Hydrostatic pressure rotary table detection method and system based on industrial vision

Abstract

The invention discloses a hydrostatic pressure rotary table detection method and system based on industrial vision, and relates to the technical field of machine tool precision detection. A camera-witness ring-working face-encoder unified coordinate chain is established, a zero position and a scale are anchored at a time, and five-degree-of-freedom priori is inverted; acquiring polarization / phase multiple frames under the condition of not invading an oil cavity, constructing disturbance fingerprints and mapping the disturbance fingerprints into equivalent film thickness, refractive index and scattering intensity under physical priori; learning a space-time continuous aberration compensation field and fusing with an encoder angle circular domain, peeling off eccentricity and tilting aliasing, and outputting unified angle and form and position quantification; standardized curves and indexes are formed according to ISO230-2 / 230-7, and rapid re-labeling and protection operation and maintenance triggered by a linkage threshold value are carried out. The method has the technical effects of cross-working-condition consistency, no drifting for a long time, audibility and traceability. The rotation precision is improved, the online time is shortened, and the maintenance cost and energy consumption are reduced.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of machine tool precision detection, in particular to a liquid static pressure rotary table detection method and system based on industrial vision. BACKGROUND

[0002] The liquid static pressure rotary table is widely used in high-end scenes such as aerospace inertial navigation docking, semiconductor wafer alignment / taking and placing, precision bearing and gear detection, and five-axis machine tool rotary shaft calibration due to high bearing stiffness, small friction and stable angular displacement. The current detection adopts a combination of high-resolution encoders, offline autocollimation / interference, contact or cavity capacitance probes and simple visual edge methods: the encoder is stable in the short term but is affected by temperature drift, indexing error and backlash; autocollimation / interference requires an air light path, is difficult to cross the oil film and is sensitive to the environment; the contact and cavity probes are invasive, require frequent maintenance and have high downtime costs; the common visual scheme relies on intensity or single-channel phase and is difficult to distinguish the polarization-phase change caused by the oil film and the assembly geometric error, and the cross-condition zero position and scale are difficult to unify. As a result, there is a lack of unified scale and closed-loop operation and maintenance in the online-operation-acceptance process, and the rechecking cost is high under the ISO230-2 / 230-7 standard, and the traceability is insufficient.

[0003] The technical problem to be solved by the application is that, under the constraints of not changing the main measurement field of view, not interrupting production and not invading the oil cavity, the visual measurement and the encoder reading remain the same angular scale and zero position for a long time, and the optical disturbance induced by the oil film is strictly decoupled from the structural geometric error, thereby stably outputting indexes that can be rechecked according to ISO230-2 and ISO230-7. This problem arises from the actual operation of the liquid static pressure rotary table: as the shift, load and environment change, the oil temperature, flow and pressure fluctuate slowly, the equivalent film thickness and refractive index of the oil film change with time, causing the coupling of the polarization state and the phase field; in the low-speed rotation or intermittent point motion state, this coupling is presented as optical drag on the micro-drift of the characteristic coordinate, which is not significant in the short term but accumulates as an implicit drift of the angular zero position and scale over a long period of time. At the same time, although the encoder is stable in the short term, it is affected by temperature drift, backlash and tooth groove harmonics, resulting in a slow deviation. If there is no unified coordinate chain of the camera-witness ring-working surface-encoder and no primary anchor, the visual and the encoder cannot be compared on the same scale; if there is no disturbance fingerprint based on polarization-phase and no physical prior agent, the film thickness / refractive index change will be mixed with geometric errors such as eccentricity and tilt; if round domain fusion and integral round harmonic stripping are not implemented, the mixed harmonic remains in the final angle; if there is no ISO mapping and threshold trigger self-correction, it is difficult to recheck and maintain in time when the oil temperature range is switched or the imaging health decreases. The direct consequences are: distortion of the rotation error evaluation, long-term drift of the unified angle, incomparable batch results, easy rejection of third-party acceptance, increased on-site downtime for rechecking, and decreased assembly / alignment accuracy and quality traceability in the key process. SUMMARY

[0004] (I) Technical problems solved

[0005] In view of the deficiencies of the prior art, the present application provides a liquid static pressure rotary table detection method and system based on industrial vision, establishes a camera-witness ring-working surface-encoder unified coordinate chain, anchors the zero position and scale once and inverses the five degrees of freedom priori; obtains polarization / phase multi-frames without invading the oil cavity, constructs the disturbance fingerprint and maps it into equivalent film thickness, refractive index and scattering intensity under physical priori; learns the spatiotemporal continuous aberration compensation field and fuses it with the encoder angle circular domain to strip the eccentricity and tilt aliasing, outputs the unified angle and geometric quantity; forms the standardized curve and index according to ISO230-2 and ISO230-7, and links the threshold triggered rapid recalibration and protection operation and maintenance. Technical effect: consistent across working conditions, long-term drift-free, auditable and traceable. Improving the rotation accuracy, shortening the online time, thereby solving the technical problems recorded in the background technology.

[0006] (II) Technical solutions

[0007] To achieve the above object, the following technical solutions are adopted: obtaining polarization / phase multi-frames without invading the oil cavity, constructing the disturbance fingerprint and mapping it into equivalent film thickness, refractive index and scattering intensity under physical priori; learning the spatiotemporal continuous aberration compensation field and fusing it with the encoder angle circular domain to strip the eccentricity and tilt aliasing, outputting the unified angle and geometric quantity; forming the standardized curve and index according to ISO230-2 / 230-7, and linking the threshold triggered rapid recalibration and protection operation and maintenance. Technical effect: consistent across working conditions, long-term drift-free, auditable and traceable. Improving the rotation accuracy, shortening the online time, reducing the maintenance cost and energy consumption, thereby solving the technical problems recorded in the background technology.

[0008] The liquid static pressure rotary table detection method based on industrial vision comprises,

[0009] The plane homography matrix is obtained based on the plane homography and the quadratic curve, the pixel-angle mapping coefficient and the zero position offset are calibrated with the standard angle event, the five degrees of freedom error vector is inversely calculated with multiple short-range poses, and the circular concentration and harmonic energy baseline are established;

[0010] The plane homography matrix is used for geometric normalization, the intensity stack is collected by the polarizer and the retarder, the disturbance fingerprint of polarization consistency, phase gradient and circular domain phase correlation is constructed, and the equivalent film thickness, equivalent refractive index, scattering intensity and working domain indication are obtained according to the thin film interference and polarization response;

[0011] The aberration compensation field is estimated to be applied to the feature, the visual angle is re-estimated according to the mapping coefficient and the zero position offset, the corresponding visual angle and the encoder angle are fused, the first-order and second-order harmonic projection is performed, and the unified angle and the harmonic coefficient are outputted;

[0012] generate indicators with the harmonic coefficients according to ISO 230-2 / 230-7, trigger microstep angle and static bias events according to the circular concentration, the harmonic energy, and the working field indicator, update the mapping coefficients, bias zero, and phase offset.

[0013] Further, adopt witness ring with the same material and coating as the working surface, solve the plane homography matrix based on the constraint of the conic, and jointly estimate the pixel-angle mapping coefficients and the zero bias in the primary standard angle event according to the von Mises likelihood, and record the mapping from the witness ring plane to the image plane as the geometric alignment operator.

[0014] Further, construct a design vector in the uniform angle domain determined by the mapping coefficients and the zero bias under short-range multi-pose acquisition, couple the visual angle and the encoder angle, and invert the five-degree-of-freedom error vector, wherein the design vector includes eccentricity, tilt, and axial constant bias components, and define the circular concentration and the first and second harmonic energies according to the complex exponential as the consistency baseline.

[0015] Further, acquire intensity stacks using discrete settings of an electrically controlled polarizer and a controllable retarder, solve Stokes vectors after geometric normalization according to the plane homography matrix, construct the disturbance fingerprint including polarization consistency, phase structure gradient, and circular domain phase correlation, and perform segmented organization within the working condition template defined by oil temperature, flow rate, pressure, and ambient temperature and humidity.

[0016] Further, according to the physical constraints of thin film interference phase delay and polarization response, establish a proxy mapping with a penalty, map the disturbance fingerprint to the equivalent film thickness, the equivalent refractive index, and the scattering intensity, and generate the working field indicator according to polarization and phase consistency, and align the three types of quantities with the time axis of the uniform angle domain for output.

