A multi-dimensional metrology system and method based on FPGA

Abstract

This invention discloses a multi-dimensional metrology system and method based on FPGA, belonging to the field of electronic data processing technology. It includes: a data acquisition module for acquiring and preprocessing multi-source sensor data; a simulator construction module for constructing a multi-physics interference simulator and loading the geometric model and physical response coefficients of a reference shadow workpiece, wherein the reference shadow workpiece is a digital representation of the actual workpiece being measured, and its geometric model and physical response coefficients are decoupled datasets; a vector field generation module for generating a pre-distorted vector field based on the output of the interference simulator and the reference shadow workpiece; and an interference identification module for extracting primary features from the multi-source sensor data. This invention enables accurate modeling of physical state drift across sensor channels by constructing a multi-physics interference simulator and generating a pre-distorted vector field, reducing multi-sensor data fusion errors, and improving data consistency and reliability in high-speed dynamic metrology scenarios.

Description

Technical Field

[0001] This invention relates to the field of electronic data processing technology, and more specifically, to a multi-dimensional measurement system and measurement method based on FPGA. Background Technology

[0002] In the industrial manufacturing and precision machining sectors, multi-dimensional metrology technology is a key support for ensuring product accuracy and performance. Especially in high-speed dynamic metrology scenarios, it is necessary to capture the geometric parameters and attitude information of moving workpieces in real time and with precision. To improve the metrology dimensions and coverage, existing multi-dimensional metrology systems typically adopt multi-sensor fusion solutions, integrating various types of sensors such as optical imaging, laser triangulation, and angle encoding. Comprehensive metrology is achieved through the complementarity of multi-source data. However, under high-speed dynamic conditions, multi-sensor fusion metrology technology still faces accuracy bottlenecks.

[0003] The key problem lies in the two defects of inconsistent spatiotemporal references and asynchronous physical state representations in multi-sensor data, which ultimately lead to significant fusion errors. Although the industry has controlled the timestamp error of data acquisition to the nanosecond level through hardware synchronization technology to achieve alignment of acquisition time points, the differences in physical characteristics of different sensors still make it impossible for such synchronization to eliminate the fundamental problem. Optical imaging, laser triangulation, angle encoding and other sensors differ in detection principles, sampling rates, installation positions and environmental sensitivity. Each sensor actually captures information slices of the same moving workpiece at different physical states at instants. This inherent asynchrony caused by the differences in sensor mechanisms cannot be eliminated by simple timestamp alignment, making it difficult for multi-source data to truly reflect the physical state of the workpiece at the same moment. The accuracy of the fused data is difficult to meet the requirements of high-speed microscale metrology.

[0004] Further analysis reveals that the root of the problem lies in the post-processing fusion mode driven by a unified clock in the existing system. This mode interprets synchronization as a time alignment problem of signal acquisition, mainly relying on post-acquisition compensation and fusion algorithm optimization, but it does not fully consider the complex dynamic characteristics under high-speed micro-scale measurement. Under high-speed motion conditions, the small displacement of the workpiece, environmental disturbances, and sensor response lag will form a complex dynamic effect, causing the physical state of the workpiece corresponding to the data from different sensor channels to have undergone small and nonlinear changes when the data is generated. The existing system lacks modeling and tracing of this cross-channel physical state drift. Relying solely on time alignment and simple compensation is insufficient to eliminate the deviation caused by the asynchrony of physical states, making it difficult to improve the fusion accuracy and meet the high-precision requirements of precision manufacturing for high-speed dynamic measurement. Summary of the Invention

[0005] To address the problems existing in the prior art, the present invention aims to provide a multi-dimensional metrology system and method based on FPGA, which can accurately model the physical state drift across sensing channels by constructing a multi-physics interference simulator and generating a pre-distortion vector field, thereby reducing the data fusion error of multi-sensor data and improving the data consistency and reliability in high-speed dynamic metrology scenarios.

[0006] To solve the above problems, the present invention adopts the following technical solution:

[0007] Firstly, a multi-dimensional metrology system based on FPGA includes:

[0008] The data acquisition module is used to acquire and preprocess multi-source sensor data;

[0009] The simulator building module is used to build a multiphysics disturbance simulator and load the geometric model and physical response coefficients of the reference shadow workpiece. The reference shadow workpiece is a digital representation of the actual workpiece under test, and its geometric model and physical response coefficients are mutually decoupled datasets.

[0010] The vector field generation module generates a pre-distorted vector field based on the output of the interference simulator and the reference shadow workpiece;

[0011] The interference identification module is used to extract primary features from multi-source sensor data, match the morphology of the primary features with the expected feature change pattern described by the pre-distortion vector field, and identify the dominant interference type and the corresponding interference quantization parameters.

[0012] The sensor calibration module performs physical quantity backtracking calculations for each sensor channel based on the dominant interference type and the pre-distortion vector field corresponding to the interference quantization parameters, generates state calibration parameters, and uses the state calibration parameters to calibrate each sensor channel.

[0013] The parameter package output module is used to perform metrological processing on the calibrated data of each sensor channel to obtain intermediate results. The intermediate results are adjusted according to the geometric constraints of the reference shadow workpiece to ensure that each intermediate result satisfies geometric self-consistency, and a unified state multi-dimensional parameter package containing geometric self-consistency deviation is generated.

[0014] The model optimization module iteratively adjusts the model parameters in the interference simulator corresponding to the dominant interference type and the corresponding interference quantization parameters, based on the geometric self-consistency deviation of the unified state multi-dimensional parameter package.

[0015] Furthermore, the simulator building module also includes:

[0016] The asymmetric coupling relationship between multi-physics interferences is decomposed in a structured manner to generate a coupling coefficient matrix and a transfer function library;

[0017] Based on the coupling coefficient matrix and transfer function library, a hardware-based interference field interaction network is constructed, which is used to generate composite waveforms.

[0018] The composite waveform is phase- and envelope-modulated using a pseudo-random sequence to generate a non-stationary interference field waveform as the output of the interference simulator.

[0019] Furthermore, the simulator building module also includes:

[0020] The geometric topology information of the baseline shadow workpiece is decoupled from the attached physical properties, and a dependency graph describing the influence rules between them is constructed to form a structured workpiece model.

[0021] Based on the structured workpiece model and dependency graph, a memory-based storage structure and a real-time query interface supporting bidirectional lookup are constructed.

[0022] Furthermore, the vector field generation module further includes:

[0023] The interference field waveform output by the interference simulator is decomposed into basic interference components. For each geometric feature of the structured workpiece model, the deformation response mode of each basic interference component is analyzed based on its attached physical properties, and a multi-dimensional set of influencing factor parameters is established.

[0024] Based on the set of influence factor parameters, the memory storage structure, and the real-time query interface, the deformation response path corresponding to the basic disturbance component is pre-calculated for each geometric feature, forming a deformation path lookup table associated with the geometric feature ID.

[0025] The interference field waveform is analyzed in real time to identify the basic interference components and their instantaneous intensity. The corresponding pre-calculated deformation response path is retrieved from the deformation path lookup table through a real-time query interface, and weighted synthesis is performed based on the instantaneous intensity to obtain the synthetic deformation vector of each geometric feature.

[0026] A constraint network is constructed using the geometric topology information in the structured workpiece model. The synthetic deformation vector is used as the initial value, and the pre-distortion vector field is generated through iterative processing using a diffusion smoothing algorithm.

[0027] Furthermore, the interference identification module also includes:

[0028] Multi-source sensor data is separated into baseline features and fluctuation features to obtain the interference feature fluctuation sequence of each sensor channel;

[0029] Calculate the time-delay cross-correlation function between the interference characteristic fluctuation sequences of each channel, and construct a cross-correlation diagram;

[0030] The pre-distorted vector field is mapped to the observation space of each sensing channel to generate a multi-channel joint observation dictionary containing characteristic wave patterns and inter-channel relationships;

[0031] The cross-correlation plot and the multi-channel joint observation dictionary are used to measure graph similarity to match the dominant interference type. The expected fluctuation pattern of the matching template is aligned with the interference characteristic fluctuation sequence to deduce the interference quantification parameters.

[0032] Furthermore, the sensing calibration module also includes:

[0033] Construct a sensor-specific positive perturbation model that incorporates nonlinear geometric and physical effects;

[0034] The displacement vectors of feature points are extracted based on the pre-distortion vector field, and the ideal expected reading and the disturbed expected reading are calculated using the positive perturbation model to obtain the reading deviation field.

