Precise metal part defect detection method and system based on multi-modal large model

Abstract

The invention provides a multi-mode large model-based precision metal part defect detection method and system. The method comprises the following steps of: acquiring laser scattering image data of a precision metal part in an electromagnetic shielding environment; performing multi-scale spatial analysis on the laser scattering image data to extract a multi-modal image feature set containing a crystal boundary topological structure of the surface of the precision metal part and scattering spot distribution gradient features of the subsurface; converting process parameters, corresponding to the precision metal part, in an existing metal heat treatment process database into heat historical feature vectors, wherein the heat historical feature vectors comprise temperature gradient distribution features and time sequence phase change features; performing fusion processing on the thermal historical feature vector and the multi-modal image feature set through a multi-modal large model to obtain a fusion processing result so as to generate defect detection data; according to the invention, the detection rate and the detection efficiency of micron-sized defects in a complex electromagnetic environment are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of defect detection, and particularly to a method and system for defect detection of precision metal parts based on a multimodal large model. Background Art

[0002] With the continuous improvement of the reliability requirements for precision metal parts in high-end manufacturing fields such as aerospace and nuclear power equipment, the on-line detection of micron-scale grain boundary cracks and subsurface defects has become a key challenge. When such components are in service under complex working conditions (such as high-cycle fatigue), defects often exhibit multi-physical-field coupling characteristics.

[0003] The current mainstream technical solution adopts a defect detection system based on laser speckle imaging and convolutional neural network. The technical implementation process includes laser speckle image acquisition, single-modal feature extraction, and defect detection by a binocular vision system. There are some defects in the existing technical solutions. For example, relying only on visual modal data leads to a high defect miss detection rate; image acquisition in an open environment results in missing data of micro-cracks (such as 5 μm), etc. Summary of the Invention

[0004] The present invention provides a method and system for defect detection of precision metal parts based on a multimodal large model to solve the problems in the prior art that only rely on visual modal data, resulting in a high defect miss detection rate and missing micro-crack data acquisition.

[0005] In a first aspect, the present invention provides a method for defect detection of precision metal parts based on a multimodal large model, including: Obtaining laser scattering image data of the precision metal part in an electromagnetic shielding environment; Performing multi-scale spatial analysis on the laser scattering image data to extract a multimodal image feature set including the grain boundary topological structure on the surface of the precision metal part and the scattering spot distribution gradient feature of the subsurface; Converting the process parameters corresponding to the precision metal part in the existing metal heat treatment process database into a thermal history feature vector, where the thermal history feature vector includes a temperature gradient distribution feature and a time-series phase change feature; Performing fusion processing on the thermal history feature vector and the multimodal image feature set through a multimodal large model to obtain a fusion processing result; Generating defect detection data based on the fusion processing result, where the defect detection data includes three-dimensional position information of grain boundary cracks and prediction data of crack propagation directions.

[0006] Optionally, performing fusion processing on the thermal history feature vector and the multimodal image feature set through a multimodal large model to obtain a fusion processing result, including: Generate a weight regulation parameter set based on the material property parameters of the precision metal part, where the weight regulation parameter set includes a thermal history weight vector and an image weight vector; In the thermal history feature encoding channel of the multi-modal large model, perform mapping processing on the thermal history feature vector according to the thermal history weight vector to generate a process feature encoding sequence with temperature gradient correlation characteristics; In the image feature encoding channel of the multi-modal large model, perform a sliding window process on the multi-modal image feature set according to the image weight vector to generate an image feature encoding sequence with grain boundary topology correlation characteristics; Perform weighted processing on the process feature encoding sequence and the image feature encoding sequence according to the weight regulation parameter set to generate a fused encoding tensor; Perform screening processing on the fused encoding tensor to generate an optimized fused processing result.

[0007] Optionally, performing weighted processing on the process feature encoding sequence and the image feature encoding sequence according to the weight regulation parameter set to generate a fused encoding tensor includes: Perform element-wise multiplication on each element value of the thermal history weight vector in the weight regulation parameter set and the corresponding element value of the process feature encoding sequence to generate a weighted process feature sequence; Perform element-wise multiplication on each element value of the image weight vector in the weight regulation parameter set and the corresponding element value of the image feature encoding sequence to generate a weighted image feature sequence; Determine a splicing ratio coefficient according to the material property parameters of the precision metal part, and perform dimensionality adjustment processing on the weighted process feature sequence and the weighted image feature sequence according to the splicing ratio coefficient to obtain an adjusted weighted process feature sequence and an adjusted weighted image feature sequence; Interleave and splice the adjusted weighted process feature sequence and the adjusted weighted image feature sequence in the order of element positions to generate initial fused encoding data; Perform target redundant dimension marking processing on the initial fused encoding data to obtain marked fused encoding data; Remove all marked target redundant dimensions in the marked fused encoding data to generate a fused encoding tensor.

[0008] In a second aspect, the present invention provides a precision metal part defect detection system based on a multi-modal large model, including: An acquisition module for acquiring laser scattering image data of a precision metal part in an electromagnetic shielding environment; An analysis module for performing multi-scale spatial analysis on the laser scattering image data to extract a multi-modal image feature set including the grain boundary topological structure on the surface of the precision metal part and the scattering spot distribution gradient feature in the subsurface; A conversion module for converting the process parameters corresponding to the precision metal part in the existing metal heat treatment process database into a thermal history feature vector, where the thermal history feature vector includes a temperature gradient distribution feature and a time series phase change feature; A fusion module for performing fusion processing on the thermal history feature vector and the multi-modal image feature set through a multi-modal large model to obtain a fusion processing result; A generation module for generating defect detection data based on the fusion processing result, where the defect detection data includes three-dimensional position information of grain boundary cracks and prediction data of crack propagation directions.

[0009] In a third aspect, the present invention provides a computing device including a processor and a memory, where a computer program is stored in the memory, and the processor is configured to run the computer program to execute any one of the methods for defect detection of precision metal parts based on a multi-modal large model in the first aspect.

[0010] In a fourth aspect, the present invention provides a computer storage medium having computer program instructions stored thereon, and when the computer program instructions are executed by a processor, they implement any one of the methods for defect detection of precision metal parts based on a multi-modal large model in the first aspect.

[0011] In the present invention, laser scattering image data of precision metal parts is acquired in an electromagnetic shielding environment; multi-scale spatial analysis is performed on the laser scattering image data to extract a multi-modal image feature set including the grain boundary topological structure on the surface of the precision metal parts and the scattering spot distribution gradient characteristics in the subsurface; the process parameters corresponding to the precision metal parts in the existing metal heat treatment process database are converted into a thermal history feature vector, and the thermal history feature vector includes temperature gradient distribution characteristics and time series phase change characteristics; the thermal history feature vector and the multi-modal image feature set are fused through a multi-modal large model to obtain a fusion processing result; based on the fusion processing result, defect detection data is generated, and the defect detection data includes three-dimensional position information of grain boundary cracks and prediction data of crack propagation directions. The technical solution provided by the present invention suppresses high-frequency interference signals in the industrial field through an electromagnetic shielding environment to ensure the stable capture of micron-level defect features, and solves the problem of missed detection caused by electromagnetic pollution of detection signals in an open environment; through multi-scale analysis, cross-dimensional features of the surface grain boundary topology (macroscopic structure) and subsurface scattering spot gradient (microscopic distribution) are synchronously extracted, breaking through the limitation of traditional single-scale feature extraction for the characterization of complex defects and enhancing the joint detection ability of grain boundary cracks and inclusions; through the spatio-temporal feature vector, a physical association between the heat treatment history and defect formation is established, providing prior knowledge of process constraints for multi-modal fusion; by fusing process history data and real-time detection data, the problem of misjudgment of multi-physical field coupling defects (such as thermal stress cracks) by traditional single-modal models is solved, and the robustness of defect recognition under complex working conditions is improved; the composite detection result breaks through the geometric distortion defects of traditional two-dimensional positioning and provides high-precision input for life prediction and maintenance decision-making.