[0017] Further, construct aberration driving terms from the equivalent film thickness, the equivalent refractive index, the scattering intensity, and the main field phase gradient, and obtain aberration compensation fields through anisotropic spatial kernel and causal temporal kernel convolution; the spatial kernel covariance is set according to polarization consistency and phase consistency, the compensation fields are applied to the characteristic coordinates, and the visual angle is re-estimated according to the mapping coefficients and the zero bias.

[0018] Further, the compensated visual angle and the encoder angle are weighted and fused in the form of complex vector coincidence in the circular domain, the fusion weight is adjusted by the working field indicator and the circular concentration and satisfies non-negativity and sum of one, the updated uniform angle is obtained, and the first and second harmonic coefficients are stripped by integral circle projection on the uniform phase.

[0019] Further, the updated uniform angle and the first-order and second-order harmonic coefficients are mapped into a standard index vector according to ISO230-2 and ISO230-7, and a witness ring radius is combined to realize angle-length conversion, and a key sequence is resampled at a preset interval on the uniform phase axis and outputted to match the curve and list of the caliber.

[0020] Further, a threshold function driven by the circular concentration, the harmonic energy, the scattering intensity and the working domain indicator and oil temperature is constructed to trigger micro-step angle events and static polarization events, and the pixel-angle mapping coefficient and the zero-point bias are updated respectively, and the polarization zero point and the phase bias are updated, and the oil-repellent layer cleaning and the air curtain pulse cleaning strategy are linked.

[0021] The liquid static pressure turntable detection system based on industrial vision comprises,

[0022] A geometric calibration module obtains a plane homography matrix based on a plane homography and a conic curve, calibrates a pixel-angle mapping coefficient and a zero-point bias at a standard angle event, inverses a five-degree-of-freedom error vector at a short-range multi-pose, and establishes a circular concentration and a harmonic energy baseline;

[0023] A polarization fingerprint module performs geometric normalization with the plane homography matrix, collects intensity stacks with a polarizer and a retarder, constructs a disturbance fingerprint of polarization consistency, a phase gradient and a circular domain phase correlation, and obtains an equivalent film thickness, an equivalent refractive index, a scattering intensity and a working domain indicator according to thin film interference and polarization response;

[0024] A compensation fusion module is driven by the equivalent film thickness, the equivalent refractive index and the scattering intensity, estimates aberration compensation fields applied to features, reestimates visual angles according to the mapping coefficient and the zero-point bias, fuses corresponding visual angles and encoder angles, performs first-order and second-order harmonic projection, and outputs uniform angles and harmonic coefficients;

[0025] An index closed-loop module generates indexes according to the uniform angles and the harmonic coefficients according to ISO230-2 / 230-7, triggers micro-step angle and static polarization events according to the circular concentration, the harmonic energy and the working domain indicator, and updates the mapping coefficient, the polarization zero point and the phase bias.

[0026] (Three) beneficial effects

[0027] The application provides a liquid static pressure turntable detection method and system based on industrial vision, and has the following beneficial effects:

[0028] By unifying the coordinate chain and the primary anchor, the camera, witness ring, working surface and encoder are placed on the same angle scale and the same origin when the equipment is online, and the five-degree-of-freedom error prior and the visual-encoder consistency baseline are given, avoiding repeated calibration and caliber drift, and ensuring that the results of different shifts, different clamping and different batches can be aligned and traced.

[0029] Without changing the main field of view and without invading the oil cavity, a disturbance fingerprint containing Stokes quantity and phase structure is constructed, and thin film interference and polarization response are used as physical priors to robustly map into equivalent film thickness, equivalent refractive index and scattering intensity, and combined with the working area to indicate shielding bubbles and high light burst anomalies, realizing the explainable and transferable characterization of optical disturbance.

[0030] When learning the spatiotemporal aberration compensation field driven by physical priors, it directly acts on the main field of view feature coordinates, and strips the optical drag and micro-drift caused by the oil film; in the unified angle domain, the compensated visual angle and the encoder angle are circularly weighted and fused, and combined with the integral circular harmonic projection to strip the eccentricity and tilt aliasing, and output the robust unified angle and rotation and repeatability quantification, realizing the combination of short-term stability and long-term drift-free.

[0031] The fusion results and harmonic quantization are mapped into statistical and harmonic indicators in accordance with relevant provisions, and curves and lists are automatically generated, which can be directly reviewed and accepted by third parties; different devices and batches can be compared horizontally according to the unified caliber, significantly reducing the report arrangement and recalculation cost, and improving the audit transparency.

[0032] According to the oil temperature interval and the imaging health degree, a threshold triggering mechanism is constructed to quickly complete the online recalibration of pixel-angle mapping coefficients, polarization zero point and phase bias, and to link the oil-repellent protection and air curtain maintenance, so that the system can be restored to a healthy state in the early reversible stage, reducing downtime and manual intervention, and maintaining consistency and stability throughout the life cycle. BRIEF DESCRIPTION OF DRAWINGS

[0033] Figure 1 The figure is a flowchart of the liquid static pressure turntable detection method based on industrial vision of the present application.

[0034] Figure 2 The figure is a structural diagram of the liquid static pressure turntable detection system based on industrial vision of the present application. DETAILED DESCRIPTION

[0035] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0036] Referring to Figure 1 The application provides an industrial vision-based liquid static pressure rotary table detection method, comprising,

[0037] ISO 230-2 is "Machine tools-Test code-Part 2: Determination of centering and alignment accuracy of numerically controlled axes". It specifies the test and evaluation method for the positioning accuracy and repeatability of linear and rotary axes of CNC machine tools (directly on the machine). The current public version is commonly ISO 230-2:2014.

[0038] ISO 230-7 is "Machine tools-Test code-Part 7: Geometric accuracy of rotary axes". It standardizes the geometric accuracy test method of rotary axes (spindles, rotary heads, rotary / swing worktables), covering indicators such as rotary axis error motion and speed-induced axis deviation. The current public version is commonly ISO 230-7:2015.

[0039] Step one, in the online scene, the camera, witness ring, work surface and encoder are brought into the same coordinate and angle scale, the zero position and scale are anchored at the same time through a standard angle / geometric event, and a short-range multi-position inverse five-degree-of-freedom assembly priori is used to generate a vision-encoder consistency baseline and residual statistics, which supplements the deficiencies of zero position drift, scale inconsistency and invisible geometric error from the source, and provides a unified and traceable reference for subsequent disturbance modeling and fusion.

[0040] Among them, the internal and external participating pixel-angle scale calibration of the camera imaging link is completed with the goal of unified geometry and optical-mechanical link, and the zero position and scale are anchored at the same time with a standard angle / geometric reference, so that the vision angle θ v and the encoder angle θ e are aligned at the same scale and origin.

[0041] In the liquid static pressure rotary table scene, the oil film perturbation and the weak anisotropy of the work surface coating will cause the edge phase and geometric features observed by the camera to slowly drift with the attitude; if the unified coordinate chain of camera-witness ring-work surface-encoder is not first established, the subsequent aberration compensation field with polarization / phase signal as input will lose the absolute scale and zero reference. Therefore, the witness ring with the same material / coating as the work surface is needed as an optical geometric reference, and the zero position and scale are anchored at the same time through a geometric-angle calibration strategy in the industrial field, and the mapping operator is recorded explicitly for cross-condition calling.

[0042] A single-chain process with planar homography calibration, conic quadratic curve constraint robust fitting and standard angle event maximum likelihood anchoring. First, with the witness ring as the observation window, the planar homography of the ring surface is solved without changing the main field of view, and the mapping from the witness ring plane to the image plane is obtained; then the constraint is constructed by using the quadratic curve preserving property of the circle in imaging, so that the deviation of the geometric center and the short axis of the projected ellipse from the main point of the main field of view is corrected; finally, the pixel-angle scale and zero offset are determined simultaneously through a standard angle event (provided by a standard block gauge or an encoder reference scale line), and the pixel-angle mapping coefficient γ and the zero offset β0 are output, providing a callable calibration pair for subsequent short-range multi-pose inversion and cross-modal fusion.

