Method for detecting defects of a device based on infrared recognition

Abstract

The application relates to the field of intelligent monitoring, and discloses a device defect detection method based on infrared identification, which comprises the steps of infrared thermal imaging data acquisition, noise reduction processing, nonlinear heat conduction modeling, high-order partial differential optimization and variational regularization inversion solving. An infrared thermal imager collects device surface temperature distribution data, after multi-scale wavelet noise reduction and boundary heat flow balance optimization, a nonlinear heat conduction model is established by using temperature-related thermal conductivity coefficients, the heat diffusion boundary is optimized in combination with a high-order Laplace operator, and the defect heat source distribution is inverted through a variational regularization method, so that the geometric characteristics, thermal parameters and position coordinates of the defects are accurately extracted. The application can adapt to complex working conditions, realize high-precision detection and characteristic analysis of device defects, overcome the problems of noise interference, insufficient boundary processing and weak defect inversion capability in the prior art, and is widely applicable to fault diagnosis and health monitoring of industrial equipment.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of intelligent monitoring, in particular to a device defect detection method based on infrared recognition. BACKGROUND

[0002] With the development of the complexity and diversity of industrial equipment, defect detection and fault monitoring during equipment operation have become one of the core technologies in the field of industrial operation and maintenance. The common device defect detection technologies at present mainly include visual detection, vibration analysis, acoustic detection and detection methods based on infrared thermal imaging. These technologies have certain effects in specific scenarios, but generally have limitations in applicability and detection accuracy under complex working conditions.

[0003] In recent years, with the development of infrared thermal imaging technology, its application in device defect detection has gradually attracted attention. Infrared thermal imaging technology can realize a certain degree of defect positioning by collecting the temperature distribution of the surface of the equipment and analyzing the abnormal area. However, the existing detection methods based on infrared thermal imaging also have many shortcomings, including the following points: first, the infrared thermal imaging data is usually disturbed by high-frequency noise, and the existing technology mostly uses simple filtering algorithm, which is difficult to realize high-precision signal-to-noise ratio improvement. Secondly, the existing detection methods generally model the temperature distribution inside the equipment based on the linear heat conduction model, without fully considering the temperature dependence and dynamic thermal diffusion characteristics of the material, resulting in insufficient description ability of the thermal diffusion behavior under complex working conditions. In addition, the boundary region data is affected by the unbalanced heat flow phenomenon due to the complex geometry of the equipment surface, and the existing method lacks effective boundary optimization technology, thereby affecting the accuracy of the overall model. Finally, in the defect area inversion aspect, the existing technology usually uses simple threshold method for region segmentation or rough fitting, which is difficult to accurately extract the geometric and thermal characteristics of the defect, especially in the multi-defect scene or complex distribution situation. SUMMARY

[0004] In view of the shortcomings of the prior art, the present application provides a device defect detection method based on infrared recognition, which solves the problems that the infrared thermal imaging data is difficult to realize high-precision signal-to-noise ratio improvement, the description ability of the thermal diffusion behavior under complex working conditions is insufficient, and the accuracy of the overall model is affected.

[0005] To achieve the above purpose, the present application realizes the following technical scheme: a device defect detection method based on infrared recognition, comprising the following steps:

[0006] Collecting infrared thermal imaging data of the surface of the equipment under running state, obtaining infrared thermal distribution sequence changing with time;

[0007] Performing noise reduction processing on the infrared thermal imaging data to eliminate high-frequency noise and retain effective signals, and using heat flow balance condition optimization on the boundary region;

[0008] A nonlinear heat conduction model is constructed to describe the dynamic changes in temperature distribution inside the device.

[0009] Based on the high-order partial differential extended model, the heat diffusion characteristics of the defect area are optimized and analyzed.

[0010] The heat conduction model is optimized and solved by using the variational regularization method, and the heat source distribution of the defect area is inverted.

[0011] According to the heat source distribution of the defect area, the defect position inside the device is located, and the geometric shape and thermal characteristics of the defect are extracted.

[0012] Preferably, the infrared thermal imaging data acquisition step comprises:

[0013] A high-resolution infrared thermal imager is used to obtain the temperature distribution of the device surface.

[0014] In the time dimension, multiple frames of thermal imaging data are collected at a fixed sampling frequency to form a thermal distribution time sequence.

[0015] The operating condition parameters of the device are recorded, including the ambient temperature, load state and running time.

[0016] Preferably, the noise reduction processing step comprises:

[0017] The collected thermal imaging data is decomposed into high-frequency and low-frequency components by two-dimensional wavelet decomposition.