[0012] These aspects or other aspects of the present invention will be more clearly understood in the following description of the embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or in the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0014] Figure 1 It is a flowchart of a method for detecting defects of precision metal parts based on a multi-modal large model provided by an embodiment of the present invention; Figure 2 It is a schematic structural diagram of a system for detecting defects of precision metal parts based on a multi-modal large model provided by an embodiment of the present invention; Figure 3 It is a schematic structural diagram of a computing device provided by an embodiment of the present invention. Specific embodiments

[0015] In order to enable those skilled in the art of the present technology to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0016] In some processes described in the specification and claims of the present invention and the above-mentioned accompanying drawings, a plurality of operations appear in a specific order. However, it should be clearly understood that these operations may not be executed in the order in which they appear in this article or may be executed in parallel. The serial numbers of the operations, such as 101, 102, etc., are only used to distinguish different operations, and the serial numbers themselves do not represent any execution order. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions such as "first" and "second" in this article are used to distinguish different messages, devices, modules, etc., do not represent a sequence, and do not limit that "first" and "second" are of different types.

[0017] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative efforts belong to the scope of protection of the present invention.

[0018] Figure 1 The flowchart of a method for detecting defects in precision metal parts based on a multimodal large model is provided for the embodiments of the present invention. As Figure 1 shown, the method includes: Aiming at the core problems such as insufficient detection accuracy of micro-defects and difficulty in identifying multi-physical-field coupling defects of precision metal parts in a complex electromagnetic interference environment, this technology breaks through the limitations of traditional single-modal detection solutions, suppresses high-frequency noise interference by constructing an electromagnetic shielding environment, innovatively integrates multi-scale features of laser scattering and historical data of heat treatment processes, designs a multimodal large model to dynamically associate process parameters with real-time detection features, solves problems such as missed detection of subsurface defects and distorted prediction of crack propagation directions caused by the lack of thermal-optical cross-modal fusion in the existing technology, and at the same time realizes precise defect positioning through three-dimensional space mapping and physical constraint modeling, overcoming the technical bottleneck of full-dimensional detection of micron-level defects in an industrial field environment, and forming a high-precision defect detection closed-loop system driven by process data and coordinated by multiple modalities. Based on this, the present invention provides a method for detecting defects in precision metal parts based on a multimodal large model, as Figure 1 follows: Step 101: Obtain laser scattering image data of precision metal parts in an electromagnetic shielding environment.

[0019] In this step, the electromagnetic shielding environment refers to a closed detection space constructed with a conductive shielding structure and wave-absorbing materials, which is used to suppress external electromagnetic interference signals (frequency range: 10 kHz - 6 GHz) and ensure the purity of the laser scattering signal acquisition. The laser scattering image data refers to the reflected spot distribution data captured by an optical sensor when a polarized laser beam irradiates the surface of a precision metal part, and it includes the surface microstructure topography and the diffraction characteristics of subsurface defects.

[0020] In this step, in the electromagnetic shielding environment, a closed detection space is constructed through a Faraday detection cavity. This cavity adopts a composite structure of multiple layers of conductive shielding layers and wave-absorbing materials to attenuate and suppress electromagnetic interference signals with a frequency range of 10 kHz - 6 GHz in the detection area. Inside the closed detection space, a laser emission device is used to project a polarized laser beam with a wavelength of 532 nm onto the surface of the precision metal part, so that the polarized laser beam undergoes a scattering effect with the surface of the alloy part and generates a reflected spot distribution. The optical sensing device is used to receive the reflected light distribution data in the closed detection space in real time to generate the laser scattering image data to be optimized, which includes surface microstructure features. Based on the electromagnetic interference suppression environment provided by the closed detection space, dark current correction and background noise filtering are performed on the laser scattering image data to be optimized, and the signal distortion introduced by external electromagnetic disturbances is eliminated, and the preprocessed laser scattering image data with a high signal-to-noise ratio is obtained as the input source for subsequent multi-scale spatial analysis.

[0021] Step 102: Perform multi-scale spatial analysis on the laser scattering image data to extract a multi-modal image feature set that includes the grain boundary topological structure on the surface of the precision metal part and the scattering spot distribution gradient feature of the subsurface.

[0022] In this step, the grain boundary topological structure refers to the spatial network morphological features formed by the grain boundaries on the surface of the metal part, including the grain boundary orientation angle, node density, and connectivity parameters, which reflect the microscopic deformation under the action of mechanical stress. The scattering spot distribution gradient feature refers to the intensity change rate of the scattering spots in space caused by subsurface defects, which characterizes the defect depth, size, and the degree of stress concentration inside the material. The multi-modal image feature set refers to a cross-dimensional feature set that fuses the surface grain boundary topological structure (macro scale) and the subsurface scattering spot distribution gradient (micro scale), forming a multi-physical field characterization for defect detection.

[0023] This step performs multi-scale spatial analysis on the preprocessed laser scattering image data, specifically including: extracting the grain boundary topological structure features by using the pixel gradient analysis method in the spatial domain to obtain the grain boundary orientation angle distribution map and the node density distribution map; performing two-dimensional wavelet decomposition on the laser scattering image data in the frequency domain to extract the scattered spot energy distribution features at different scales and generate the subsurface scattered spot frequency spectrum feature map; aligning the grain boundary topological structure features with the subsurface scattered spot frequency spectrum feature map in the feature space to construct a multi-modal image feature set containing surface-subsurface joint features as the feature input for multi-modal fusion processing.

[0024] Step 103: Convert the process parameters corresponding to the precision metal part in the existing metal heat treatment process database into a thermal history feature vector, where the thermal history feature vector includes temperature gradient distribution features and time series phase change features.

[0025] In this step, the thermal history feature vector refers to encoding the heat treatment process parameters (quenching temperature gradient, tempering time series) into continuous vector data, including the quantitative features of the spatial distribution of the temperature field and the temporal phase change process. The temperature gradient distribution features refer to the rate and direction features of temperature change in three-dimensional space during the heat treatment process, reflecting the distribution law of thermal stress during the material cooling process. The time series phase change features refer to the dynamic characteristics of the metal phase change process in the time dimension, including parameters such as the duration of the phase change critical point and the temperature fluctuation frequency.

[0026] This step extracts the process parameter set corresponding to the complex alloy part from the metal heat treatment process database. The process parameter set includes temperature sampling point data, time series data, and phase change critical point records; expands the temperature sampling points in the temperature sampling point data to generate temperature gradient distribution features, and at the same time performs time series modeling on the time series data and the phase change critical point records to generate time series phase change features; performs tensor splicing on the temperature gradient distribution features and the time series phase change features to form a thermal history feature vector as the process data input for the multi-modal large model.

[0027] Step 104: Perform fusion processing on the thermal history feature vector and the multi-modal image feature set through the multi-modal large model to obtain the fusion processing result.

[0028] This step constructs a multi-modal large model based on the attention mechanism. The multi-modal large model includes a dual-channel architecture of a thermal history feature encoder and an image feature encoder; dynamically configures cross-modal association weights according to the material property parameters of precision metal parts, and the material property parameters include the proportion of alloy element composition and the type of crystal structure; performs a non-linear transformation on the thermal history feature vector through the thermal history feature encoder to generate a process feature embedding vector; performs convolutional feature extraction on the multi-modal image feature set through the image feature encoder to generate a visual feature embedding vector; performs cross-modal attention fusion on the process feature embedding vector and the visual feature embedding vector to generate a fusion processing result containing material process constraints.

[0029] Step 105: Based on the fusion processing result, generate defect detection data, where the defect detection data includes the three-dimensional position information of grain boundary cracks and the prediction data of crack propagation direction.

[0030] This step inputs the fusion processing result into a defect detection decoder, and the defect detection decoder includes a three-dimensional space reconstruction module and a crack propagation prediction module; performs voxel-level positioning on the grain boundary cracks through the three-dimensional space reconstruction module to generate the three-dimensional position information of the cracks including X-Y-Z coordinates; analyzes the stress field distribution characteristics at the crack tip through the crack propagation prediction module, combines with the temperature gradient distribution characteristics in the thermal history feature vector, and calculates the crack propagation direction angle and potential propagation path; integrates the results of the above steps to generate defect detection data with spatial topological relationships.

[0031] In the embodiment of the present invention, by constructing an electromagnetic shielding environment to suppress industrial site interference, combining laser scattering multi-scale feature analysis with heat treatment process data fusion, breaking through the identification bottleneck of traditional single-modal detection for subsurface defects and multi-physical field coupling defects, realizing the three-dimensional precise positioning of grain boundary cracks and the prediction of propagation direction, improving the detection rate and detection efficiency of micron-level defects in a complex electromagnetic environment, and systematically solving the problems of missed detection and misjudgment caused by data fragmentation and environmental sensitivity in the prior art.