[0043] In the witness ring field of view, the witness ring edge can be regarded as a quadratic curve on the image obtained by planar homography transformation. Considering the homography matrix from the witness ring plane to the image plane, the invariants of quadratic curve transformation are used for robust estimation, so that the geometric center and scale can still be stably recovered in the presence of local highlights and edge micro-gaps. The core constraint is written as:

[0044]

[0045] where: the image quadratic curve matrix represents the symmetric matrix of the projected ellipse, which is real symmetric and rank 2, taking real number domain and satisfying the non-negative definite geometric constraint; the planar homography matrix represents the projective mapping from the witness ring plane to the camera image plane, with real elements and being invertible;

[0046] the ring plane quadratic curve matrix is constructed by the witness ring geometric radius r w >0 and the unit conic quadratic curve matrix constructed by the orthogonal basis of the ring plane, which is used to describe the geometric ideal model of the ring in its own plane; the witness ring geometric radius is the witness ring physical radius, which is consistent with the material and coating and the working surface, used to ensure the homogeneity of spectral response and geometric edge.

[0047] Based on the above constraints, combined with the process conditions of the same material / same coating, the system fixes the imaging process with fixed exposure time and aperture, and locks the light transmission direction of the polarizer. First, the planar homography matrix H cw is roughly estimated, and then the image quadratic curve matrix C c is obtained by robust fitting with generalized extreme value suppression, which is consistent with the planar homography matrix H cw ; therefore, the deviation of the witness ring center from the coordinate origin of the main field of view is quantified, so that the pixel geometry and physical geometry are aligned in the same coordinate system, and then the reversible affine approximation is prepared for pixel-angle mapping.

[0048] When used, even if there are local reflections and oil film glint halo, the projected ellipse can still be recovered stably by the quadratic curve constraint; the witness ring and working surface are treated equally in the unified coordinate chain, reducing the system error introduced by material / paint mismatch; the homography level geometric closure is provided for subsequent angle mapping and zero position anchoring.

[0049] To ensure the homography matrix H from the witness ring plane to the image plane cw Robust joint estimation of the projected quadratic curve C c and the robust loss function, a two-stage algorithm is proposed as follows:

[0050] Stage I (coarse estimation / RANSAC): fit the quadratic curve initial value C0 with the sub-pixel edge sampling point set {p i} Outliers with distance > 2σ are filtered out with the minimum algebraic distance with a normalization strategy; the four-point correspondence estimation is used Stage II (IRLS consistent estimation): the target function is minimized with the initial value C0 and the edge point set {p

[0051]

[0052] Iterative reweighted least squares (IRLS) is performed.

[0053] where: robust loss : Huber loss, threshold set by edge signal-to-noise ratio;

[0054] Point-to-quadratic curve orthogonal distance : geometric distance; regular weight : penalty factor for the constraint homography consistent with the ideal conic; Frobenius norm ||·|| F : matrix norm; ring plane ideal quadratic curve C w : explicitly constructed with the homogeneous form of the unit circle .

[0055] After geometric alignment, the visual angle θ v and the encoder angle θ e need to be mapped to the unified angle domain. To avoid the destruction of the periodicity of the angle variable by the common linear regression, the von Mises likelihood of circular statistics is used to solve the pixel-angle mapping coefficient γ and the zero position bias β0 simultaneously in a standard angle event:

[0056]

[0057] In the formula: pixel-angle mapping coefficient Scale factor mapping encoder angle domain to vision angle domain, positive value, engineering corresponding to the pixel phase gain per radian; zero bias β0∈(-π, π]: unified angle domain reference zero, defined with encoder reference scale or standard block rule triggered angle as reference;

[0058] Sample count : Standard angle event sampling frame number, at least covering one short arc to ensure robust estimation;

[0059] von Mises concentration parameter Control the preference of likelihood to phase consistency, set according to image SNR and edge clarity, the larger the weight is more concentrated; vision angle sequence H cw Normalized and phase-solved angle samples; encoder angle sequence Angle samples output by the encoder during the standard angle event.

[0060] Under the von Mises likelihood of circular statistics, the joint solution of pixel-angle mapping coefficient γ and zero bias β0 can use the gradient-Newton hybrid method, and the identifiable condition is given: sampling phase At least cover two non-symmetric sectors (avoid π symmetric degeneration), N≥8. Incremental update (convenient for on-site iteration):

[0061]

[0062] Where: step size η∈(0, 1]; gradient Jacobi approximation : von Mises consistency gradient and approximate Hessian calculated

[0063] The maximum problem is given by the maximum likelihood condition to give a closed or semi-closed update formula, which can complete the joint anchoring of pixel-angle mapping coefficient γ and zero bias β0 through one event; Therefore, the vision angle domain establishes a consistent affine relationship with the encoder angle domain, thereby unifying the zero and scale, and then all subsequent angle measurements can be directly projected into the unified angle domain for comparison and fusion.

[0064] When used, the phase wrap-around problem of angle regression is avoided by using the maximum likelihood of circular statistics; one event simultaneously solves the zero and scale, reducing the time cost of repeated calibration and working condition disturbance alignment; The obtained γ and β0 have traceable statistical meaning, which is convenient for cross-batch review.

[0065] ​On the basis of unified scale and zero position, short-range multi-position data are collected to inverse five-degree-of-freedom error priors and construct a consistency baseline and residual statistics of vision-encoder angle, which provide initial values and criteria with physical meaning for subsequent disturbance fingerprint mapping and aberration compensation.

[0066] Hydrostatic rotary tables may still be affected by eccentricity, tilt and assembly bias in the small-angle segment, which is manifested as non-ideal circumferential motion and phase micro-distortion of feature points. Without explicit five-degree-of-freedom error priors and consistency baseline, the physical agent of disturbance fingerprint will be mixed with geometric errors. Therefore, the dominant harmonic term of geometric error and the slowly varying term induced by oil film need to be distinguished under short-range multi-position collection, so that subsequent aberration compensation only deals with optical disturbances without repeated compensation of structural errors.

[0067] Firstly, the vision angle θ v and the encoder angle θ e are projected into the unified angle domain and resampled according to the rotation phase φ ; then the harmonic design vector is constructed for small-angle approximation, and the five-degree-of-freedom error vector is inverted; finally, the circular concentration degree and low-order harmonic energy represented by complex exponent are taken as the consistency baseline to quickly detect drift or assembly changes during online operation.

[0068] In the small-angle segment, the effects of eccentricity and tilt on the vision angle θ v can be represented as a linear combination of the finite harmonic basis. Let the five-degree-of-freedom error vector be The rotation phase φ is linearly coupled with the error:

[0069]

[0070] In the formula: vision angle distortion δθ : angle offset caused by geometric error in the unified angle domain, used to strip the structural term from the vision angle; design vector : defined as carries the main harmonic components of eccentricity and tilt and the axial constant bias, respectively;

[0071] Rotation phase φ : phase parameter in the unified angle domain, obtained by normalizing the encoder angle θ e ; eccentricity error : represents the equivalent eccentricity in the x / y direction of the working surface, corresponding to the component; tilt error : represents the equivalent tilt around the x / y axis, corresponding to the component; axial constant bias ​represents a constant term related to the clamping reference, used to absorb mean bias.

[0072] By minimizing the inconsistency of the residual on the unit circle (with γ, β0as fixed mapping), the estimation of the five-degree-of-freedom error vector ξ is obtained, thus explicitly modeling the geometric dominant term; therefore, the disturbance fingerprinting stage can treat optical drags and micro-shifts as explained quantities, thus avoiding the misattribution of geometric errors to physical priors; further, this prior is used in the integral harmonic analysis to initialize the filter state, providing a physically consistent initial state for the fusion in the third step.

[0073] In use, the explicitization of decentration / tilt / constant bias is enhanced by harmonic decoupling, enhancing interpretability; only short-range multi-pose data is required for stable inversion, improving online efficiency; the obtained five-degree-of-freedom error vector ξ serves as a shared geometric prior across steps, making subsequent aberration compensation only act on optical disturbances, with clear synergistic effects.

[0074] To judge the measurement health in subsequent running periods, baseline indicators need to be constructed, defining the circular concentration indicator ρ and the low-order harmonic energy E m , respectively reflecting instantaneous consistency and harmonic aliasing strength:

[0075]

[0076] In the formula: the circular concentration ρ ∈ [0, 1]: represents the aggregation degree of the residual phase on the unit circle, the closer to 1, the better the consistency; the complex imaginary unit i: satisfies i 2 = -1, used to map the angle residual to the unit circle of the complex plane; the angle residual ε k ∈ (-π, π]: unified angle domain under a single frame residual, with the five-degree-of-freedom error term stripped out;

[0077] The wrap (·) function: an operator that maps real angles to the principal value interval (-π, π], avoiding ±π jumps;

[0078]

[0079] In the formula: the input The output falls in [a, b); is the floor function. The equivalent writing is:

[0080] wrap [a,b) (x) = a + ((x-a) mod (b-a)), where mod returns the remainder in [0, b-a).