[0035] The reading deviation field is fitted with an inverse function to generate a dynamic calibration mapping table;

[0036] The original data stream is permuted point by point using a dynamic calibration mapping table to output calibration data.

[0037] Furthermore, the parameter packet output module also includes:

[0038] Based on the geometric topology information in the structured workpiece model, identify the geometric feature set and trigger metrological processing related to the geometric feature set to generate an initial metrological result set with confidence level labels;

[0039] A consistency constraint equation is established based on the initial set of measurement results and geometric topology information, conflict detection is performed, and a geometric contradiction relationship graph is constructed.

[0040] Weights based on confidence level are assigned to nodes in the geometric contradiction graph, and elasticity coefficients are assigned to edges. The adjusted econometric results are obtained by calculating the elasticity coefficients through an iterative relaxation method.

[0041] The difference between the adjusted measurement results and the initial measurement result set is calculated to obtain the local adjustment amount. The final residual deviation of the consistency constraint equation is obtained as the global constraint residual. Correlation analysis is performed in combination with the dominant disturbance type and the corresponding disturbance quantification parameter. The adjusted measurement results, geometric self-consistency deviation and correlation analysis results are encapsulated into a unified state multi-dimensional parameter package.

[0042] Furthermore, the model optimization module also includes:

[0043] The unified state multidimensional parameter package is parsed to obtain a summary of the correlation analysis between geometric self-consistency deviation and interference deviation. Based on the correlation analysis summary, the geometric self-consistency deviation is decomposed into principal deviation components and residual deviation components, and the target model parameter location identifier is determined.

[0044] Inversion calculations are performed based on the principal deviation components, the target model parameter location identifiers, and the pre-stored parameter deviation sensitivity relationships to generate the target model parameter update vector.

[0045] Furthermore, the model optimization module also includes:

[0046] Based on the residual bias component, a matching interference effect is selected from the library of unactivated candidate interference effects, and the interference effect is added as a new coupling term to the interaction network between the coupling coefficient matrix and the hardware interference field to expand the model parameter set.

[0047] The target model parameter update vector is applied to the verification simulator copy, and the verification is performed using the historical interference field waveform. Based on the verification results, the target model parameter update vector is selectively fused into the main simulator.

[0048] Secondly, a multi-dimensional metrology method based on FPGA includes:

[0049] Step 1: Acquire and preprocess multi-source sensor data;

[0050] Step 2: Construct a multiphysics disturbance simulator and load the geometric model and physical response coefficients of the reference shadow workpiece;

[0051] Step 3: Based on the output of the interference simulator and the reference shadow workpiece, generate a pre-distortion vector field;

[0052] Step 4: Extract primary features from multi-source sensor data, match the morphology of the primary features with the expected feature change pattern described by the pre-distortion vector field, and identify the dominant interference type and the corresponding interference quantization parameters.

[0053] Step 5: Based on the dominant interference type and the pre-distortion vector field corresponding to the interference quantization parameter, perform physical quantity backtracking calculation for each sensing channel to generate state calibration parameters, and use the state calibration parameters to calibrate each sensing channel.

[0054] Step 6: Perform metrological processing on the calibrated data of each sensing channel to obtain intermediate results. Adjust the intermediate results according to the geometric constraints of the reference shadow workpiece to ensure that each intermediate result satisfies geometric self-consistency, and generate a unified state multi-dimensional parameter package containing geometric self-consistency deviation.

[0055] Step 7: Based on the geometric self-consistency deviation of the unified state multi-dimensional parameter package, iteratively adjust the model parameters in the interference simulator that correspond to the dominant interference type and the corresponding interference quantization parameters.

[0056] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0057] (1) This scheme constructs a multi-physics interference simulator and generates a pre-distorted vector field to accurately model the physical state drift across the sensing channels, reduce the data fusion error of multi-sensor data, and improve the data consistency and reliability in high-speed dynamic metering scenarios.

[0058] (2) This solution relies on FPGA hardware to realize core modules such as interference field interaction network and real-time query interface. Combined with pre-calculated deformation path and dynamic calibration mapping table, it greatly improves the real-time response speed of interference identification and sensor calibration, which can match the needs of micro-scale and high sampling rate metrology and ensure the dynamic processing performance of the system.

[0059] (3) This scheme optimizes the measurement results through the geometric self-consistency adjustment mechanism and the iterative relaxation method, eliminates the geometric conflict of measurement data in each sensor channel, and combines correlation analysis to achieve accurate attribution of deviations, significantly improving the accuracy and reliability of multi-dimensional measurement results, and providing reliable data support for high-precision industrial measurement scenarios.

[0060] (4) This scheme has the ability to dynamically iterate and optimize the model. It can expand the set of interference model parameters through residual deviation decomposition, adapt to the changes in interference types under different scenarios, enhance the system's adaptability to complex multi-physics interference environments, reduce measurement deviations caused by environmental disturbances, and expand the system's application scope. Attached Figure Description

[0061] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0062] Figure 1 This is a flowchart illustrating the relationships between various modules in a multi-dimensional metrology system based on FPGA according to the present invention.

[0063] Figure 2 This is a flowchart of a multi-dimensional measurement method based on FPGA according to the present invention. Detailed Implementation

[0064] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0065] Please refer to Example 1 Figure 1 A multi-dimensional metrology system based on FPGA, comprising:

[0066] The data acquisition module is used to acquire and preprocess multi-source sensor data;

[0067] The simulator building module is used to build a multiphysics disturbance simulator and load the geometric model and physical response coefficients of the reference shadow workpiece. The reference shadow workpiece is a digital representation of the actual workpiece under test, and its geometric model and physical response coefficients are mutually decoupled datasets.

[0068] The vector field generation module generates a pre-distorted vector field based on the output of the interference simulator and the reference shadow workpiece;

[0069] The interference identification module is used to extract primary features from multi-source sensor data, match the morphology of the primary features with the expected feature change pattern described by the pre-distortion vector field, and identify the dominant interference type and the corresponding interference quantization parameters.

[0070] The sensor calibration module performs physical quantity backtracking calculations for each sensor channel based on the dominant interference type and the pre-distortion vector field corresponding to the interference quantization parameters, generates state calibration parameters, and uses the state calibration parameters to calibrate each sensor channel.

[0071] The parameter package output module is used to perform metrological processing on the calibrated data of each sensor channel to obtain intermediate results. The intermediate results are adjusted according to the geometric constraints of the reference shadow workpiece to ensure that each intermediate result satisfies geometric self-consistency, and a unified state multi-dimensional parameter package containing geometric self-consistency deviation is generated.

[0072] The model optimization module iteratively adjusts the model parameters in the interference simulator that correspond to the dominant interference type and the corresponding interference quantization parameters, based on the geometric self-consistency deviation of the unified state multi-dimensional parameter package.

[0073] The simulator build module also specifically performs the following steps:

[0074] Step 11: Perform a structured decomposition of the asymmetric coupling relationship between multi-physics interferences to generate a coupling coefficient matrix and a transfer function library. The specific operations are as follows:

[0075] First, it is necessary to identify the types of physical field interference that may exist in the multi-dimensional measurement scenario, typically including vibration interference, temperature field interference, electromagnetic interference, etc., and clarify the physical nature and characterization parameters of each interference field. Based on this, a quantitative analysis is conducted on the interaction between any two interference fields. Due to the differences in the action mechanisms of different physical fields, their coupling relationships are not symmetrically distributed. For example, the amplitude enhancement effect of temperature field changes on vibration interference differs from the influence of vibration interference on the temperature field diffusion rate. This asymmetric characteristic needs to be accurately captured through structured decomposition. During the decomposition process, multiple sets of interference coupling experiments are constructed using the controlled variable method, and interaction data under different combinations of interference intensities are collected. The coupling strength parameters between each interference field are calculated based on methods such as least squares or neural network fitting, thereby generating a coupling coefficient matrix. Each element in the matrix corresponds to the coupling coefficient between two specific interference fields. For example, the element in the coupling coefficient matrix represents the coupling coefficient of the i-th type of interference to the j-th type of interference, and its value reflects the strength of the coupling effect. At the same time, for the signal transmission process between each interference field, a functional relationship describing the conversion and transmission law of interference signals from one physical field to another is established, and a transfer function library is compiled. Each transfer function in the function library corresponds to the transmission process between specific interference fields. For example, the transfer function is used to characterize the transmission characteristics of the conversion of the x-type interference signal to the y-type interference signal.

