Pump body heat treatment uniformity detection method based on infrared imaging

Abstract

The invention discloses a pump body heat treatment uniformity detection method based on infrared imaging. The method comprises the following steps that 1, a pump body three-dimensional structure model is obtained, and detection points are calibrated; 2, applying thermal disturbance to each detection area; 3, carrying out dynamic multi-channel temperature image acquisition to obtain a temperature-time sequence; 4, constructing a non-Fourier thermoelastic inertial response model of each detection point, and extracting thermal response characteristic parameters; 5, constructing a thermal response memory map; step 6, inputting the thermal response memory map into an improved TimeSFormer model, introducing a thermal potential field tensor guide mechanism into the model, and obtaining a thermal treatment uniformity score in combination with the spatial position information of the detection point and the structural topological coding vector; and 7, identifying a heat treatment deviation concentration area and outputting a detection report. The intelligent detection and evaluation method for the heat treatment uniformity of the pump body is realized by combining infrared imaging and the improved TimeSFormer model.

Description

Technical Field

[0001] The present invention relates to the technical field of heat treatment quality assessment, and in particular to a method for detecting uniformity of heat treatment of a pump body based on infrared imaging. Background Art

[0002] As the requirements for heat treatment quality consistency for large and complex components continue to increase, how to accurately monitor and identify deviations in the temperature response behavior of complex pump bodies during heat treatment has become a research focus in the field of industrial quality control. Currently, commonly used heat treatment quality assessment methods mainly rely on thermocouple point measurement, single-channel infrared imaging, or post-processing statistical analysis methods. However, the following problems are common in practical applications: Existing thermocouple methods have few sampling points and are unable to cover the entire pump surface. Single-channel infrared imaging systems are susceptible to surface emissivity differences and optical interference, resulting in large errors in temperature measurement results and difficulty reflecting the true thermal diffusion process. Most methods only evaluate steady-state temperature or maximum temperature rise, ignoring the dynamic response characteristics during thermal disturbances, such as response delay, thermal hysteresis, and local thermal elastic echo behavior. At the same time, the lack of spatial analysis methods combined with three-dimensional structural information makes it difficult to identify uneven heat treatment within complex structural areas. In addition, traditional clustering or classification methods lack the ability to model time series and spatial structural information, making it impossible to effectively extract spatiotemporal correlation features and conduct comparative analysis between regions. This limits the ability to accurately locate and interpret areas where heat treatment deviations are concentrated.

[0003] Therefore, how to provide a method for detecting uniformity of heat treatment of a pump body based on infrared imaging is an urgent problem that those skilled in the art need to solve. Summary of the Invention

[0004] One purpose of the present invention is to propose a method for detecting the uniformity of heat treatment of a pump body based on infrared imaging. The present invention combines infrared multi-band thermal image acquisition with an improved TimeSFormer model, constructs a thermal response memory map, and introduces a thermal potential field tensor guidance mechanism. Combined with the topological encoding of the detection point structure, the present invention improves the modeling accuracy of the physical essence of the thermal diffusion process and the spatial resolution capability of the regional heat treatment uniformity assessment. It has the advantages of strong analytical robustness, strong adaptability to complex geometric structures, and intuitive and interpretable detection results.

[0005] A method for detecting uniformity of heat treatment of a pump body based on infrared imaging according to an embodiment of the present invention includes the following steps: Step 1: Obtain the three-dimensional structural model of the pump body, divide the pump body surface into several detection areas according to the structural characteristics, and calibrate the detection points; Step 2: Apply thermal disturbance in each detection area by laser pulse heating; Step 3: Use short-wave, medium-wave, and long-wave infrared thermal imagers to collect dynamic multi-channel temperature images of the pump surface after thermal disturbance, and obtain the temperature-time series corresponding to each detection point; Step 4: Based on the temperature-time series, construct a non-Fourier thermoelastic inertial response model for each detection point and extract thermal response characteristic parameters; Step 5: Construct the thermal response characteristic parameters of each detection point into a temperature-response hysteresis matrix to form a thermal response memory map; Step 6: Input the thermal response memory map into the improved TimeSFormer model. The improved TimeSFormer model introduces a thermal potential field tensor guidance mechanism, combines the spatial position information of the detection point with the structural topology encoding vector, and obtains the heat treatment uniformity score of each detection area; Step 7: Perform a cluster analysis of the inspection area based on the heat treatment uniformity score and the pump body structure information, identify the area where the heat treatment deviation is concentrated, and output a test report.

[0006] Optionally, the step 1 is specifically as follows: Obtaining a three-dimensional structural model of the pump body, wherein the three-dimensional structural model of the pump body is a digital model established by industrial-grade three-dimensional laser scanning equipment and has a geometric accuracy corresponding to the actual pump body structure; The pump body surface is divided into regions according to structural features, including ribs, cavities, flanges, flow channels, and wall thickness mutation boundaries, to generate several detection regions; At least one detection point is calibrated in each detection area. The detection point is set at the geometric center or edge corner of the detection area and corresponds to a unique spatial coordinate identifier in the three-dimensional structural model of the pump body.

[0007] Optionally, the step 2 is specifically as follows: In each detection area, laser pulse heating is performed by a near-infrared laser with a wavelength of 980 nanometers to 1550 nanometers, the pulse width of the near-infrared laser is 100 milliseconds to 500 milliseconds, and the pulse frequency is 5 Hz to 20 Hz; The laser is focused by the collimating lens and acts on the surface of the pump body, forming a thermal disturbance spot with a diameter of 1 mm to 5 mm; The near-infrared laser controls the action position through a two-dimensional displacement platform and a scanning galvanometer, and performs positioning heating on a designated detection point in each detection area.