[0081] Sample count The number of short-range multi-pose sampling frames; harmonic energy denotes the energy of the residual on the mth harmonic, used to identify the assembly bias and eccentricity-induced aliasing; harmonic order m∈{1,2}: first and second order are taken to correspond to the dominant frequencies of eccentricity and tilt; rotation phase sequence : uniform phase corresponding to the sample;

[0082] pixel-angle mapping coefficient zero bias β0∈(-π,π], visual angle encoder angle visual angle distortion : as defined above.

[0083] Thus, the baseline takes ρ as the main indicator of consistency, and E1, E2 as the secondary indicators of harmonic aliasing; therefore, once the on-line roundness concentration indicator ρ drops or the energy E m of a certain low-order harmonic rises, it is determined that there is a decrease in imaging health or a change in assembly bias, thereby triggering the on-line self-correction of the fourth step; finally, the baseline-threshold-triggered closed loop is made explicit and parameterized output, facilitating third-party review.

[0084] When used, the circular domain statistics replace the linear variance measure, improving the robustness to phase wrapping and multi-peak distribution; the harmonic energy separates the effects of assembly geometry from optical disturbances, providing more indicative information; the parameters can be directly fed into the subsequent operation and maintenance closed loop, forming a traceable chain of detection-review-trigger.

[0085] Step two, under the condition of not changing the main field of view and not interrupting the process, jointly collect multiple frames of polarization / phase, construct a disturbance fingerprint containing Stokes quantity and phase structure, and under the prior constraints of thin film interference and polarization response, robustly map it to equivalent film thickness, refractive index and scattering intensity, define a reasonable working domain combined with soft gating, make up for the shortcomings of single statistical aperture and non-migratable across working conditions, and provide interpretable input for aberration compensation.

[0086] Based on the unified angle domain and unified coordinate chain, the polarization / phase multi-frame synchronous acquisition and Stokes vector calculation are completed, and under the driving of engineering working condition variables, a reusable disturbance fingerprint is constructed, so that the subsequent physical prior agent has sufficient information supply and cross-condition migratability.

[0087] The oil film of the hydrostatic turntable will produce thickness slow change and micro zone refractive index disturbance under different oil temperature and pressure conditions, superimposed with the weak anisotropy of the witness ring plating layer, causing low-frequency dragging and micro-drifting of the main field of view characteristic trajectory. If only single intensity or single channel phase observation is used, it will not be able to distinguish the change of reflection component caused by the change of polarization state from the change of interference phase caused by the change of film thickness. Therefore, based on the unified coordinate chain and unified angle domain (pixel-angle mapping coefficient γ, zero bias β0, plane homography matrix H cw, five-degree-of-freedom error vector ξ) is designed, so that the disturbance information has the consistency of coordinates and scales when it is passed back; therefore, the subsequent proxy mapping is no longer confused by geometric errors and phase rotation, thereby ensuring the uniqueness of physical interpretation.

[0088] First, keep the main field of view and the witness ring field of view unchanged, and only switch the azimuth angle of the external electrically controlled polarizer and the setting of the controllable phase retarder; second, use the plane homography matrix H cw Map the witness ring area to the unified plane coordinates, remove the perspective distortion and align the main field of view coordinate origin; then, according to the unified angle phase Collect the intensity stack in sequence, solve the Stokes vector and align it with the main field of view phase field; then, construct the discrete working condition template and threshold cluster according to oil temperature, flow, pressure and environmental temperature and humidity for segmentation; finally, in each working condition segment, the disturbance fingerprint is formed by the polarization consistency, phase structure gradient and circular domain phase correlation, and the time series consistent with the unified angle domain is output, providing input for the physical priori agent.

[0089] Through the discrete setting of the electrically controlled polarizer and the controllable retarder, multiple intensity stacks are obtained, and the Stokes vector is solved in the form of weighted least squares after geometric normalization, so that the polarization state can be dimensionless under the disturbance of the oil film. Considering the pixel points on the witness ring normalized coordinate system, the relationship between the intensity stack and the Stokes vector is written as:

[0090]

[0091] In the formula: the Stokes vector A four-dimensional quantity describing the polarization state, a real number, is the polarization component of the disturbance fingerprint; the intensity stack The intensity observation vector under J≥4 polarization-delay combinations is obtained through the plane homography matrix H cw After geometric normalization; the analysis matrix The association matrix determined by the azimuth angle of the polarizer and the phase of the retarder of each sampling frame, the elements are known real numbers; the weighting matrix The diagonal matrix is weighted by the signal-to-noise ratio and the cosine of the incident angle for each frame of intensity, and is positive definite; further,

[0092]

[0093] Where: the incident angle θ i ∈[0, π / 2); the standard deviation of the intensity noise Numerical stability term

[0094] Polarization solving operator Given by the right side closed expression, which ensures robustness when the light fluctuates slightly.

[0095] In use, the comparable s is obtained on the unified coordinate chain and the unified angle domain, avoiding the influence of the viewing angle and the magnification change; the specular highlight frame caused by the oil film is suppressed by the weight matrix W, so that the solution is not sensitive to local abnormalities; and a high signal-to-noise polarization description is provided for subsequent working condition segmentation and physical prior agent.

[0096] Without changing the main field of view, a unified angle phase For the time parameter, the oil temperature, flow, pressure and environmental temperature and humidity are mapped into a discrete working condition template, so that the disturbance statistics in the same template are approximately stationary. Then, a disturbance fingerprint vector is constructed, which simultaneously represents the polarization consistency, phase gradient intensity and circular domain phase correlation:

[0097]

[0098] In the formula: disturbance fingerprint vector In order, containing three statistical quantities, as the input of the subsequent physical prior agent;

[0099] Phase field φ ∈ (-π, π]: main field phase, has been projected to the unified angle domain according to the pixel-angle mapping coefficient γ and the zero position offset β0 defined as before;

[0100] Polarization consistency : Measure the order of polarization; phase structure intensity The normalized path integral of the phase gradient on the witness ring arc length l>0; circular domain phase correlation Where Δφ m The main value of the phase difference of the adjacent pixel pair, the sample number Visual angle sequence And pixel-angle mapping coefficient γ, zero position offset β0: defined as before, used for solving and resampling.

[0101] In use, the three-element fingerprint is complementary in information, which can simultaneously reflect the polarization order, phase roughness and phase consistency; f is constructed under the unified angle domain to ensure cross-segment alignment and comparability; the fingerprint dimension is low and the dimension is clear, which is convenient for the training and deployment of subsequent agent mapping.

[0102] The disturbance fingerprint is mapped into the equivalent film thickness fluctuation, refractive index fluctuation and scattering intensity under the physical prior constraint of film interference and polarization response, and then a reasonable working domain is defined and bubble clusters and other sudden abnormalities are shielded, so as to provide a stable and interpretable intermediate quantity for aberration compensation field learning.

[0103] Directly entering aberration compensation with fingerprint quantity only will mix and superimpose the contributions of film thickness fluctuation and refractive index fluctuation, and it is difficult to ensure consistency when migrating across working conditions.

[0104] Therefore, it is necessary to introduce a physical prior agent to make the mapping follow the phase delay law of thin film interference and the energy distribution law of polarization response, and to take the consistency baseline indicators ρ, E1, and E2 given in step one as external constraints of operation health degree; thus, the coupling path of disturbance to optical aberration is disassembled, and subsequent filtering fusion can work in a lower-dimensional and more physically coupled space.

[0105] First, the phase delay formula of the equivalent thin film is used to depict the decisive influence of film thickness and refractive index on phase; then, a proxy mapping with a penalty is constructed under the constraint of the forward relationship, so that data-driven and physical consistency are compatible; subsequently, a smooth working domain indicator function is constructed based on polarization consistency and circular domain phase consistency, and abnormal soft shielding is realized in combination with the consistency baseline indicators; finally, the equivalent film thickness fluctuation, refractive index fluctuation, and scattering intensity sequence are output and aligned with the unified angle domain time axis as the driving variables of the next step of aberration compensation field.