[0076] Step 12: Based on the coupling coefficient matrix and transfer function library, construct a hardware-based interference field interaction network. This hardware-based network is used to generate composite waveforms. The specific operations are as follows:

[0077] Considering the stringent real-time requirements of FPGA-based multi-dimensional metrology systems, the interference field interaction network is implemented in hardware rather than through software simulation. This ensures the real-time response speed of the interference field simulation and meets the needs of high-speed dynamic metrology scenarios. During construction, the connection relationships and interaction strengths between each interference field module are first determined based on the coupling coefficient matrix. Each interference field is mapped to a hardware functional module. Connection channels are established between modules based on the non-zero elements in the coupling coefficient matrix. The weight of each connection channel is determined by the corresponding coupling coefficient. For example, when the coupling coefficient matrix is ​​non-zero, a connection channel is established between the i-th type of interference hardware module and the j-th type of interference hardware module. The channel weight... The system is configured to implement hardware mapping of the coupling relationship between interference fields. Simultaneously, the transfer functions in the transfer function library are solidified using a hardware description language such as Verilog and integrated into each connection channel. This allows the interference signals transmitted through the channels to be converted and transmitted according to the corresponding transfer functions. The initial interference signals generated by each interference field hardware module interact, superimpose, and convert signals according to the coupling relationship through the aforementioned hardware-based connection channels. The final output is a composite waveform that comprehensively reflects the synergistic effect of multi-physics interference. This composite waveform can simulate the complex interference environment in actual metrology scenarios where multiple interferences coexist and influence each other, providing a realistic interference input for the subsequent generation of the pre-distorted vector field.

[0078] Step 13: Use a pseudo-random sequence to perform phase and envelope modulation on the composite waveform to generate a non-stationary modulation interference field waveform as the output of the interference simulator. The specific operation is as follows:

[0079] In actual industrial metrology scenarios, multi-physics interference is not a stationary signal. Its phase, amplitude, and other parameters often exhibit non-stationary characteristics with random fluctuations over time. The composite waveform generated in step 12 cannot completely replicate this non-stationary characteristic. If it is directly used for interference simulation, it will lead to deviations in the accuracy of subsequent interference identification and calibration. Therefore, a pseudo-random sequence needs to be introduced to modulate the composite waveform. The pseudo-random sequence has statistical characteristics similar to random signals, and also has repeatability and controllability. It can accurately achieve phase and envelope modulation of the composite waveform. During the modulation process, a pseudo-random sequence of preset length and frequency is first generated. The generation parameters of this sequence can be configured according to the interference characteristics of the actual metrology scenario. For example, the period T of the pseudo-random sequence can be set to any value between 10μs and 100μs to match the fluctuation period of interference in different scenarios. Then, the pseudo-random sequence is divided into two paths, one of which is used to modulate the phase of the composite waveform. This is achieved by changing the phase offset of each frequency component in the composite waveform. The phase offset is determined by the instantaneous value of the pseudo-random sequence. For example, when the instantaneous value of the pseudo-random sequence is k, the phase offset of the composite waveform is set to θ = k × 10 ...

[0080] Step 14: Decouple the geometric topology information and attached physical properties of the reference shadow workpiece, and construct a dependency graph describing the influence rules between them to form a structured workpiece model. The specific operations are as follows:

[0081] The reference shadow workpiece is a digital representation of the actual workpiece being measured. Its geometric topology information and attached physical properties are related but have different mechanisms of action. Directly fusing them into a model would increase the complexity of subsequent deformation response analysis and make it difficult to accurately determine the influence weight of different properties on deformation. Therefore, decoupling is a key prerequisite for achieving accurate modeling. First, the geometric topology information of the reference shadow workpiece is extracted and structured. The geometric topology information includes the workpiece's outline, dimensional parameters, and various geometric features, such as the spatial position and connection relationship of holes, slots, and edges. It is represented in the form of a topology graph, where nodes correspond to the geometric features of the workpiece and edges correspond to the connection relationships between geometric features. Simultaneously, the attached physical properties of the workpiece are extracted, including the density, elastic modulus, coefficient of thermal expansion, and thermal conductivity of the workpiece material. These properties directly determine the deformation response characteristics of the workpiece under different disturbance fields and are recorded in the form of a parameterized list, with each parameter associated with a corresponding workpiece part. After decoupling and extraction, a dependency graph describing the influence rules between geometric topology information and attached physical properties is constructed. In this graph, nodes correspond to geometric topology features and physical property parameters, respectively, and edges correspond to the influence relationship between the two. At the same time, an influence weight coefficient is labeled for each edge. For example, when the thermal expansion deformation of a certain geometric feature is mainly determined by the coefficient of thermal expansion of the material, a directed edge is established between the geometric feature node and the coefficient of thermal expansion node, with the edge weight coefficient set to 0.8. The weight coefficient ranges from 0 to 1, with a larger value indicating a higher degree of influence. Through the above decoupling and dependency modeling, a structured workpiece model is finally formed. This model can clearly distinguish the geometric and physical properties of the workpiece and accurately represent the association rules between the two.

[0082] Step 15: Based on the structured workpiece model and dependency graph, construct a memoized storage structure and real-time query interface that support bidirectional lookups. The specific operations are as follows:

[0083] In the process of multi-dimensional metrology, generating the pre-distorted vector field requires frequent queries of the attached physical properties corresponding to specific geometric features in the structured workpiece model, as well as the influence rules of specific physical property parameters on each geometric topological feature. If traditional linear storage and query methods are used, the query latency will be too high, failing to meet the real-time requirements of high-speed dynamic metrology. Therefore, a memory-based storage structure needs to be constructed. This structure uses a key-value pair storage method, pre-caching frequently queried geometric feature IDs and their corresponding physical property parameters, dependency edge weights, etc., as key data in a high-speed storage unit, such as the BRAM built into an FPGA. Simultaneously, an index table is established to enable rapid location of key data. For example, the geometric feature ID can be used as the key, and the corresponding elastic modulus, thermal expansion coefficient, and related influence weights can be stored as values. The index table records the storage address corresponding to each key. To achieve bidirectional query functionality, a bidirectional indexing mechanism needs to be constructed. On one hand, it supports querying the corresponding attached physical attributes and influence rules through geometric feature IDs. On the other hand, it supports querying all geometric topological features affected by physical attribute parameters. For example, the thermal expansion coefficient parameter can be used to quickly query a list of all geometric features affected by this parameter and their corresponding influence weights. Based on this, a real-time query interface is constructed. This interface adopts a hardware design and interacts with other functional modules of the FPGA, such as the vector field generation module, through a standardized bus protocol, such as the AXI4-Lite protocol. The interface integrates functions such as query request parsing, index table lookup, data reading, and return, ensuring that query requests can be responded to in a very short time, such as query latency controlled within 10ns, thereby guaranteeing the real-time processing performance of the entire multi-dimensional metering system.

[0084] In a preferred embodiment of the present invention, the vector field generation module is further configured to perform the following steps:

[0085] Step 21: Decompose the interference field waveform output by the interference simulator into basic interference components. For each geometric feature of the structured workpiece model, analyze the deformation response mode of each basic interference component based on its attached physical properties, and establish a multi-dimensional influencing factor parameter set. The specific operations are as follows:

[0086] First, for the non-stationary modulated interference field waveform output by the interference simulator, a signal decomposition algorithm is used to decompose it into multiple basic interference components. These basic interference components correspond to single-type physical field interference, such as pure vibration components, pure temperature gradient components, and pure electromagnetic interference components. The decomposition process must ensure the signal integrity of each basic component and avoid introducing distortion due to the decomposition operation. For example, wavelet decomposition algorithm is used to decompose the composite interference field waveform into basic components of different frequency bands. The frequency bandwidth of each basic component is controlled between 1kHz and 10MHz to match the frequency range of typical interference in actual measurement scenarios. Subsequently, focusing on each geometric feature in the structured workpiece model, deformation response mode analysis is carried out in combination with its decoupled attached physical properties. The attached physical properties of the geometric features include the material elastic modulus E, Key parameters such as Poisson's ratio μ and the coefficient of thermal expansion α, with different physical properties, determine the differences in the response of geometric features to different basic disturbance components. For example, geometric features with high elastic modulus have smaller deformation response amplitudes to vibration basic components, while geometric features with high thermal conductivity have faster deformation response speeds to temperature gradient basic components. In the analysis process, by establishing a correlation model between physical properties and deformation response, the influence of each basic disturbance component on the deformation of geometric features is quantified, ultimately forming a multi-dimensional set of influencing factor parameters. Each factor in the parameter set corresponds to the response characteristics of a specific geometric feature to a specific basic disturbance component. For example, the influencing factor represents the deformation response sensitivity of the i-th geometric feature to the i-th basic disturbance component, with a value range of 0.1 to 1.0. The larger the value, the more sensitive the geometric feature is to the corresponding basic disturbance component.

