Residual stress detection method based on dual-scale contour fusion and three-dimensional stress constraint

Abstract

The invention discloses a residual stress detection method and system based on dual-scale contour fusion and three-dimensional stress constraint, and belongs to the technical field of material mechanical property testing. The method comprises the following steps: preparing a micro-pit array with a nano-scale corrugated structure and a reinforced coating on the surface of a sample by adopting a femtosecond laser double-beam interference technology as a high-precision reference mark; stress freezing release is achieved through a liquid nitrogen atomization spraying auxiliary low-temperature micro-cutting technology, and cutting heat-force damage is effectively restrained; the white light interferometer and the laser confocal microscope are combined for multi-mode scanning, and high-signal-to-noise-ratio surface topography data are obtained; adopting a deep learning algorithm to realize sub-pixel-level data registration and fusion; a cube is cut from the interior of a sample innovatively through an electrolysis-assisted low-speed wire cutting technology to serve as a three-dimensional stress constraint body, and the accurate geometric contour of the cube serves as a physical calibration reference; and finally, establishing an anisotropic inversion model based on a crystal plasticity theory, training a neural network agent model by taking cube data as a priori constraint condition, carrying out iterative calculation in combination with multi-modal fusion data, and reconstructing a three-dimensional residual stress field with a spatial resolution of 50 microns. According to the method, by introducing the physical calibration reference, the precision and reliability of stress field reconstruction are remarkably improved, and the problems that a traditional method is large in model error and lacks three-dimensional constraint are solved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of material mechanical property testing and precision measurement, and particularly relates to a residual stress intelligent detection method and system based on double-scale profile fusion and three-dimensional stress constraint. BACKGROUND

[0002] Residual stress is a key factor affecting the fatigue life, dimensional stability and stress corrosion resistance of mechanical components. Currently, the detection methods of residual stress are mainly divided into destructive detection (such as drilling method, cutting method) and non-destructive detection (such as X-ray diffraction method, ultrasonic method). Among them, the cutting method based on stress release principle is widely used because it can measure the macro residual stress inside the component. However, the traditional cutting method has the following significant limitations: (1) The cutting heat and mechanical force in the cutting process will introduce new additional stress and plastic deformation, which seriously interferes with or even covers the original stress field to be measured, resulting in a significant reduction in measurement accuracy; (2) The surface deformation mark usually adopts mechanical indentation or simple laser dotting, which has insufficient precision and single feature, and it is difficult to realize accurate deformation tracking and matching at the micron level; (3) Data acquisition mainly relies on single measurement technology (such as strain gauge or ordinary optical microscope), and there is a lack of effective fusion means for multi-scale and multi-modal data, so it is difficult to obtain high transverse resolution and high longitudinal resolution topographic data at the same time; (4) Most importantly, the existing methods rely heavily on idealized constitutive models (such as isotropic elastic model) for stress inversion, and ignore the real influence of actual material microstructure (such as grain orientation, anisotropy) on stress field. This leads to a lack of reliable physical constraints in the inversion process, and the reconstructed stress field is usually two-dimensional or approximately two-dimensional, which is difficult to truly reflect the complex three-dimensional stress state and has limited spatial resolution and questionable reliability. SUMMARY

[0003] In view of the deficiencies in the prior art, the present application provides a residual stress intelligent detection method and system based on double-scale profile fusion and three-dimensional stress constraint. The core innovation of the present application is to introduce a "three-dimensional stress constraint body" as a physical calibration reference, which fundamentally solves the problems of model error and constraint deficiency in the stress inversion process. Specifically, the present application uses electrolytic slow-wire cutting technology to accurately cut a 5-8mm edge length cube sample from the inside of the sample to be tested. The cube is considered as a "stress-free" geometric reference block during the cutting process, and its accurate three-dimensional profile data records the true geometric characteristics of the material in the stress-free state.

[0004] The present application first provides a residual stress detection method based on double-scale profile fusion and three-dimensional stress constraint, which can realize high-precision three-dimensional residual stress field reconstruction with a spatial resolution of 50μm and a root mean square error of the reconstructed stress field ≤0.8%.