[0008] Optionally, the step three is specifically as follows: A multi-channel infrared imaging system consisting of short-wave, medium-wave and long-wave infrared thermal imagers is used to synchronously capture temperature images of the pump surface after thermal disturbance. The operating band of the short-wave infrared thermal imager is 0.9 microns to 1.7 microns, the operating band of the medium-wave infrared thermal imager is 3 microns to 5 microns, and the operating band of the long-wave infrared thermal imager is 8 microns to 14 microns. The infrared thermal imager has an image acquisition frame rate of 60 to 200 frames per second, a thermal sensitivity greater than 40 millikelvin, and a spatial resolution of not less than 640×512 pixels; Each infrared channel records thermal images of the same detection area, and uses a multi-scale gradient directional histogram algorithm for image registration. The temperature-time series of each detection point after heating excitation is extracted from the registered thermal image sequence.

[0009] Optionally, the multi-scale gradient direction histogram algorithm specifically includes: Perform grayscale normalization on short-wave, medium-wave and long-wave infrared thermal images to form a unified brightness scale; A multi-scale image set is constructed based on an image pyramid. At each scale, a sliding window of a set size is used to extract image blocks. The gradient directional histogram features are calculated for the image blocks, where the gradient is obtained by the Sobel operator and the directional histogram is discretely encoded using a preset angular interval. A long-wave infrared thermal image is selected as a reference image, and cross-band image block matching is performed in the short-wave infrared image and the medium-wave infrared thermal image using the directional histogram feature. The position of the corresponding image block in the image to be matched is determined by minimizing the L2 norm of the pixel difference between the image blocks; The image to be matched is subjected to pixel-level position correction, and image resampling is completed using B-spline interpolation to obtain a registered image that is consistent with the reference image in spatial position, and a registered thermal image sequence is generated.

[0010] Optionally, the step 4 is specifically as follows: Preprocessing the temperature-time series of the detection points includes reducing high-frequency noise using a sliding mean filter, normalizing the initial temperature reference using a signal baseline drift correction method, and aligning the starting points of each series according to the laser excitation time; During the excitation phase, a two-phase hysteresis model is used to fit the temperature variation process at the initial stage of excitation based on the non-Fourier heat conduction theory. The two-phase hysteresis model introduces heat flow hysteresis time and temperature gradient hysteresis time to describe the relative hysteresis behavior between heat flow and temperature response. During the peak response phase, a thermal stress-driven model caused by temperature gradients is established based on the thermoelastic properties of the material. This thermal stress-driven model couples the instantaneous temperature gradient field with the local thermal expansion coefficient, calculates the stress distribution caused by the non-uniform thermal field through finite element discretization, and fits the main peak time and amplitude in the temperature response. In the decay phase, a recovery model that expresses the response hysteresis is constructed through the time deconvolution kernel function to clearly distinguish the nonlinear process of thermal diffusivity changing with time; Continuously splicing the two-phase hysteresis model, the thermal stress driving model and the recovery model to form a complete non-Fourier thermoelastic inertial response model; Thermal response characteristic parameters are extracted based on the non-Fourier thermoelastic inertial response model, and the thermal response characteristic parameters include: Temperature change slope: the average rate of change of the temperature rise section after the start of the excitation; Response delay time: the time interval from the start of excitation to the maximum slope change of the temperature curve; Thermal hysteresis characteristics: the delay difference between the peak response time and the excitation termination time; Thermal rebound echo morphology: the rebound temperature fluctuation amplitude and symmetry distribution characteristics that appear in the attenuation stage.

[0011] Optionally, the step five is specifically as follows: The thermal response characteristic parameters extracted from each detection point are arranged according to the preset time nodes to construct the corresponding temperature-response hysteresis matrix. The rows of the temperature-response hysteresis matrix represent the spatial positions of different detection points, the columns represent the time evolution dimensions after the thermal disturbance, and the matrix elements are the values ​​of the thermal response characteristic parameters at the corresponding time. The constructed temperature-response hysteresis matrix is ​​two-dimensionally encoded and mapped to generate a thermal response memory map.

[0012] Optionally, the step six is ​​specifically as follows: The thermal response memory map is input into an improved TimeSFormer model, which adopts a Transformer architecture with a time-space separation structure, including a spatial encoding module, a temporal encoding module, and a regional aggregation module; The spatial encoding module introduces a thermal potential field tensor guidance mechanism, which specifically includes: Constructing a three-dimensional thermal potential field tensor, wherein the three components of the three-dimensional thermal potential field tensor are the time derivative of the temperature change rate, the normalized numerical distribution of the local heat capacity of the material, and the thermal hysteresis distance calculated based on the thermal response delay between detection points; The three-dimensional thermal potential field tensor is converted into a guided vector field through a convolutional encoder at the input stage, and is weightedly fused with the linear transformation results of the query and key in the spatial encoding module to construct the guided perception attention weight in the spatial attention mechanism; Based on the guided perceptual attention weight, feature extraction in the spatial dimension is performed on each frame of the thermal response memory map to generate a spatial encoding feature; The temporal coding module is constructed in the form of a multi-layer Transformer encoder stack, which models the spatial coding features in the temporal dimension and introduces a structural topological coding vector of the detection point at each time step. The structural topological coding vector is obtained by mapping the spatial coordinates of the detection point in the three-dimensional structural model of the pump body through an MLP embedding network. The MLP embedding network includes two hidden layers with ReLU activation functions and one output layer, which are respectively used to encode the three-dimensional spatial coordinates into embedded representations of preset dimensions and fuse them with the spatial coding features of the corresponding time step to obtain the temporal coding features. The regional aggregation module aggregates the time coding features of multiple detection points in each detection area and generates a regional level thermal response representation vector by weighted averaging; The region-level thermal response representation vector is input into a fully connected layer, which uses a Sigmoid activation function to output a thermal treatment uniformity score between 0 and 1.