[0018] The high-frequency component is filtered by a soft threshold filtering method to filter random noise signals.

[0019] The low-frequency component retains the effective signal, and the processed thermal imaging data is reconstructed.

[0020] Preferably, the boundary region uses a heat flow balance condition optimization comprising:

[0021] For the complex geometric boundary of the device surface, the boundary condition is set as the heat flow balance condition in the normal direction of the device surface.

[0022] The Neumann boundary condition based on the shape of the device surface is used to ensure that the heat flow of the device surface does not flow out of the boundary.

[0023] Preferably, the nonlinear heat conduction model is:

[0024] According to the thermal conductivity characteristics of the device material, a nonlinear heat conduction model is established to describe the dynamic behavior of temperature distribution changes over time.

[0025] In the model, the thermal conductivity coefficient is related to the temperature, and the heat diffusion behavior is adjusted by the heat source characteristics of the device defect area.

[0026] Preferably, the higher-order partial differential extended model includes:

[0027] A fourth-order partial differential term is introduced into the heat conduction model, and the thermal diffusion characteristics of the defect region boundary are described by the higher-order form of the Laplace operator.

[0028] By introducing higher-order terms, the clarity of the thermal boundary in the defect region is enhanced, and noise interference in the boundary region is reduced.

[0029] Preferably, the variational regularization method includes:

[0030] Define an objective function based on the residuals of the heat conduction equation, and balance the fitting accuracy and data smoothness of the heat distribution through a regularization term;

[0031] The regularization term includes a second-order partial differential regularization term and a fourth-order partial differential regularization term;

[0032] The objective function is solved by numerical optimization methods, and the temperature distribution on the surface and inside of the equipment is optimized.

[0033] Preferably, the step of retrieving the heat source distribution in the inversion defect region includes:

[0034] Calculate the heat source intensity distribution inside the equipment using the optimized temperature distribution;

[0035] Based on the spatial characteristics of the heat source distribution, the location of the defect area is identified, and the geometric characteristics and thermal intensity of the defect are extracted.

[0036] Preferably, the defect detection results obtained from the inversion of the heat source distribution in the defect area are used to generate a three-dimensional thermal distribution map of the defect area, and further used to provide a report on the defect location, geometry and thermal characteristics for reference in equipment operation and maintenance decisions.

[0037] Infrared recognition-based equipment defect detection systems include:

[0038] Infrared thermal imagers are used to collect temperature distribution data on the surface of equipment.

[0039] The data processing module is used to perform noise reduction and boundary optimization on the acquired infrared thermal imaging data;

[0040] The model building module is used to establish a temperature distribution model based on the nonlinear heat conduction equation and the higher-order partial differential extended model.

[0041] The optimization and inversion module is used to optimize the temperature distribution using variational regularization methods and invert the defect heat source.

[0042] The output module is used to generate defect detection results, including the geometric characteristics of the defect, thermal intensity distribution, and location coordinates.

[0043] The application provides an equipment defect detection method based on infrared identification.

[0044] 1、The equipment defect detection method based on infrared identification, through the technical schemes of infrared thermal imaging data acquisition, data noise reduction processing, nonlinear heat conduction model construction, high-order partial differential expansion model optimization and variational regularization method optimization solving, achieves the technical effects of accurately positioning the equipment defects and extracting the geometric characteristics. Compared with the methods in the prior art which rely on visual detection, vibration analysis or acoustic detection, the problems of low detection accuracy and poor adaptability in complex working conditions (such as high temperature, dynamic operation or noise interference environment) are solved.

[0045] 2、The combination of the nonlinear heat conduction model and the high-order partial differential expansion model, the use of temperature-dependent thermal conductivity and the fourth-order Laplacian operator to enhance the description of heat diffusion characteristics, achieves the technical effects of accurately simulating the dynamic changes of the internal temperature distribution of the equipment and clearly defining the boundaries of the defect area. Compared with the detection scheme based on the linear heat conduction model in the prior art, the problems of being unable to depict complex heat diffusion behavior and fuzzy defect boundaries are solved.

[0046] 3、The variational regularization method is used to solve the optimized heat conduction model, and the heat source distribution of the defect area is inverted, the geometric characteristics and thermal characteristics of the defect are extracted, and the technical effects of high-precision inversion of the defect heat source and suppression of noise interference are achieved. Compared with the simple direct processing scheme of infrared data in the prior art, the problem of unstable defect inversion caused by noise and ill-conditioned model is solved.