[0032] The present invention provides a specific embodiment. In step 104, the thermal history feature vector and the multi-modal image feature set are fused through a multi-modal large model to obtain a fusion processing result, which specifically includes the following steps: Step 401: Based on the material property parameters of the precision metal parts, generate a weight regulation parameter set, and the weight regulation parameter set includes a thermal history weight vector and an image weight vector.

[0033] In this step, the material property parameters refer to the set of parameters characterizing the physical properties of precision metal parts, including the mass percentage composition of the main alloying elements (such as 85% titanium, 10% aluminum, and 5% vanadium) and the digital encoding of the crystal structure type (such as the body-centered cubic encoding is 001, and the face-centered cubic encoding is 010). The weight regulation parameter set refers to the set of parameters including the thermal history weight vector and the image weight vector, which are generated from the material property parameters and used to control the cross-modal feature fusion. The thermal history weight vector is used to adjust the mapping intensity of the temperature gradient features in the thermal history feature encoding channel, and the vector length is consistent with the dimension of the thermal history feature vector. The image weight vector is used to control the extraction ratio of the grain boundary topological features in the image feature encoding channel, and the vector length matches the number of types of multi-modal image features.

[0034] In this step, the composition ratio parameters of the main alloying elements and the encoding parameters of the crystal structure type are extracted from the material property database of precision metal parts to generate a composition ratio vector and a crystal structure encoding vector. Among them, the composition ratio parameters are the vector of the mass percentage composition of each main alloying element; the encoding parameters are the one-hot encoding vectors classified according to the crystal structure symmetry; the composition ratio vector and the crystal structure encoding vector are connected in sequence to form a combined attribute vector; the combined attribute vector is input into a preset weight generation function to generate a thermal history weight vector and an image weight vector. The weight generation function includes independent fully connected operation branches; among them, the length of the thermal history weight vector is equal to the number of input nodes of the thermal history feature encoding channel and is used to adjust the mapping intensity of the process features; the length of the image weight vector is equal to the number of input channels of the image feature encoding channel and is used to control the extraction weight of the image features; the thermal history weight vector and the image weight vector together constitute the cross-modal weight regulation parameter set.

[0035] Step 402: In the thermal history feature encoding channel of the multi-modal large model, perform mapping processing on the thermal history feature vector according to the thermal history weight vector to generate a process feature encoding sequence with temperature gradient correlation characteristics.

[0036] In this step, the thermal history feature encoding channel refers to the data processing channel in the multi-modal large model that specifically processes the thermal history feature vector. Its input interface receives the thermal history feature vector, and its output interface generates a process feature encoding sequence. The number of internal nodes of the channel is equal to the dimension value of the thermal history feature vector. The temperature gradient correlation characteristic refers to the degree of correlation between each element value in the process feature encoding sequence and the temperature gradient distribution feature in the thermal history feature vector, specifically manifested as the characteristic that the element value increases with the increase of the temperature difference between adjacent temperature sampling points. The process feature encoding sequence refers to the feature data sequence output by the thermal history feature encoding channel. Each element value in the sequence reflects the comprehensive influence degree of the temperature gradient at a specific spatial position on the material phase change process, and the element arrangement order is consistent with the dimension order of the thermal history feature vector.

[0037] This step's thermal history feature encoding channel is one of the two-channel feature encoding architectures of the multimodal large model. The number of its input nodes is equal to the total dimension value of the thermal history feature vector. The total dimension value is the sum of the dimensions of the temperature gradient distribution feature and the time series phase change feature contained in the thermal history feature vector. Its input interface is directly connected to the data storage location of the thermal history feature vector. When performing non-linear mapping processing in the thermal history feature encoding channel, each element value of the thermal history weight vector is multiplied by the corresponding dimension value of the thermal history feature vector to generate an initial weighted eigenvalue. Non-linear transformation processing is performed on this initial eigenvalue, and the output value of the non-linear transformation processing is added to the temperature gradient change amount output by the previous layer to generate the current element value of the process feature encoding sequence. Finally, a process feature encoding sequence with temperature gradient correlation characteristics is formed based on the current element value, where each element value linearly increases as the temperature gradient change amount increases.

[0038] Step 403: In the image feature encoding channel of the multimodal large model, perform a sliding window process on the multimodal image feature set according to the image weight vector to generate an image feature encoding sequence with grain boundary topology correlation characteristics.

[0039] In this step, the image feature encoding channel refers to the data processing channel in the multimodal large model that specifically processes the multimodal image feature set. Its input interface receives the grain boundary topological structure feature and the sub-surface scattering spot distribution feature, and its output interface generates an image feature encoding sequence. The sliding window size of the channel is inversely proportional to the average grain boundary spacing. The grain boundary topology correlation characteristic refers to the degree of correlation between each element value in the image feature encoding sequence and the spatial distribution feature of the grain boundary structure, specifically manifested as the characteristic that the element value increases as the grain boundary density increases. The image feature encoding sequence refers to the feature data sequence output by the image feature encoding channel. Each element value in the sequence reflects the combined feature intensity of the grain boundary topology and the scattering spot distribution in a specific window area, and the element arrangement order is consistent with the image acquisition spatial coordinate order.

[0040] This step's image feature encoding channel is one of the two-channel feature encoding architectures of the multimodal large model. The number of its input channels is equal to the total number obtained by adding the number of types of grain boundary topological structure features and the number of types of subsurface scattering spot distribution features in the multimodal image features. Its input interface is directly connected to the data storage location of the multimodal image features. When performing a sliding window process in the image feature encoding channel, the window step size is calculated based on the average grain boundary spacing. The window step size is the reciprocal of the average grain boundary spacing multiplied by a preset proportionality coefficient. Slide on the grain boundary topological structure feature map with the window step size, and the area covered by each slide is the square of the window step size. Extract the local statistics of the subsurface scattering spot distribution features at each window position. The local statistics include the mean number of scattering spots and the variance of the distribution density. Multiply the local statistics by the corresponding channel values of the image weight vector to generate the current element value of the image feature encoding sequence. Finally, form an image feature encoding sequence with grain boundary topological correlation characteristics based on the current element value.

[0041] Step 404: Perform a weighting process on the process feature encoding sequence and the image feature encoding sequence according to the weight regulation parameter set to generate a fused encoding tensor.

[0042] In this step, the fused encoding tensor refers to a multi-dimensional data set formed by weighting and splicing the process feature encoding sequence and the image feature encoding sequence after aligning them by element position. The number of its dimensions is the sum of the dimensions of the two input sequences. The value at each position in the tensor is generated by superimposing the values of the two input sequences at the corresponding position according to the weights.

[0043] In this step, each element value of the thermal history weight vector in the weight regulation parameter set is multiplied point by point with the corresponding element value of the process feature encoding sequence to obtain the weighted process feature component. At the same time, each element value of the image weight vector in the weight regulation parameter set is multiplied point by point with the corresponding element value of the image feature encoding sequence to obtain the weighted image feature component. The weighted process feature component and the weighted image feature component are aligned by element position and then spliced to form a fused encoding tensor, and its data dimension is the sum of the dimensions of the process feature encoding sequence and the image feature encoding sequence.

[0044] Step 405: Perform a screening process on the fused encoding tensor to generate an optimized fused processing result.

[0045] This step extracts the data distribution of each feature dimension from the fused encoded tensor, where the feature dimension is an independent data channel in the fused encoded tensor representing different feature types; calculates the spatial distribution correlation of each feature dimension, which is obtained by calculating the correlation between the data sequence of this feature dimension at all detection positions and the pre-established spatial distribution map of grain boundary cracks. Specifically, it is the average value of the product of the corresponding position values of the two data sequences minus the product of their respective average values; at the same time, calculates the direction sensitivity of each feature dimension, which is obtained by dividing the data variance value of this feature dimension in the main crack propagation direction by the data variance value in the direction perpendicular to the main direction. The main crack propagation direction is determined according to the stress field distribution model of the precision metal part; sets the spatial distribution correlation threshold and the direction sensitivity threshold, and retains the feature dimensions that simultaneously satisfy the condition that the correlation is greater than the threshold and the direction sensitivity is greater than 1; reorganizes the retained feature dimensions in the original arrangement order in the fused encoded tensor to generate an optimized fused processing result.