[0106] In order to make the mapping have physical constraints, a joint form of forward physical relationship + agent regression + consistency penalty is adopted. The phase delay physical model of the equivalent thin film under the incident angle is defined as follows:

[0107]

[0108] In the formula, the physical phase delay is the phase delay generated by the equivalent thin film; the equivalent refractive index is the equivalent refractive index of the oil film, which is positive; the equivalent film thickness is the equivalent thickness of the oil film, which is positive; the incident angle θ i ∈ [0, π / 2): the effective incident angle of the main field of view, determined by the device geometry and H cw is determined; the working wavelength is the center wavelength of illumination, which is positive.

[0109] On this basis, a proxy mapping with a physical consistency penalty is constructed with the disturbance fingerprint as the independent variable:

[0110]

[0111] In the formula, the equivalent film thickness is one of the optimization variables, corresponding to the film thickness estimate; the equivalent refractive index is one of the optimization variables, corresponding to the refractive index estimate; the scattering intensity is one of the optimization variables, representing the order of magnitude of non-coherent scattering induced by microparticles and surface roughness; the robust penalty is a Cauchy-type penalty with a scale parameter is the phase equivalent delay is the equivalent delay extracted from the main field of view phase φ (processed by the rotation function defined as before); the penalty weight : Coefficients balancing physical consistency and proxy regression residuals; coefficient matrix Linear regression coefficients, computed offline; ||·||1: l1-norm, to promote sparsity and improve generalization across conditions;

[0112] Feature mapping Polynomial-rational function expansion on perturbation fingerprint vector f, e.g. containing DoLP, η l , C Δ and their rational combinations and logarithm;

[0113]

[0114] where Approximated by polarization inconsistency 1-DoLP, solver uses alternating direction method of multipliers (ADMM) to convexly approximate Cauchy consistency + l1-regularization objective, coordinate-wise iteration.

[0115] In use, physical phase retardation is used as consistency constraint to make film thickness and refractive index estimation not deviate from interference law; robust penalty is used to suppress outlier influence brought by sporadic bright spots and local scattering surge; output Directly used as physical driving quantity of aberration compensation field, to improve interpretability and stability of compensation.

[0116] To avoid bubble cluster, oil film rupture boundary and mirror highlight and other sudden abnormal pollution of subsequent estimation, construct continuous soft gating indicator function between polarization consistency, circular domain phase consistency and harmonic baseline given in step one, neither exclude effective data too much nor miss significant abnormality:

[0117]

[0118] where: working domain indicator χ∈(0,1): soft gating weight, the closer to 1, the more reliable the sample; logistic function Smooth gating function, monotonically differentiable;

[0119] Polarization consistency : consistent with DoLP, used to measure polarization order degree; phase consistency : circular domain phase concentration; gating slope : positive value parameter controlling sharpness of logistic function switch;

[0120] Threshold π min ,η min ∈(0,1)、 : respectively, lower limit of polarization and phase consistency, upper limit of harmonic energy;

[0121] π min =Q 0.2 (π),ηmin = Q 0.2 (η), E * = Q 0.8 (E1 + E2)

[0122] where: quantile operation Q q (·): q-quantile of sample distribution.

[0123] Consistency baseline indicator As defined above, used to indicate the residual strength of eccentricity and tilt.

[0124] In use, time jitter and information break caused by hard threshold are avoided by continuous gating; geometric harmonic baseline and optical consistency are jointly constrained, which can give more robust reservation / shielding decisions when assembly bias changes and optical abnormalities coexist; soft weight χ can be directly used for time filtering of and weighting of aberration compensation field samples, forming a seamless interface.

[0125] Step three, learn the spatiotemporal continuous aberration compensation field driven by disturbance fingerprint + physical prior, directly correct the optical drag and micro-drift of the main field feature; in the unified angle domain, the compensated visual angle and the encoder angle are circularly weighted and fused, and the eccentricity and tilt aliasing are stripped by the integral circular harmonic, and the robust unified angle and the rotation / repeatability are outputted, which makes up for the contradiction between short-term noise and long-term drift.

[0126] Under the premise of witnessing ring geometry normalization and unified angle domain, equivalent film thickness, equivalent refractive index and scattering intensity are used as physical driving to construct spatiotemporal continuous aberration compensation field and apply it to the main field feature coordinates, and obtain the compensated visual angle, which provides high-fidelity, low-drift observation for subsequent filtering fusion.

[0127] Step two has outputted disturbance fingerprint and physical prior in the unified angle domain, including equivalent film thickness Equivalent refractive index Scattering intensity And working domain indicator χ. These quantities and the main field phase φ together characterize the spatiotemporal structure of optical drag caused by the oil film. If direct compensation is performed at the pixel level smoothing or rigid registration, it will inevitably couple the real small pose change with the optical aberration, leading to angle drift difficult to converge.

[0128] Therefore, the strategy of spatiotemporal filtering-field driven displacement driven by physical quantities is needed, which estimates the aberration compensation field on the witness ring normalized coordinates first, and then projects and acts on the main field feature, so as to strip the optical displacement induced by thin film interference and polarization response from the source; therefore, the compensated visual angle can maintain the same scale comparability with the encoder angle in the unified coordinates and unified angle domain, thereby laying the foundation for subsequent fusion. ​

[0129] First, the main visual field phase gradient and physical prior aberration driving term are constructed; then the anisotropic spatial kernel and causal time kernel are convolved and integrated to obtain a time-space continuous aberration compensation field; then the compensation field is applied to the feature coordinates after step one geometric normalization; finally, the compensated visual angle is re-estimated in the unified angle domain, and the five-degree-of-freedom error prior is used for geometric stripping to make it a high-quality visual measurement of the fusion link.

[0130] To avoid over-compensation of the structure pose, the physical prior is pressed into the driving term and integrated in time with the causal kernel, and the working domain indication is used for soft gating to make the compensation field only effective for trusted samples. The construction of the aberration compensation field is written as:

[0131]

[0132] In the formula: aberration compensation field Witness the two-dimensional displacement field on the ring-normalized coordinates, the components are Ψ x ,Ψ y , used to correct the main visual field feature coordinates;

[0133] Spatial anisotropic kernel : Elliptical Gaussian kernel, whose covariance is modulated by polarization consistency and phase consistency along the principal axis. In the implementation, σ || ∝(1-DoLP) -1 ,σ ⊥ ∝(1-C Δ ) -1 are set; further: is a two-dimensional Gaussian kernel

[0134]

[0135] And let the kernel direction be determined by the phase gradient direction , and the scale is:

[0136] σ || = s0 / (∈+DoLP),σ ⊥ = s0 / (∈+C Δ )

[0137] Discrete implementation uses FFT convolution or separate kernel expansion. Where: rotation matrix Scale constant Stabilizing term

[0138] Spatial convolution operator★ x : Convolution operator for spatial variable x, linear and translationally invariant;

[0139]

[0140] where: anisotropic Gaussian kernel (covariance modulated with direction and consistency), vector field g = (g x ,g y ). Result is still homogenous vector, convolution per component.

[0141] time forgetting kernel exp(-(t-τ) / τ c ): causal exponential kernel, time constant control memory length; working domain indicates χ(τ) ∈ (0, 1): soft gating weight defined in step two, used to suppress outliers; aberration driving term : two-dimensional vector field defined in the following formula;

[0142] witness annulus normalized coordinates time : spatial and temporal arguments. Aberration driving term takes the form of linear combination with physical interpretation:

[0143]

[0144] where: gain coefficients respectively measure the contributions of equivalent film thickness, equivalent refractive index and scattering intensity to optical drag, positive values determined by offline calibration; equivalent film thickness equivalent refractive index scattering intensity physical prior defined in step two; reference refractive index : nominal refractive index at device calibration; phase gradient : spatial gradient of phase in main field, direction consistent with optical drag direction.

[0145] In use, the aberration compensation field is directly driven by the physical prior, making the compensation consistent with the interference law and scattering characteristics in the same direction; the causal time kernel and anisotropic spatial kernel simultaneously constrain the compensation to extend smoothly in space and converge smoothly in time; the soft gating weight ensures that the influence of abnormal samples on the compensation is suppressed, thereby improving the spatiotemporal stability.