[0047] 4、The application combines data processing, model construction, optimization solving and result output organically through modular system design, generates complete detection results containing defect position, geometric characteristics and heat intensity distribution, and achieves the technical effects of efficient, comprehensive and automatic defect detection. Compared with the decentralized detection process in the prior art, the problems of low work efficiency, poor detection consistency and non-intuitive results are solved. BRIEF DESCRIPTION OF DRAWINGS

[0048] Figure 1 The steps of the equipment defect detection method based on infrared identification of the application are shown in the figure.

[0049] Figure 2 The framework of the equipment defect detection system based on infrared identification of the application is shown in the figure. DETAILED DESCRIPTION

[0050] The technical solutions in the embodiments of the present application will be clearly and completely described with reference to the drawings in the specification of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0051] Please refer to the drawings in the specification of the present application Figure 1 The embodiment of the present application provides an equipment defect detection method based on infrared identification, comprising the following steps:

[0052] S1, infrared thermal imaging data of the equipment surface in a running state is collected, and an infrared thermal distribution sequence changing with time is obtained.

[0053] Specifically, step S1 continuously monitors the infrared thermal distribution in the dynamic running process of the equipment, forms a thermal distribution sequence changing with time, and provides basic data support for subsequent modeling analysis and defect positioning. This step is closely related to subsequent noise reduction processing, nonlinear heat conduction model construction and optimization calculation, and is the basic link of the whole method, directly determining the accuracy and reliability of subsequent detection.

[0054] In the embodiment, for the collection of equipment surface thermal distribution data, a high-resolution infrared thermal imager is used for monitoring, which specifically includes the following contents:

[0055] Under the running state of the equipment, the high-resolution infrared thermal imager records the temperature distribution of different spatial positions of the equipment surface in real time. The resolution of the infrared thermal imager is set to 640x480 pixels, and the frame rate is set to 30fps, which ensures that the rapidly changing thermal distribution during the running of the equipment can be captured.

[0056] The running state of the equipment includes various working conditions, such as running under different loads, external environment temperature changes, dynamic processes such as equipment acceleration and deceleration, etc. When collecting, the infrared thermal imaging data needs to be recorded for these working conditions respectively, so as to analyze the thermal distribution characteristics of the equipment under different working conditions subsequently.

[0057] After the data collected by the infrared thermal imager is stored, time sequence data is formed, where: represents the thermal distribution of the spatial position of the equipment surface at the frame, is the number of sampling frames. Through the continuous time sequence, the dynamic change of the temperature of the equipment surface with time can be reflected.

[0058] Each frame of the infrared thermal imaging time sequence data represents the thermal distribution of the spatial position of the equipment surface at time The temperature distribution at time t can be represented as a function:

[0059] wherein: The sequence of thermal distributions is inputted as initial boundary conditions in subsequent modeling.

[0060] Generally, in order to reduce measurement errors, the environmental parameters of the infrared thermal imager need to be corrected, including background temperature, device surface emissivity, and thermal imager calibration coefficient. The calibrated infrared thermal imager can output accurate surface temperature data, providing a reliable data basis for subsequent processing.

[0061] For the calibration of the infrared thermal imager, the emissivity correction formula is:

[0062] wherein: is the corrected true temperature, is the measured radiation temperature, is the emissivity of the device surface, is the environmental background temperature.

[0063] Through the above formula correction and multi-angle acquisition strategy, the accuracy and comprehensiveness of the data can be effectively improved, laying a foundation for subsequent processing.

[0064] As an option, the present application can also combine the environmental information of the device operation, such as external environmental temperature, humidity, and air flow speed, to compensate for the infrared thermal imaging data. Specifically, in the case of significant fluctuations in external environmental temperature, the temperature data of multiple reference regions can be collected as a compensation benchmark for environmental parameters, further improving the accuracy of thermal distribution data.

[0065] In some embodiments, in order to capture the possible slight temperature changes of specific operating parts of the device (such as gearboxes, bearings, etc.), the present application adopts a local high-resolution sampling strategy. In this case, the sampling area of the infrared thermal imager can be limited to specific key parts, further improving the spatial resolution of the data.

[0066] In order to ensure the time continuity of the thermal distribution sequence, the present application also sets the time sampling step. Specifically, the time step should be determined according to the dynamic characteristics of the device operation, for example, for rapidly changing device components, the time step can be set to 0.03 seconds to avoid missing critical temperature changes.

[0067] In one possible implementation, the present invention collects equipment operating parameters (such as rotational speed, vibration signals, etc.) simultaneously with the heat distribution data on the surface of the equipment, and synchronizes these parameters with the heat distribution sequence in time, providing support for multi-parameter coupling analysis in subsequent defect modeling.