[0087] Step 22: Based on the influencing factor parameter set, the memorized storage structure, and the real-time query interface, pre-calculate the deformation response path corresponding to the basic disturbance component for each geometric feature, forming a deformation path lookup table associated with the geometric feature ID. The specific operations are as follows:

[0088] First, for each geometric feature, based on its corresponding multi-dimensional influencing factor parameter set, the deformation law of the geometric feature under the action of each single basic disturbance component is analyzed one by one. Combining the basic theories of structural mechanics and thermodynamics, the complete deformation trajectory of the geometric feature from the initial state to the disturbed stable state is calculated, that is, the deformation response path. The description of the deformation response path needs to include key information such as the displacement change process of each feature point on the geometric feature, the deformation rate, and the final stable position. During the pre-calculation process, the geometric topology information and attached physical property data in the structured workpiece model need to be frequently called. At this time, the memory storage structure can be accessed through the real-time query interface to quickly obtain the required data, such as querying the elastic modulus of a certain geometric feature and the connection relationship with adjacent geometric features, effectively reducing the data retrieval delay and ensuring the efficient progress of the pre-calculation process. After the deformation response paths of all geometric features under the action of each basic disturbance component are pre-calculated, these paths are associated and bound with the corresponding geometric feature IDs to form a deformation path lookup table. The lookup table is constructed using a key-value pair storage mode, with the geometric feature ID and the basic disturbance component ID as the joint key and the corresponding deformation response path as the value. At the same time, an index structure is built for the lookup table, with the geometric feature ID as the key, which can quickly locate the deformation response paths of all basic disturbance components corresponding to the geometric feature. Finally, the constructed deformation path lookup table is stored in the high-speed storage unit of the FPGA and works in conjunction with the memory storage structure.

[0089] Step 23: Analyze the interference field waveform in real time to identify the fundamental interference components and their instantaneous intensity. Retrieve the corresponding pre-calculated deformation response path from the deformation path lookup table through a real-time query interface, and perform weighted synthesis based on the instantaneous intensity to obtain the composite deformation vector of each geometric feature. The specific operations are as follows:

[0090] First, the real-time interference field waveform output by the interference simulator is dynamically analyzed. A real-time signal recognition algorithm is used to quickly identify the types of basic interference components and their corresponding instantaneous intensities in the current waveform. The instantaneous intensity is quantified by detecting the signal amplitude of each basic interference component. For example, the instantaneous amplitude of a certain vibration basic component is A=5V, corresponding to its instantaneous intensity I=0.8, with the instantaneous intensity ranging from 0 to 1.0, calculated by amplitude normalization. Subsequently, for each geometric feature, based on the identified basic interference component type, the deformation path lookup table constructed in step 22 is accessed through a real-time query interface to retrieve the corresponding instantaneous intensity of each basic interference component. The pre-calculated deformation response paths of the existing basic disturbance components are used as the weighting coefficients. Multiple deformation response paths are then weighted and synthesized. The synthesis process follows the principle of vector superposition, superimposing the deformation displacement vectors corresponding to each path according to their weights. This yields the comprehensive deformation result of the geometric feature under the current real-time disturbance field, i.e., the synthesized deformation vector. The calculation of the synthesized deformation vector must be real-time; therefore, the multiplication of weighting coefficients and the superposition of vectors are implemented using FPGA hardware. Parallel computing architecture improves computational efficiency, ensuring that the calculation delay of the synthesized deformation vector for each geometric feature is controlled within microseconds. For example, if a geometric feature is simultaneously affected by both vibration and temperature components, and the corresponding deformation response paths and their corresponding instantaneous intensities are found to be 0.6 and 0.4 respectively, then the synthesized deformation vector V = × + ×, where and are the final deformation displacement vectors corresponding to the paths and respectively. Through this process, the synthesized deformation vectors of all geometric features under the real-time disturbance field can be generated one by one.

[0091] Step 24: Construct a constraint network using the geometric topology information in the structured workpiece model. Use the synthesized deformation vector as the initial value and iterate through the diffusion smoothing algorithm to generate a pre-distortion vector field. The specific operations are as follows:

[0092] First, a constraint network is constructed based on the geometric topology information in the structured workpiece model. This network uses geometric features as nodes and the connection and positional constraints between geometric features as edges. For example, there is a constraint that the relative positions of two adjacent geometric features remain unchanged, and the corresponding edge weight is set to 0.9. The weight value ranges from 0 to 1.0, with a larger value indicating a stronger constraint. The purpose of the constraint network is to ensure that the deformation of each geometric feature in the pre-distortion vector field conforms to the geometric structure of the actual workpiece, avoiding situations where geometric features conflict with each other or deviate from the overall structure of the workpiece after deformation. Subsequently, the composite deformation vector of each geometric feature obtained in step 23 is used as the initial value and input into the constraint network. The diffusion smoothing algorithm is then started for iterative optimization. During the iteration process, the deformation vector of each geometric feature will be adjusted according to its adjacent geometric features. The deformation vectors of the features and the constraint strength between them are dynamically adjusted. For example, when the initial deformation vector of a certain geometric feature causes its relative position with adjacent geometric features to exceed the constraint range, the algorithm will adjust the deformation vector of the geometric feature based on the constraint edge weights so that the relative position of the two features returns to a reasonable range. The termination condition of the iterative optimization is that the average deviation of all geometric feature deformation vectors obtained in two adjacent iterations is less than a preset threshold, for example, the threshold is set to 0.01μm, to ensure that the optimized deformation vector has sufficient stability and accuracy. After multiple iterations of optimization, the deformation vectors of all geometric features form a unified vector set that conforms to geometric self-consistency. This set is the pre-distortion vector field. This vector field can accurately characterize the expected deformation state of each geometric feature of the reference shadow workpiece under the action of the current interference field.

[0093] In a preferred embodiment of the present invention, the interference identification module is further configured to perform the following steps:

[0094] Step 31: Separate the multi-source sensor data into baseline features and fluctuation features to obtain the interference feature fluctuation sequence of each sensor channel. The specific operation is as follows:

[0095] Multi-source sensor data originates from various sensor channels, including optical imaging, laser triangulation, and angle encoding. It contains two parts of information: baseline features corresponding to the actual state of the workpiece under interference-free conditions, and fluctuation features caused by multi-physics interference. Extracting the interference feature fluctuation sequence requires accurate separation of these two types of features. First, for the raw data from each sensor channel, an adaptive smoothing filter algorithm is used to construct baseline features. The size of the filter window can be dynamically adjusted according to the sampling rate of the sensor channel. For example, for a laser triangulation channel with a sampling rate of 1MHz, the filter window is set to 50 data points to ensure... Preserving baseline features can accurately characterize the true basic state of the workpiece while avoiding the loss of true features due to excessive filtering. Subsequently, the initial fluctuation data of each sensing channel is obtained by calculating the point-by-point difference between the original sensing data and the baseline features. Then, outlier removal processing is performed on the initial fluctuation data. The removal threshold is set to the sum of the mean of the fluctuation data and 3 times the standard deviation σ, i.e., +3σ, to eliminate the interference of random noise during data acquisition. Finally, the stable fluctuation data sequence obtained after the above processing is the interference feature fluctuation sequence of each sensing channel. This sequence can accurately reflect the change law of the signal of each sensing channel under the action of interference.

