Collaborative detection method of pressure sensing based on multi-field coupling dynamic correction and deep learning

Abstract

This invention provides a collaborative pressure sensor detection method based on multi-field coupling dynamic correction and deep learning, including: S1. Data acquisition; S2. Preprocessing; S3. Physical field correction; S4. Surface fitting; S5. Deep learning optimization; S6. Defect detection; S7. Dynamic adjustment; S8. Report generation. By synergizing multi-field modeling, dynamic optimization, and deep learning technologies, this invention significantly improves the accuracy, efficiency, and reliability of pressure sensor detection.

Description

Technical Field

[0001] The present invention relates to the field of pressure sensor detection technology, and in particular to a pressure sensor collaborative detection method based on multi-field coupling dynamic correction and deep learning. Background Art

[0002] A pressure sensor is a type of mechanical sensor. It is a device or apparatus that can sense pressure signals and convert them into usable output electrical signals according to certain rules.

[0003] Pressure sensors are the most commonly used sensors in industrial practice. They are widely used in various industrial automation environments, involving many industries such as water conservancy and hydropower, railway transportation, intelligent buildings, production automation, aerospace, military industry, petrochemicals, oil wells, electricity, ships, machine tools, pipelines, etc. There are many different detection methods for pressure sensors, mainly including the following:

[0004] 1. Appearance inspection: Before using the pressure sensor, first check the appearance of the sensor to see if there are cracks, scratches, dents or deformations on the sensor surface. Confirm that the threaded interface, pins, terminals, etc. are not oxidized, loose or damaged. Observe whether the sensor housing (such as IP67 / IP69K protection level) is well sealed and there is no leakage or dust seepage.

[0005] 2. Electrical Performance Testing: Electrical performance testing is a crucial step in ensuring the proper functioning of pressure sensors, encompassing four key dimensions: static accuracy, dynamic response, power supply stability, and safety compliance. Key testing areas include zero offset, full-scale, sensitivity, and linearity testing to ensure long-term reliability.

[0006] 3. Accuracy testing: In order to ensure that the measurement accuracy of the pressure sensor meets the requirements, accuracy testing is required. The main steps of accuracy testing include:

[0007] Select a standard pressure source of known accuracy, connect it to the pressure sensor, and place it under standard atmospheric pressure.

[0008] Apply pressure to the standard pressure source and record the voltage value output by the standard pressure source.

[0009] At the same time, record the output voltage value of the pressure sensor to be tested under the same pressure.

[0010] Calculate the error between the output voltage of the pressure sensor under test and the output voltage of the standard pressure source. The smaller the error, the higher the measurement accuracy of the pressure sensor under test.

[0011] 4. Other testing methods: pressure test, zero point test, bridge test.

[0012] In the existing technology, the fixed sensor layout cannot adapt to the non-uniform pressure field, and the environmental interference leads to a decrease in data credibility. Summary of the Invention

[0013] The present invention provides a collaborative detection method for pressure sensing based on multi-field coupling dynamic correction and deep learning. The method improves spatial resolution by using a dynamic spacing sensor array (S1), combines adaptive wavelet denoising with local normalization (S2), generates a high-confidence preprocessing matrix P', introduces a thermal-mechanical-humidity three-field coupling finite element model (S3), eliminates nonlinear interference of environmental factors on the pressure field, and outputs a physically corrected physical field correction matrix P".

[0014] To achieve the above object, the present invention adopts the following technical solutions:

[0015] The collaborative detection method based on multi-field coupling dynamic correction and deep learning pressure sensing includes:

[0016] S1. Data acquisition: The spatiotemporal correlation matrix {P(t), T(t)} is acquired using an array of dual-mode pressure-temperature sensors with adjustable spacing. The initial sensor spacing d0 = 10 / (2κ_max), where κ_max is the maximum curvature of the measured surface.

[0017] S2. Preprocessing: Perform sym4 wavelet denoising and regional normalization on the pressure data matrix P(t) to generate the preprocessing matrix P';

[0018] S3. Physical field correction: Construct a thermal-mechanical-humidity three-field coupled finite element model and calculate the temperature gradient field based on the temperature data matrix T(t) Use the model to correct P' and output the physical field correction matrix P" and the confidence map M;

[0019] S4. Surface fitting: Based on the confidence map M, valid data points are selected and the genetic-particle swarm optimization algorithm is used to optimize the NURBS surface control point parameter set θ* to generate the initial fitting surface A1±σ;

[0020] S5. Deep learning optimization: Input A1±σ, P” and M into the deep residual network, and output the optimized surface A2 and the M-weighted defect probability map D_pre;

[0021] S6. Defect detection: Based on the geometric features of D_pre⊙M, A2 and Adaptive threshold segmentation and cascade SVM classification are performed on the spatial distribution of the defect to generate the defect coordinate set C;

[0022] S7. Dynamic adjustment: According to the spatial density ρ_C of C, the distribution of low confidence regions of M and The local extreme value of the sensor is dynamically adjusted to d_new, and the thermal conductivity coefficient κ_new and humidity diffusion coefficient D_new of the thermal-mechanical-humidity three-field coupled finite element model are updated;

[0023] S8. Report generation: Generates a report containing the surface error thermal map ΔS = ‖A2-P”‖ and the temperature gradient field Enhanced detection of distribution reports R+.