[0068] During the data acquisition process described above, the influence of the device surface geometry on infrared imaging can also be considered. Generally, for curved or complex geometrically structured device surfaces, a multi-angle acquisition strategy can be used to stitch together thermal distribution data from multiple perspectives into a unified surface thermal distribution, thereby achieving more comprehensive detection.

[0069] Through the above steps, the present invention can obtain the heat distribution sequence of the equipment surface under different operating conditions, providing a reliable data foundation for subsequent noise reduction processing, modeling optimization and defect location analysis.

[0070] S2. Denoise the infrared thermal imaging data to eliminate high-frequency noise, retain the effective signal, and optimize the boundary area using heat flow balance conditions.

[0071] Specifically, in step S2, to improve the reliability and accuracy of thermal imaging data during equipment defect detection, this invention further performs noise reduction processing on the acquired infrared thermal imaging data to eliminate interference from high-frequency noise signals while retaining effective low-frequency signals to accurately reflect the true thermal distribution characteristics of the equipment during operation. Furthermore, since equipment surfaces typically have complex geometric boundary shapes, which may lead to distortion or imbalance in temperature data in the boundary areas, this invention, based on noise reduction processing, introduces heat flow balance conditions to optimize the boundary areas, ensuring spatial smoothness and physical consistency of the data. This step is directly related to subsequent modeling steps, and its results directly affect the accuracy of the heat conduction model and the reliability of subsequent defect area inversion.

[0072] In this embodiment, a multi-scale wavelet transform method is used to denoise the acquired thermal imaging sequence data for infrared thermal imaging data. Generally, infrared thermal imaging data... The data includes both effective temperature signals generated during equipment operation and random noise signals. High-frequency components primarily manifest as noise interference, while low-frequency components reflect the temperature distribution characteristics of the equipment. Therefore, noise reduction can be achieved by effectively suppressing high-frequency components and preserving low-frequency components.

[0073] Specifically, firstly, each frame of thermal imaging data... Perform two-dimensional wavelet decomposition using wavelet basis functions. Perform multi-scale decomposition on the data. The decomposition result can be expressed as:

[0074] wherein: represents the scale of decomposition, represents the direction, is the wavelet coefficient after decomposition, and are the total number of decomposition scales and directions, respectively. High-frequency noise is mainly concentrated in the coefficients of smaller scales , while low-frequency information is retained in the coefficients of larger scales.

[0075] After the decomposition is completed, the wavelet coefficients of the high-frequency components are processed using a soft threshold filtering method to suppress high-frequency noise. The filtered wavelet coefficients are calculated by the following formula:

[0076] wherein: is a threshold parameter set to control the strength of the filter. If is less than or equal to , the point is considered as noise and set to zero; if it exceeds the threshold, the value will be adjusted according to the soft threshold rule. The choice of threshold can be adjusted according to the operating conditions of the device and the noise level. In one possible implementation, the threshold can be selected by cross-validation method to balance the noise suppression effect and the preservation of effective signals.

[0077] After filtering, the coefficients of the low-frequency components are unchanged, and the data is subjected to two-dimensional inverse wavelet transform combined with the filtered wavelet coefficients to reconstruct the thermal imaging data after noise reduction. The reconstruction formula is:

[0078] As an option, to further improve the noise reduction effect, three-dimensional wavelet transform can be performed on the time series data . By including the time dimension in the wavelet decomposition, it is possible to better suppress time-dependent noise, especially strong random temperature fluctuations.

[0079] After completing the data noise reduction, the present application performs heat flow balance optimization processing on the boundary region of the device surface. Generally, due to the complex geometry of the device surface, the infrared thermal imaging data may exhibit gradient abnormalities or non-smoothness in the boundary region. Therefore, the present application modifies the data by introducing Neumann boundary conditions:

[0080] wherein: represents the boundary of the device surface, This represents the normal vector of the boundary. Introducing boundary conditions ensures heat flow balance in the boundary region, preventing anomalous data that does not conform to physical laws from affecting subsequent modeling and analysis.

[0081] In one possible implementation, spatial interpolation can be performed on the denoised thermal imaging data. For areas with missing data or local anomalies, data can be supplemented using interpolation methods (such as bicubic interpolation) to further improve data continuity.

[0082] Through the above steps, this invention can obtain high-quality, denoised thermal imaging data, providing a reliable data input foundation for subsequent nonlinear heat conduction model construction and defect region inversion. The denoising process not only effectively eliminates high-frequency noise interference but also ensures the physical consistency of the data through boundary optimization, thereby significantly improving the accuracy and robustness of subsequent detection steps.