[0146] After the aberration compensation field is obtained, it needs to be robustly transferred from the witness annulus normalized coordinates to the main field feature coordinates, and the visual angle is re-estimated in the unified angular domain. First, the plane homography matrix H cw completes geometric projection consistency; then, the five-degree-of-freedom error vector ξ is used to a priori strip the geometric distortion term; finally, the post-compensation visual angle is calculated in the unified angular domain:

[0147]

[0148] where: post-compensation visual angle : Visual angle measurement in uniform angle domain; Pixel-angle mapping coefficient Zero bias β0∈(-π,π]: Joint anchoring result of step one; Characteristic coordinate H cw Main field feature position after geometric normalization and deduction of five-degree-of-freedom prior; Aberration compensation component : Value of aberration compensation field at corresponding position and time;

[0149] Complex argument operator arg(·): Complex argument operator that maps two-dimensional coordinates to circular domain angles, outputs principal value, and is continuous and differentiable for any

[0150] arg(z)=atan2(y,x)∈(-π,π]

[0151] Also equivalent to:

[0152]

[0153] Where atan2 is the four-quadrant inverse tangent, ensuring that the angle falls within the principal value interval (-π,π].

[0154] In use, through the series connection of geometric normalization-prior stripping-angle re-estimation, the remaining error of visual angle mainly comes from optical disturbance rather than structural error; After compensation, the visual angle and the encoder angle are naturally comparable in the uniform angle domain, reducing the data reorganization overhead before fusion; The mapping from the aberration compensation field to the angle re-estimation is continuous and differentiable, providing conditions for the sensitivity analysis of subsequent filters.

[0155] In the uniform angle domain, the compensated visual angle and the encoder angle are fused by circular domain weighted filtering to obtain a robust unified angle, and the eccentricity, tilt and assembly bias aliasing are stripped by integral circular harmonic projection, and the rotary and repetitive shape and position quantities are output.

[0156] The encoder angle has excellent quantitative stability in the short term, but will be slowly affected by temperature drift and mechanical backlash in the long term; The compensated visual angle has stronger awareness of slow drift in the long term, but is more affected by shot noise and local reflection in the instantaneous frame. Therefore, a fusion filter that can dynamically adjust the weight needs to be constructed on the circular domain, so that the short-term is biased towards the encoder and the long-term trusts the vision, and the circular concentration and harmonic energy given in step one are used as the basis for weight modulation; Subsequently, the integrated circular harmonic projection is performed on the fused angle to explicitly strip the residual first and second order aliasing terms, so that the final unified angle and shape and position quantization are immune to assembly and geometric bias.

[0157] First, the encoder angular velocity is taken as the state model for circle domain prediction; then, the circle vector of the compensated visual angle and the encoder angle is combined to realize updating, and the weight is adjusted by the working domain indication and the circle concentration; then, the updated angle is projected and peeled off on the first and second harmonics in the unified phase; finally, the unified angle sequence and the indexes such as rotation accuracy and repeatability are output, which are used for the ISO mapping and operation and maintenance closed loop in step four.

[0158] The fusion update adopts the form of circle domain complex vector combination, avoiding the linear filtering distortion caused by angle rotation. The update equation is:

[0159]

[0160] In the formula: the updated unified angle : the unified angle output in this step; the predicted unified angle : the predicted value obtained by extrapolating the last time update angle and the encoder angular velocity; the prediction term The encoder angular velocity ω e and the last estimation form:

[0161]

[0162] In the formula: the encoder angular velocity ω ; the time step Δt

[0163] The prediction weight α t ∈(0, 1): the time smoothing coefficient, which can be taken as α t = exp(-Δt / τ s ) in engineering; the time constant τ controls the inertial strength; the imaginary unit i: satisfies i 2 =-1, which is used for circle domain embedding;

[0164] The compensated visual angle θ The encoder angle ω : the encoder reading after the scale alignment of the unified angle domain; the visual weight α The encoder weight α is a non-negative weight, which satisfies α t +w v +w e >0; further,

[0165] w v = λ v χρ, w e = λ e (1-χρ) is supplemented as the boundary: to ensure w v +w e ≥ w min >0; when χρ<∈ w , we ←w e +w safe . Wherein: minimum weight

[0166] To achieve adaptive allocation of short-term encoder-long-term vision, the weight can be set as: w v = λ v χρ, w e = λ e (1-χρ), where the weight coefficient is a tuning constant, the circular concentration degree ρ∈[0,1] and the working domain indication χ∈(0,1) are from the previous steps.

[0167] In use, the complex vector is naturally processed in the circular domain to avoid the rotation problem and angle jump; the weight is jointly modulated by χ and ρ, so that the filter is biased towards the encoder in the abnormal section and more trusted vision in the healthy section, thereby achieving the balance between short-term stability and long-term drift-free; the parameter form is fully disclosed and can be directly adjusted on site, which has engineering implementability.

[0168] The updated unified angle may still contain low-order aliasing caused by eccentricity and inclination. Therefore, a harmonic projection operator is constructed on the unified rotation phase to strip the first and second order components from the angle sequence, and the final unified angle is obtained:

[0169]

[0170] In the formula: the final unified angle : the unified angle after stripping low-order aliasing; the updated unified angle : the angle function with the unified phase; the harmonic coefficient : the mth complex coefficient, the amplitude angle corresponds to the phase offset, and the modulus corresponds to the aliasing intensity; the unified phase : the rotation phase with the same scale as the encoder angle;

[0171] The real part operator The real part of a complex number is an operation that takes the real part of a complex number or a complex-valued expression.

[0172] The harmonic coefficient adopts a weighted closed-form estimate:

[0173]

[0174] In the formula: sample count integer circular sampling number; soft weight χ k ∈(0,1): working domain indication corresponding to the sample; phase sample discrete unified phase; angle sample Updated uniform angle at corresponding phase; imaginary unit i: same as before.

[0175] In use, the first-order / second-order harmonic is removed by explicit projection, so that the eccentricity and tilt aliasing terms are quantified and stripped; the soft weight participates in the harmonic estimation, ensuring that the abnormal section does not dominate the projection; the final uniform angle obtained and the low-order harmonic coefficient α m At the same time, it is output, providing direct input for the ISO harmonic aperture mapping of step four.

[0176] Step four, map the fusion result according to ISO230-2 / 230-7 into standardized statistics and harmonic indicators, automatically generate curves and lists for third-party review; according to the oil temperature interval and health degree, trigger the polarization zero point, phase zero point and pixel-angle scale for rapid recalibration, and link the maintenance of oil and gas curtain, to make up for the shortcomings of inconsistent acceptance aperture, lagging maintenance and difficult tracking of parameter drift, and realize the auditable operation and maintenance closed loop.

[0177] Among them, the final uniform angle of the compensation fusion output and the low-order harmonic coefficient α1, α2 are mapped into ISO230-2 / 230-7 indicators according to the standard aperture, and curves and tables are generated with the uniform phase as the parameter, and then the one-key third-party review material output is realized.

[0178] Step three has obtained the final uniform angle in the uniform angle domain and the explicit coefficient α1, a2 of the first-order / second-order aliasing term. If the original sequence is directly provided for acceptance, it will face the problems of inconsistent aperture, inconsistent dimension and high review cost. Therefore, on the premise of not losing physical meaning, the rotary angle domain measurement needs to be converted into the indicator set described in ISO230-2 (numerical control shaft positioning performance) and ISO230-7 (rotary shaft accuracy); therefore, the subsequent review no longer depends on the on-site context, but can complete the consistency determination through standardized curves and lists; further, the standard mapper should also take into account the robustness to avoid distortion of the indicators caused by rare abnormal samples; finally, the output indicators will become a direct reference for the threshold triggering and recalibration strategy in step four.

[0179] Firstly, define the angle error function in the uniform phase domain; then calculate the one-way positioning deviation, reverse error and repeatability according to the ISO230-2 aperture; then construct the radial / tilt aperture of ISO230-7 with the witness ring radius and harmonic coefficient; finally, synthesize the curve and list under the same phase scale to generate the auditable report.

[0180] To ensure full disclosure and implementation, define the angle error function for rotary shaft and the standard indicator vector; and replace the point state sensitivity of extreme value with the essential bound to improve robustness:

[0181]

[0182] ISO230-2 indicator vector : one-way positioning error E P , reverse error E BR , repeatability R ε ;

[0183] one-way positioning error : the essential supremum of the uniform phase domain angle error function, reflecting the one-way driving positioning error envelope; reverse error : the essential supremum of the positive / negative difference represented by the phase difference π; repeatability : the median absolute deviation measure in the circular domain of two repeated measurements; uniform phase domain Φ = [0, 2π): parameter interval of the uniform angle domain;

[0184] wrap(·) : an operator that maps real angles to the principal value interval (-π, π], fixed as a monotone piecewise continuous function;

[0185] The wrap function refers to an operator that folds any real number (mostly angles) to a specified principal value interval to eliminate the 2π multi-valuedness, facilitating comparison, filtering, and statistics. It is denoted as wrap(·). [a,b)

[0186] The general form (to any half-open interval [a, b)) is:

[0187]

[0188] where: interval endpoints a < b; floor operator The output must fall within [a, b).