[0013] Optionally, the step seven is specifically as follows: Combine the heat treatment uniformity score of each inspection area with the geometric center coordinates in the three-dimensional structural model of the pump body to construct a feature vector set containing the heat treatment uniformity score and spatial position information; The characteristic vector set is analyzed using a density-based spatial clustering algorithm, with a density threshold condition set such that the difference in heat treatment uniformity score does not exceed 0.05 and the spatial distance does not exceed 10 mm; Dividing the detection areas that meet the density threshold condition into the same cluster; In all clusters, the mean heat treatment uniformity score is calculated, and the clusters whose mean heat treatment uniformity score is lower than the mean heat treatment uniformity score of all the detection areas by more than 30% are identified as the heat treatment deviation concentration areas; Output a test report, which includes the spatial distribution of each cluster, the statistical results of the heat treatment uniformity score, and the location information of the heat treatment deviation concentration area.

[0014] The beneficial effects of the present invention are: The present invention integrates multi-band infrared imaging and non-Fourier thermoelastic inertial response modeling to solve the problems of low temperature response coverage, lack of dynamic features and weak thermal deviation recognition ability during the heat treatment of complex pump structures. It uses near-infrared laser pulses to impose short-term thermal disturbances on multi-region detection points, combines short-wave, medium-wave and long-wave infrared thermal imagers to construct a dynamic multi-channel temperature image sequence, and realizes high spatial resolution and comprehensive acquisition of multi-dimensional temperature response through image registration and temperature time series extraction. In the response modeling link, a non-Fourier thermoelastic inertial response model containing three stages of excitation, peak response and attenuation is constructed, integrating the two-phase Hysteresis heat conduction, temperature gradient driven stress response and thermal diffusion deconvolution mechanism are used to precisely extract thermal response characteristic parameters; in the uniformity assessment link, a thermal response memory map is constructed and input into the improved TimeSFormer model that introduces the thermal potential field tensor guidance mechanism. The three-dimensional thermal potential field tensor and the detection point structure topology encoding are used to jointly guide the spatial attention distribution and time series modeling, significantly improving the physical relevance of thermal diffusion behavior modeling and the accuracy of regional scoring; finally, density space clustering is used to identify spatial deviations of the scoring feature vector set, accurately locate the concentrated areas of heat treatment deviations, and realize intelligent detection output of the heat treatment quality of the pump body. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings: Figure 1 This is an overall flow chart of a method for detecting uniformity of heat treatment of a pump body based on infrared imaging proposed by the present invention; Figure 2 This is a flow chart for constructing a non-Fourier thermoelastic inertial response model of the detection point and extracting thermal response characteristic parameters for a method for detecting uniformity of heat treatment of a pump body based on infrared imaging proposed by the present invention; Figure 3 This is a schematic diagram of the improved TimeSFormer model structure that introduces the thermal potential field tensor guidance mechanism for the infrared imaging-based pump body heat treatment uniformity detection method proposed in the present invention. DETAILED DESCRIPTION

[0016] The present invention will now be described in further detail with reference to the accompanying drawings, which are simplified schematic diagrams that illustrate the basic structure of the present invention in a schematic manner.

[0017] refer to Figure 1-Figure 3 A method for detecting uniformity of heat treatment of a pump body based on infrared imaging comprises the following steps: Step 1: Obtain the three-dimensional structural model of the pump body, divide the pump body surface into several detection areas according to the structural characteristics, and calibrate the detection points; Step 2: Apply thermal disturbance in each detection area by laser pulse heating; Step 3: Use short-wave, medium-wave, and long-wave infrared thermal imagers to collect dynamic multi-channel temperature images of the pump surface after thermal disturbance, and obtain the temperature-time series corresponding to each detection point; Step 4: Based on the temperature-time series, construct a non-Fourier thermoelastic inertial response model for each detection point and extract thermal response characteristic parameters; Step 5: Construct the thermal response characteristic parameters of each detection point into a temperature-response hysteresis matrix to form a thermal response memory map; Step 6: Input the thermal response memory map into the improved TimeSFormer model. The improved TimeSFormer model introduces a thermal potential field tensor guidance mechanism, combines the spatial position information of the detection point with the structural topology encoding vector, and obtains the heat treatment uniformity score of each detection area; Step 7: Perform a cluster analysis of the inspection area based on the heat treatment uniformity score and the pump body structure information, identify the area where the heat treatment deviation is concentrated, and output a test report.

[0018] In this embodiment, the step 1 is specifically as follows: Obtaining a three-dimensional structural model of the pump body, wherein the three-dimensional structural model of the pump body is a digital model established by industrial-grade three-dimensional laser scanning equipment and has a geometric accuracy corresponding to the actual pump body structure; The pump body surface is divided into regions according to structural features, including ribs, cavities, flanges, flow channels, and wall thickness mutation boundaries, to generate several detection regions; At least one detection point is calibrated in each detection area. The detection point is set at the geometric center or edge corner of the detection area and corresponds to a unique spatial coordinate identifier in the three-dimensional structural model of the pump body.

[0019] In this embodiment, the step 2 is specifically as follows: In each detection area, laser pulse heating is performed by a near-infrared laser with a wavelength of 980 nanometers to 1550 nanometers, the pulse width of the near-infrared laser is 100 milliseconds to 500 milliseconds, and the pulse frequency is 5 Hz to 20 Hz; The laser is focused by the collimating lens and acts on the surface of the pump body, forming a thermal disturbance spot with a diameter of 1 mm to 5 mm; The near-infrared laser controls the action position through a two-dimensional displacement platform and a scanning galvanometer, and performs positioning heating on a designated detection point in each detection area.

[0020] In this embodiment, the step three is specifically as follows: A multi-channel infrared imaging system consisting of short-wave, medium-wave and long-wave infrared thermal imagers is used to synchronously capture temperature images of the pump surface after thermal disturbance. The operating band of the short-wave infrared thermal imager is 0.9 microns to 1.7 microns, the operating band of the medium-wave infrared thermal imager is 3 microns to 5 microns, and the operating band of the long-wave infrared thermal imager is 8 microns to 14 microns. The infrared thermal imager has an image acquisition frame rate of 60 to 200 frames per second, a thermal sensitivity greater than 40 millikelvin, and a spatial resolution of not less than 640×512 pixels; Each infrared channel records thermal images of the same detection area, and uses a multi-scale gradient directional histogram algorithm for image registration. The temperature-time series of each detection point after heating excitation is extracted from the registered thermal image sequence.