[0083] S3. Construct a nonlinear heat conduction model to describe the dynamic changes in temperature distribution inside the equipment.

[0084] Specifically, in step S3, to accurately describe the dynamic changes in temperature distribution inside the equipment over time and space, and to further provide basic data support for defect region inversion, this invention constructs a physical modeling method based on a nonlinear heat conduction equation after completing data denoising and boundary optimization. This model, by describing the heat diffusion behavior inside the equipment and combining temperature-related thermal conductivity characteristics and heat source distribution patterns, can achieve accurate simulation of the heat flow transfer process inside the equipment. The model construction directly relies on the aforementioned denoised infrared thermal imaging data, and also provides theoretical support for subsequent high-order model expansion and defect inversion.

[0085] In this embodiment, the nonlinear heat conduction model is constructed based on Fourier heat conduction theory, and is specifically expressed in the following physical and mathematical ways.

[0086] In general, heat conduction behavior can be described by the following partial differential equation:

[0087] in: Indicates the internal time of the device and spatial location Temperature distribution; Thermal conductivity represents the material's ability to conduct heat. This is the heat source distribution function, representing the heat generated or absorbed inside the equipment; This represents the heat diffusion term, used to characterize the spatial distribution characteristics of heat; This represents the rate of temperature change over time.

[0088] As an improvement, in order to be closer to the complex heat conduction characteristics in the actual equipment operation, the present application adopts a temperature-dependent nonlinear thermal conductivity coefficient model:

[0089] wherein: is the thermal conductivity coefficient of the material at the standard temperature; is a nonlinear correction parameter, used to reflect the dependence of the thermal conductivity coefficient on the temperature.

[0090] By replacing with the dynamic thermal conductivity coefficient, the nonlinear heat conduction equation can be rewritten as:

[0091] This equation can more accurately describe the change of the material's thermal conductivity performance at different temperature levels, and is particularly suitable for the thermal diffusion behavior of high-temperature equipment or dynamic working conditions.

[0092] Specifically, in this embodiment, the infrared thermal imaging data after noise reduction and the boundary optimization condition, the initial boundary condition and the initial temperature field of the model are set as follows:

[0093] The initial temperature field is set according to the initial data in the infrared thermal distribution sequence after noise reduction;

[0094] The boundary condition adopts the Neumann boundary condition: wherein: denotes the boundary of the equipment surface, denotes the normal vector of the boundary.

[0095] In a possible implementation, in order to improve the stability of modeling and the accuracy of numerical solution, the present application performs discretization processing on the above nonlinear heat conduction equation. Generally, the time dimension is discretized by using the implicit difference method, and the space dimension is discretized by using the finite element method. The discretized equation is represented as:

[0096] wherein: and respectively denote the temperature distribution vectors at the current time step and the next time step; is the time step; denotes the nonlinear thermal conductivity coefficient matrix, which is dynamically related to the temperature distribution ; denotes the heat source distribution vector.

[0097] As an alternative, the present application employs the Newton-Raphson method to iteratively solve the discretized nonlinear equations. The Newton-Raphson method approximates the nonlinear equations by linearization, converting them into multiple linear equation systems for solving, thereby improving the numerical convergence speed.

[0098] In some embodiments, to further adapt to the modeling needs under complex working conditions, the present application also dynamically adjusts the heat source distribution function . Specifically, the dynamic change law of the heat source term is set according to the equipment operating parameters (such as speed, vibration amplitude), thereby enhancing the adaptability of the model to dynamic operating environment.

[0099] Through the construction of the above nonlinear heat conduction model, the present application can realize the accurate description of the temperature distribution inside the equipment, providing a physical basis for the high-order model expansion and defect area inversion in the subsequent steps. The model is particularly suitable for equipment detection scenarios with complex heat conduction characteristics, such as defect detection in high-temperature or dynamic load environments. The accuracy of the model is determined by the temperature-dependent thermal conductivity coefficient and the numerical discretization method, and the physical consistency and robustness of the results are ensured through reasonable setting of boundary conditions and initial temperature field.

[0100] In a possible extended application, the model can also be combined with multi-modal data (such as vibration signals or acoustic signals) for joint modeling, further improving the comprehensive accuracy of defect detection. Through these improvements, the adaptability and detection capability of the model have been significantly improved, providing strong support for equipment maintenance and operation monitoring.

[0101] S4, optimizing the thermal diffusion characteristics of the defect area based on the high-order partial differential expansion model.