[0024] In this specification, the collaborative detection method based on multi-field coupling dynamic correction and deep learning pressure sensing also includes:

[0025] S9. Closed-loop feedback: Feedback θ* to the initial population in step S4, C to the sensor deployment logic in step S1, and the coordinates of the abnormal area M to the model boundary conditions in step S3, forming a closed-loop iterative optimization.

[0026] In this specification, the construction of the thermal-mechanical-humidity three-field coupled finite element model described in S3 includes:

[0027] Governing equations:

[0028] Heat conduction equation:

[0029] Mechanical equilibrium equation:

[0030] Humidity diffusion equation:

[0031] Boundary conditions: Dirichlet conditions P″ are imposed in the region M < 0.4 boundary =P' boundary +2.3×10 -5 ·ΔT.

[0032] In this specification, the enhanced detection report R+ described in S8 also includes a defect spatiotemporal evolution map, which is generated by:

[0033] Extract historical detection cycle data t-1 ,C t-2 ,...,C t-n};

[0034] Calculate the Hausdorff distance between adjacent periodic defect sets

[0035] Visualization method: Use arrows to represent the center of mass moving speed v c =‖C t -C t-1 The fast diffusion path with ‖ / Δt>0.2mm / s is filled with contour lines to represent the H value distribution, with a color scale range of 0 to 0.5mm.

[0036] In this manual, the deployment logic of the sensor array in S1 is:

[0037] According to the feedback defect coordinate set C, redundant micro-arrays are deployed within 3mm around C with a spacing of 0.2d_new;

[0038] The timestamp alignment accuracy is ≤10μs, the pressure sampling rate is 1kHz, and the temperature sampling rate is 10Hz.

[0039] In this specification, the preprocessing of S2 includes:

[0040] The sym4 wavelet basis function is used for 5-layer decomposition and denoising;

[0041] The regional normalization formula is P' ij =(P ij -μ local ) / σ local ;

[0042] where μ local and σ local is the mean and standard deviation of the 5×5 neighborhood centered at point ij.

[0043] In this specification, the optimization goal of the genetic-particle swarm optimization algorithm in S4 is:

[0044]

[0045] Where σ is the fitting standard deviation, and the optimization termination condition is that the fitness change is <1e-4 for 10 consecutive generations.

[0046] In this specification, the structure of the deep residual network in S5 is:

[0047] Input layer: 128×128 grid data of three channels A1±σ, P”, and M;

[0048] Network architecture: Improved ResNet-50, with the last layer replaced by a dual-branch output, namely A2 regression branch and D_pre classification branch;

[0049] Loss function: The A2 regression branch uses Huber loss, and the D_pre classification branch uses Focal Loss.

[0050] In this specification, the features of the cascaded SVM classification in S6 include:

[0051] Geometric features: A2 surface curvature κ, normal vector deviation Δn;

[0052] Physical characteristics: Amplitude, weighted probability value of D_pre⊙M;

[0053] Classification rules: The first-level SVM screens suspected defect areas, and the second-level SVM distinguishes cracks, holes, and delamination defects.

[0054] In this manual, the formula for dynamically adjusting the sensor spacing in S7 is:

[0055]

[0056] And when When the local extreme value is greater than 10℃ / mm, the spacing is forced to be increased to 0.5d_new.

[0057] In summary, the present invention has at least the following beneficial effects:

[0058] This invention significantly improves the accuracy, efficiency and reliability of pressure sensor detection through the collaboration of multi-physics field modeling, dynamic optimization and deep learning technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0060] Figure 1 This is a schematic diagram of the collaborative detection method based on multi-field coupling dynamic correction and deep learning pressure sensing involved in the present invention.

[0061] Figure 2 This is a schematic diagram of data transmission and feedback of the collaborative detection method based on multi-field coupling dynamic correction and deep learning pressure sensing involved in the present invention. DETAILED DESCRIPTION

[0062] Hereinafter, only certain exemplary embodiments are briefly described. As will be appreciated by those skilled in the art, the described embodiments may be modified in various ways without departing from the spirit or scope of the embodiments of the present invention. Therefore, the drawings and description are to be regarded as illustrative in nature and not restrictive.

