked with the exposure start time to generate a stroboscopic-shutter phase-locked pulse width modulation coding sequence; The requirements for thermal deformation suppression and spectral fingerprint modulation are extracted. With the goal of maximizing the thermal deformation threshold and spectral discriminability, and combined with minimizing the melatonin suppression rate, the spectral power distribution curve and the mixing ratio strategy are solved by multi-objective particle swarm optimization. The polarization degree requirement and color temperature spatial consistency requirement are extracted. Based on the light field polarization degradation tensor, the predicted polarization degree of each spatial grid under the current dust concentration is calculated. With the predicted polarization degree of each grid not lower than the preset lower limit and the maximum color temperature deviation within the same working surface not exceeding the preset consistency threshold as hard constraints, the target polarization output angle and target light output direction deflection angle of each lighting execution device are solved to generate polarization-direction control parameters.

6. The method of claim 5, wherein, The extraction of the stroboscopic-shutter shooting frequency avoidance requirement and the shutter scanning frequency, constrained by an integer multiple of the scanning frequency, and phase-locked with the exposure start time, generates a stroboscopic-shutter phase-locked pulse width modulation (PWM) coded sequence, specifically including: The flicker-shutter frequency avoidance requirement is extracted from the lighting control constraint feature tensor, and the rolling shutter scanning frequency and the exposure start time of the current frame are obtained from the high-speed visual perception node. A set of candidate stroboscopic frequencies is defined by positive integer multiples of the shutter scanning frequency, and the exposure time difference between two adjacent rows of pixels at each candidate frequency is calculated. With lighting fluctuation cycle ratio The fundamental frequency whose ratio is closest to an integer is selected as the optimal stroboscopic fundamental frequency; Align the rising edge of the pulse width modulation coding sequence to the exposure start time of the current frame, and determine the period length and initial duty cycle of the coding sequence according to the optimal stroboscopic base frequency to generate a stroboscopic-shutter phase-locked pulse width modulation coding sequence. The generated encoded sequence is synchronized with the shutter scanning frequency to verify the frequency deviation rate between the fundamental harmonic component of the encoded sequence and the shutter scanning frequency. In the formula: The actual output fundamental frequency of the encoded sequence. For optimal harmonic coefficients, The shutter scanning frequency is used; when the frequency deviation rate exceeds the preset synchronization tolerance threshold, the phase offset of the encoded sequence is automatically adjusted until the deviation rate is lower than the threshold.

7. The lighting control decision-making method integrating multi-source sensing and edge computing according to claim 6, characterized in that, The extraction of thermal deformation suppression and spectral fingerprint modulation requirements aims to maximize the thermal deformation threshold and spectral discriminability, combined with minimizing the melatonin suppression rate. A multi-objective particle swarm optimization method is used to solve for the spectral power distribution curve and the mixing ratio strategy. Specifically, this includes: The key measurement base thermal deformation threshold in the thermal deformation suppression requirement and the spectral identification weight vector of the target material in the spectral fingerprint modulation requirement are extracted from the lighting control constraint feature tensor. At the same time, the upper limit of melatonin inhibition rate in the photobiological safety constraint is extracted. Initialize the spectral power distribution particle swarm of the multi-channel LED. The position vector of each particle is composed of the peak wavelength, half width at half maximum (WHM), and driving current amplitude of each channel LED. The velocity vector of the particle controls the iteration step size of each parameter. A multi-objective fitness function is constructed with three parallel optimization objectives: minimizing the predicted value of base thermal deformation, maximizing the weighted spectral fingerprint recognition, and minimizing the predicted value of melatonin inhibition rate. The historical optimal position and global optimal position of the particle swarm are updated by Pareto dominance relation, and the non-dominated solution set is solved iteratively. The solution closest to the ideal point is selected from the Pareto front, and the corresponding multi-channel LED spectral power distribution curve and the mixing ratio strategy of each channel are output. At the same time, the predicted value of thermal deformation, spectral fingerprint recognition and melatonin suppression rate are recorded.

8. The method of claim 7, wherein, The polarization-direction control parameters and spectral mixing ratio strategy are issued and executed, and the actual polarization degree, actual color temperature, and eye-tracking data are collected in real time to construct a lighting performance evaluation matrix. The feature fusion weights of the lighting constraint model are adaptively corrected, specifically including: The strobe-shutter phase-locked pulse width modulation coding sequence, multi-channel LED spectral mixing ratio strategy, polarization-direction control parameters and thermal radiation power upper limit constraint parameters are sent to the lighting drive execution unit to trigger each lighting execution device to perform dimming, color adjustment, polarization adjustment and light direction adjustment operations. Real-time acquisition of stroboscopic artifact rate of industrial camera output images, real-time micro-variables of precision measurement base, confidence level of material surface spectral fingerprint recognition algorithm, actual polarization feedback value and actual color temperature feedback value of each spatial grid, as well as actual eye movement feature data set and actual melatonin secretion phase change data of operators; The collected feedback data are compared with the corresponding target values ​​in the lighting control constraint feature tensor item by item to calculate the flicker artifact rate deviation, micro-variable deviation, spectral fingerprint confidence deviation, polarization deviation, color temperature deviation and eye movement feature deviation, and construct the lighting multi-physics coupling effectiveness evaluation matrix. Based on the evaluation matrix bias, gradient descent is used to correct the multi-physics feature fusion weight coefficients, while fine-tuning the phase shift of the coding sequence, the duty cycle of the light mixing ratio, and the polarization output angle.