[0102] Specifically, after the construction of the nonlinear heat conduction model, to further optimize the analysis accuracy of the equipment defect area, the present application proposes a method based on a high-order partial differential expansion model to enhance and optimize the thermal diffusion characteristics of the defect area. By introducing a high-order Laplace operator into the traditional heat conduction model, the changing trend of the thermal diffusion boundary can be more accurately described, and the influence of noise interference on the results can be smoothed, thereby improving the spatial resolution and analysis accuracy of defect positioning. This step is extended based on the nonlinear heat conduction model, closely connected to the previous steps in terms of model data and physical mechanisms, and provides an optimized basis for the subsequent defect heat source inversion.

[0103] In this embodiment, the high-order partial differential expansion model is constructed by introducing a fourth-order Laplace operator into the nonlinear heat conduction equation, and the specific equation form is as follows:

[0104] wherein: is the temperature distribution; is the traditional second-order heat diffusion term, describing the spatial distribution of heat and heat conduction behavior; is the newly added fourth-order partial derivative term, used to enhance the clarity of the heat diffusion boundary and noise suppression ability; is the heat source distribution function; is the high-order term smoothing coefficient, used to adjust the action strength of the fourth-order Laplacian term.

[0105] Generally, the traditional heat conduction model has certain limitations in the boundary clarity of the heat diffusion region, especially under high noise or complex boundary conditions, which may cause the boundary of the heat diffusion region to appear blurred or discontinuous. By introducing the fourth-order partial derivative term , the boundary smoothness can be effectively enhanced, and the optimization strength can be controlled by adjusting the coefficient , so as to more accurately identify the boundary characteristics of defects.

[0106] Specifically, in this embodiment, for the numerical solution of the above high-order partial derivative extended model, the finite element method is used to discretize the spatial dimension, and the implicit difference method is used to discretize the time dimension. The discretized model equation is represented as:

[0107] wherein: is the mass matrix, representing the spatial integral coefficient; and represent the temperature distribution vectors of the current time step and the next time step, respectively; is the second-order Laplacian operator discrete matrix, used to describe the second-order heat diffusion term; is the fourth-order Laplacian operator discrete matrix, which realizes high-order boundary enhancement by squaring the second-order operator; is the time step, used to control the numerical stability in the time direction; is the heat source distribution vector.

[0108] As a possible implementation manner, in order to ensure the stability and calculation accuracy of the numerical solution, the present application adopts a mixed finite element method to discretize the fourth-order partial derivative term . Specifically, by introducing auxiliary variables , the high-order equation is decomposed into two second-order equations for joint solution, thereby reducing the calculation complexity and improving the convergence performance.

[0109] In some embodiments, to further enhance the optimization capability of the high-order model on the boundary characteristics, the present application also combines a multi-scale boundary decomposition method. Specifically, by performing multi-scale decomposition on the boundary characteristics of the heat diffusion region, the boundary region is divided into sub-regions of different scales, and different high-order partial differential term weights are introduced in each sub-region, thereby achieving accurate optimization of the complex boundary region.

[0110] Generally, the performance of the high-order partial differential extended model directly depends on the settings of the parameters and . As an option, the present application can automatically adjust the values of and by a parameter optimization algorithm (such as Bayesian optimization or genetic algorithm) to ensure that the model can obtain the best boundary clarity and noise suppression effect under different devices and operating conditions.

[0111] In this embodiment, to verify the effectiveness of the high-order partial differential extended model, through simulation analysis of real device operation data, it is found that the fourth-order Laplace term can significantly improve the smoothness of the heat diffusion boundary, while effectively reducing the pseudo-boundary signal caused by noise. For example, in the thermal imaging data of a certain device gear box, the high-order term can clearly identify the heat distribution boundary of the gear contact area, while the traditional second-order model produces a more obvious fuzzy phenomenon in this area.

[0112] Through the high-order partial differential extended model of the present application, the accuracy of defect region analysis can be significantly improved under complex devices and operating environments, providing a more optimized input basis for subsequent defect heat source inversion. This model is particularly suitable for device defect detection scenarios with complex boundaries, strong noise interference, or severe heat diffusion gradient changes, such as large-scale rotating machinery, high-temperature furnace equipment, or precision machining devices for operation monitoring. The applicability and universality of the model are fully demonstrated through various tests.

[0113] S5, using a variational regularization method to optimize and solve the heat conduction model, and inverting the heat source distribution of the defect region.

[0114] Specifically, after the high-order partial differential extended model is constructed, to further improve the solving accuracy and stability of the heat conduction model and simultaneously achieve accurate inversion of the heat source distribution of the defect region, the present application introduces a variational regularization method to optimize the model solving process. The variational regularization can effectively handle ill-posed problems caused by boundary noise, incomplete data, or heat diffusion complexity, and achieve a reasonable balance between model fitting and data smoothing. Based on the solving results of the nonlinear heat conduction model and the high-order partial differential model, this step further optimizes the calculation accuracy of the heat source distribution, laying a foundation for subsequent positioning and characteristic extraction of the defect region.