[0063] The disclosure below provides many different embodiments or examples for implementing different structures of the embodiments of the present invention. In order to simplify the disclosure of the embodiments of the present invention, the components and configurations of specific examples are described below. Of course, these are merely examples and are not intended to limit the embodiments of the present invention. In addition, the embodiments of the present invention may repeat reference numerals and / or reference letters in different examples. Such repetition is for the purpose of simplicity and clarity and does not in itself indicate the relationship between the various embodiments and / or configurations discussed.

[0064] The embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0065] like Figure 1 As shown, this embodiment provides a collaborative detection method based on multi-field coupling dynamic correction and deep learning pressure sensing, including:

[0066] S1. Data acquisition: The spatiotemporal correlation matrix {P(t), T(t)} is collected by an array of pressure-temperature dual-mode sensors with adjustable spacing. The initial sensor spacing d0 = 10 mm / (2κ_max), where κ_max is the maximum curvature of the measured surface (unit: mm -1 );

[0067] S2. Preprocessing: Perform sym4 wavelet denoising and regional normalization on the pressure data matrix P(t) to generate the preprocessing matrix P';

[0068] S3. Physical field correction: Construct a thermal-mechanical-humidity three-field coupled finite element model and calculate the temperature gradient field based on the temperature data matrix T(t) Use this model to correct P' and output the physical field correction matrix P" and the confidence map M (M∈[0,1], the higher the M value, the stronger the data reliability);

[0069] S4. Surface fitting: Based on the confidence map M, valid data points are selected and the genetic-particle swarm optimization algorithm is used to optimize the NURBS surface control point parameter set θ* to generate the initial fitting surface A1±σ (σ is the fitting standard deviation, unit: MPa);

[0070] S5. Deep learning optimization: Input A1±σ, P” and M into the deep residual network, and output the optimized surface A2 and the M-weighted defect probability map D_pre (D_pre∈[0,1]);

[0071] S6. Defect detection: Based on the geometric features of D_pre⊙M, A2 and Adaptive threshold segmentation and cascade SVM classification are performed on the spatial distribution of to generate a defect coordinate set C (i.e., a defect position coordinate set);

[0072] S7. Dynamic adjustment: According to the spatial density of C ρ_C (ρ_C = number of defects / unit area), the low confidence region distribution of M and The local extreme value of the sensor is dynamically adjusted to d_new, and the thermal conductivity coefficient κ_new and humidity diffusion coefficient D_new of the thermal-mechanical-humidity three-field coupled finite element model are updated;

[0073] S8. Report generation: Generates a thermal map containing the surface error ΔS = ‖A2-P”‖ (unit: MPa) and the temperature gradient field Enhanced detection of distribution reports R+.

[0074] In this specification, the collaborative detection method based on multi-field coupling dynamic correction and deep learning pressure sensing also includes:

[0075] S9. Closed-loop feedback: Feedback θ* to the initial population in step S4, C to the sensor deployment logic in step S1, and the coordinates of the abnormal area M to the model boundary conditions in step S3, forming a closed-loop iterative optimization.

[0076] In some embodiments, the construction of the thermal-mechanical-humidity three-field coupled finite element model in S3 includes:

[0077] Governing equations:

[0078] Heat conduction equation:

[0079] Mechanical equilibrium equation:

[0080] Humidity diffusion equation:

[0081] ρ is the material density, c p is the specific heat capacity at constant pressure, is the temperature-time change rate, is the divergence operator, k is the thermal conductivity, Q ext is the external heat source term, σ is the stress tensor, α is the thermal expansion coefficient, E is the elastic modulus, is, D is the humidity diffusivity, is the Laplace operator of the humidity field, β is the heat-humidity coupling coefficient;

[0082] ‖▽T‖ is the temperature gradient norm;

[0083] Boundary conditions: Dirichlet conditions P″ are imposed in the region M < 0.4 boundary =P' boundary +2.3×10 -5 ΔT, P″ boundary is the corrected boundary pressure value (output data after physical field correction in S3), P' boundary is the original value of the boundary pressure after preprocessing (the boundary point data in the matrix P' after preprocessing in S2), and ΔT is the temperature change in the boundary area (the boundary area time difference calculation of the temperature signal T(t) in S1 (ΔT = T(t) - T(t - Δt))).

[0084] In some embodiments, the enhanced detection report R+ in S8 further includes a defect spatiotemporal evolution map, which is generated by:

[0085] Extract historical detection cycle data t-1 ,C t-2 ,...,C t-n};

[0086] Calculate the Hausdorff distance between adjacent periodic defect sets

[0087] Visualization method: Use arrows to represent the center of mass moving speed v c =‖C t -C t-1 The fast diffusion path with ‖ / Δt>0.2mm / s is filled with contour lines to represent the H value distribution, with a color scale range of 0 to 0.5mm.

[0088] In some embodiments, the deployment logic of the sensor array in S1 is:

[0089] According to the feedback defect coordinate set C, redundant micro-arrays are deployed within 3mm around C with a spacing of 0.2d_new;

[0090] The timestamp alignment accuracy is ≤10μs, the pressure sampling rate is 1kHz, and the temperature sampling rate is 10Hz.