[0046] The embodiment of the present invention realizes the physical consistency fusion of thermal history process data and real-time image detection data, solves the problem of inaccurate feature association caused by fixed weights in traditional multimodal fusion; enhances the coupling characterization ability of subsurface defects and thermal stress cracks; improves the recognition accuracy and model generalization performance of multi-physical field coupling defects in complex electromagnetic environments.

[0047] The present invention provides a specific embodiment, step 404, performs weighted processing on the process feature encoding sequence and the image feature encoding sequence according to the weight adjustment parameter set to generate a fused encoded tensor, which specifically includes the following steps: Step 411: Perform an element-by-element multiplication operation on each element value of the thermal history weight vector in the weight adjustment parameter set and the corresponding element value of the process feature encoding sequence to generate a weighted process feature sequence.

[0048] In this step, the weighted process feature sequence refers to the process feature data sequence adjusted by the thermal history weight vector, and each element value reflects the weight strengthening result of the temperature gradient feature.

[0049] This step sequentially extracts each element value from the thermal history weight vector, and at the same time extracts the corresponding element value from the same position of the process feature encoding sequence, that is, the temperature gradient correlation value; multiplies the thermal history weight element value at each position by the process feature element value to generate the current element value of the weighted process feature sequence, and combines the current element values to form a weighted process feature sequence. The number of dimensions of the weighted process feature sequence is the same as that of the original process feature encoding sequence.

[0050] Step 412: Perform an element-wise multiplication operation on each element value of the image weight vector in the weight regulation parameter set and the corresponding element value of the image feature coding sequence to generate a weighted image feature sequence.

[0051] In this step, the weighted image feature sequence refers to the image feature data sequence adjusted by the image weight vector, and each element value reflects the weight strengthening result of the grain boundary topological feature.

[0052] In this step, each element value is extracted from the image weight vector in channel order, and at the same time, the grain boundary topological correlation value is extracted from the corresponding channel position of the image feature coding sequence; the image weight element value of each channel is multiplied by the image feature element value to generate the current channel value of the weighted image feature sequence; the current channel values are combined to form a weighted process feature sequence, and the number of channels of the weighted image feature sequence is the same as that of the original image feature coding sequence. Step 413: Determine the splicing ratio coefficient according to the material property parameters of the precision metal part, and perform dimensionality adjustment processing on the weighted process feature sequence and the weighted image feature sequence according to the splicing ratio coefficient to obtain an adjusted weighted process feature sequence and an adjusted weighted image feature sequence.

[0053] In this step, the splicing ratio coefficient refers to the fusion ratio control parameter generated by the material composition and the crystal structure, which determines the dimensionality adjustment ratio of the process and the image feature.

[0054] In this step, the main alloy element composition ratio parameter is extracted from the material property parameters, and a preset weight value is assigned according to the sensitivity of each element to crack propagation, and the weighted sum of all elements is calculated; the weighted sum is multiplied by the crystal structure coding parameter (numerical coding, such as body-centered cubic coding is 1.2, face-centered cubic coding is 1.5) to generate the splicing ratio coefficient, which is used to control the fusion ratio of the process and the image feature. Multiply the original dimensionality of the weighted process feature sequence by the splicing ratio coefficient to expand its dimensionality to a new dimensionality value; at the same time, divide the original dimensionality of the weighted image feature sequence by the splicing ratio coefficient to compress its dimensionality to a new dimensionality value; the dimensionality of the two sequences after adjustment is equal to ensure the feasibility of subsequent splicing.

[0055] Step 414: Interleave and splice the adjusted weighted process feature sequence and the adjusted weighted image feature sequence in the order of element positions to generate initial fusion coding data.

[0056] In this step, the order of element positions refers to the order of elements arranged in the spatial coordinates of the detection area in the data sequence, ensuring a one-to-one correspondence between features and physical positions. The initial fusion coding data refers to the original fusion data formed by interleaving and splicing the weighted process and image features, including effective features and redundant features.

[0057] When performing interleaved splicing in this step, the elements of the adjusted weighted process feature sequence and the elements of the adjusted weighted image feature sequence are alternately arranged in the order of element positions to generate initial fusion-encoded data. Specifically, process feature elements are filled in odd positions, and image feature elements are filled in even positions; the alternating arrangement order is consistent with the physical space distribution of the detection area.

[0058] Step 415: Perform target redundant dimension marking processing on the initial fusion-encoded data to obtain marked fusion-encoded data.

[0059] In this step, the position correlation between the data of each dimension of the initial fusion-encoded data and the spatial distribution map of grain boundary cracks is calculated: if the numerical fluctuation range of the data of a certain dimension in the high crack-occurrence area is less than a preset threshold, it is marked as a redundant dimension; the spatial distribution map of grain boundary cracks is the output of a pre-established defect distribution statistical model.

[0060] Step 416: Remove all marked target redundant dimensions from the marked fusion-encoded data to generate a fusion-encoded tensor.

[0061] In this step, all dimensions of the initial fusion-encoded data are traversed, and all dimension data marked as redundant are deleted; the remaining dimension data are reorganized in the original spatial order to generate a fusion-encoded tensor, and the number of its dimensions is the total number of remaining effective dimensions.

[0062] The embodiment of the present invention solves the problems in the traditional method, such as unreasonable weight allocation of process features and detection features, poor adaptability caused by a fixed multi-modal feature fusion ratio, destruction of spatial correlation by traditional serial splicing, and increased false detection rate caused by feature redundancy; it breaks through the feature extraction deviation caused by static weights.

[0063] The present invention provides a specific embodiment. Step 101: Obtain laser scattering image data of a precision metal part in an electromagnetic shielding environment, which specifically includes the following steps: Step 111: Construct a closed detection space with a multi-layer conductive shielding structure, and the multi-layer conductive shielding structure includes alternately arranged metal grid layers and absorbing material layers.

[0064] In this step, the multi-layer conductive shielding structure refers to a composite barrier formed by alternately stacking more than two layers of materials with different conductive properties. Each layer of material has specific electromagnetic wave reflection and absorption characteristics, and is used to block the interference of external electromagnetic fields on the detection process. The enclosed detection space refers to a physically isolated area surrounded by the multi-layer conductive shielding structure, and the internal environment and external electromagnetic fields are attenuated and isolated by at least 40 dB, providing a controlled electromagnetic environment for laser detection. The metal mesh layer refers to a mesh structure layer formed by weaving conductive metal wires, and the geometric parameters of the mesh are matched with the frequency band of the electromagnetic waves to be shielded, and the high-frequency electromagnetic waves are blocked from penetrating through reflection. The wave-absorbing material layer refers to a polymer composite material layer filled between the metal mesh layers, and the energy of the transmitted electromagnetic waves is converted into heat energy through dielectric loss to suppress the multiple reflections of electromagnetic waves within the shielding structure.

[0065] In this step, first, the metal mesh layer and the wave-absorbing material layer are stacked in an alternating arrangement to assemble the multi-layer structure and form an enclosed detection space. A laser emission device and an optical sensing device are arranged within the enclosed detection space. The laser emission device projects a polarized laser beam onto the surface of the precision metal part at a preset incident angle, and the preset incident angle is dynamically adjusted according to the surface curvature of the precision metal part; among them, the surface roughness of the inner wall surface of the enclosed detection space is controlled within 0.4 μm to reduce the interference of internal light reflection; the aperture size of each mesh in the metal mesh layer is determined according to the maximum wavelength of the electromagnetic waves to be suppressed, specifically, the maximum wavelength value is divided by 5 to obtain the upper limit value of the aperture size; the thickness of the wave-absorbing material layer is adjusted according to the conductivity gradient of the adjacent metal mesh layers to ensure that the dielectric constant of each layer of wave-absorbing material matches the impedance characteristics of the adjacent metal mesh layers.

[0066] Step 112: Adjust the laser incident angle of the laser emission device within the enclosed detection space according to the surface curvature distribution data of the precision metal part to obtain the adjusted laser incident angle.