[0096] Step 32: Calculate the time-delay cross-correlation function between the interference characteristic fluctuation sequences of each channel and construct a cross-correlation plot. The specific operations are as follows:

[0097] The effects of different physical field interferences on each sensing channel vary over time. This time delay characteristic is one of the key features for distinguishing different types of interference. Therefore, it is necessary to quantify the degree of temporal correlation between the fluctuation sequences of each channel through a time delay cross-correlation function. First, select the interference characteristic fluctuation sequences of any two sensing channels as the analysis objects and set a reasonable time delay range. The maximum value of the time delay range is usually taken as 100 times the sampling period of the sensing data. For example, for a channel with a sampling period of 1μs, the time delay range is set to -50μs to 50μs to fully cover possible time delay situations. Subsequently, the cross-correlation coefficients of the two wave sequences under different time delays are calculated. The cross-correlation coefficients range from -1 to 1. A value close to 1 indicates a strong positive correlation between the two sequences under that time delay, a value close to -1 indicates a strong negative correlation, and a value close to 0 indicates no obvious correlation. The cross-correlation coefficients corresponding to all time delays are summarized, and a cross-correlation curve between the two channels is constructed with time delay as the horizontal axis and cross-correlation coefficient as the vertical axis. Then, by traversing all sensor channel combinations, the cross-correlation curves of each channel pair are integrated to form a cross-correlation graph. The cross-correlation graph clearly marks the peak correlation strength of each channel pair under different time delays. For example, the optical imaging channel and the angle coding channel show a correlation coefficient peak of 0.85 when the time delay is 5μs. These peak values ​​can accurately reflect the transmission timing characteristics of interference between different channels.

[0098] Step 33: Map the pre-distortion vector field to the observation space of each sensing channel to generate a multi-channel joint observation dictionary containing characteristic wave patterns and inter-channel relationships. The specific operations are as follows:

[0099] The pre-distortion vector field characterizes the expected deformation state of each geometric feature of the reference shadow workpiece under different disturbances. The observed signals from each sensing channel are an indirect reflection of the workpiece's deformation state. Therefore, a mapping transformation is needed to convert the deformation features into observable signal features for each channel. First, the observation space attributes of each sensing channel must be clarified. Different sensing channels have different observation dimensions and characterization methods. For example, the observation space of the optical imaging channel is a two-dimensional pixel grayscale distribution, the observation space of the laser triangulation channel is a distance measurement value sequence, and the observation space of the angle encoding channel is an angle change value sequence. The mapping process requires establishing a transformation model based on parameters such as the imaging principle and installation position of each channel. Subsequently, based on the transformation model, the deformation vectors of each geometric feature in the pre-distortion vector field are converted into corresponding sensing channel signals. The signal characteristic change pattern in the observation space is analyzed. For example, the deformation vector of a certain edge of a workpiece is converted into the gray-level gradient change pattern of edge pixels in the optical imaging channel. During the conversion, the temporal and amplitude characteristics of the deformation must be preserved. Finally, the characteristic fluctuation pattern of each channel corresponding to all interference types and the time delay correlation characteristics between channels are summarized to construct a multi-channel joint observation dictionary. The dictionary adopts a structured storage method with the interference type as the core index. Each index corresponds to a set of features containing the expected characteristic fluctuation pattern of each channel, the peak value of the cross-correlation coefficient between channels, and the time delay position. For example, when the dominant interference type is vibration interference, the dictionary will record its expected fluctuation frequency, amplitude range, and the peak value of the correlation coefficient between each channel and the time delay parameter in the optical, laser, and angle coding channels.

[0100] Step 34: Perform graph similarity measurement between the cross-correlation graph and the multi-channel joint observation dictionary to match the dominant interference type, and align the expected fluctuation pattern of the matching template with the interference characteristic fluctuation sequence to retrieve the interference quantification parameters. The specific operations are as follows:

[0101] First, a graph similarity metric algorithm is used to match the cross-correlation graph constructed in step 32 with the inter-channel correlation feature graphs corresponding to each interference type in the multi-channel joint observation dictionary. The cosine similarity algorithm is used to calculate the similarity value between the cross-correlation graph to be matched and each template graph in the dictionary. The similarity value ranges from 0 to 1; the closer the value is to 1, the more similar the correlation features are. A similarity threshold of 0.8 is set. When the similarity value of a certain interference type template is greater than or equal to 0.8, the interference type is determined to be a candidate dominant interference type. If multiple candidate types exist, the type with the highest similarity value is selected as the final dominant interference type. For example, if the similarity value of the vibration interference template is 0.92 and the similarity value of the temperature interference template is 0.65, then the dominant interference type is determined to be vibration interference. Subsequently, for the matched dominant interference type, the corresponding expected fluctuation pattern template is extracted from the multi-channel joint observation dictionary. This template is then time-delay aligned with the interference feature fluctuation sequences of each sensing channel obtained in step 31. The alignment process uses the cross-correlation graph as a reference. Based on the peak time delay of each channel pair, the timing of the expected fluctuation pattern template is adjusted to ensure that the timing characteristics of the template are consistent with those of the actual fluctuation sequence. Finally, based on the deviation analysis between the aligned expected fluctuation pattern and the actual interference characteristic fluctuation sequence, the interference quantification parameters are obtained. The quantification parameters include indicators such as the amplitude, frequency, and duration of the interference. For example, by calculating the amplitude deviation between the actual fluctuation sequence and the expected template, and combining it with the preset amplitude conversion coefficient, the actual amplitude of the vibration interference is obtained. When the amplitude conversion coefficient is set, if the deviation amplitude is 2V, the actual amplitude of the vibration interference is obtained as 1mm. Finally, the dominant interference type and the corresponding complete interference quantification parameter set are output.

[0102] In a preferred embodiment of the present invention, the sensing calibration module is further configured to perform the following steps:

[0103] Step 41: Construct a sensor-specific positive perturbation model that includes nonlinear geometric and physical effects. The specific steps are as follows:

[0104] Different types of sensing channels, such as optical imaging, laser triangulation, and angle encoding, will exhibit specific response deviations under the same interference due to differences in their hardware structure, imaging, or measurement principles. These deviations often involve nonlinear geometric and physical effects. For example, lens distortion in optical imaging channels can cause nonlinear imaging deviations, while laser scattering effects in laser triangulation channels can cause nonlinear distance measurement deviations. Therefore, the model must possess sensor-specific characteristics. During the modeling process, the physical characteristics and operating parameters of the components of each sensing channel are first analyzed. For example, the focal length f and pixel size s of the optical imaging channel, and the laser wavelength λ and detection angle of the laser triangulation channel. These parameters are used as the basic input variables of the model. Then, nonlinear geometric effect factors and physical effect factors are introduced. The nonlinear geometric effect factor is used to characterize the nonlinear mapping relationship between the sensor measurement coordinate system and the real physical coordinate system. The physical effect factor is used to quantify the impact of disturbances such as temperature and vibration on the performance of the internal electronic components of the sensor. For example, the temperature attenuation factor of laser emission power is dynamically adjusted between 0.95 and 1.0 with temperature changes. By integrating the above parameters and factors, a positive disturbance model specific to each sensing channel is established. This model can accurately describe the expected output reading law of the sensor under the action of a given physical quantity input and disturbance.

[0105] Step 42: Extract the displacement vector of feature points based on the pre-distortion vector field, and calculate the ideal expected reading and the disturbed expected reading using the positive perturbation model to obtain the reading deviation field. The specific operation is as follows:

[0106] First, the displacement vectors of workpiece feature points corresponding to the observation range of each sensing channel are extracted from the pre-distortion vector field. These displacement vectors represent the expected deformation results under the influence of the dominant interference type and corresponding quantization parameters, directly related to the sensor's measurement object. Subsequently, two types of reading calculations are performed based on the positive disturbance model: one is the ideal expected reading calculation, which assumes no interference and inputs the actual displacement vector of the feature point into the model to obtain the sensor output reading. This reading can characterize the sensor's accurate measurement results under ideal working conditions; the other is the disturbed expected reading calculation, which compares the expected deformation displacement vector of the feature point with the identified interference... The disturbance quantization parameters are input into the model to obtain the expected output reading of the sensor under the disturbance. By performing point-by-point difference calculation on the ideal expected reading and the disturbed expected reading of the same feature point, the reading deviation value corresponding to that point is obtained. Then, the reading deviation values ​​of all feature points are arranged according to their positions in the sensor observation space to finally form a reading deviation field. The reading deviation field can intuitively and comprehensively reflect the deviation distribution law of the disturbance throughout the entire sensor observation range. For example, in the reading deviation field of the laser triangulation channel, the deviation value is larger in the region near the edge of the workpiece, which is directly related to the larger deformation amplitude of the edge features in the pre-distortion vector field.