[0091] In some embodiments, the pre-processing of S2 includes:

[0092] The sym4 wavelet basis function is used for 5-layer decomposition and denoising;

[0093] The regional normalization formula is P' ij =(P ij -μ local ) / σ local ;

[0094] where μ local and σ local is the mean and standard deviation of the 5×5 neighborhood centered at point ij.

[0095] In some embodiments, the optimization objective function of the genetic-particle swarm optimization algorithm in S4 is:

[0096]

[0097] Where σ is the fitting standard deviation, and the optimization termination condition is that the fitness change is <1e-4 for 10 consecutive generations.

[0098] In some embodiments, the structure of the deep residual network in S5 is:

[0099] Input layer: 128×128 grid data of three channels A1±σ, P”, and M;

[0100] Network architecture: Improved ResNet-50, with the last layer replaced by a dual-branch output, namely A2 regression branch and D_pre classification branch;

[0101] Loss function: The A2 regression branch uses Huber loss, and the D_pre classification branch uses Focal Loss.

[0102] In this specification, the features of the cascaded SVM classification in S6 include:

[0103] Geometric features: A2 surface curvature κ, normal vector deviation Δn;

[0104] Physical characteristics: Amplitude, weighted probability value of D_pre⊙M;

[0105] Classification rules: The first-level SVM screens suspected defect areas, and the second-level SVM distinguishes cracks, holes, and delamination defects.

[0106] In some embodiments, the formula for dynamically adjusting the sensor spacing in S7 is:

[0107]

[0108] And when When the local extreme value is greater than 10℃ / mm, the spacing is forced to be increased to 0.5d_new.

[0109] In some embodiments, the specific operation of the closed-loop feedback in step S9 is:

[0110] θ* feedback: The optimized NURBS parameter set θ* is used as the initial population center of the next iteration cycle step S4;

[0111] C feedback: Based on the spatial distribution density of C, increase the sampling points within a radius of 0.5 mm around C to 3 times the original density in step S1;

[0112] M feedback: The coordinates of the region M<0.3 are used as mandatory boundary conditions for the thermal-mechanical-moisture three-field coupled finite element model in step S3.

[0113] In some embodiments, the maximum curvature κ_max of the measured surface is obtained by one of the following two technical solutions, which is selected based on the completeness of the prior information of the measured object:

[0114] Solution 1 (when the design model of the object under test is known)

[0115] Step S0 (pre-step): CAD model analysis

[0116] Input the 3D CAD model of the object to be measured and extract its parametric surface equation S(u,v).

[0117] Calculate the principal curvature field κ1(u,v),κ2(u,v) of the surface using the formula:

[0118]

[0119] H is the mean curvature, K is the Gaussian curvature;

[0120] The maximum absolute value of all principal curvatures is taken as κ_max=max(|κ1|,|κ2|).

[0121] Output κ_max to step S1 for calculating the initial sensor distance d0.

[0122] Solution 2 (when the model of the object under test is unknown)

[0123] Step S1 Additional Operation: Dynamic Curvature Estimation

[0124] First iteration:

[0125] Use default sensor spacing Conduct preliminary data collection.

[0126] The curvature field κ(u,v) of the initial fitting surface A1 in step S4 is calculated.

[0127] Take the maximum absolute value of the curvature to update κ_max.

[0128] Subsequent iterations:

[0129] Data is collected again according to d_new dynamically adjusted in step S7.

[0130] A more accurate κ_max is calculated using the surface A2 optimized in step S5 and fed back to the next cycle S1.

[0131] Parameter closed-loop logic

[0132] Solution 2 dynamically updates κ_max through the surface fitting results of steps S4 / S5, forming the following closed loop:

[0133] S1 collects data → S4 / S5 calculates κ_max → S7 adjusts d_new → S1 updates the data, forming a joint optimization with the formula for dynamically adjusting the sensor spacing.

[0134] The core data and their technical meanings used to illustrate the pressure sensor test results in the enhanced test report R+ are as follows:

[0135] 1. Surface error heat map (ΔS = ‖A2-P”‖)

[0136] Technical Definition:

[0137] The absolute value of the point-by-point residual between the optimized surface A2 and the measured pressure matrix P” after physical field correction reflects the measurement accuracy and spatial consistency of the pressure sensor array.

[0138] Interpretation of test results:

[0139] Qualification judgment: If ΔS < 0.1MPa (preset threshold), it indicates that the measurement error of the sensor in the corresponding area meets the accuracy requirements;

[0140] Abnormal positioning: Areas with ΔS>0.3MPa (highlighted in red) indicate poor sensor probe contact or hardware failure;

[0141] Data reliability association: Transparency α = 1-M. The ΔS in the low confidence region (low M value) needs to be comprehensively judged in combination with the confidence mapping in step S3.