[0067] In this step, the surface curvature distribution data refers to a set of quantization parameters characterizing the bending degree of each point on the surface of the precision metal part, which is obtained by a non-contact three-dimensional contour measurement device, and the data form is a distribution map of the curvature radius value changing with the surface coordinates. The laser incident angle refers to the angle between the laser beam projected by the laser emission device and the normal direction of the detection point on the surface of the precision metal part, and its value is dynamically adjusted according to the surface curvature distribution data to optimize the signal acquisition efficiency.

[0068] In this step, first, surface curvature distribution data of the precision metal part is obtained by a surface profile measuring device. According to the curvature radius of each measurement point in the surface curvature distribution data, the corresponding laser incident angle compensation value is calculated, and this compensation value is the product of the reciprocal of the curvature radius and a preset coefficient. The laser emitting device dynamically adjusts the projection direction of the polarized laser beam according to this compensation value, so that the deviation angle between the laser beam incident direction and the surface normal direction of the detection point is always less than 2 degrees, and finally the adjusted laser incident angle is obtained. At the same time, the field of view angle range of the optical sensing device is dynamically synchronized according to the incident angle of the laser emitting device to ensure that the coincidence degree between the reflected light signal receiving coverage area and the laser projection area is greater than 95%.

[0069] Step 113: Scan the surface of the precision metal part based on the adjusted laser incident angle to obtain reflected light signal data.

[0070] In this step, the reflected light signal data refers to the set of original electrical signals generated by the photoelectric conversion of the laser signals reflected from the surface of the precision metal part, and includes the signal modulation characteristics caused by the surface topography features and internal defects.

[0071] In this step, in a closed detection space, the surface of the precision metal part is scanned by a polarized laser beam with an adjusted laser incident angle, and the optical sensing device is used to receive the reflected light signal data scattered by the surface of the precision metal part. The reflected light signal data includes the light intensity distribution data modulated by the surface grain boundary characteristics and the spot morphology data modulated by the subsurface defects. Specifically, the photoelectric conversion unit in the optical sensing device is used to convert the optical signal into an electric current signal. Among them, the part of the electric current signal modulated by the surface grain boundary characteristics shows periodic intensity fluctuations, and the fluctuation frequency is proportional to the grain boundary density; the part modulated by the subsurface defects shows topological distortion of the spot morphology, and the degree of topological distortion has an exponential relationship with the defect depth. The signal transmission channel in the optical sensing device adopts a coaxial shielding structure, and the grounding impedance of the shielding layer is less than 0.1Ω to isolate the electromagnetic interference outside the closed detection space.

[0072] Step 114: Eliminate the signal distortion components of the reflected light signal data to generate optical response data, and perform correction processing on the optical response data to generate laser scattering image data to be optimized.

[0073] In this step, the signal distortion component refers to the signal components in the reflected light signal that are not related to the detection target introduced by external electromagnetic field coupling or device noise, manifested as random fluctuations or periodic interferences that have no association with the surface physical characteristics. The optical response data refers to the subset of the reflected light signal remaining after eliminating the signal distortion component, and there is a definite corresponding relationship between its signal amplitude and frequency characteristics and the surface physical properties of the precision metal part. The laser scattering image data to be optimized refers to the initial image data set generated by mapping the optical response data according to the spatial coordinates, which contains the mixed information of the effective detection signal and the residual interference signal.

[0074] In this step, first, a model of the electromagnetic field intensity distribution within the closed detection space is established. The inputs of the model include the aperture distribution data of the metal mesh layer and the wave absorption efficiency parameters of the wave-absorbing material layer. According to the spatial electromagnetic field intensity gradient map output by this model, the signal distortion components related to the distribution of this gradient map are separated from the reflected light signal data. The signal distortion components are eliminated through a signal subtractor, and only the light intensity change data directly caused by the surface physical characteristics of the precision metal part is retained in the remaining signal components to generate the optical response data. According to the proportional relationship between the physical size of the pixel unit in the optical sensing device and the actual size of the surface of the precision metal part, a mapping rule between the pixel coordinates and the actual spatial coordinates is established, and a material refractive index parameter correction factor is introduced in this mapping rule for correction. This correction factor is the ratio of the material refractive index to the standard reference refractive index. The corrected mapping rule is applied to the optical response data to generate the gray value of each pixel point in the initial laser scattering image data. The size of the gray value is linearly positively correlated with the light intensity value at the corresponding position, and finally, the laser scattering image data to be optimized is formed.

[0075] Step 115: Perform interference component separation processing on the laser scattering image data to be optimized to extract the target scattering data and generate the laser scattering image data.

[0076] In this step, the interference component refers to the invalid image information in the laser scattering image data to be optimized that comes from non-target features such as the background reflection of the detection environment and device thermal noise. The target scattering data refers to the subset of the scattering signals directly related to the surface of the precision metal part extracted from the data to be optimized through physical space filtering, and its data distribution characteristics reflect the true physical state of the surface and subsurface defects.

[0077] In this step, a three-dimensional coordinate system with the surface of the precision metal part as the reference plane is established within the closed detection space, and the vertical distance from each scattering signal source to the reference plane is calculated. A fixed distance threshold is set as 10 times the value of the laser wavelength, and the scattering signal components with a vertical distance exceeding the fixed distance threshold are filtered out through a distance filter. Spatial clustering analysis is performed on the remaining signal components, and the data in the clustering regions with a signal density greater than the preset threshold is retained, and finally, the laser scattering image data is generated, where each pixel point corresponds to a selected scattering signal source.

[0078] The embodiments of the present invention effectively suppress the influence of electromagnetic interference in the industrial environment on the scattered signal; improve the imaging quality of surface and subsurface defects of complex curved specimens; and systematically solve the problem of missed detection of micron-level defects caused by electromagnetic pollution, curved surface geometric distortion, and noise coupling in the existing open environment detection scheme.

[0079] The present invention provides a specific embodiment. In step 114, the signal distortion component of the reflected light signal data is eliminated to generate optical response data, and the optical response data is corrected to generate laser scattering image data to be optimized, which specifically includes the following steps: Step 121: Detect the spectral distribution pattern of electromagnetic interference in the closed detection space and extract the target interference frequency components.

[0080] In this step, electromagnetic interference refers to the remaining electromagnetic noise signal in the closed detection space after shielding, with a frequency range of 10 kHz - 6 GHz. The spectral distribution pattern refers to the energy distribution characteristics of the electromagnetic interference signal at different frequencies, including fundamental waves, harmonics, and random noise components. The target interference frequency components refer to the electromagnetic signal components that are irrelevant to the surface defects of precision metal parts and need to be suppressed or eliminated.

[0081] When detecting the spectral distribution pattern of the residual electromagnetic interference in the closed detection space, a broadband spectral analysis device is used to scan the electromagnetic signals in the frequency band of 10 kHz - 6 GHz, and the distribution curve of the signal amplitude varying with frequency is recorded; the periodic spike components synchronized with the laser scanning timing sequence and the broadband noise components related to the environmental electromagnetic field in the distribution curve are extracted as the target interference frequency components.

[0082] Step 122: Generate a compensation signal waveform with a phase opposite to that of the target interference frequency components based on the aperture size of the metal mesh layer and the attenuation coefficient of the wave-absorbing material layer.

[0083] In this step, the attenuation coefficient of the wave-absorbing material layer refers to the attenuation ability of the material per unit thickness for electromagnetic waves (dB / m), which is related to the dielectric constant and magnetic permeability of the material. The compensation signal waveform refers to a cancellation signal with a phase opposite to that of the interference signal waveform and an amplitude adaptation, and its duration is synchronized with the laser scanning period.

[0084] In this step, the cut-off frequency of the metal mesh layer is calculated according to its aperture size, and the calculation formula is: the speed of light / (5 × aperture size); based on the attenuation coefficient of the wave-absorbing material layer at the cut-off frequency (dB / m), according to: the original interference amplitude × = the compensation signal amplitude, the compensation signal amplitude is calculated, and based on this, a compensation signal waveform with the same waveform and opposite phase as the interference frequency components is generated, and its waveform duration is strictly synchronized with the laser scanning period.

[0085] Step 123: Superimpose the compensation signal waveform and the reflected light signal data to obtain compensated reflected light signal data.

[0086] In this step, the compensated reflected light signal data refers to the reflected light signal after superimposing the compensation signal. The periodic interference is suppressed, and the effective defect signal is retained.