[0107] Step 43: Perform inverse function fitting on the reading deviation field to generate a dynamic calibration mapping table. The specific operations are as follows:

[0108] The reading deviation field reflects the distribution of deviations caused by interference. However, directly using the deviation field for calibration will result in large computational load and poor real-time performance. Therefore, it is necessary to transform the deviation field into a quickly searchable mapping relationship through inverse fitting. First, the input and output variables of the fitting are defined. The input variable is the original reading of the sensor or the feature point position corresponding to the original reading. The output variable is the corresponding reading deviation value. The fitting process needs to cover the entire measurement range of the sensor to ensure accurate calibration within the full range. Then, an algorithm suitable for fitting nonlinear relationships is used to fit the reading deviation field data. For example, a piecewise polynomial fitting algorithm is used to divide the measurement range into multiple intervals. A cubic polynomial is used to fit each interval to balance the fitting accuracy and computational complexity. During the fitting process, the residual between the fitting result and the actual data of the deviation field is minimized by adjusting the polynomial coefficients. The residual threshold is set to 1 / 10 of the measurement accuracy. For example, when the sensor measurement accuracy is 0.1μm, the residual threshold is set to 0.01μm to ensure that the fitting accuracy meets the calibration requirements. Finally, the input-output correspondence obtained by fitting is stored in the form of key-value pairs to form a dynamic calibration mapping table. The boundary parameters of the fitting interval are also recorded in the mapping table to quickly locate the fitting interval and deviation value corresponding to the original reading.

[0109] Step 44: Use the dynamic calibration mapping table to perform point-by-point permutation on the original data stream and output calibration data. The specific operation is as follows:

[0110] Because the multi-dimensional metrology system operates at high speed and in a dynamic state, the calibration process must meet real-time requirements. Therefore, a point-by-point replacement calibration method is adopted to avoid complex real-time calculations. First, the raw data streams from each sensor channel are received. These raw data streams contain continuous measurement readings and corresponding timestamps, feature point identifiers, and other information. For each raw reading in the data stream, the corresponding reading deviation value is quickly retrieved based on the value of the raw reading by consulting a dynamic calibration mapping table. The retrieval process leverages the high-speed storage and parallel processing architecture of the FPGA to ensure that the retrieval latency is controlled at the microsecond level; for example, the retrieval latency for a single reading does not exceed 2μ. To match the high sampling rate requirements of the sensor, calibration is then performed by inversely calculating the original reading and the corresponding deviation value. That is, if the deviation value is positive, the original reading is subtracted from the deviation value to obtain the calibrated reading; if the deviation value is negative, the original reading is added to the deviation value to obtain the calibrated reading. The specific calculation method is determined by the deviation representation logic of the positive perturbation model. After calibration, the validity of the calibration data is verified. The verification standard is whether the variation range of the calibration data is within a reasonable range. For example, if the variation range exceeds 3 times the fluctuation range of the original data, it is judged as invalid data and discarded. Finally, stable and accurate calibration data is output.

[0111] In a preferred embodiment of the present invention, the parameter packet output module is further configured to perform the following steps:

[0112] Step 51: Identify the geometric feature set based on the geometric topology information in the structured workpiece model, and trigger metrological processing related to the geometric feature set to generate an initial metrological result set with confidence level labels. The specific operations are as follows:

[0113] First, geometric topological information is extracted from the structured workpiece model. Based on metrological requirements, a core set of geometric features is selected. These features typically include key dimensions of the workpiece, hole centers, edge contours, and angular relationships. The selection criteria must be combined with the accuracy requirements of the metrological scenario. For example, in high-precision assembly scenarios, geometric features with tolerance levels of IT5 and above are selected for inclusion in the feature set. Subsequently, for different types of geometric features in the feature set, corresponding dedicated metrological processing algorithms are triggered. For example, laser triangulation is used to measure key dimensions, angle encoding and parsing algorithms are used to calculate angular relationships, and optical imaging edge extraction and fitting algorithms are used for contour features. In the characterization process, a confidence assessment mechanism is introduced simultaneously. The confidence calculation comprehensively considers factors such as the stability of the calibrated sensor data, the fitting residual of the metrology algorithm, and the signal-to-noise ratio of the feature points. For example, when the signal-to-noise ratio of the calibration data is greater than 30dB and the fitting residual is less than 0.02μm, the confidence index is set to 0.9, and the confidence index ranges from 0 to 1. The larger the value, the more reliable the metrology result. When the signal-to-noise ratio is less than 20dB or the fitting residual is greater than 0.05μm, the confidence index is set to less than 0.3. Finally, the metrology results of all geometric features are summarized with the corresponding confidence index to form the initial metrology result set.

[0114] Step 52: Based on the initial set of measurement results and geometric topology information, establish consistency constraint equations, perform conflict detection, and construct a geometric inconsistency graph. The specific operations are as follows:

[0115] First, based on the geometric topology information in the structured workpiece model, the inherent constraint relationships between various geometric features are identified. These constraints include parallel relationships, perpendicular relationships, coaxial relationships, and dimensional chain constraints, such as the distance constraint between two parallel planes or the perpendicular distance constraint between the hole center and the datum plane. Based on these inherent constraint relationships, corresponding consistency constraint equations are established. The variables of the equations are the measurement values ​​of each geometric feature in the initial measurement result set, and the constant term of the equations is the preset theoretical constraint value in the structured workpiece model. Subsequently, the initial measurement results are substituted into the consistency constraint equations to calculate the constraint satisfaction degree of each equation. The constraint satisfaction threshold is set to 0.03μm. When the deviation between the equation calculation result and the theoretical constraint value exceeds this threshold, it is determined that there is a conflict in the measurement results of the corresponding geometric feature. Finally, a geometric conflict relationship graph is constructed with geometric features as nodes and conflict relationships between features as edges. The nodes in the graph are labeled with the identifier of the corresponding geometric feature and the magnitude of the conflict deviation, and the edges are labeled with the type of the corresponding constraint equation, such as parallel constraint conflict and dimensional chain constraint conflict. This graph can intuitively present the distribution and associated features of conflicts in the initial measurement results.

[0116] Step 53: Assign weights based on confidence level to the nodes in the geometric contradiction graph and assign elasticity coefficients to the edges. Calculate the adjusted econometric results using an iterative relaxation method. The specific operations are as follows:

[0117] First, weights and elasticity coefficients are assigned. Node weights are directly associated with the confidence level markers generated in step 51. Geometric feature nodes with higher confidence levels have larger weights. For example, a node with a confidence level of 0.9 has a weight of 0.9, and a node with a confidence level of 0.3 has a weight of 0.3. A larger weight indicates higher reliability of the node's measurement results and a smaller adjustment range during the adjustment process. The elasticity coefficients of edges are set according to the importance of the corresponding constraints. For example, for constraints related to the reference surface, the elasticity coefficient is set to a smaller value, such as 0.1, indicating that the constraint is relatively rigid and should be prioritized during the adjustment process. The elasticity coefficients of non-core constraints are set to a larger value, such as 0.5, indicating that the constraint has a certain degree of adjustment flexibility. Subsequently, the iterative relaxation method is started for adjustment calculation. In the initial stage, the initial measurement results of each node are used as a benchmark. According to the constraint requirements of the edges in the contradiction relationship graph, the conflict nodes with smaller weights are initially adjusted. Then, the satisfaction degree of each constraint equation is recalculated based on the adjusted results. During the iteration process, the magnitude of each adjustment is related to the node weight and the elasticity coefficient of the edge. The smaller the weight and the larger the elasticity coefficient, the larger the adjustment magnitude, until the deviation of all constraint equations is less than the preset convergence threshold. For example, the convergence threshold is set to 0.01μm. At this point, the iteration stops and the adjusted measurement result is obtained. This result has eliminated the conflict between various geometric features and meets the geometric self-consistency requirement.

[0118] Step 54: Calculate the difference between the adjusted measurement results and the initial measurement result set to obtain the local adjustment amount. Obtain the final residual deviation of the consistency constraint equation as the global constraint residual. Combine the dominant disturbance type and the corresponding disturbance quantification parameter to perform correlation analysis. Encapsulate the adjusted measurement results, geometric self-consistency deviation, and correlation analysis results into a unified state multi-dimensional parameter package. The specific operations are as follows:

[0119] First, the difference between the adjusted measurement result and the corresponding feature value in the initial measurement result set is calculated. This difference is the local adjustment amount. The magnitude of the local adjustment amount reflects the correction range of a single geometric feature measurement result. By summarizing all local adjustment amounts, the overall deviation distribution of the initial measurement results can be intuitively understood. Simultaneously, the final residual deviation of the consistency constraint equation after iterative convergence is extracted and used as the global constraint residual. The magnitude of the global constraint residual reflects the degree to which overall geometric self-consistency is satisfied. For example, a mean global constraint residual of 0.008 μm indicates that the consistency deviation among geometric features after overall adjustment is at a low level. Subsequently, the local adjustment amount, the global constraint residual, and the previous identification are compared... The dominant interference types and their corresponding interference quantification parameters are correlated to explore the mapping relationship between interference parameters and geometric self-consistency deviation. For example, the correlation between the amplitude of vibration interference and the local adjustment amount of a certain type of geometric feature is analyzed. If the correlation coefficient reaches 0.8 or above, it indicates that the deviation of this type of geometric feature is mainly caused by vibration interference. Finally, the adjusted measurement results, the geometric self-consistency deviation data composed of the local adjustment amount and the global constraint residual, and the above correlation analysis results are integrated and packaged to form a unified state multi-dimensional parameter package. The parameter package will also include auxiliary information such as data generation timestamp, sensor channel identifier, and structured workpiece model version to ensure the integrity and traceability of the parameter package.