[0021] In this embodiment, the multi-scale gradient directional histogram algorithm specifically includes: Perform grayscale normalization on short-wave, medium-wave and long-wave infrared thermal images to form a unified brightness scale; A multi-scale image set is constructed based on an image pyramid. At each scale, a sliding window of a set size is used to extract image blocks. The gradient directional histogram features are calculated for the image blocks, where the gradient is obtained by the Sobel operator and the directional histogram is discretely encoded using a preset angular interval. A long-wave infrared thermal image is selected as a reference image, and cross-band image block matching is performed in the short-wave infrared image and the medium-wave infrared thermal image using the directional histogram feature. The position of the corresponding image block in the image to be matched is determined by minimizing the L2 norm of the pixel difference between the image blocks; The matching cost function is defined as: ; in, express and The L2 norm of the pixel difference between the two is used to measure the similarity. The smaller the value, the higher the degree of match. Indicates the width of the image block; Indicates the height of the image block; Represents the image patch of the reference image, with a size of , the coordinates are Grayscale normalized value at ; Indicates the image to be matched with The image block with the upper left corner coordinates, relative coordinates Grayscale normalized value at ; In the sliding window search area, select The smallest upper left corner coordinate is used to determine the corresponding image block position in the image to be matched.

[0022] The image to be matched is subjected to pixel-level position correction, and image resampling is completed using B-spline interpolation to obtain a registered image that is consistent with the reference image in spatial position, and a registered thermal image sequence is generated.

[0023] The introduction of a multi-scale gradient directional histogram algorithm in this invention significantly improves the accuracy and robustness of multi-channel infrared image registration. Due to the inherent differences in imaging mechanisms, spatial mismatch, and response inconsistency between short-wave, medium-wave, and long-wave infrared images, traditional image registration methods, such as those based on grayscale or template matching, struggle to achieve high-precision alignment across different bands. The multi-scale gradient directional histogram algorithm employed in this invention, by extracting local gradient directional distribution features on a multi-scale image pyramid, effectively captures the structural edge response of thermally disturbed regions in images of different bands, enabling stable cross-modal image matching. Furthermore, by combining the L2 norm matching criterion with B-spline interpolation correction, the accumulation of registration errors in nonlinear structural regions caused by affine transformations is avoided. This integration of the algorithm not only improves the registration accuracy of thermal imaging data but also enhances the reliability of subsequent thermal response feature extraction, thereby improving the accuracy of pump heat treatment uniformity assessment. Compared to existing technologies, this invention extends the gradient directional histogram to multi-scale and multi-band applications, realizing the engineering feasibility of cross-band thermal image registration in industrial-grade infrared testing.

[0024] In this embodiment, the step 4 is specifically as follows: Preprocessing the temperature-time series of the detection points includes reducing high-frequency noise using a sliding mean filter, normalizing the initial temperature reference using a signal baseline drift correction method, and aligning the starting points of each series according to the laser excitation time; During the excitation phase, a two-phase hysteresis model is used to fit the temperature variation process at the initial stage of excitation based on the non-Fourier heat conduction theory. The two-phase hysteresis model introduces heat flow hysteresis time and temperature gradient hysteresis time to describe the relative hysteresis behavior between heat flow and temperature response. During the peak response phase, a thermal stress-driven model caused by temperature gradients is established based on the thermoelastic properties of the material. This thermal stress-driven model couples the instantaneous temperature gradient field with the local thermal expansion coefficient, calculates the stress distribution caused by the non-uniform thermal field through finite element discretization, and fits the main peak time and amplitude in the temperature response. In the decay phase, a recovery model that expresses the response hysteresis is constructed through the time deconvolution kernel function to clearly distinguish the nonlinear process of thermal diffusivity changing with time; The temporal deconvolution kernel is a time-domain function that describes the dynamic response characteristics of a thermal system. It reflects the time-delayed pattern of the system's temperature response after a unit thermal stimulus is applied at a given moment. In the heat conduction inversion process, the system's received temperature response can be viewed as the point-by-point temporal superposition of the historical stimulus signal and the kernel function. The goal of deconvolution is to derive the original thermal stimulus signal distribution or identify the system's thermal conduction characteristics by combining the known temperature response curve with the temporal deconvolution kernel.

[0025] In practical applications, the temporal deconvolution kernel function is used to separate the causal relationship between thermal disturbance signals and actual responses, thereby more accurately restoring the true temporal characteristics of the thermal input. This function typically exhibits certain hysteresis and diffusion characteristics, and can represent the non-instantaneous response strength of the system to thermal stimulation at different time points. It is an important component of non-Fourier heat conduction modeling. Continuously splicing the two-phase hysteresis model, the thermal stress driving model and the recovery model to form a complete non-Fourier thermoelastic inertial response model; Thermal response characteristic parameters are extracted based on the non-Fourier thermoelastic inertial response model, and the thermal response characteristic parameters include: Temperature change slope: the average rate of change of the temperature rise section after the start of the excitation; Response delay time: the time interval from the start of excitation to the maximum slope change of the temperature curve; Thermal hysteresis characteristics: the delay difference between the peak response time and the excitation termination time; Thermal rebound echo morphology: the rebound temperature fluctuation amplitude and symmetry distribution characteristics that appear in the attenuation stage.

[0026] In this embodiment, the step five is specifically as follows: The thermal response characteristic parameters extracted from each detection point are arranged according to the preset time nodes to construct the corresponding temperature-response hysteresis matrix. The rows of the temperature-response hysteresis matrix represent the spatial positions of different detection points, the columns represent the time evolution dimensions after the thermal disturbance, and the matrix elements are the values ​​of the thermal response characteristic parameters at the corresponding time. The constructed temperature-response hysteresis matrix is ​​two-dimensionally encoded and mapped to generate a thermal response memory map.