[0189] Final uniform angle : defined in step three; repeated measurement angle : the angle function obtained by repeating the measurement twice under the same working conditions; essential supremum esssup: the supremum ignoring zero-measure abnormal points, improving robustness; median med: the median operation equivalent to the circular domain, avoiding extreme value dominance.

[0190] Further, according to the harmonic aperture of the rotation axis in ISO230-7, the low-order aliasing coefficients and the witness ring radius r w are mapped to the dimensionless amplitudes of radial and tilt:

[0191]

[0192] ISO230-7 indicator vector ​Includes the amplitude of the first-order radial synchronization component H1, the amplitude of the second-order radial component H2, and the amplitude of the sway peak-valley T. PV Witnessing the radius of the ring Defined in step one, used for angle-to-length dimension conversion; harmonic coefficient : First-order and second-order complex coefficients defined in step three; complex modulus |·|: amplitude calculation; sway peak-valley amplitude The oscillation peak-valley index is approximated by the amplitude of the second harmonic.

[0193] In use, by replacing the fragile extreme values ​​and linear variance, the accuracy and robustness of the indicators are significantly improved; this is demonstrated by the witnessing ring radius r. w For harmonic coefficient α m The dimensionality is transformed to achieve angle-to-length conversion, which is consistent with the ISO230-7 standard; the indicators are vectorized and can be directly used for report generation and threshold determination, reducing the cost of human-computer interpretation.

[0194] To facilitate third-party review under a unified standard, robust smoothing and resampling of key sequences were performed on a unified phase axis, generating a data list and judgment conclusions. Variational robust splines were used. Smooth out:

[0195]

[0196] Where: robust spline The optimal function in the first-order Sobolev space serves as the baseline for the reporting curve; on the circular parameter interval Φ = [0, 2π), For all:

[0197]

[0198] The set of functions that hold true (values ​​integrable with the square of the first weak derivative) usually fits the periodic boundary ζ(0)=ζ(2π); integrable variational half-norm : Total variation term to suppress spike oscillations; Weights of data fit terms : Balancing smoothness and fidelity; unifying the phase domain Φ = [0, 2π), and the objective function derivative Differentiable functions and their derivatives within the same domain.

[0199] : Refers to the candidate report curve (an angle as a function of phase) on the unified phase domain Φ = [0, 2π). It belongs to the first-order Sobolev space. Used to approximate the final unified angle A smoothed version; the optimal curve is selected through optimization. :refer to The first order (weak) derivative of phase, which characterizes the local slope / rate of change of the curve. The integral of its absolute value constitutes the total variation term, which is used to suppress jitter, preserve kinks and microsteps.

[0200] The typical variational form (used to derive ) is:

[0201]

[0202] where the uniform phase is balanced by the weighting coefficient In engineering, the periodic boundary (ζ(0) = ζ(2π)) is taken. In discrete implementation, if the sampling points are the differential approximation can be used with circular difference. The output is the smooth real-valued angle sequence used to generate the ISO curve and table.

[0203] In use, the total variation-l1 composite regularization suppresses isolated noise points while preserving step / microstep information, and the curve, list and caliber threshold are integrated into the output after uniform phase resampling, so third parties do not need to do coordinate alignment again; the synthesizer retains the parameter λ to facilitate fast reuse and playback at different acceptance calibers.

[0204] On the basis of landing standard indicators, a threshold triggering mechanism driven by oil temperature interval and imaging health degree is constructed, and the polarization zero point ζ pol , the phase offset δ φ , and the pixel-angle mapping coefficient γ are quickly recalibrated, while the oil-repellent protection and air curtain maintenance strategies are linked, forming a closed-loop operation and maintenance.

[0205] Temperature drift, lens contamination and bubble clusters during operation can cause the polarization reference and phase zero point to shift, and then gradually degrade the scale and zero position of the uniform angle domain. If only manual intervention is made when the anomaly is significant, it will often have an impact on production rhythm and report effectiveness. Therefore, the consistency baseline ρ, E1, E2 of step one and the working domain indicator χ of step two need to be included in the threshold criterion, with oil temperature T o as an external parameter to trigger lightweight recalibration and link physical protection;

[0206] Therefore, self-correction can be completed in the early, slight and reversible stage; then, when the health degree enters the sub-healthy interval, the oil-repellent layer cleaning and air curtain strategy are triggered to restore the optical-mechanical link to a steady state; finally, the recalibration parameters are fed back to the operators of steps one / two / three to ensure consistency throughout the link.

[0207] First, the health degree and oil temperature are used to construct an operation and maintenance index and determine whether to trigger; then, in the short-term event without changing the main field of view, γ, ζ pol , δ φInstantaneous recalibration; then tasking of oil / gas repellent curtain according to pollution and scattering intensity; finally reevaluation with standard mapper, closed loop verification and parameter solidification.

[0208] To ensure that recalibration does not interrupt production, two types of very short events are designed: one type is microstep angle event, used to update γ; the other type is static polarization event, used to update ζ pol ,δ φ . The corresponding update formulas are as follows. First, the ratio of pixel-angle mapping coefficients is quickly updated:

[0209]

[0210] In the formula: pixel-angle mapping coefficient after recalibration This fast recalibration output; current pixel-angle mapping coefficient : last valid calibration value; encoder angle increment Encoder angle main value increment of microstep angle event; compensated visual angle increment Visual angle main value increment within the same event, from step three output; compensated visual angle Encoder angle Reading at the start and end of the event; wrap(·): same as the above definition, to ensure that the angle difference is within the main value domain.

[0211] Second, the synchronous update of polarization zero point and phase offset (use the static polarization event to calculate the reference difference between azimuth and phase on the witness ring):

[0212]

[0213] In the formula: polarization zero point after recalibration : updated polarization azimuth zero point; phase offset after recalibration Updated phase zero point; original polarization zero point ζ pol ∈(-π,π]、 original phase offset δ φ ∈(-π,π]: last valid calibration value; Stokes component Synchronous step two definition, current value under static polarization event; reference Stokes component Reference value when the device is shipped or last overhauled; phase median Reference phase median Circular domain median phase within the witness ring region; complex argument arg(·), wrap(·): same as the above definitions.

[0214] The closed-form update of the angle increment ratio eliminates the effect of velocity drift on γ, and the complex index is available in real time. The azimuth and median residual are used to correct the polarization / phase zero point, which is insensitive to local reflections and thermal noise. Both types of events are short, light, and non-intrusive, and do not change the main measurement field of view and production rhythm.

[0215] Early intervention is particularly critical in the early stages of health degradation. Constructing the operation and maintenance index and linking it with the threshold function related to the oil temperature interval, outputting the maintenance control signal, and taking the scattering intensity and oil-repellent layer state as auxiliary decision quantities:

[0216]

[0217] In the formula: operation and maintenance index : Weighted indicator of comprehensive health, harmonic energy, scattering, and working domain;

[0218] Weight coefficient : Non-negative weight set by acceptance criteria and working condition priority; circular concentration ρ(t) ∈ [0, 1], harmonic energy Scattering intensity Working domain indicator χ(t) ∈ (0, 1): Real-time quantity from steps one / two / three; time : Running time.

[0219] τ op (T o )=τ0+κ T (T o -T ref ) +

[0220] And define the air curtain pulse schedule

[0221]

[0222] Where: baseline threshold ; temperature coefficient reference temperature positive function (·) + ; duty cycle D duty ∈ [0, 1]; coefficient / upper and lower limits

[0223] When ζ op (t) > τ op (T o ) (where the oil temperature The threshold function τ op (T o ) is monotonically increasing with temperature to suppress false triggering in the high temperature section), the system sequentially issues:

[0224] First, micro-cleaning of the oil-hating layer (low-energy plasma / UV-cured refresh, maintaining the contact angle Θ ∈ (90°, 120°] in the target interval); second, pulse cleaning of the air curtain (pulse width and duty cycle proportional to the working condition); third, if it still does not recover, trigger the fast recalibration process and update the extended working condition template.