[0120] In a preferred embodiment of the present invention, the model optimization module is further configured to perform the following steps:

[0121] Step 61: Parse the unified state multi-dimensional parameter package to obtain a summary of the correlation analysis between geometric self-consistency deviation and interference deviation. Based on the correlation analysis summary, decompose the geometric self-consistency deviation into principal deviation components and residual deviation components, and determine the target model parameter location identifiers. The specific operations are as follows:

[0122] First, the unified state multi-dimensional parameter package is structured and analyzed, focusing on extracting the geometric self-consistency deviation data and the correlation analysis summary of the disturbance deviation. The geometric self-consistency deviation includes local adjustment amount and global constraint residual, while the correlation analysis summary clarifies the correspondence between different disturbance parameters and deviation components. Based on the association analysis summary, a deviation attribution decomposition algorithm is used to decompose geometric self-consistency deviation into principal deviation components and residual deviation components. The principal deviation component refers to the deviation portion that can be explained by the currently identified dominant disturbance type and its corresponding quantization parameters. For example, when the correlation coefficient between the vibration disturbance amplitude and the local adjustment amount of a certain type of geometric feature reaches 0.85, the corresponding deviation is classified as a principal deviation component. The residual deviation component is the deviation portion that cannot be explained by the existing dominant disturbance. Its value reflects the degree of inadequacy of the coverage of the existing disturbance model. During the decomposition process, an attribution threshold of 0.7 is set. When the correlation coefficient between the deviation and the dominant disturbance is higher than this threshold, it is determined to be a principal deviation component; otherwise, it is classified as a residual deviation component. At the same time, according to the mapping relationship between the dominant disturbance and the model parameters in the association analysis summary, the target model parameter location identifier is determined. This identifier corresponds to a specific element of the coupling coefficient matrix in the disturbance simulator or a specific function parameter in the transfer function library. For example, when the dominant disturbance is vibration-temperature coupled disturbance, the location identifier points to the parameters of the coupling coefficient and related transfer function corresponding to vibration and temperature in the coupling coefficient matrix.

[0123] Step 62: Based on the principal bias components, target model parameter location identifiers, and pre-stored parameter bias sensitivity relationships, perform inversion calculations to generate the target model parameter update vector. The specific operations are as follows:

[0124] First, the pre-stored parameter deviation sensitivity relationship data is retrieved. This data records the influence of the unit change of each model parameter in the interference simulator on the geometric self-consistency deviation. For example, the sensitivity coefficient represents the change of the geometric self-consistency deviation of the i-th type when the i-th model parameter changes by a unit, and its value ranges from 0.01μm to 0.1μm. The larger the value, the more significant the influence of the parameter on the deviation. Then, with the elimination of the principal deviation component as the goal, an inversion optimization objective function is constructed. The purpose of the objective function is to minimize the difference between the expected deviation corresponding to the adjusted model parameter and the principal deviation component. Based on this objective function and combined with the parameter deviation sensitivity relationship, the gradient descent method is used for inversion calculation. The model parameter adjustment amount that makes the objective function converge is iteratively solved. The convergence threshold is set to the difference between the adjusted expected deviation and the principal deviation component being less than 0.005μm. Finally, the obtained target model parameter adjustment amounts are sorted by position identifier to form a target model parameter update vector. Each element in the vector corresponds to the adjustment magnitude of a target model parameter. For example, a vector element = 0.02 means that the coupling coefficient needs to be increased by 0.02.

[0125] Step 63: Based on the residual bias components, select matching interference effects from the unactivated candidate interference effect library, and add the interference effects as new coupling terms to the interaction network between the coupling coefficient matrix and the hardware-based interference field to expand the model parameter set. The specific operations are as follows:

[0126] The existence of residual bias components indicates that the model parameters in the current interference simulator fail to fully represent the interference effects in the actual scenario, and there are potential interference factors that have not been activated. Therefore, it is necessary to screen matching interference effects from the library of inactive candidate interference effects. First, feature extraction is performed on the residual bias components to analyze their fluctuation patterns, frequency characteristics, and temporal correlation with existing dominant interference. For example, feature parameters such as the peak frequency and fluctuation period of the residual bias are extracted. Then, the extracted residual bias features are matched with the typical features of each interference effect in the library of inactive candidate interference effects. The matching adopts the feature similarity calculation method, and the similarity threshold is set to 0. 8. When the similarity between the typical features of a candidate interference effect and the residual deviation features is higher than the threshold, it is determined to be a matched interference effect. Finally, the matched interference effect is added as a new coupling term to the coupling coefficient matrix of the interference simulator and the hardware interference field interaction network. For example, the matched humidity interference effect is added as a new coupling term. The coupling coefficient matrix is ​​updated with the coupling coefficients of humidity and existing interference types. The signal generation and transmission module of humidity interference is added to the hardware interference field interaction network. At the same time, the transfer function library is updated to complete the expansion of the model parameter set, so that the optimized model can cover more potential interference effects.

[0127] Step 64: Apply the target model parameter update vector to the verification simulator copy, verify using the historical interference field waveform, and selectively fuse the target model parameter update vector to the main simulator based on the verification results. The specific operations are as follows:

[0128] First, based on the parameter configuration of the current master interference simulator, a replica of the verification simulator is generated. This replica has the same model structure and initial parameters as the master simulator and is used only for parameter update verification; it does not participate in the actual interference simulation process of the metering system. Then, the target model parameter update vector generated in step 62 is applied to the replica of the verification simulator. Simultaneously, historical interference field waveform data is loaded to drive the replica in interference simulation, and the expected geometric self-consistency deviation corresponding to the simulation results is obtained. During the verification process, the evaluation index is the ratio of the expected geometric self-consistency deviation obtained by the replica simulation after applying the update vector to the actual unified state multidimensional... The degree of agreement of the geometric self-consistency deviation in the degree parameter package is set with a agreement threshold of 0.9. When the agreement is higher than this threshold, it indicates that the target model parameter update vector is effective and can improve the accuracy of the interference simulation. When the agreement is lower than the threshold, it indicates that the update vector is invalid and the inversion calculation needs to be repeated in step 62. Finally, selective fusion is performed based on the verification results. If the update vector is effective, it is gradually fused into the corresponding model parameters of the main interference simulator. The fusion process adopts a smooth transition strategy to avoid fluctuations in the interference simulation caused by parameter abrupt changes. If the update vector is invalid, the update vector is discarded and the verification failure information is recorded.

[0129] Please refer to Example 2 Figure 2 Based on Example 1, this example provides a multi-dimensional measurement method based on FPGA, including:

[0130] Step 1: Acquire and preprocess multi-source sensor data;

[0131] Step 2: Construct a multiphysics disturbance simulator and load the geometric model and physical response coefficients of the reference shadow workpiece;

[0132] Step 3: Based on the output of the interference simulator and the reference shadow workpiece, generate a pre-distortion vector field;

[0133] Step 4: Extract primary features from multi-source sensor data, match the morphology of the primary features with the expected feature change pattern described by the pre-distortion vector field, and identify the dominant interference type and the corresponding interference quantization parameters.

[0134] Step 5: Based on the dominant interference type and the pre-distortion vector field corresponding to the interference quantization parameter, perform physical quantity backtracking calculation for each sensing channel to generate state calibration parameters, and use the state calibration parameters to calibrate each sensing channel.

[0135] Step 6: Perform metrological processing on the calibrated data of each sensing channel to obtain intermediate results. Adjust the intermediate results according to the geometric constraints of the reference shadow workpiece to ensure that each intermediate result satisfies geometric self-consistency, and generate a unified state multi-dimensional parameter package containing geometric self-consistency deviation.