[0142] 2. Defect probability diagram D_pre⊙M (M-weighted defect probability)

[0143] Technical Definition:

[0144] The point-wise product of the defect probability map D_pre output by the deep residual network and the confidence map M quantifies the possibility (0 to 1) of the structural defects detected by the sensor array.

[0145] Interpretation of test results:

[0146] Defect distribution: The area where D_pre⊙M>0.7 indicates that there are high-probability defects within the sensor coverage area;

[0147] Sensor performance verification: If D_pre⊙M still shows a high value in a known defect-free area (such as a calibration block), it indicates that the sensor has a false detection (calibration is required);

[0148] Dynamic sensitivity: The defect probability is negatively correlated with the sensor spacing d_new in step S7 (the smaller the spacing, the higher the detection resolution).

[0149] 3. Defect coordinate set C and classification results

[0150] Technical Definition:

[0151] The defect location coordinate set C output by the cascade SVM classification includes defect types (cracks, holes, delamination) and geometric features (length, area).

[0152] Interpretation of test results:

[0153] Sensor coverage capability: The distribution density of C reflects the detection integrity of the sensor array for surface defects being measured;

[0154] Missed detection analysis: If historical data comparison shows that the defect has spread but C has not been updated, it indicates that the sensor deployment logic (step S1) needs to be optimized;

[0155] Type affinity:

[0156] Cracks: mostly caused by pressure overload, it is necessary to check whether the sensor range matches;

[0157] Holes / delamination: This indicates that the sensor mounting surface is not properly fitted and that the array spacing or pressure application method needs to be adjusted.

[0158] 4. Defect spatiotemporal evolution map

[0159] Technical Definition:

[0160] Based on the visualization of Hausdorff distance H and center of mass moving speed, the spatiotemporal evolution of defects in multi-cycle detection is displayed.

[0161] Interpretation of test results:

[0162] Sensor stability assessment:

[0163] If the H value continues to increase, it indicates that the sensor's positioning consistency for the same defect has decreased;

[0164] Center of mass moving speed v c >0.2mm / s indicates that the defect is rapidly expanding, and it is necessary to verify whether the sensor sampling rate (1kHz) meets the dynamic detection requirements;

[0165] Environmental interference analysis: defect diffusion direction and When the gradient directions coincide, it indicates that the temperature field interference causes the sensor measurement drift.

[0166] 5. Temperature gradient field distributed

[0167] Technical Definition:

[0168] The temperature gradient amplitude in the thermal-mechanical-humidity coupled field calculated in step S3 reflects the interference intensity of the ambient temperature on the pressure sensor measurement.

[0169] Interpretation of test results:

[0170] Temperature drift compensation effect:

[0171] Wakata The ΔS in the region (>10°C / mm) is still below the threshold, indicating that the physical field correction model in step S3 is effective;

[0172] On the contrary, the thermal conductivity coefficient κ_new needs to be optimized (step S7) to improve the temperature drift suppression capability;

[0173] Sensor selection basis: High temperature resistant sensors (such as MEMS piezoresistive) are required in extreme value areas.

[0174] 6. Dynamically adjust parameter records (d_new,κ_new,D_new)

[0175] Technical Definition:

[0176] Step S7 outputs the sensor distance d_new, the updated thermal conductivity κ_new, and the humidity diffusion coefficient D_new.

[0177] Interpretation of test results:

[0178] Adaptive capability verification:

[0179] The reduction factor of d_new reflects the response sensitivity of the sensor array to local defects / temperature gradients;

[0180] The update amplitude of κ_new / D_new reflects the online learning ability of the multi-field coupling model;

[0181] System robustness: If d_new converges to a stable value during iteration, it indicates that the detection system has reached the optimal configuration.

[0182] Comprehensive detection conclusion generation logic

[0183] Report R+ outputs the final conclusion through the following rules:

[0184] Eligibility criteria:

[0185] The global mean of ΔS is less than 0.1 MPa and there are no type I defects (cracks) in C;

[0186] The extreme value area ΔS<0.15MPa and D_pre⊙M<0.4.

[0187] Calibration suggestion trigger conditions:

[0188] At the same location, D_pre⊙M is continuously > 0.6 in three iterations or the local peak value of ΔS is > 0.3 MPa;

[0189] System optimization suggestions:

[0190] If v c >0.5mm / s, it is recommended to increase the temperature sampling rate to 20Hz;

[0191] If the H value increases by >30% across cycles, the sensor array reference plane needs to be recalibrated.

[0192] The specific process of the technical solution of the present invention is as follows ( Figure 1 and Figure 2 shown):

[0193] S1. Data Collection

[0194] Input parameters:

[0195] Initial / feedback sensor distance d0 or d_new (unit: mm);

[0196] Historical defect coordinate set C (unit: mm coordinate).