9. The method of claim 8, wherein, The method of correcting the multi-physics feature fusion weight coefficients based on the evaluation matrix bias using gradient descent, while simultaneously fine-tuning the phase shift of the encoded sequence, the duty cycle of the light mixing ratio, and the polarization output angle, specifically includes: Calculate the cumulative error integral of each deviation value in the lighting multiphysics coupling effectiveness evaluation matrix within a preset time window. In the formula: Indicates the preset time window length. Indicates time No. The deviation values ​​in each dimension are used to update the multiphysics feature fusion weight coefficients of the lighting-biology-thermal-visual coupling constraint model using the gradient descent method. Based on the sign and amplitude of the stroboscopic artifact rate deviation, the phase offset of the stroboscopic-shutter phase-locked pulse width modulation coding sequence is adjusted in the opposite direction. If the sign is positive, it indicates that the artifact rate is too high. Then, the phase offset is finely adjusted in the positive direction by one step unit until the deviation converges. Based on the weighted sum of the spectral fingerprint confidence deviation and the color temperature deviation, the duty cycle of each channel in the multi-channel LED spectral mixing strategy is adjusted; at the same time, the polarization output angle in the polarization-direction control parameter is fine-tuned according to the sign of the polarization degree deviation.

10. A lighting control decision system integrating multi-source sensing and edge computing, characterized in that, A lighting control decision-making method integrating multi-source sensing and edge computing as described in any one of claims 1-9, comprising a multi-source heterogeneous sensing module, an edge computing node, a lighting drive execution unit, and a cloud management platform, wherein: The multi-source heterogeneous sensing module includes a biological rhythm sensing node, a high-speed visual sensing node, a precision mechanical deformation sensing node, a material surface sensing node, a polarized light field sensing node, an eye-tracking device, and a multispectral sensor network. These are used to collect data on the melatonin secretion phase and core body temperature rhythm of workers, the rolling shutter scanning frequency and exposure start time data of industrial cameras, the thermoelastic micro-deformation and thermal expansion coefficient data of key measurement bases, the surface reflectance spectral fingerprint data of the workpiece under test, the multi-directional polarization attenuation sequence and dust particle size distribution and material classification signals of various lighting actuators, the eye movement feature data set of workers, and the color temperature difference sequence and color temperature drift rate of various spatial locations. An edge computing node, connected to the multi-source heterogeneous sensing module, includes: The coupling constraint model construction module is used to perform spatiotemporal alignment and multiphysics field fusion on the above heterogeneous data, and output lighting control constraint feature tensors that characterize the requirements for stroboscopic-shutter frequency avoidance, thermal deformation suppression, spectral fingerprint modulation, polarization degree preservation, color temperature consistency and biological rhythm adaptation. The multiphysics optimization solution module is used to decouple multiple constraints and solve the spectral power distribution and driving waveform that satisfy shutter synchronization conditions, thermal deformation threshold, maximization of spectral discriminability, lower limit of polarization degree, upper limit of color temperature difference and minimum melatonin suppression rate. It generates strobe-shutter phase-locked PWM encoding sequence, spectral mixing ratio strategy, polarization-direction control parameters and thermal radiation upper limit constraint parameters. The performance verification and weight evolution module is used to send control parameters to the lighting drive execution unit, collect flicker artifact rate, micro-variables, spectral recognition confidence, actual polarization degree, actual color temperature, eye movement characteristics and melatonin phase change data in real time, construct a multi-physics field coupling performance evaluation matrix, adaptively correct the fusion weight coefficient of the coupling constraint model based on deviation analysis, and fine-tune the PWM encoding sequence phase shift, light mixing ratio duty cycle and polarization output angle online. The lighting drive execution unit is connected to the edge computing node and is used to receive and execute the control parameters to drive the lighting execution device to perform dimming, color adjustment, polarization adjustment and light direction adjustment operations. The cloud management platform communicates with the edge computing nodes to receive uploaded data and perform incremental training on the multiphysics optimization algorithm, and then sends the updated parameters to the edge computing nodes. The edge computing nodes and the cloud management platform use a mechanism to resume transmission after network interruption.

Images