[0115] In this embodiment, the variational regularization method takes the residual of the heat conduction model as the optimization objective, and introduces a regularization term to enhance the smoothness of the heat diffusion region and the clarity of the boundary, while suppressing the influence of noise on the solving process. The definition of the objective optimization function is as follows:

[0116] Wherein: is the optimization objective function; the first term represents the residual of the heat conduction equation; the second term is a second-order regularization term for enhancing the boundary characteristics of the heat field; the third term is a fourth-order regularization term for smoothing the noise of the heat diffusion region; and are regularization weight parameters that control the strength of the regularization term.

[0117] Specifically, in this embodiment, the solution of the objective optimization function is calculated by using the conjugate gradient method for iterative calculation. In each iteration, the gradient of the objective function is calculated and the temperature distribution is updated to gradually approach the optimal solution. The gradient calculation formula is:

[0118] Wherein: is the heat diffusion term; is the heat source distribution function; and are the gradients of the second-order and fourth-order regularization terms, respectively.

[0119] In general, the initial value of the conjugate gradient method is taken as the solution of the high-order partial differential model. The iteration stopping condition can be set as the Euclidean norm of the gradient being less than a certain threshold , for example:

[0120] As an option, this embodiment can also combine a multi-resolution variational method to solve the objective optimization function. Specifically, by performing regularization solving on spatial grids of different resolutions, the computational complexity can be effectively reduced, while the ability to describe the details of the defect region can be improved. For example, after preliminary solving on a low-resolution grid, the result is taken as the initial value of the high-resolution grid to gradually improve the fineness of the solution.

[0121] In some embodiments, to further enhance the robustness of the solution, the present application combines a heat source dynamic updating strategy. Specifically, based on the results of the previous iteration, the heat source distribution is dynamically adjusted as follows:

[0122] wherein: is the update step size used to control the rate of adjustment of the heat source.

[0123] Through the above optimization solving process, the present application can accurately invert the heat source distribution of the defect region inside the equipment. Generally, the intensity and spatial distribution of the heat source distribution can directly reflect the position and characteristics of the defect inside the equipment. For example, for surface crack defects, the heat source distribution usually shows a linear local heat concentration, while for wear or fatigue defects, it may show a scattered point or surface heat concentration.

[0124] As a possible extension application, the present application can further extract the geometric and thermal characteristics of the defect by clustering analysis of the heat source distribution. For example, the K-Means clustering method is used to divide the heat source distribution into different sub-regions, and the specific morphology of the defect is analyzed based on the shape parameters (such as area, length and direction) of the sub-regions. In addition, for dynamic defect regions, time series analysis methods (such as LSTM model) can be combined to predict the defect expansion trend, providing early warning for equipment maintenance.

[0125] By introducing the variational regularization method, the present application can significantly improve the accuracy and stability of the heat source inversion of the defect region. Compared with the traditional heat conduction model solving method, the present application not only can better handle complex working conditions and noise interference, but also can extract more defect characteristic information through the optimization solving process, thereby providing strong support for the accurate detection and predictive maintenance of equipment defects. The completeness of the model solving process and the accuracy of the calculation results have been fully proved by the verification of multiple test scenarios.

[0126] S6, according to the heat source distribution of the defect region, positioning the defect position inside the equipment, and extracting the geometric shape and thermal characteristics of the defect.

[0127] Specifically, after completing the optimization solving of the heat conduction model and the inversion of the heat source distribution of the defect region, the present application further accurately locates the specific position of the defect inside the equipment according to the inverted heat source distribution , and extracts the geometric shape and thermal characteristics of the defect in combination with the spatial distribution characteristics of the heat source. Through these analysis results, the defect characteristics inside the equipment can be fully characterized, providing key data support for subsequent equipment health assessment, maintenance and fault prediction.

[0128] ​This step directly depends on the defect heat source distribution data obtained by inversion in the previous step. The heat source distribution not only reflects the spatial position of the defect, but also contains relevant information of the defect thermal characteristics, such as heat intensity, temperature rise rate, and distribution mode, etc. Therefore, this step extracts multi-dimensional defect characteristic information from the heat source distribution through multi-level data analysis method.