[0027] In this embodiment, the step six is ​​specifically as follows: The thermal response memory map is input into an improved TimeSFormer model, which adopts a Transformer architecture with a time-space separation structure, including a spatial encoding module, a temporal encoding module, and a regional aggregation module; The spatial encoding module introduces a thermal potential field tensor guidance mechanism, which specifically includes: Constructing a three-dimensional thermal potential field tensor, wherein the three components of the three-dimensional thermal potential field tensor are the time derivative of the temperature change rate, the normalized numerical distribution of the local heat capacity of the material, and the thermal hysteresis distance calculated based on the thermal response delay between detection points; During the construction of the thermal potential field tensor, the normalized numerical distribution of the material's local heat capacity is used to characterize the differences in heat storage and transfer capabilities across different structural locations. Specifically, the equivalent heat capacity of each test point in the three-dimensional pump body structural model is calculated based on the heat capacity (the amount of heat stored per unit mass) parameters of the material at each test point, combined with its volume and density information. Due to significant differences in heat capacity distributions due to variations in wall thickness and material type across different test areas, the heat capacity values ​​of all test points are normalized using the Z-score method. This ultimately results in a normalized local heat capacity distribution covering the entire pump body test point area, which serves as one of the one-dimensional channels in the thermal potential tensor.

[0028] On the other hand, the thermal hysteresis distance calculated based on the thermal response delay between detection points is used to reflect the time delay characteristics of thermal disturbance propagation within the structure. The specific steps are as follows: first, the time difference between the inflection point of temperature change under the action of thermal disturbance is calculated for each pair of detection points. Based on this time difference and the spatial distance between the two points in the three-dimensional structural model, the thermal response propagation delay per unit distance, i.e., the thermal hysteresis distance, is calculated by constructing a joint spatiotemporal distribution. As an important component of the thermal potential field tensor, it is embedded in the generation of the spatial attention weight of the model, thereby achieving in-depth modeling of the physical mechanism of structural heat diffusion. The three-dimensional thermal potential field tensor is converted into a guided vector field through a convolutional encoder at the input stage, and is weightedly fused with the linear transformation results of the query and key in the spatial encoding module to construct the guided perception attention weight in the spatial attention mechanism; Based on the guided perceptual attention weight, feature extraction in the spatial dimension is performed on each frame of the thermal response memory map to generate a spatial encoding feature; The temporal coding module is constructed in the form of a multi-layer Transformer encoder stack, which models the spatial coding features in the temporal dimension and introduces a structural topological coding vector of the detection point at each time step. The structural topological coding vector is obtained by mapping the spatial coordinates of the detection point in the three-dimensional structural model of the pump body through an MLP embedding network. The MLP embedding network includes two hidden layers with ReLU activation functions and one output layer, which are respectively used to encode the three-dimensional spatial coordinates into embedded representations of preset dimensions and fuse them with the spatial coding features of the corresponding time step to obtain the temporal coding features. The regional aggregation module aggregates the time coding features of multiple detection points in each detection area and generates a regional level thermal response representation vector by weighted averaging; The region-level thermal response representation vector is input into a fully connected layer, which uses a Sigmoid activation function to output a thermal treatment uniformity score between 0 and 1.

[0029] Compared with the existing standard TimeSFormer architecture, the improved TimeSFormer model introduces a thermal potential field tensor guidance mechanism in the spatial encoding module, realizing the physical perception modeling of the non-uniform thermal diffusion behavior in the pump structure. Specifically, the improvement introduces a three-dimensional thermal potential field tensor, integrates the derivative of the temperature change rate, the local heat capacity distribution of the material, and the thermal lag distance, which are closely related to the thermal diffusion dynamics, into the spatial attention mechanism, and constructs an attention weight generation process with thermal physical prior guidance capabilities. Compared with the traditional Transformer that constructs attention based only on the content similarity of the query and key, the present invention guides the model to focus on the physically sensitive areas on the heat diffusion path through the fusion of physical field information, thereby improving the matching degree and recognition ability of the feature representation to the heat conduction behavior.

[0030] In addition, this improved model also introduces a structural topology coding vector in the time coding module, and uses the relative position relationship of the detection points in the three-dimensional space of the pump body to assist in modeling the temporal correlation of thermal response characteristics, thereby enhancing the generalization ability of the model in long-term thermal response pattern recognition.

[0031] Through the above improvements, the model not only has the ability to extract spatiotemporal features based on data-driven, but also integrates the knowledge priors of pump structure and thermal diffusion physical mechanism, significantly improving the accuracy and interpretability of identifying non-uniformity in complex heat treatment processes. It has high engineering adaptability and model versatility, and is particularly suitable for situations in industrial equipment where local thermal responses are difficult to be accurately modeled by pure data models.

[0032] In this embodiment, the step seven is specifically as follows: Combine the heat treatment uniformity score of each inspection area with the geometric center coordinates in the three-dimensional structural model of the pump body to construct a feature vector set containing the heat treatment uniformity score and spatial position information; The characteristic vector set is analyzed using a density-based spatial clustering algorithm, with a density threshold condition set such that the difference in heat treatment uniformity score does not exceed 0.05 and the spatial distance does not exceed 10 mm; Dividing the detection areas that meet the density threshold condition into the same cluster; In all clusters, the mean heat treatment uniformity score is calculated, and the clusters whose mean heat treatment uniformity score is lower than the mean heat treatment uniformity score of all the detection areas by more than 30% are identified as the heat treatment deviation concentration areas; Output a test report, which includes the spatial distribution of each cluster, the statistical results of the heat treatment uniformity score, and the location information of the heat treatment deviation concentration area.