[0005] Further, the method specifically comprises the following steps: (1) Reference mark preparation: using femtosecond laser double-beam interference processing technology to form a micro-pit array of sub-wavelength grating structure on the sample surface, the pit depth is 20-50 μm and the inner wall is plated with a 200-500 nm thick titanium dioxide enhanced reflection film, the adjacent pits are arranged in a hexagonal close-packed arrangement and the pitch is 1.25-1.4 times the width of the cutting path; (2) Low-temperature stress freezing cutting: micro-cutting assisted by liquid nitrogen atomization spraying in a closed cavity, using a 30-50 μm diameter molybdenum alloy electrode wire, cutting speed 0.05-0.2 mm / s, the cutting zone is maintained at a low temperature brittle transition temperature interval of-40 ℃ to-10 ℃ by a vortex cooling system controlled by PID, temperature fluctuation ≤±1 ℃; (3) Multi-modal dynamic scanning: combined use of a phase shift white light interferometer (vertical resolution 0.05 μm) and a polarization-sensitive laser confocal microscope (lateral resolution 0.02 μm), three repeated scans on a vibration-proof platform and use of an adaptive weighted average algorithm to improve the signal-to-noise ratio; (4) Intelligent data fusion: automatic identification of three-dimensional coordinates of marker points based on deep learning feature matching algorithm, sub-pixel level registration of global and local data by non-rigid ICP algorithm, and application of improved second-generation wavelet transform to eliminate frequency band aliasing noise; (5) Three-dimensional stress constraint body extraction: using electrolytic assisted slow wire cutting technology to obtain a 5-8 mm edge length cubic sample, real-time monitoring of temperature field and strain field changes near the cut during cutting; (6) Double-stage stress field reconstruction: first, establishing an anisotropic inversion model based on crystal plasticity theory, using the cubic profile data as a prior constraint condition to train a neural network proxy model, then using an improved L-M algorithm (introducing Hessian matrix preprocessing) for iterative calculation, when the root mean square error is ≤0.8%, inputting the fusion data as displacement boundary conditions, finally outputting a three-dimensional residual stress field with a spatial resolution of 50 μm.

[0006] Preferably, the micro-pit array of step (1) has: an asymmetric structure with a parabolic bottom, a nanoscale corrugated structure (period 300-500 nm) on the sidewall, a graphene quantum dot enhancement coating on the surface, and a positioning accuracy of ≤0.5 μm for a single pit.

[0007] Preferably, the micro-pit array of step (1) has: a depth gradient change rate of the nanoscale corrugated structure of 5-8 nm / μm, and a corrugated axis at an angle of 45°-60° to the cutting path, which can enhance the signal-to-noise ratio of the laser scattering signal by 1.8-2.5 times.

[0008] Preferably, the low-temperature cutting system of step (2) comprises: a multi-layer thermal insulation structure of a low-temperature processing cabin, a vortex nozzle array with two-degree-of-freedom adjustment, a high-response rate micro-thermocouple temperature measurement system (sampling frequency ≥ 1 kHz), and an online monitoring module of incision quality based on machine vision.

[0009] Preferably, the low-temperature cutting system of step (2) is characterized in that the vortex nozzle array adopts a bionic shark skin surface microstructure, and the inner wall is provided with a spiral flow guide groove (groove depth 100-200 μm, spiral angle 15°-25°), which can reduce the liquid nitrogen flow resistance by 23%-35%.

[0010] Preferably, the anti-vibration platform used in step (3) comprises a three-stage active vibration isolation system: the first stage is an air floating vibration isolation base (natural frequency ≤ 2 Hz), the second stage is a magnetorheological damper (response time ≤ 5 ms), and the third stage is a piezoelectric ceramic active compensation mechanism (displacement resolution 10 nm).

[0011] Preferably, the improved second-generation wavelet transform of step (4) adopts a non-uniform sampling strategy, sets the sampling density in the area with a larger displacement gradient to be 3-5 times that in the conventional area, and introduces a direction-sensitive filter group to improve the anisotropic feature recognition rate.