[0197] Execution process:

[0198] Sensor deployment: If C is not empty, deploy redundant micro-arrays within 3 mm around C with a spacing of 0.2d_new; in the remaining areas, deploy arrays evenly according to d_new or d0.

[0199] Synchronous acquisition: The pressure signal P(t) is sampled at 1kHz, and the temperature signal T(t) is sampled at 10Hz; the timestamp alignment error is ≤10μs.

[0200] Output data: spatiotemporal correlation matrix {P(t), T(t)}.

[0201] S2. Preprocessing

[0202] Input data: P(t) (from S1)

[0203] Execution process:

[0204] Wavelet denoising: Use sym4 wavelet basis function for 5-layer decomposition and soft threshold denoising.

[0205] Regional normalization: For each data point P ij , calculate the mean μ of its 5×5 neighborhood local and standard deviation σ local ;

[0206] Output data: preprocessing matrix P'.

[0207] S3. Physics correction

[0208] Input data: P' (from S2), T(t) (from S1)

[0209] Execution process:

[0210] Temperature gradient calculation:

[0211] T(t) is calculated using the central difference method (Unit: ℃ / mm).

[0212] Finite element model solution:

[0213] Solve the thermal-mechanical-humidity three-field coupled model based on the control equations and boundary conditions;

[0214] Output the corrected pressure matrix P".

[0215] Confidence map generation:

[0216] Normalized to [0,1].

[0217] Output data: P", M,

[0218] S4. Surface fitting

[0219] Input data: P (from S3), M (from S3), θ* (feedback from S9)

[0220] Execution process:

[0221] Data Filtering:

[0222] Data points with M < 0.3 are eliminated, and points with M ≥ 0.6 are given a weight of 3 times.

[0223] Hybrid optimization algorithm:

[0224] According to the optimization objective function, genetic algorithm (crossover rate 0.8) and particle swarm algorithm (inertia weight 0.6) are used for alternating optimization;

[0225] Generate NURBS surface A1±σ (σ is the standard deviation of the fitting residual).

[0226] Output data: initial fitting surface A1±σ.

[0227] S5. Deep Learning Optimization

[0228] Input data: A1±σ (from S4), P (from S3), M (from S3)

[0229] Execution process:

[0230] Data input:

[0231] Splice A1±σ, P”, and M into three-channel 128×128 grid data.

[0232] Network Reasoning:

[0233] Use the improved ResNet-50 network to output the optimized surface A2 and defect probability map D_pre.

[0234] Weighted processing:

[0235] Generate an M-weighted defect probability map D_pre⊙M.

[0236] Output data: A2, D_pre⊙M.

[0237] S6. Defect Detection

[0238] Input data: A2 (from S5), D_pre⊙M (from S5), (From S3)

[0239] Execution process:

[0240] Feature extraction:

[0241] Extract the curvature κ and normal vector deviation Δn of the A2 surface;

[0242] extract Amplitude and D_pre⊙M value.

[0243] Cascade classification:

[0244] First-level SVM screening or suspected areas with D_pre⊙M>0.7;

[0245] The secondary SVM distinguishes cracks, holes, and delamination defects.

[0246] Output data: defect coordinate set C.

[0247] S7. Dynamic Adjustment

[0248] Input data: C (from S6), M (from S3), (From S3)

[0249] Execution process:

[0250] Calculate the adjustment parameters:

[0251] Calculate d_new according to the formula for dynamically adjusting the sensor spacing;

[0252] according to Extreme value update κ_new=1.2k_old(when ).

[0253] Model Update:

[0254] Write κ_new and D_new into the thermal-mechanical-humidity three-field coupled finite element model.

[0255] Output data: d_new, κ_new, D_new.

[0256] S8. Report Generation

[0257] Input data: A2 (from S5), ΔS = ‖A2-P"‖ (from S5), (from S3), C (from S6)

[0258] Execution process:

[0259] Heatmap generation:

[0260] The transparency of the ΔS heat map is α=1-M (M comes from S3).

[0261] Defect evolution analysis:

[0262] Calculate the Hausdorff distance H and the center of mass moving speed v c ;

[0263] Generate a spatiotemporal defect evolution map with arrows and contour lines.

[0264] Output data: Enhanced detection report R+.

[0265] S9. Closed-loop feedback

[0266] Input data: θ* (from S4), C (from S6), M (from S3)

[0267] Execution process:

[0268] Parameter feedback:

[0269] θ*→S4: serves as the initial population center of the next cycle of genetic algorithm;

[0270] C→S1: Intensify sampling within 0.5 mm around C to 3 times the density;

[0271] M→S3: Set the coordinates in the region M<0.3 as mandatory boundary conditions for the model.

[0272] Output data: closed loop completion signal.

[0273] The above embodiments are intended to illustrate the present invention, not to limit the present invention. Therefore, changes in illustrative values ​​or substitutions of equivalent components should still fall within the scope of the present invention.