[0225] In use, the operation index integrates geometric harmonics, optical scattering and working domain consistency into a unified scale, with clear and adjustable triggering conditions; the temperature zone self-adaptive threshold avoids ineffective maintenance of high-temperature and high-flow sections; the physical protection of the oil-hating and air curtain in the closed loop cooperates with recalibration, both treating symptoms (contamination / scattering) and correcting root causes (zero point / scale).

[0226] Referring to Figure 2 The application provides an industrial vision-based liquid static pressure turntable detection system, comprising,

[0227] A geometric calibration module obtains a plane homography matrix based on a plane homography and a conic, calibrates pixel-angle mapping coefficients and a zero position offset at a standard angle, inverses a five-degree-of-freedom error vector from a short-range multi-pose, and establishes a circular concentration degree and a harmonic energy baseline;

[0228] A polarization fingerprint module performs geometric normalization with the plane homography matrix, collects intensity stacks with a polarizer and a retarder, constructs a disturbance fingerprint of polarization consistency, phase gradient and circular domain phase correlation, and obtains equivalent film thickness, equivalent refractive index, scattering intensity and a working domain indicator according to thin film interference and polarization response;

[0229] A compensation fusion module is driven by the equivalent film thickness, the equivalent refractive index and the scattering intensity, estimates aberration compensation fields applied to features, re-estimates visual angles according to the mapping coefficients and the zero position offset, fuses corresponding visual angles and encoder angles, projects first and second order harmonics, and outputs unified angles and harmonic coefficients;

[0230] An index closed loop module generates indexes according to the unified angles and the harmonic coefficients according to ISO 230-2 / 230-7, triggers microstep angles and static polarization events according to the circular concentration degree, the harmonic energy and the working domain indicator, and updates the mapping coefficients, the polarization zero point and the phase offset.

[0231] Those skilled in the art can appreciate that the units and algorithm steps of the examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether the functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Professionals can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the application.

[0232] ​Those skilled in the art can clearly understand the specific working process of the system, the device and the unit described above can refer to the corresponding process in the foregoing method embodiments for description convenience and brevity, and details are not described herein.

[0233] In several embodiments provided in the present application, it should be understood that the disclosed system, device and method can be implemented in other ways. For example, the device embodiments described above are merely illustrative, for example, the division of the units is only some logical function division, and actual implementation can have another division manner, for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some interfaces, devices or units, and can be electrical, mechanical or other forms.

[0234] The units described as separate components can or can not be physically separated, and the components shown as units can or can not be physical units, that is, they can be located in one place, or can be distributed on a plurality of network units. Part or all of the units can be selected to achieve the purpose of the embodiment scheme according to actual needs.

[0235] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of changes or replacements within the technical range disclosed in the present application, which should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A liquid static pressure turntable detection method based on industrial vision, characterized in that: Comprising, Based on the plane homography and the conic section, the plane homography matrix is obtained, the pixel-angle mapping coefficient and the zero position bias are calibrated in the standard angle event, the five-degree-of-freedom error vector is inversely calculated in the short-range multi-position, and the circular concentration and the harmonic energy baseline are established; The plane homography matrix is used for geometric normalization, the intensity stack is collected by the polarizer and the retarder, the disturbance fingerprint of the polarization consistency, the phase gradient and the circular domain phase correlation is constructed, and the equivalent film thickness, the equivalent refractive index, the scattering intensity and the working domain indication are obtained according to the thin film interference and the polarization response; Driven by the equivalent film thickness, the equivalent refractive index and the scattering intensity, the aberration compensation field is applied to the feature, the visual angle is re-estimated according to the mapping coefficient and the zero position bias, the corresponding visual angle and the encoder angle are fused, the first-order and second-order harmonic projection is made, and the unified angle and the harmonic coefficient are output.

2. The liquid static pressure turntable detection method based on industrial vision according to claim 1, wherein: The witness ring and the working surface are made of the same material and have the same coating, the plane homography matrix is solved based on the conic section constraint, and the pixel-angle mapping coefficient and the zero position bias are jointly estimated in the standard angle event based on the von Mises likelihood, and the mapping from the witness ring plane to the image plane is recorded as a geometric alignment operator.

3. The liquid static pressure turntable detection method based on industrial vision according to claim 2, wherein: In the unified angle domain determined according to the mapping coefficient and the zero position bias under the short-range multi-position acquisition, a design vector is constructed, the visual angle and the encoder angle are coupled and fitted, and the five-degree-of-freedom error vector is inversely calculated, wherein the design vector includes eccentric, tilt and axial constant bias components, and the circular concentration and the first-order and second-order harmonic energy are defined according to the complex exponential as the consistency baseline.

4. The liquid static pressure turntable detection method based on industrial vision according to claim 2 or 3, wherein: The intensity stack is collected by the discrete setting of the electrically controlled polarizer and the controllable retarder, the Stokes vector is calculated after geometric normalization by the plane homography matrix, the disturbance fingerprint including the polarization consistency, the phase structure gradient and the circular domain phase correlation is constructed, and the segmented organization is performed within the working condition template defined by the oil temperature, the flow, the pressure and the environmental temperature and humidity.

5. The liquid static pressure turntable detection method based on industrial vision according to claim 3, wherein: According to the physical constraints of the thin film interference phase delay and the polarization response, a proxy mapping with punishment is established, the disturbance fingerprint is mapped to the equivalent film thickness, the equivalent refractive index and the scattering intensity, the working domain indication is generated according to the polarization and the phase consistency, and the three types of quantities are aligned with the time axis of the unified angle domain for output.

6. The liquid static pressure turntable detection method based on industrial vision according to claim 4, wherein: ​ Anisotropic spatial kernel and causal temporal kernel are used to obtain aberration compensation field, and the spatial kernel covariance is set according to polarization consistency and phase consistency; the compensation field is applied to the characteristic coordinates and the visual angle is re-estimated according to the mapping coefficient and the zero position bias.

7. The industrial vision-based liquid static pressure turntable detection method according to claim 5, wherein: The compensated visual angle and the encoder angle are fused in the form of complex vectors in the circular domain, the fusion weight is adjusted by the working domain indicator and the circular concentration, and the fusion weight satisfies the non-negative and the sum of one, the updated unified angle is obtained, and the first-order and second-order harmonic coefficients are stripped by the whole circle projection in the unified phase.

8. The industrial vision-based liquid static pressure turntable detection method according to claim 6, wherein: According to ISO230-2 and ISO230-7, the updated unified angle and the first-order and second-order harmonic coefficients are mapped into the standard index vector, the angle-length conversion is realized by combining the witness ring radius, the key sequence is resampled at a preset interval on the unified phase axis, and the curve and the list matched with the corresponding caliber are output.

9. The industrial vision-based liquid static pressure turntable detection method according to claim 7, wherein: A threshold function driven by the circular concentration, the harmonic energy, the scattering intensity, the working domain indicator and the oil temperature is constructed to trigger the micro-step angle event and the static polarization event, respectively update the pixel-angle mapping coefficient and the zero position bias, and the polarization zero point and the phase bias, and the oil-repellent layer cleaning and the air curtain pulse cleaning strategy are linked.

10. A liquid hydrostatic turntable detection system based on industrial vision, characterized in that: Including, A geometric calibration module obtains a plane homography matrix based on plane homography and conic, calibrates the pixel-angle mapping coefficient and the zero position bias with a standard angle event, inverses a five-degree-of-freedom error vector with a short-range multi-pose, and establishes a circular concentration and a harmonic energy baseline; A polarization fingerprint module performs geometric normalization with the plane homography matrix, collects intensity stacks with polarizers and retarders, constructs a disturbance fingerprint of polarization consistency, phase gradient and circular domain phase correlation, and obtains equivalent film thickness, equivalent refractive index, scattering intensity and working domain indicator according to thin film interference and polarization response; A compensation fusion module is driven by the equivalent film thickness, the equivalent refractive index and the scattering intensity, estimates an aberration compensation field applied to the characteristic, re-estimates the visual angle according to the mapping coefficient and the zero position bias, fuses the corresponding visual angle and the encoder angle, performs first-order and second-order harmonic projection, and outputs the unified angle and the harmonic coefficient; An index closed loop module generates indexes according to the unified angle and the harmonic coefficient according to ISO230-2 and ISO230-7, triggers micro-step angle and static polarization events according to the circular concentration, the harmonic energy and the working domain indicator, and updates the mapping coefficient, the polarization zero point and the phase bias.

Images