[0136] Step 7: Based on the geometric self-consistency deviation of the unified state multi-dimensional parameter package, iteratively adjust the model parameters in the interference simulator that correspond to the dominant interference type and the corresponding interference quantization parameters.

[0137] The above description is merely a preferred embodiment of the present invention; however, the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and its improved concepts, should be covered within the scope of protection of the present invention.

Claims

1. A multi-dimension metrology system based on FPGA, characterized in that, include: The data acquisition module is used to acquire and preprocess multi-source sensor data; The simulator building module is used to build a multiphysics disturbance simulator and load the geometric model and physical response coefficients of the reference shadow workpiece. The reference shadow workpiece is a digital representation of the actual workpiece under test, and its geometric model and physical response coefficients are decoupled datasets. The geometric topology information and attached physical properties of the reference shadow workpiece are decoupled, and a dependency graph describing the influence rules between the geometric topology information and attached physical properties is constructed to form a structured workpiece model. The vector field generation module is used to generate a pre-distorted vector field based on the output of the interference simulator and the reference shadow workpiece. Specifically, it includes: The interference field waveform output by the interference simulator is decomposed into basic interference components. For each geometric feature of the structured workpiece model, the deformation response mode of each basic interference component is analyzed based on the attached physical properties of the geometric features, and a multi-dimensional influencing factor parameter set is established. Based on the set of influence factor parameters, the memory storage structure, and the real-time query interface, the deformation response path corresponding to the basic disturbance component is pre-calculated for each geometric feature, forming a deformation path lookup table associated with the geometric feature ID. The interference field waveform is analyzed in real time to identify the basic interference components and their instantaneous intensity. The corresponding pre-calculated deformation response path is retrieved from the deformation path lookup table through a real-time query interface, and weighted synthesis is performed based on the instantaneous intensity to obtain the synthetic deformation vector of each geometric feature. A constraint network is constructed using the geometric topology information in the structured workpiece model. The synthetic deformation vector is used as the initial value, and the pre-distortion vector field is generated through iterative processing using a diffusion smoothing algorithm. The interference identification module is used to extract primary features from multi-source sensor data, match the morphology of the primary features with the expected feature change pattern described by the pre-distortion vector field, and identify the dominant interference type and the corresponding interference quantization parameters. The sensor calibration module is used to perform physical quantity backtracking calculations for each sensor channel based on the dominant interference type and the pre-distortion vector field corresponding to the interference quantization parameter, generate state calibration parameters, and use the state calibration parameters to calibrate each sensor channel. The parameter package output module is used to perform metrological processing on the calibrated data of each sensor channel to obtain intermediate results. The intermediate results are adjusted according to the geometric constraints of the reference shadow workpiece to ensure that each intermediate result satisfies geometric self-consistency, and a unified state multi-dimensional parameter package containing geometric self-consistency deviation is generated. The model optimization module is used to iteratively adjust the model parameters in the interference simulator that correspond to the dominant interference type and the corresponding interference quantization parameters, based on the geometric self-consistency deviation of the unified state multi-dimensional parameter package.

2. The FPGA-based multi-dimensional metrology system of claim 1, wherein, The simulator building module also includes: The asymmetric coupling relationship between multi-physics interferences is decomposed in a structured manner to generate a coupling coefficient matrix and a transfer function library; Based on the coupling coefficient matrix and transfer function library, a hardware-based interference field interaction network is constructed, which is used to generate composite waveforms. The composite waveform is phase- and envelope-modulated using a pseudo-random sequence to generate a non-stationary interference field waveform as the output of the interference simulator.

3. The FPGA-based multi-dimensional metrology system of claim 2, wherein, The simulator building module also includes: Based on the structured workpiece model and dependency graph, a memory-based storage structure and a real-time query interface supporting bidirectional lookup are constructed.

4. The FPGA-based multi-dimensional measurement system according to claim 3, characterized in that, The interference identification module further includes: Multi-source sensor data is separated into baseline features and fluctuation features to obtain the interference feature fluctuation sequence of each sensor channel; Calculate the time-delay cross-correlation function between the characteristic fluctuation sequences of interference in each sensor channel, and construct a cross-correlation diagram; The pre-distorted vector field is mapped to the observation space of each sensing channel to generate a multi-channel joint observation dictionary containing characteristic wave patterns and inter-channel relationships; The cross-correlation plot and the multi-channel joint observation dictionary are used to measure graph similarity to match the dominant interference type. The expected fluctuation pattern of the matching template is aligned with the interference characteristic fluctuation sequence to deduce the interference quantification parameters.

5. A multi-dimensional measurement system based on FPGA according to claim 4, characterized in that, The sensing calibration module further includes: Construct a sensor-specific positive perturbation model that incorporates nonlinear geometric and physical effects; The displacement vectors of feature points are extracted based on the pre-distortion vector field, and the ideal expected reading and the disturbed expected reading are calculated using the positive perturbation model to obtain the reading deviation field. The reading deviation field is fitted with an inverse function to generate a dynamic calibration mapping table; The original data stream is permuted point by point using a dynamic calibration mapping table to output calibration data.

6. The FPGA-based multi-dimensional measurement system according to claim 5, characterized in that, The parameter package output module further includes: Based on the geometric topology information in the structured workpiece model, identify the geometric feature set and trigger metrological processing related to the geometric feature set to generate an initial metrological result set with confidence level labels; A consistency constraint equation is established based on the initial set of measurement results and geometric topology information, conflict detection is performed, and a geometric contradiction relationship graph is constructed. Weights based on confidence level are assigned to nodes in the geometric contradiction graph, and elasticity coefficients are assigned to edges. The adjusted econometric results are obtained by calculating the elasticity coefficients through an iterative relaxation method. The difference between the adjusted measurement results and the initial measurement result set is calculated to obtain the local adjustment amount. The final residual deviation of the consistency constraint equation is obtained as the global constraint residual. Correlation analysis is performed in combination with the dominant disturbance type and the corresponding disturbance quantification parameter. The adjusted measurement results, geometric self-consistency deviation and correlation analysis results are encapsulated into a unified state multi-dimensional parameter package.

7. A multi-dimensional measurement system based on FPGA according to claim 6, characterized in that, The model optimization module also includes: The unified state multidimensional parameter package is parsed to obtain a summary of the correlation analysis between geometric self-consistency deviation and interference deviation. Based on the correlation analysis summary, the geometric self-consistency deviation is decomposed into principal deviation components and residual deviation components, and the target model parameter location identifier is determined. Inversion calculations are performed based on the principal deviation components, the target model parameter location identifiers, and the pre-stored parameter deviation sensitivity relationships to generate the target model parameter update vector.

8. A multi-dimensional measurement system based on FPGA according to claim 7, characterized in that, The model optimization module also includes: Based on the residual bias component, a matching interference effect is selected from the library of unactivated candidate interference effects, and the interference effect is added as a new coupling term to the interaction network between the coupling coefficient matrix and the hardware interference field to expand the model parameter set. The target model parameter update vector is applied to the verification simulator copy, and the verification is performed using the historical interference field waveform. Based on the verification results, the target model parameter update vector is selectively fused into the main simulator.

9. A multi-dimensional metrology method based on FPGA, applied to the multi-dimensional metrology system based on FPGA as described in any one of claims 1-8, characterized in that, include: Step 1: Acquire and preprocess multi-source sensor data; Step 2: Construct a multiphysics disturbance simulator and load the geometric model and physical response coefficients of the reference shadow workpiece; Step 3: Based on the output of the interference simulator and the reference shadow workpiece, generate a pre-distortion vector field; Step 4: Extract primary features from multi-source sensor data, match the morphology of the primary features with the expected feature change pattern described by the pre-distortion vector field, and identify the dominant interference type and the corresponding interference quantization parameters. Step 5: Based on the dominant interference type and the pre-distortion vector field corresponding to the interference quantization parameter, perform physical quantity backtracking calculation for each sensing channel to generate state calibration parameters, and use the state calibration parameters to calibrate each sensing channel. Step 6: Perform metrological processing on the calibrated data of each sensing channel to obtain intermediate results. Adjust the intermediate results according to the geometric constraints of the reference shadow workpiece to ensure that each intermediate result satisfies geometric self-consistency, and generate a unified state multi-dimensional parameter package containing geometric self-consistency deviation. Step 7: Based on the geometric self-consistency deviation of the unified state multi-dimensional parameter package, iteratively adjust the model parameters in the interference simulator that correspond to the dominant interference type and the corresponding interference quantization parameters.

Images