[0012] Preferably, the process parameters of electrolytic auxiliary slow wire cutting in step (5) are as follows: pulse voltage 8-12 V, electrolyte 5%-8% sodium nitrate solution, wire speed 2-5 mm / s, electrode wire tension 15-20 N, and processing gap maintained at 50-80 μm.

[0013] Preferably, step (6) comprises a material parameter dynamic feedback system, which is specifically implemented as follows: the hardness / modulus distribution of a 9×9 grid is obtained by in-situ nanoindentation testing, a spatial variable material parameter field is established by Kriging interpolation, real-time parameter updating based on GPU acceleration is implemented in finite element inversion, and the stress field reconstruction accuracy is verified by combining digital image correlation technology.

[0014] The precise three-dimensional profile data of the cut cube is input into the anisotropic inversion model based on the crystal plasticity theory as a prior constraint condition. This physical calibration benchmark provides an absolutely reliable geometric boundary constraint for the training of the neural network surrogate model and the iterative calculation of stress inversion, forcing the inversion result not only to satisfy the mechanical law, but also to match the real, stress-free geometric state. Finally, combined with multi-modal fusion displacement data, a high-resolution (up to 50 μm) and high-precision (root mean square error ≤ 0.8%) three-dimensional residual stress field that satisfies both the mechanical constitutive and the real geometric constraints is reconstructed, significantly surpassing traditional methods.

[0015] Compared with the prior art, the beneficial effects of the present application include: (1) The physical calibration reference is introduced: by cutting the "stress-free" cube, reliable geometric constraints are provided for stress inversion, greatly reducing the error caused by model assumptions, which is not available in the prior art.

[0016] (2) True three-dimensional stress field reconstruction: based on the three-dimensional constraints provided by the cube, a more realistic and accurate three-dimensional stress distribution can be reconstructed, rather than an approximate two-dimensional plane stress.

[0017] (3) The accuracy and reliability are significantly improved: the combination of geometric reality and mechanical inversion makes the spatial resolution of the final stress field reach 50μm, and the calculation root mean square error is ≤0.8%, with a much higher confidence than traditional methods.

[0018] (4) Systematic innovation: from high-precision mark preparation, low-interference stress release, multi-scale data fusion to intelligent inversion under physical constraints, a complete and complementary technical system is formed, which systematically solves the shortcomings of the prior art. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 The stress field cloud chart obtained in Example 1 of the present application. DETAILED DESCRIPTION

[0020] The present application will be further described in detail below in conjunction with the drawings and specific examples, but the embodiments of the present application are not limited thereto. Example 1

[0021] The present application method is used to detect the internal residual stress of a nickel-based high-temperature alloy (GH4169) sample for a certain aero-engine turbine disc.

[0022] (1) Reference mark preparation: polish the surface of the sample to be tested (Ra < 0.05μm) and ultrasonic clean. A hexagonal close-packed micro-pit array is prepared on the surface of the sample using a femtosecond laser double-beam interference processing system (wavelength 800nm, pulse width 100fs, repetition frequency 1kHz). The processing parameters are: single pulse energy 0.5mJ, scanning speed 50mm / s. The obtained pit depth is 35±2μm, and the inner wall of the pit is coated with a 300nm thick titanium dioxide reflective film by magnetic sputtering technology. The center distance between adjacent pits is 60μm (about 1.3 times the width of the subsequent cutting path). The micro-pit has a parabolic bottom asymmetric structure, and the sidewall is processed with a nanoscale corrugated structure with a period of 400nm (depth gradient change rate of 6nm / μm), and the corrugated axis is at an angle of 50° with the preset cutting path. Finally, a layer of graphene quantum dot enhancement coating is spin-coated on the surface of the array. It is measured that the positioning accuracy of a single pit is better than 0.4μm.

[0023] (2) Cryogenic stress freeze cutting: The prepared sample with markers was fixed in a multi-layer insulation structure of the low-temperature processing cabin. A 40-μm-diameter molybdenum electrode wire was used for cutting. The cutting path passed through the center of the marker area. The cutting speed was set to 0.1 mm / s. A vortex cooling system (equipped with a bionic shark skin microstructure nozzle, groove depth 150 μm, spiral angle 20°) controlled by PID was used to spray liquid nitrogen atomized flow to the cutting area, so that the temperature of the cutting area was stably maintained at the low-temperature brittle transition interval of -25°C±0.8°C. A micro-thermocouple array with a sampling frequency of 1.2 kHz was used to monitor the temperature in real time, and a machine vision-based monitoring module was used to observe the cut quality online.