[0087] In this step, the compensation signal waveform and the reflected light signal data are input into a differential amplifier to adjust the compensation signal delay to align it with the interference component in time; the amplitude of the superimposed compensation signal is adjusted by a gain controller to obtain the superimposed signal, and the amplitude of its periodic interference component is attenuated to less than 5% of the original signal, thereby generating the compensated reflected light signal data.

[0088] Step 124: Eliminate the periodic interference component in the compensated reflected light signal data, retain the target light signal, and process the target light signal to generate optical response data including light intensity distribution and spot morphology information.

[0089] In this step, the periodic interference component refers to a sinusoidal or pulsed interference signal homologous to the laser scanning frequency, manifested as discrete spikes in the spectrum. The target light signal refers to the effective optical signal modulated by the surface grain boundaries and subsurface defects, carrying defect characteristic information. The light intensity distribution refers to the spatial distribution data of the reflected light intensity at each point in the detection area, reflecting the surface microstructure differences. The spot morphology information refers to the geometric characteristic data of the reflected spot, including parameters such as diameter, ellipticity, and edge sharpness.

[0090] When eliminating the periodic interference component in this step, perform sliding window energy analysis on the compensated reflected light signal data, and the window width is 3 times the laser pulse width; if the standard deviation of the signal energy fluctuation within the window is less than the set threshold, it is determined as residual random noise and filtered out, and the target light signal modulated by the surface grain boundary scattering and subsurface defect diffraction is retained. Use a photodiode array to convert the target light signal into a current signal, and the current value I and the light intensity P satisfy I = kP (k is the photoelectric conversion coefficient); the current signal is quantized by a 16-bit analog-to-digital converter at a sampling rate of 1 MHz to generate optical response data including light intensity distribution (P values at each point) and spot morphology information (spot diameter, ellipticity).

[0091] Step 125: According to the adjusted laser incident angle and the surface refractive index parameter of the precision metal part, perform correction processing on the optical response data to generate laser scattering image data to be optimized.

[0092] In this step, the surface refractive index parameter refers to the refractive characteristic value of the precision metal part surface for the laser wavelength.

[0093] When this step corrects and processes the generated image data, the surface normal direction correction factor C = cosθ / n² (n is the surface refractive index parameter) is calculated according to the adjusted laser incident angle θ; the light intensity value P in the optical response data is multiplied by C and arranged in a two-dimensional pixel array in the order of the X-Y coordinates of the scanning path; through the non-linear mapping function generate pixel gray values, thereby forming the laser scattering image data to be optimized.

[0094] In the embodiment of the present invention, by solving the problem that the traditional method has incomplete detection of residual electromagnetic interference, the recognition accuracy is improved, and the interference frequency band coverage is large; it can effectively eliminate periodic interference and the amplitude distortion rate of micro-crack signals; it can eliminate most of the geometric distortions of curved workpieces and reduce the spatial positioning error.

[0095] The present invention provides a specific embodiment. In step 103, the process parameters corresponding to the precision metal part in the existing metal heat treatment process database are converted into a heat history feature vector, and the heat history feature vector includes a temperature gradient distribution feature and a time series phase change feature, which specifically include the following steps: Step 301: Extract the process parameter set corresponding to the precision metal part from the existing metal heat treatment process database. The process parameter set includes temperature sampling point data in the high-temperature treatment stage higher than the preset temperature value and time series data in the low-temperature holding stage lower than the preset temperature value.

[0096] In this step, the preset temperature value refers to the critical temperature threshold for dividing the high-temperature treatment stage and the low-temperature holding stage, which is dynamically set according to the material melting point characteristics, and the calculation formula is: material melting point × 0.6. The temperature sampling point data refers to the set of temperature measurement values recorded at specific spatial positions and time points during the heat treatment process, including three-dimensional coordinates (X, Y, Z) and temperature values (T). The time series data refers to the set of temperature values recorded in chronological order during the low-temperature holding stage, and the data structure is a time-temperature pair (t, T).

[0097] When this step extracts the process parameter set corresponding to the precision metal part from the existing metal heat treatment process database, first match the process record table in the database according to the unique identification code of the precision metal part; the extraction conditions for the process parameter set are: all temperature sampling point data with temperature values exceeding the preset temperature value in the high-temperature treatment stage, and all time series data with temperature values lower than the preset temperature value in the low-temperature holding stage, where the preset temperature value is set according to the material melting point characteristics of the precision metal part.

[0098] Step 302: Based on the spatial position information of each temperature sampling point in the temperature sampling point data and the temperature change amount of adjacent points, perform an expansion process on the temperature sampling points to generate a temperature gradient distribution feature.

[0099] In this step, the spatial position information refers to the coordinate data of the temperature sampling points in the three-dimensional space, which is used to construct the spatial correlation relationship of the temperature distribution. The temperature change amount between adjacent points refers to the temperature difference between two adjacent temperature sampling points in the same grid divided by the distance between their nodes, which is used to calculate the temperature gradient direction. The temperature gradient distribution feature refers to the continuous temperature distribution data generated by interpolation expansion, which characterizes the temperature gradient change characteristics in the three-dimensional space.

[0100] The temperature sampling point data in this step includes the spatial position information in the three-dimensional space and the corresponding temperature values recorded at each sampling moment; a three-dimensional grid coordinate system is constructed according to the spatial position information, and the temperature change amount between each grid node and its adjacent nodes is calculated. The calculation formula for the temperature change amount is: the difference between the temperature values of adjacent nodes divided by the node spacing; the temperature values in the areas that are not directly sampled are estimated based on the temperature change amount, and the discrete temperature sampling points are extended to a continuously distributed temperature gradient surface through linear interpolation; the temperature values of the grid nodes on the temperature gradient surface are arranged in the order of spatial coordinates to construct the temperature gradient distribution feature.

[0101] Step 303: Extract the critical duration and temperature fluctuation frequency at the phase change critical point in the time series data to generate time series phase change features.

[0102] In this step, the phase change critical point refers to the starting time point when the temperature fluctuation tends to be stable during the material phase change process, which is determined by detecting the change in the slope of the temperature curve. The critical duration refers to the duration of the stable state corresponding to the phase change critical point, and the calculation method is the end time minus the start time. The temperature fluctuation frequency feature refers to the number of temperature fluctuations per unit time, and the calculation method is the number of fluctuations divided by the duration. The time series phase change feature refers to the feature vector formed by encoding the duration feature and the temperature fluctuation frequency feature in chronological order, which characterizes the dynamic characteristics of the phase change process.

[0103] In this step, the phase change critical point is extracted from the time series data. The phase change critical point is the starting time point when the absolute value of the slope in the temperature fluctuation curve is first less than the preset threshold; the duration of the stable interval corresponding to each phase change critical point is statistically calculated, and its feature is the difference between the end time and the start time of the critical point; the number of temperature fluctuations is divided by the duration to calculate the temperature fluctuation frequency feature within this stable interval; the duration and the temperature fluctuation frequency feature are encoded in chronological order to form the time series phase change feature.

[0104] Step 304: Concatenate the temperature gradient distribution feature and the time series phase change feature to generate a thermal history feature vector, and the dimensionality of the thermal history feature vector is proportional to the number of material component types of the precision metal part.

[0105] This step concatenates the one-dimensional array of temperature gradient distribution features and the one-dimensional array of time-series phase change features in the head-to-tail order, and inserts a dimension separation identifier at the connection, generating a thermal history feature vector. The dimension separation identifier contains the information on the number of material component types of the precision metal part, specifically the binary encoding of the number of component types. The total number of dimensions of the thermal history feature vector is equal to the sum of the dimensions of the temperature gradient distribution features and the dimensions of the time-series phase change features, and is proportional to the number of material component types of the precision metal part.

[0106] The embodiments of the present invention solve the problem of insufficient defect-process association modeling caused by discrete input of traditional process data; realize the accurate mapping of the heat treatment history of different alloy systems and the defect formation mechanism, and improve the detection sensitivity of multi-modal large models to process-related defects such as thermal stress cracks and phase change-induced defects.

[0107] The present invention provides a specific embodiment, step 105, generating defect detection data based on the fusion processing result. The defect detection data includes the three-dimensional position information of grain boundary cracks and the prediction data of the crack propagation direction, and specifically includes the following steps: Step 501: Convert the detection positions corresponding to the feature dimension data related to the spatial distribution of grain boundary cracks in the fusion processing result into three-dimensional coordinate values in a preset physical space coordinate system, generating spatial coordinate data.