[0033] Example 1

[0034] To verify the feasibility of this invention, we applied it to a heat treatment quality inspection scenario in a pump casting workshop. This workshop primarily produces high-pressure cast steel pump bodies with complex structures, including multiple cavities, ribs, thick-walled sections, and slender flow channels. During the heat treatment process, these pumps are prone to uneven temperature zones, concentrated thermal stresses, and incomplete annealing.

[0035] First, a high-precision laser 3D scanning system was used to establish a digital structural model of the pump body. The pump body surface was divided into 52 inspection areas according to the geometric boundaries and heat-sensitive areas (cavity corners, rib roots, and flow channel outlets). Several inspection points were calibrated in each area, for a total of 317 calibrated inspection points.

[0036] After the heat treatment, each test area was heated with a laser pulse (wavelength 1064nm, pulse width 300ms, frequency 10Hz). A multi-channel, synchronized imaging system consisting of short-wave, medium-wave, and long-wave infrared thermal imagers captured a complete dynamic thermal response spectrum. The data acquisition frequency was set at 150 frames per second, covering the entire process from the onset of excitation to temperature decay and stabilization, a duration of 15 seconds.

[0037] After preprocessing, the collected temperature-time series were modeled using a non-Fourier thermoelastic-inertial response model. Four key thermal response characteristic parameters were extracted for each test point: temperature rise slope, response delay time, thermal hysteresis difference, and thermoelastic echo morphology. To further analyze the uniformity of thermal treatment across regions, these parameters were combined to construct a temperature-response hysteresis matrix and generate a thermal response memory map.

[0038] The thermal response memory map is fed into the improved TimeSFormer model proposed in this paper. This model incorporates a thermal potential field tensor guidance mechanism, integrating spatial coordinate encoding and structural topology to enhance the accuracy of temporal modeling. During the inference phase, the model outputs a thermal treatment uniformity score for each inspection area, ranging from 0 to 1, with higher values ​​indicating more uniform thermal treatment.

[0039] Table 1 Statistics of heat treatment uniformity scores in some test areas Area Number Number of detection points Average rating Lowest Rating Standard deviation Structural characteristics A03 6 0.89 0.85 0.013 upper rib A14 5 0.47 0.42 0.023 Wall thickness mutation A19 4 0.52 0.44 0.031 Middle cavity corners A25 5 0.68 0.63 0.021 Runner outlet area A31 4 0.39 0.35 0.018 Flange root inner cavity corner The data in Table 1 show significant differences in uniformity during the heat treatment process across regions with different structural features. Among the higher-scoring areas, the A03 test area performed the most consistently, with an average score of 0.89 across six test points, a minimum score of 0.85, and a standard deviation of only 0.013, indicating an overall uniform heat treatment process with minimal fluctuations. This area, corresponding to the upper ribs of the pump body, typically exhibits regular geometry and a clear heat conduction path, making it easier to achieve a stable thermal response.

[0040] In contrast, areas with lower scores, such as A31, have significantly lower heat treatment scores, with an average of 0.39, a minimum score of 0.35, and a standard deviation of 0.018. Although the fluctuations are not the most dramatic, the overall heat treatment effect is clearly insufficient. This area corresponds to the inner cavity corner at the root of the flange, which has a complex structure and a restricted heat flow path. There may be cold spot effects or delayed heat accumulation. The present invention effectively reveals such weak heat treatment areas through multi-source infrared spectrum registration and thermal response modeling, providing a key basis for subsequent process optimization.

[0041] Region A14 also exhibits poor uniformity, with an average score of 0.47 and a standard deviation of 0.023. This corresponds to a sudden change in wall thickness. The heat capacity distribution and thermal diffusion rate in this region are uneven, making it prone to localized overheating or insufficient cooling. The delay and slope features extracted by the non-Fourier thermoelastic-inertial response model precisely identify these response anomalies. Regions A19 (mid-cavity corner) and A25 (channel outlet) scored 0.52 and 0.68, respectively. While not as high as A03, they are improvements over A14 and A31, indicating that the heat treatment quality at the edges and flow intersections is moderate.

[0042] By analyzing the combined indicators of the score mean, extreme value and standard deviation, it is possible to accurately identify local areas with defects in heat treatment, guide the adjustment of heat source parameters and structural optimization, and has good engineering application value.

[0043] To more intuitively identify problematic heat treatment areas, the present invention further introduces a structure-guided clustering algorithm to construct a set of spatial feature vectors of the heat treatment state based on the region's scores and its geometric coordinates in the 3D model. The DBSCAN clustering algorithm was used to analyze all test areas, identifying three spatially close clusters with low scores.

[0044] Table 2 Cluster analysis of heat treatment deviation Cluster number Number of regions Average rating Deviation margin Distribution structure characteristics Recommended adjustment measures C1 5 0.43 35.2% Lower corner of cavity A19~A23 Improve heating uniformity and preheating time C2 4 0.41 37.1% Narrow inner channel A30~A33 Add auxiliary heat source and reduce thermal resistance C3 3 0.44 33.8% Rib thickness transition zone A13~A15 Reduce temperature rise gradient and reduce cooling rate The average scores of the clusters in Table 2 are significantly lower than the overall average score (0.66), indicating that these areas are at risk of uneven heat treatment, high residual thermal stress, or insufficient annealing, and require targeted structural optimization during the heat treatment process.

[0045] Cluster C1 covers five areas, with an average score of 0.43 and a deviation of 35.2%. This cluster is primarily located in the lower corners of the pump cavity (A19-A23). Due to the complex geometry and structural transitions in this area, heat concentration and uneven cooling rates can occur, leading to significant heat treatment non-uniformity. For this area, it is recommended to optimize the heat conduction path and thermal diffusion time by improving heating uniformity and appropriately extending the preheating time to alleviate the low score.