[0274] From the above detailed description, it will be clear to those skilled in the art that the present invention can indeed achieve the aforementioned objectives and is in compliance with the provisions of the Patent Law.

[0275] Although preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they become aware of the basic inventive concepts. Therefore, the appended claims are intended to be interpreted as covering the preferred embodiments and all changes and modifications that fall within the scope of the invention. The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. It should be noted that any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention are intended to be included within the scope of protection of the present invention.

[0276] It should be noted that the above description of the relevant processes is for illustration and purpose only and does not limit the scope of application of this specification. For those skilled in the art, various modifications and changes can be made to the processes under the guidance of this specification. However, such modifications and changes are still within the scope of this specification.

[0277] The basic concepts have been described above. It will be apparent to those skilled in the art after reading this application that the above disclosures are merely illustrative and do not constitute limitations on this application. Although not explicitly stated herein, those skilled in the art may make various modifications, improvements, and amendments to this application. Such modifications, improvements, and amendments are suggested in this application and remain within the spirit and scope of the exemplary embodiments of this application.

[0278] At the same time, this application uses specific terms to describe the embodiments of this application. For example, "one embodiment," "an embodiment," and / or "some embodiments" refer to a certain feature, structure, or characteristic related to at least one embodiment of this application. Therefore, it should be emphasized and noted that "one embodiment," "an embodiment," or "an alternative embodiment" mentioned twice or more in different places in this specification does not necessarily refer to the same embodiment. In addition, certain features, structures, or characteristics in one or more embodiments of this application may be appropriately combined.

[0279] In addition, it will be understood by those skilled in the art that various aspects of the present application can be illustrated and described by a number of patentable categories or situations, including any new and useful combination of processes, machines, products or substances, or any new and useful improvements thereto. Therefore, various aspects of the present application can be implemented entirely by hardware, entirely by software (including firmware, resident software, microcode, etc.), or by a combination of hardware and software. The above hardware or software can all be referred to as "units", "modules" or "systems". In addition, various aspects of the present application can take the form of a computer program product embodied in one or more computer-readable media, wherein computer-readable program code is contained therein.

[0280] The computer program code required for the operation of each part of the application can be written in any one or more programming languages, including object-oriented programming languages ​​such as Java, Scala, Smalltalk, Eiffel, JADE, Emerald, C++, C#, VB.NET, Python, conventional procedural programming languages ​​such as C programming language, VisualBasic, Fortran2103, Perl, COBOL2102, PHP, ABAP, dynamic programming languages ​​such as Python, Ruby and Groovy or other programming languages. The program code can be run completely on the user's computer, or run on the user's computer as an independent software package, or run partly on the user's computer and partly on a remote computer, or run completely on a remote computer or server. In the latter case, the remote computer can be connected to the user's computer through any network form, such as a local area network (LAN) or a wide area network (WAN), or be connected to an external computer (such as by the Internet), or in a cloud computing environment, or used as a service such as software as a service (SaaS).

[0281] In addition, unless expressly stated in the claims, the order of the processing elements and sequences described in this application, the use of alphanumeric characters, or the use of other names are not intended to limit the order of the processes and methods of this application. Although the above disclosure discusses some embodiments of the invention that are currently considered useful through various examples, it should be understood that such details are only for illustrative purposes, and the attached claims are not limited to the disclosed embodiments. On the contrary, the claims are intended to cover all modifications and equivalent combinations that are consistent with the essence and scope of the embodiments of this application. For example, although the implementation of the various components described above can be embodied in a hardware device, it can also be implemented as a pure software solution, for example, installation on an existing server or mobile device.

[0282] Similarly, it should be noted that in order to simplify the presentation of this disclosure and thereby facilitate understanding of one or more of the invention's embodiments, the foregoing descriptions of the embodiments of this disclosure sometimes combine multiple features into a single embodiment, figure, or description thereof. However, this approach should not be interpreted as reflecting an intention that the claimed subject matter requires more features than expressly recited in each claim. Rather, the subject matter of the invention may possess fewer features than the single embodiment described above.