[0108] In this step, the spatial distribution of grain boundary cracks refers to the position and orientation characteristics of grain boundary cracks in the three-dimensional space of the precision metal part, including parameters such as length, depth, and bifurcation number. The feature dimension data refers to the data channels directly related to the detection of grain boundary cracks in the fusion processing result, and each dimension corresponds to a defect-associated feature. The preset physical space coordinate system refers to the coordinate system established based on the calibration parameters of the laser scattering system, with the origin at the geometric center of the detection area and the coordinate axes aligned with the scanning direction. The spatial coordinate data refers to a structured data set containing the three-dimensional positions (X, Y, Z) of all detection points and the corresponding feature dimension values.

[0109] This step first extracts the detection position information corresponding to each feature dimension in the feature dimension data; combines the calibration parameters of the preset physical space coordinate system, which includes the horizontal resolution and the vertical resolution, and converts the detection position information into three-dimensional coordinate values; the conversion formula is: X coordinate = horizontal pixel index × horizontal resolution, Y coordinate = vertical pixel index × vertical resolution, Z coordinate = feature dimension value × interlayer resolution; generates spatial coordinate data containing the three-dimensional coordinates of all detection points; where the horizontal pixel index is the column number of the pixel point in the laser scattering image data, and the vertical pixel index is the row number, which is automatically generated by the image acquisition system according to the scanning timing.

[0110] Step 502: Identify the numerical mutation regions and the coordinates of the target gradient regions in the spatial coordinate data to generate an initial crack position set, where the initial crack position set includes crack tip positions and branch node positions.

[0111] In this step, the numerical mutation region refers to the spatial position where the characteristic dimension value changes drastically in a local area (standard deviation > 3 times the mean). The target gradient region refers to the spatial position where the characteristic dimension value changes stably in a specific direction (change rate > 10% / point). The crack tip position refers to the stress concentration point at the crack propagation front, where the characteristic dimension value mutates most significantly. The branch node position refers to the key point where the crack path bifurcates or turns, and the characteristic dimension value shows continuous gradient characteristics.

[0112] In this step, in the spatial coordinate data, calculate the standard deviation of the characteristic dimension values within a region with a radius of 3 times the grain size around each coordinate point; if the standard deviation exceeds 3 times the average value of the adjacent region, mark it as a numerical mutation region; detect the change rate of the characteristic dimension values of adjacent coordinate points, if the change rates of 5 consecutive points are in the same direction and exceed 10% / point, mark it as a target gradient region; merge the coordinates of all marked points to form an initial crack position set including crack tip positions and branch node positions.

[0113] Step 503: Calculate the stress field distribution vector of the crack tip position according to the temperature gradient distribution characteristic in the thermal history feature vector and the subsurface scattering spot distribution gradient characteristic in the multi-modal image feature set.

[0114] In this step, the stress field distribution vector refers to a three-dimensional vector representing the force direction and intensity at the crack tip, and its direction is synthesized by the temperature gradient and the subsurface scattering spot density gradient.

[0115] In this step, extract the temperature gradient direction vector of the region corresponding to the crack tip position from the temperature gradient distribution characteristic; extract the scattering spot density gradient direction vector of the same region from the subsurface scattering spot distribution characteristic; synthesize the two direction vectors according to the weight, according to: Stress field distribution vector = Temperature gradient direction vector × ln(peak temperature / 800) + Density gradient direction vector × , generate the stress field distribution vector.

[0116] Step 504: Bind the coordinates of the crack tip position with the corresponding stress field distribution vector to generate the bound data, and combine the change rate of the characteristic dimension value of the connection path between the branch node position and the adjacent crack tip position to generate the primary defect data.

[0117] In this step, the primary defect data refers to the set of unoptimized original defect information, including position, direction, and preliminary topological relationship.

[0118] This step binds the three-dimensional coordinate data at the crack tip position with the corresponding stress field distribution vector in the following format: [X coordinate, Y coordinate, Z coordinate, vector X component, vector Y component, vector Z component]. For each branch node position, record the change rate of the connection path characteristic dimension value between it and the adjacent tip position, and generate primary defect data containing the spatial topological relationship.

[0119] Step 505: Perform topological connection verification on the primary defect data to generate defect detection data.

[0120] This step performs topological connection verification on the primary defect data. Specifically, it detects isolated coordinate points in the primary defect data (no other crack points within 5 times the grain size around), and deletes them if any; for the fracture paths in the primary defect data, according to the vector direction consistency of adjacent paths and the continuity of characteristic dimension changes (change rate difference < 5%), insert and complete points; output the optimized defect detection data.

[0121] The embodiments of the present invention solve the problem of deviation between virtual coordinates and physical positions in traditional detection methods, achieve sub-millimeter-level precise positioning through physical calibration parameters; break through the problem of missed detection of microcracks and realize automatic identification of microcracks; solve the problem of inaccurate prediction of crack propagation direction; eliminate pseudo-cracks caused by detection noise, and ensure that the crack propagation direction conforms to the laws of material fracture mechanics.

[0122] Figure 2 The following is a schematic structural diagram of a precision metal part defect detection system provided by an embodiment of the present invention, as Figure 2 shown, the system includes: An acquisition module 21, configured to acquire laser scattering image data of a precision metal part in an electromagnetic shielding environment; An analysis module 22, configured to perform multi-scale spatial analysis on the laser scattering image data to extract a multi-modal image feature set including the grain boundary topological structure on the surface of the precision metal part and the scattering spot distribution gradient feature in the subsurface; A conversion module 23, configured to convert the process parameters corresponding to the precision metal part in the existing metal heat treatment process database into a heat history feature vector, where the heat history feature vector includes a temperature gradient distribution feature and a time series phase change feature; A fusion module 24, configured to perform fusion processing on the heat history feature vector and the multi-modal image feature set through a multi-modal large model to obtain a fusion processing result; A generation module 25, configured to generate defect detection data based on the fusion processing result, where the defect detection data includes three-dimensional position information of grain boundary cracks and prediction data of crack propagation directions.

[0123] Figure 2The described precision metal part defect detection system based on a multimodal large model can execute Figure 1 The described precision metal part defect detection method according to the illustrated embodiment, the implementation principle and technical effects will not be elaborated. For the precision metal part defect detection system based on a multimodal large model in the above embodiment, the specific ways for each module and unit to execute operations have been described in detail in the embodiment related to the method, and will not be elaborated here.

[0124] In a possible design, Figure 2 The precision metal part defect detection system according to the illustrated embodiment can be implemented as a computing device, such as Figure 3 As shown, the computing device may include a storage component 31 and a processing component 32; The storage component 31 stores one or more computer instructions, where the one or more computer instructions are called and executed by the processing component 32.

[0125] The processing component 32 is configured to: obtain laser scattering image data of a precision metal part in an electromagnetic shielding environment; perform multi-scale spatial analysis on the laser scattering image data to extract a multimodal image feature set including the grain boundary topological structure on the surface of the precision metal part and the scattering spot distribution gradient feature in the subsurface; convert the process parameters corresponding to the precision metal part in the existing metal heat treatment process database into a thermal history feature vector, where the thermal history feature vector includes a temperature gradient distribution feature and a time series phase change feature; perform fusion processing on the thermal history feature vector and the multimodal image feature set through a multimodal large model to obtain a fusion processing result; generate defect detection data based on the fusion processing result, where the defect detection data includes three-dimensional position information of grain boundary cracks and prediction data of crack propagation directions.

[0126] Among them, the processing component 32 may include one or more processors to execute computer instructions to complete all or part of the steps in the above method. Of course, the processing component may also be implemented by one or more application specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors or other electronic components for executing the above method.

[0127] The storage component 31 is configured to store various types of data to support the operation of the terminal. The storage component can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0128] Of course, the computing device may also necessarily include other components, such as input / output interfaces, display components, communication components, etc.

[0129] The input / output interface provides an interface between the processing component and the peripheral interface module, and the above peripheral interface module can be an output device, an input device, etc.

[0130] The communication component is configured to facilitate communication between the computing device and other devices in a wired or wireless manner, etc.

[0131] Among them, the computing device can be a physical device or an elastic computing host provided by a cloud computing platform, etc. At this time, the computing device can refer to a cloud server, and the above processing component, storage component, etc. can be basic server resources leased or purchased from a cloud computing platform.