[0046] Cluster C2 involves four test areas, with an average score of 0.41 and the highest deviation of 37.1%, mainly distributed in the narrow channels of the inner cavity (A30-A33). This type of structure has a narrow space and a single and unobstructed heat flow channel, which easily forms local thermal resistance and convection bottlenecks. This makes the heat distribution in this area extremely uneven during heating and cooling, with severe thermal hysteresis and significantly low scores. Therefore, it is recommended to increase the layout of auxiliary heat sources during the heat treatment process, or adjust the heating path and angle, while reducing the cooling rate and thermal resistance configuration to improve the consistency of heat treatment in this area.

[0047] Cluster C3 comprises three regions with an average score of 0.44 and a deviation of 33.8%. These regions are concentrated in the rib-thickness transition region (A13-A15). This type of structure exhibits significant thermal capacity abrupt changes and geometric transitions, leading to sudden hysteresis fluctuations in the thermal response. During heat treatment, these regions are prone to overheating or cold spots due to uneven stress and unstable heat conduction. It is recommended that a stepped heating control strategy with a low temperature rise rate be adopted for these structures, while appropriately reducing cooling efficiency to minimize thermal stress accumulation and improve overall thermal stability.

[0048] This embodiment achieves high-precision modeling and intelligent analysis of the non-steady-state temperature response during the heat treatment of the pump body by introducing multi-channel infrared imaging, multi-physics field modeling and improving the TimeSFormer model deep coding architecture. In particular, in typical areas with complex spatial structures and non-uniform heat diffusion, by constructing a thermal response memory map and combining it with the thermal potential field tensor guidance mechanism, the model's ability to identify local heat treatment deviations is effectively improved. At the same time, by utilizing structural topology coding and regional aggregation mechanisms, efficient feature fusion between the detection point level and the regional level is achieved, enhancing the model's robust representation of heat treatment inconsistency patterns. Furthermore, by using a density clustering algorithm to identify concentrated areas of deviations with low scores and close spatial proximity, it provides a basis for accurately proposing process optimization suggestions. The present invention not only improves the automation and intelligence level of detection, but also provides a generalizable technical path for the evaluation of heat treatment uniformity of complex structural parts, and has good engineering practical value and industrialization prospects.

[0049] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A method for detecting uniformity of heat treatment of a pump body based on infrared imaging, characterized in that: The steps include: Step 1: Obtain the three-dimensional structural model of the pump body, divide the pump body surface into several detection areas according to the structural characteristics, and calibrate the detection points; Step 2: Apply thermal disturbance in each detection area by laser pulse heating; Step 3: Use short-wave, medium-wave, and long-wave infrared thermal imagers to collect dynamic multi-channel temperature images of the pump surface after thermal disturbance, and obtain the temperature-time series corresponding to each detection point; Step 4: Based on the temperature-time series, construct a non-Fourier thermoelastic inertial response model for each detection point and extract thermal response characteristic parameters; Step 5: Construct the thermal response characteristic parameters of each detection point into a temperature-response hysteresis matrix to form a thermal response memory map; Step 6: Input the thermal response memory map into the improved TimeSFormer model. The improved TimeSFormer model introduces a thermal potential field tensor guidance mechanism, combines the spatial position information of the detection point with the structural topology encoding vector, and obtains the heat treatment uniformity score of each detection area; Step 7: Perform a cluster analysis of the inspection area based on the heat treatment uniformity score and the pump body structure information, identify the area where the heat treatment deviation is concentrated, and output a test report.

2. The method for detecting uniformity of heat treatment of a pump body based on infrared imaging according to claim 1, characterized in that: The step 1 is specifically as follows: Obtaining a three-dimensional structural model of the pump body, wherein the three-dimensional structural model of the pump body is a digital model established by industrial-grade three-dimensional laser scanning equipment and has a geometric accuracy corresponding to the actual pump body structure; The pump body surface is divided into regions according to structural features, including ribs, cavities, flanges, flow channels, and wall thickness mutation boundaries, to generate several detection regions; At least one detection point is calibrated in each detection area. The detection point is set at the geometric center or edge corner of the detection area and corresponds to a unique spatial coordinate identifier in the three-dimensional structural model of the pump body.

3. The method for detecting uniformity of heat treatment of a pump body based on infrared imaging according to claim 1, characterized in that: The step 2 is specifically as follows: In each detection area, laser pulse heating is performed by a near-infrared laser with a wavelength of 980 nanometers to 1550 nanometers, the pulse width of the near-infrared laser is 100 milliseconds to 500 milliseconds, and the pulse frequency is 5 Hz to 20 Hz; The laser is focused by the collimating lens and acts on the surface of the pump body, forming a thermal disturbance spot with a diameter of 1 mm to 5 mm; The near-infrared laser controls the action position through a two-dimensional displacement platform and a scanning galvanometer, and performs positioning heating on a designated detection point in each detection area.

4. The method for detecting uniformity of heat treatment of a pump body based on infrared imaging according to claim 1, characterized in that: The step three is specifically as follows: A multi-channel infrared imaging system consisting of short-wave, medium-wave and long-wave infrared thermal imagers is used to synchronously capture temperature images of the pump surface after thermal disturbance. The operating band of the short-wave infrared thermal imager is 0.9 microns to 1.7 microns, the operating band of the medium-wave infrared thermal imager is 3 microns to 5 microns, and the operating band of the long-wave infrared thermal imager is 8 microns to 14 microns. The infrared thermal imager has an image acquisition frame rate of 60 to 200 frames per second, a thermal sensitivity greater than 40 millikelvin, and a spatial resolution of not less than 640×512 pixels; Each infrared channel records thermal images of the same detection area, and uses a multi-scale gradient directional histogram algorithm for image registration. The temperature-time series of each detection point after heating excitation is extracted from the registered thermal image sequence.