[0024] (3) Multi-modal dynamic scanning: Before and after cutting, the sample was moved to a three-stage active vibration isolation platform (first stage: air floating vibration isolation base, natural frequency 1.5 Hz; second stage: magnetorheological damper, response time 4 ms; third stage: piezoelectric ceramic active compensation mechanism, displacement resolution 10 nm). A phase-shifted white light interferometer (vertical resolution 0.05 μm) and a polarization-sensitive laser confocal microscope (lateral resolution 0.02 μm, NA=0.95) were combined to perform three repeated scans of the complete cutting surface containing the marker area. An adaptive weighted average algorithm was used to fuse the three scan data, effectively improving the signal-to-noise ratio.

[0025] (4) Intelligent data fusion: The three-dimensional topography data obtained by scanning was input into a convolutional neural network (CNN) based on attention mechanism (Attention Mechanism) for feature extraction and matching, automatically identifying and matching the three-dimensional coordinates of all marker points before and after cutting. Through the non-rigid iterative closest point (Non-rigid ICP) algorithm, sub-pixel level registration (accuracy better than 0.01 pixels) of global and local data was achieved. Subsequently, an improved second-generation wavelet transform was applied to process the displacement field data: the sampling density in the area with large displacement gradient (such as the cut edge) was set to 4 times that of the regular area, and a direction-sensitive filter bank was introduced, effectively eliminating frequency band aliasing noise and improving the recognition rate of anisotropic features.

[0026] (5) Three-dimensional stress constraint body extraction: An electrolytic slow wire cutting technology was used to accurately cut a 6 mm×6 mm×6 mm cubic sample from the inside of the sample that had undergone stress release as a "stress-free" three-dimensional constraint body. The cutting process parameters were: pulse voltage 10 V, electrolyte 6% sodium nitrate solution, wire speed 3 mm / s, electrode wire tension 18 N, and processing gap maintained at 60±5 μm. During the cutting process, an infrared thermal imager and a microscopic digital image correlation (DIC) system were used to monitor the temperature field and strain field changes near the cut in real time, ensuring that no additional thermal stress and mechanical stress was introduced during the entire process.

[0027] (6) Two-stage stress field reconstruction: Primary reconstruction (model training and constraints): First, an anisotropic inversion model based on crystal plasticity finite element (CPFEM) is established. A 9x9 grid test is performed on the upper surface of the cubic constraint body using a nanoindenter to obtain the spatial distribution data of hardness and modulus, and a spatially varying material parameter field is established by Kriging interpolation. The precise three-dimensional profile data of the cubic body (from step (5)) are input into the model as absolute prior geometric constraints to train the neural network proxy model.

[0028] Secondary reconstruction (stress inversion): The fused high-precision displacement field data obtained in step (4) are used as displacement boundary conditions. An improved Levenberg-Marquardt (L-M) algorithm pre-processed by a Hessian matrix is used for iterative calculation. During the iteration process, the material parameter field is updated in real time based on the GPU accelerated computing architecture. When the root mean square error (RMSE) of the inversion calculation reaches 0.75%, the iteration is terminated. Finally, a three-dimensional residual stress field distribution cloud chart with a spatial resolution of 50 μm is output.

[0029] Comparative Example 1

[0030] The same batch of GH4169 alloy samples were cut using the traditional wire electrical discharge machining method (room temperature, emulsion cooling), then scanned and data matched using the same white light interferometer, and finally the stress field was inverted using the classical isotropic elastic theory.

[0031] It was found that due to the influence of cutting heat effect and mechanical force, a clear recast layer and heat affected zone appeared near the cut, and microhardness tests showed that softening occurred in this area. The inverted stress field showed abnormally high values near the cutting path, with a large deviation from the results of the X-ray diffraction method, with a maximum deviation of more than 28%, which could not truly reflect the original residual stress state inside the sample.