[0132] The embodiment of the present invention also provides a computer storage medium storing a computer program, and when the computer program is executed by a computer, it can implement the above Figure 1 A method for detecting defects of precision metal parts based on a multi-modal large model shown in the embodiment.

[0133] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the above-described systems, devices, and units can refer to the corresponding processes in the foregoing method embodiments, and will not be described herein again.

[0134] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place, or may be distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative labor.

[0135] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on such an understanding, the essence of the above technical solution, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0136] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features. These modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A precision metal part defect detection method based on a multimodal large model, characterized in that, Including: Obtaining laser scattering image data of precision metal parts in an electromagnetic shielding environment; Performing multi-scale spatial analysis on the laser scattering image data to extract a multi-modal image feature set including the grain boundary topological structure on the surface of the precision metal parts and the scattering spot distribution gradient characteristics in the subsurface; Converting the process parameters corresponding to the precision metal parts in the existing metal heat treatment process database into thermal history feature vectors, where the thermal history feature vectors include temperature gradient distribution characteristics and time series phase change characteristics; Performing fusion processing on the thermal history feature vectors and the multi-modal image feature set through a multi-modal large model to obtain a fusion processing result; Generating defect detection data based on the fusion processing result, where the defect detection data includes three-dimensional position information of grain boundary cracks and prediction data of crack propagation directions.

2. The method according to claim 1, wherein Performing fusion processing on the thermal history feature vectors and the multi-modal image feature set through a multi-modal large model to obtain a fusion processing result, including: Generating a weight regulation parameter set based on the material property parameters of the precision metal parts, where the weight regulation parameter set includes a thermal history weight vector and an image weight vector; In the thermal history feature encoding channel of the multi-modal large model, performing mapping processing on the thermal history feature vectors according to the thermal history weight vector to generate a process feature encoding sequence with temperature gradient correlation characteristics; In the image feature encoding channel of the multi-modal large model, performing a sliding window process on the multi-modal image feature set according to the image weight vector to generate an image feature encoding sequence with grain boundary topology correlation characteristics; Performing weighted processing on the process feature encoding sequence and the image feature encoding sequence according to the weight regulation parameter set to generate a fusion encoding tensor; Performing screening processing on the fusion encoding tensor to generate an optimized fusion processing result.

3. The method according to claim 2, characterized in that, Performing weighted processing on the process feature encoding sequence and the image feature encoding sequence according to the weight regulation parameter set to generate a fusion encoding tensor, including: Performing element-wise multiplication operations on each element value of the thermal history weight vector in the weight regulation parameter set and the corresponding element value of the process feature encoding sequence to generate a weighted process feature sequence; Performing element-wise multiplication operations on each element value of the image weight vector in the weight regulation parameter set and the corresponding element value of the image feature encoding sequence to generate a weighted image feature sequence; Determining a splicing ratio coefficient according to the material property parameters of the precision metal parts, and performing dimension adjustment processing on the weighted process feature sequence and the weighted image feature sequence according to the splicing ratio coefficient to obtain an adjusted weighted process feature sequence and an adjusted weighted image feature sequence; Staggeredly splicing the adjusted weighted process feature sequence and the adjusted weighted image feature sequence in the order of element positions to generate initial fusion encoding data; Performing target redundant dimension marking processing on the initial fusion encoding data to obtain marked fusion encoding data; Removing all marked target redundant dimensions in the marked fusion encoding data to generate a fusion encoding tensor.

4. The method according to claim 1, characterized in that, Obtaining laser scattering image data of precision metal parts in an electromagnetic shielding environment, including: Constructing a closed detection space with a multi-layer conductive shielding structure, where the multi-layer conductive shielding structure includes alternately arranged metal mesh layers and wave-absorbing material layers; Adjusting the laser incident angle of the laser emission device in the closed detection space according to the surface curvature distribution data of the precision metal part to obtain the adjusted laser incident angle; Scanning the surface of the precision metal part based on the adjusted laser incident angle to obtain reflected light signal data; Eliminating the signal distortion components of the reflected light signal data to generate optical response data, and performing correction processing on the optical response data to generate laser scattering image data to be optimized; Performing interference component separation processing on the laser scattering image data to be optimized to extract target scattering data and generate laser scattering image data.

5. The method according to claim 4, characterized in that, Eliminating the signal distortion components of the reflected light signal data to generate optical response data, and performing correction processing on the optical response data to generate laser scattering image data to be optimized, including: Detecting the spectral distribution pattern of electromagnetic interference in the closed detection space and extracting target interference frequency components; Generating a compensation signal waveform with a phase opposite to that of the target interference frequency component based on the aperture size of the metal mesh layer and the attenuation coefficient of the wave-absorbing material layer; Performing superposition processing on the compensation signal waveform and the reflected light signal data to obtain compensated reflected light signal data; Eliminating the periodic interference components in the compensated reflected light signal data, retaining the target light signal, and processing the target light signal to generate optical response data including light intensity distribution and spot morphology information; Performing correction processing on the optical response data according to the adjusted laser incident angle and the surface refractive index parameters of the precision metal part to generate laser scattering image data to be optimized.

6. The method according to claim 1, characterized in that Converting the process parameters corresponding to the precision metal part in the existing metal heat treatment process database into a thermal history feature vector, where the thermal history feature vector includes temperature gradient distribution features and time series phase change features, including: Extracting a set of process parameters corresponding to the precision metal part from the existing metal heat treatment process database, where the set of process parameters includes temperature sampling point data in the high-temperature treatment stage above a preset temperature value and time series data in the low-temperature holding stage below the preset temperature value; Based on the spatial position information of each temperature sampling point in the temperature sampling point data and the temperature change amount of adjacent points, performing expansion processing on the temperature sampling points to generate temperature gradient distribution features; Extracting the critical duration and temperature fluctuation frequency at the phase change critical point in the time series data to generate time series phase change features; Performing splicing processing on the temperature gradient distribution features and the time series phase change features to generate a thermal history feature vector, where the dimensionality of the thermal history feature vector is proportional to the number of material component types of the precision metal part.

7. The method according to claim 1, wherein Generating defect detection data based on the fusion processing result, where the defect detection data includes three-dimensional position information of grain boundary cracks and prediction data on the crack propagation direction, including: Convert the detection positions corresponding to the feature dimension data related to the spatial distribution of grain boundary cracks in the fusion processing result into three-dimensional coordinate values in a preset physical space coordinate system to generate spatial coordinate data; Identify the coordinate values of the numerical mutation regions and target gradient regions of the spatial coordinate data to generate an initial crack position set, where the initial crack position set includes crack tip positions and branch node positions; Calculate the stress field distribution vector of the crack tip position according to the temperature gradient distribution feature in the thermal history feature vector and the subsurface scattering spot distribution gradient feature in the multi-modal image feature set; Bind the coordinates of the crack tip position with the corresponding stress field distribution vector to generate bound data, and combine the change rate of the connection path feature dimension value between the branch node position and the adjacent crack tip position to generate primary defect data; Perform topological connection verification on the primary defect data to generate defect detection data.

8. A precision metal part defect detection system based on a multimodal large model, characterized in that, Comprising: An acquisition module for acquiring laser scattering image data of a precision metal part in an electromagnetic shielding environment; An analysis module for performing multi-scale spatial analysis on the laser scattering image data to extract a multi-modal image feature set including the grain boundary topological structure on the surface of the precision metal part and the subsurface scattering spot distribution gradient feature; A conversion module for converting the process parameters corresponding to the precision metal part in an existing metal heat treatment process database into a thermal history feature vector, where the thermal history feature vector includes a temperature gradient distribution feature and a time series phase change feature; A fusion module for fusing the thermal history feature vector and the multi-modal image feature set through a multi-modal large model to obtain a fusion processing result; A generation module for generating defect detection data based on the fusion processing result, where the defect detection data includes three-dimensional position information of grain boundary cracks and prediction data of crack propagation directions.

9. A computing device, characterized in that, Comprising a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement a method for detecting defects in precision metal parts based on a multi-modal large model as described in any one of claims 1 to 7.

10. A computer storage medium, characterized in that, Stores a computer program, and when the computer program is executed by a computer, it implements a method for detecting defects in precision metal parts based on a multi-modal large model as described in any one of claims 1 to 7.

Images