5. The method for detecting uniformity of heat treatment of a pump body based on infrared imaging according to claim 4, characterized in that: The multi-scale gradient direction histogram algorithm specifically includes: Perform grayscale normalization on short-wave, medium-wave and long-wave infrared thermal images to form a unified brightness scale; A multi-scale image set is constructed based on an image pyramid. At each scale, a sliding window of a set size is used to extract image blocks. The gradient directional histogram features are calculated for the image blocks, where the gradient is obtained by the Sobel operator and the directional histogram is discretely encoded using a preset angular interval. A long-wave infrared thermal image is selected as a reference image, and cross-band image block matching is performed in the short-wave infrared image and the medium-wave infrared thermal image using the directional histogram feature. The position of the corresponding image block in the image to be matched is determined by minimizing the L2 norm of the pixel difference between the image blocks; The image to be matched is subjected to pixel-level position correction, and image resampling is completed using B-spline interpolation to obtain a registered image that is consistent with the reference image in spatial position, and a registered thermal image sequence is generated.

6. The method for detecting uniformity of heat treatment of a pump body based on infrared imaging according to claim 1, characterized in that: The step 4 is specifically as follows: Preprocessing the temperature-time series of the detection points includes reducing high-frequency noise using a sliding mean filter, normalizing the initial temperature reference using a signal baseline drift correction method, and aligning the starting points of each series according to the laser excitation time; During the excitation phase, a two-phase hysteresis model is used to fit the temperature variation process at the initial stage of excitation based on the non-Fourier heat conduction theory. The two-phase hysteresis model introduces heat flow hysteresis time and temperature gradient hysteresis time to describe the relative hysteresis behavior between heat flow and temperature response. During the peak response phase, a thermal stress-driven model caused by temperature gradients is established based on the thermoelastic properties of the material. This thermal stress-driven model couples the instantaneous temperature gradient field with the local thermal expansion coefficient, calculates the stress distribution caused by the non-uniform thermal field through finite element discretization, and fits the main peak time and amplitude in the temperature response. In the decay phase, a recovery model that expresses the response hysteresis is constructed through the time deconvolution kernel function to clearly distinguish the nonlinear process of thermal diffusivity changing with time; Continuously splicing the two-phase hysteresis model, the thermal stress driving model and the recovery model to form a complete non-Fourier thermoelastic inertial response model; Thermal response characteristic parameters are extracted based on the non-Fourier thermoelastic inertial response model, and the thermal response characteristic parameters include: Temperature change slope: the average rate of change of the temperature rise section after the start of the excitation; Response delay time: the time interval from the start of excitation to the maximum slope change of the temperature curve; Thermal hysteresis characteristics: the delay difference between the peak response time and the excitation termination time; Thermal rebound echo morphology: the rebound temperature fluctuation amplitude and symmetry distribution characteristics that appear in the attenuation stage.

7. The method for detecting uniformity of heat treatment of a pump body based on infrared imaging according to claim 1, characterized in that: The step five is specifically as follows: The thermal response characteristic parameters extracted from each detection point are arranged according to the preset time nodes to construct the corresponding temperature-response hysteresis matrix. The rows of the temperature-response hysteresis matrix represent the spatial positions of different detection points, the columns represent the time evolution dimensions after the thermal disturbance, and the matrix elements are the values ​​of the thermal response characteristic parameters at the corresponding time. The constructed temperature-response hysteresis matrix is ​​two-dimensionally encoded and mapped to generate a thermal response memory map.

8. The method for detecting uniformity of heat treatment of a pump body based on infrared imaging according to claim 1, characterized in that: The step six is ​​specifically as follows: The thermal response memory map is input into an improved TimeSFormer model, which adopts a Transformer architecture with a time-space separation structure, including a spatial encoding module, a temporal encoding module, and a regional aggregation module; The spatial encoding module introduces a thermal potential field tensor guidance mechanism, which specifically includes: Constructing a three-dimensional thermal potential field tensor, wherein the three components of the three-dimensional thermal potential field tensor are the time derivative of the temperature change rate, the normalized numerical distribution of the local heat capacity of the material, and the thermal hysteresis distance calculated based on the thermal response delay between detection points; The three-dimensional thermal potential field tensor is converted into a guided vector field through a convolutional encoder at the input stage, and is weightedly fused with the linear transformation results of the query and key in the spatial encoding module to construct the guided perception attention weight in the spatial attention mechanism; Based on the guided perceptual attention weight, feature extraction in the spatial dimension is performed on each frame of the thermal response memory map to generate a spatial encoding feature; The temporal coding module is constructed in the form of a multi-layer Transformer encoder stack, which models the spatial coding features in the temporal dimension and introduces a structural topological coding vector of the detection point at each time step. The structural topological coding vector is obtained by mapping the spatial coordinates of the detection point in the three-dimensional structural model of the pump body through an MLP embedding network. The MLP embedding network includes two hidden layers with ReLU activation functions and one output layer, which are respectively used to encode the three-dimensional spatial coordinates into embedded representations of preset dimensions and fuse them with the spatial coding features of the corresponding time step to obtain the temporal coding features. The regional aggregation module aggregates the time coding features of multiple detection points in each detection area and generates a regional level thermal response representation vector by weighted averaging; The region-level thermal response representation vector is input into a fully connected layer, which uses a Sigmoid activation function to output a thermal treatment uniformity score between 0 and 1.

9. The method for detecting uniformity of heat treatment of a pump body based on infrared imaging according to claim 1, characterized in that: The step seven is specifically as follows: Combine the heat treatment uniformity score of each inspection area with the geometric center coordinates in the three-dimensional structural model of the pump body to construct a feature vector set containing the heat treatment uniformity score and spatial position information; The characteristic vector set is analyzed using a density-based spatial clustering algorithm, with a density threshold condition set such that the difference in heat treatment uniformity score does not exceed 0.05 and the spatial distance does not exceed 10 mm; Dividing the detection areas that meet the density threshold condition into the same cluster; In all clusters, the mean heat treatment uniformity score is calculated, and the clusters whose mean heat treatment uniformity score is lower than the mean heat treatment uniformity score of all the detection areas by more than 30% are identified as the heat treatment deviation concentration areas; Output a test report, which includes the spatial distribution of each cluster, the statistical results of the heat treatment uniformity score, and the location information of the heat treatment deviation concentration area.

Images