[0032] Comparative Example 2

[0033] The data were obtained using the method of steps (1)-(4) of the present application, but in step (6) stress inversion, an isotropic elastic model was used instead of a crystal plasticity model, and a three-dimensional cubic constraint body and material parameter spatial distribution feedback were not introduced.

[0034] The inversion results show that the macroscopic distribution trend of the stress field is similar to that of Example 1, but the detailed resolution of the stress values near the microstructure features such as grain boundaries, twin boundaries and strengthening phases is significantly reduced, with a difference of 12-18% compared to the results of the crystal plasticity-based model in Example 1. This proves the extreme importance of introducing an anisotropic model, material parameter feedback and the most critical three-dimensional physical calibration constraint body for improving the reconstruction accuracy and reliability. Example 2

[0035] The process parameters of the electrolysis-assisted slow wire cutting in step (5) are adjusted to be: pulse voltage 12 V, electrolyte 8% sodium nitrate solution, wire speed 2 mm / s, electrode wire tension 20 N, and machining gap maintained at 50 μm. The stress field resolution obtained finally is about 55 μm, and the RMSE is 0.78%, which still meets the accuracy requirement. Example 3

[0036] The heat treatment process in step (6) is basically the same as that in Example 1, except that the holding time is extended to 3 hours. The stress field distribution obtained finally is consistent with that in Example 1, indicating that the method of the application has good process stability.

[0037] The above examples are preferred embodiments of the application, but the application is not limited to the above embodiments. Any obvious improvement, replacement or modification made by those skilled in the art without departing from the essential content of the application shall fall within the protection scope of the application.

Claims

1. A residual stress detection method based on dual-scale contour fusion and three-dimensional stress constraints, characterized in that, Includes the following steps: (1) Preparation of reference marks: A micro-dimple array of subwavelength grating structure is formed on the sample surface using femtosecond laser dual-beam interference processing technology. The depth of the pits is 20-50 μm and the inner wall is coated with a titanium dioxide enhanced reflective film with a thickness of 200-500 nm. Adjacent pits are arranged in a hexagonal close-packed pattern and the spacing is 1.25-1.4 times the width of the cutting path. (2) Low temperature stress freezing cutting: Micro-cutting assisted by liquid nitrogen atomization spray in a closed cavity, using a molybdenum alloy electrode wire with a diameter of 30-50μm, a cutting speed of 0.05-0.2 mm / s, and maintaining the cutting zone in the low temperature brittle transition temperature range of -40 ℃ to -10 ℃ through a PID controlled eddy current cooling system, with temperature fluctuation ≤±1 ℃; (3) Multimodal dynamic scanning: A phase-shifting white light interferometer (vertical resolution 0.05 μm) and a polarization-sensitive laser confocal microscope (lateral resolution 0.02 μm) were used in combination. Three repeated scans were performed on a vibration-proof platform and an adaptive weighted average algorithm was used to improve the signal-to-noise ratio. (4) Intelligent data fusion: Based on deep learning, the feature matching algorithm automatically identifies the three-dimensional coordinates of the marker points, and the non-rigid ICP algorithm achieves sub-pixel level registration of global and local data. The improved second-generation wavelet transform is applied to eliminate frequency band aliasing noise. (5) Extraction of three-dimensional stress-constrained bodies: 5-8 mm side length cubic specimens were obtained by electrolytic-assisted slow wire cutting technology. During the cutting process, the temperature field and strain field changes near the cut were monitored in real time. (6) Two-level stress field reconstruction: First, an anisotropic inversion model based on crystal plasticity theory is established. The cube contour data is used as a priori constraint to train the neural network surrogate model. Then, the improved LM algorithm (introducing Hessian matrix preprocessing) is used for iterative calculation. When the root mean square error is ≤0.8%, the fused data is used as the displacement boundary condition input. Finally, a three-dimensional residual stress field with a spatial resolution of 50μm is output.

2. The method according to claim 1, characterized in that, The micro-pit array in step (1) has the following characteristics: the bottom is a parabolic asymmetric structure, the sidewalls are provided with a nanoscale corrugated structure (period 300-500nm), the surface is covered with a graphene quantum dot enhanced coating, and the positioning accuracy of a single pit is ≤0.5μm.