Claims

1. A collaborative detection method based on multi-field coupling dynamic correction and deep learning pressure sensing, characterized in that: include: S1. Data acquisition: The spatiotemporal correlation matrix {P(t), T(t)} is acquired using an array of dual-mode pressure-temperature sensors with adjustable spacing. The initial sensor spacing d0 = 10 / (2κ_max), where κ_max is the maximum curvature of the measured surface. S2. Preprocessing: Perform sym4 wavelet denoising and regional normalization on the pressure data matrix P(t) to generate the preprocessing matrix P'; S3. Physical field correction: Construct a thermal-mechanical-humidity three-field coupled finite element model and calculate the temperature gradient field based on the temperature data matrix T(t) Use the model to correct P' and output the physical field correction matrix P" and the confidence map M; S4. Surface fitting: Based on the confidence map M, valid data points are selected and the genetic-particle swarm optimization algorithm is used to optimize the NURBS surface control point parameter set θ* to generate the initial fitting surface A1±σ; S5. Deep learning optimization: Input A1±σ, P” and M into the deep residual network, and output the optimized surface A2 and the M-weighted defect probability map D_pre; S6. Defect detection: Based on the geometric features of D_pre⊙M, A2 and Adaptive threshold segmentation and cascade SVM classification are performed on the spatial distribution of the defect to generate the defect coordinate set C; S7. Dynamic adjustment: According to the spatial density ρ_C of C, the distribution of low confidence regions of M and The local extreme value of the sensor is dynamically adjusted to d_new, and the thermal conductivity coefficient κ_new and humidity diffusion coefficient D_new of the thermal-mechanical-humidity three-field coupled finite element model are updated; S8. Report generation: Generates a report containing the surface error thermal map ΔS = ‖A2-P”‖ and the temperature gradient field Enhanced detection of distribution reports R+.

2. The collaborative detection method based on multi-field coupling dynamic correction and deep learning pressure sensing according to claim 1 is characterized in that: Also includes: S9. Closed-loop feedback: Feedback θ* to the initial population in step S4, C to the sensor deployment logic in step S1, and the coordinates of the abnormal area M to the model boundary conditions in step S3, forming a closed-loop iterative optimization.

3. The collaborative detection method based on multi-field coupling dynamic correction and deep learning pressure sensing according to claim 1 is characterized in that: The construction of the thermal-mechanical-humidity three-field coupled finite element model described in S3 includes the following control equations: Heat conduction equation: Mechanical equilibrium equation: Humidity diffusion equation: Boundary conditions: Dirichlet conditions P″ are imposed in the region M < 0.4 boundary =P' boundary +2.3×10 -5 ·ΔT.

4. The collaborative detection method based on multi-field coupling dynamic correction and deep learning pressure sensing according to claim 1 is characterized in that: The enhanced detection report R+ described in S8 also includes a defect spatiotemporal evolution map, which is generated by: Extract historical detection cycle data t-1 ,C t-2 ,...,C t-n }; Calculate the Hausdorff distance between adjacent periodic defect sets Visualization method: Use arrows to represent the velocity v of the center of mass c =‖C t -C t-1 The fast diffusion path with ‖ / Δt>0.2mm / s is filled with contour lines to represent the H value distribution, with a color scale range of 0 to 0.5mm.

5. The collaborative detection method based on multi-field coupling dynamic correction and deep learning pressure sensing according to claim 1 is characterized in that: The deployment logic of the sensor array in S1 is: According to the feedback defect coordinate set C, redundant micro-arrays are deployed within 3mm around C with a spacing of 0.2d_new; The timestamp alignment accuracy is ≤10μs, the pressure sampling rate is 1kHz, and the temperature sampling rate is 10Hz.

6. The collaborative detection method based on multi-field coupling dynamic correction and deep learning pressure sensing according to claim 1 is characterized in that: S2 preprocessing includes: The sym4 wavelet basis function is used for 5-layer decomposition and denoising; The regional normalization formula is P′ ij =(P ij -μ local ) / σ local ; where μ local and σ local is the mean and standard deviation of the 5×5 neighborhood centered at point ij.

7. The collaborative detection method based on multi-field coupling dynamic correction and deep learning pressure sensing according to claim 1 is characterized in that: The optimization goal of the genetic-particle swarm hybrid algorithm in S4 is: Where σ is the fitting standard deviation, and the optimization termination condition is that the fitness change is <1e-4 for 10 consecutive generations.

8. The collaborative detection method based on multi-field coupling dynamic correction and deep learning pressure sensing according to claim 1 is characterized in that: The structure of the deep residual network in S5 is: Input layer: 128×128 grid data of three channels A1±σ, P”, and M; Network architecture: Improved ResNet-50, with the last layer replaced by a dual-branch output, namely A2 regression branch and D_pre classification branch; Loss function: The A2 regression branch uses Huber loss, and the D_pre classification branch uses Focal Loss.

9. The collaborative detection method based on multi-field coupling dynamic correction and deep learning pressure sensing according to claim 1, characterized in that: The characteristics of the cascade SVM classification in S6 include: Geometric features: A2 surface curvature κ, normal vector deviation Δn; Physical characteristics: Amplitude, weighted probability value of D_pre⊙M; Classification rules: The first-level SVM screens suspected defect areas, and the second-level SVM distinguishes cracks, holes, and delamination defects.

10. The collaborative detection method based on multi-field coupling dynamic correction and deep learning pressure sensing according to claim 1, characterized in that: The formula for dynamically adjusting the sensor spacing in S7 is: And when When the local extreme value is greater than 10℃ / mm, the spacing is forced to be increased to 0.5d_new.

Images