3. The micro-pit array according to claim 2, characterized in that, The depth gradient change rate of the nanoscale corrugated structure is 5-8 nm / μm, and the corrugated axis forms an angle of 45°-60° with the cutting path, which can enhance the signal-to-noise ratio of laser scattering signal by 1.8-2.5 times.

4. The method according to claim 1, characterized in that, The cryogenic cutting system in step (2) includes: The low-temperature processing chamber features a multi-layered insulation structure, a two-degree-of-freedom adjustable vortex nozzle array, a high-response-rate miniature thermocouple temperature measurement system (sampling frequency ≥ 1 kHz), and a machine vision-based online monitoring module for cut quality.

5. The cryogenic cutting system according to claim 4, characterized in that, The vortex nozzle array adopts a biomimetic shark skin surface microstructure, and its inner wall is provided with a spiral guide groove (groove depth 100-200μm, spiral angle 15°-25°), which can reduce the flow resistance of liquid nitrogen by 23%-35%.

6. The method according to claim 1, characterized in that, The vibration isolation platform used in step (3) includes a three-level active vibration isolation system: the first level is an air-floating vibration isolation base (natural frequency ≤ 2Hz), the second level is a magnetorheological damper (response time ≤ 5ms), and the third level is a piezoelectric ceramic active compensation mechanism (displacement resolution 10nm).

7. The method according to claim 1, characterized in that, The improved second-generation wavelet transform in step (4) adopts a non-uniform sampling strategy, setting the sampling density in the region with a large displacement gradient to be 3-5 times that of the conventional region, and introducing a direction-sensitive filter bank to improve the anisotropic feature recognition rate.

8. The method according to claim 1, characterized in that, The process parameters for electrolytic-assisted slow wire cutting in step (5) are: pulse voltage 8-12V, electrolyte 5%-8% sodium nitrate solution, wire feed speed 2-5mm / s, electrode wire tension 15-20N, and processing gap maintained at 50-80μm.

9. The method according to claim 1, characterized in that, Step (6) includes a dynamic feedback system for material parameters. Specifically, the system is implemented by using in-situ nanoindentation testing to obtain the hardness / modulus distribution of a 9×9 grid, establishing a spatially variable material parameter field through Kriging interpolation, implementing real-time parameter updates based on GPU acceleration in finite element inversion, and verifying the accuracy of stress field reconstruction by combining digital image correlation technology.

10. An intelligent testing system for implementing the methods of claims 1-9, characterized in that, The low-temperature precision machining unit includes an integrated ultra-low temperature environment module (minimum -196℃), a six-axis linkage cutting platform, and an online quality monitoring system.

11. An intelligent testing system for implementing the methods of claims 1-9, characterized in that, The multi-scale optical measurement unit includes a wide-field scanning white light interferometer (equipped with a PZT nanostage), a super-resolution confocal microscope (NA≥0.95), and an active vibration damping optical platform (vibration isolation frequency≤1Hz).

12. An intelligent testing system for implementing the methods of claims 1-9, characterized in that, The artificial intelligence processing unit is equipped with a multimodal data fusion algorithm library, an attention-based deep learning model, and three-dimensional stress field visualization software.

13. An intelligent testing system for implementing the methods of claims 1-9, characterized in that, The high-performance computing unit is equipped with a materials genome engineering database, a hybrid precision parallel computing architecture, and an adaptive grid optimization module.

14. The intelligent testing system according to claim 10, characterized in that, The six-axis linkage cutting platform is made of carbon fiber reinforced ceramic composite material, with an axial stiffness ≥200N / μm and a coefficient of thermal expansion ≤0.5×10⁻ 6 / ℃, equipped with a nanometer-level grating ruler feedback system (resolution 1nm).

15. The multi-scale optical measurement unit according to claim 11, characterized in that, The super-resolution confocal microscope integrates a tunable laser source (wavelength range 405-780nm) and is equipped with a high-speed galvanometer scanning system (maximum scanning speed 500Hz), which can simultaneously acquire surface morphology and fluorescence lifetime information.

Images