[0129] In this embodiment, the positioning and characteristic extraction process of the defect includes the following contents:

[0130] According to the heat source distribution obtained by inversion , the position of the defect is determined by judging the local high value area of the heat source. Generally, the high value area of the heat source distribution corresponds to the core part or heat concentration area of the defect. In order to realize accurate positioning, the following positioning criteria are adopted in the present application:

[0131] In the spatial heat source distribution, the points meeting the following conditions are selected as the core points of the defect:

[0132] Among them: is a threshold coefficient, and the value range is usually , which is used to filter out the heat source area which is significantly higher than the surrounding background.

[0133] For the case of multiple defects with complex distribution, the present application further combines the spatial clustering algorithm (such as K-Means clustering) to divide the heat source high value area into multiple sub-areas. Each sub-area corresponds to the core position of a defect, and the specific spatial coordinates are determined by calculating the centroid position .

[0134] As a possible implementation manner, the present application can also combine the time variation characteristics of multi-frame thermal imaging data in the positioning process. For example, by calculating the time sequence change rate of each frame of heat source distribution , the dynamic tracking ability of the defect core position can be further improved.

[0135] After the positioning of the defect position is completed, the present application extracts the geometric shape information of the defect based on the spatial form of the heat source distribution. Generally, the geometric shape of the defect can be described by the following formula:

[0136] Among them: represents the spatial area of the defect.

[0137] By calculating the geometric parameters of the area , the following features can be extracted:

[0138] Volume: the spatial volume of the defect area , calculated by integral formula:

[0139] Boundary shape: three-dimensional boundary shape of the defect region , extracted by boundary tracking algorithm;

[0140] Maximum size: maximum size of the defect region in each axial direction , , , calculated by coordinate range difference:

[0141] As an option, for more complex defect morphology (such as cracks or planar defects), the present application combines principal component analysis (PCA) method to extract the main direction and shape characteristics of the defect. For example, by performing principal component decomposition on the spatial coordinate matrix of the defect region, the principal axis direction and aspect ratio of the defect can be obtained.

[0142] The present application also extracts the thermal characteristics of the defect based on the heat source distribution data, mainly including the following: thermal intensity:

[0143] Total thermal intensity of the defect region , calculated by integrating the heat source distribution function:

[0144] Thermal gradient: thermal gradient within the defect region , calculated by numerical differentiation;

[0145] Temperature rise rate: combined with multiple frames of thermal imaging data, the time variation rate of the defect region is calculated , used to judge the dynamic expansion trend of the defect.

[0146] Generally, the thermal intensity can be used to measure the severity of the defect, while the thermal gradient reflects the non-uniformity of heat diffusion within the defect region. For dynamic defect regions, the temperature rise rate can be used as an important indicator for early fault warning.

[0147] In order to facilitate the understanding of the results and subsequent decision support, the present application visualizes the positioning results, geometric characteristics and thermal characteristics of the defect. Specifically, by constructing three-dimensional thermal distribution map and defect space contour map, the spatial position, shape and thermal distribution characteristics of the defect can be intuitively presented.

[0148] In some embodiments, the application also outputs the analysis results as a structured report through a data export function. The report content includes the key parameters of the defect location coordinates, volume, thermal intensity, extension trend, etc., providing scientific basis for equipment maintenance personnel.

[0149] Through the implementation of this step, the application can accurately locate the defects inside the equipment from the heat source distribution data and extract the geometric shape and thermal characteristic information of the defects. These information not only provides an important reference for the fault analysis and maintenance of the equipment, but also lays a data foundation for predictive maintenance and operation state monitoring. The positioning accuracy and characteristic extraction capability of the model have been verified through actual tests of various industrial equipment, and can effectively meet the equipment defect detection needs under complex working conditions.

[0150] Please refer to the attached Figure 2 The application also provides an equipment defect detection system based on infrared identification, comprising:

[0151] an infrared thermal imager for collecting temperature distribution of the equipment surface;

[0152] a data processing module for denoising and boundary optimization of the collected infrared thermal imaging data;

[0153] a model construction module for establishing a temperature distribution model based on a nonlinear heat conduction equation and a high-order partial differential extension model;

[0154] an optimization and inversion module for optimizing and solving the temperature distribution by using a variational regularization method and inverting the defect heat source;

[0155] an output module for generating defect detection results, including the geometric characteristics, thermal intensity distribution and location coordinates of the defect.

[0156] Specifically, the infrared thermal imager:

[0157] The infrared thermal imager is used to collect the temperature distribution of the equipment surface in the running state in real time, and can obtain high-resolution infrared thermal imaging data to generate a time series of dynamic temperature distribution graphs.

[0158] Generally, the infrared thermal imager supports multi-frame sampling and dynamic adjustment of the sampling frequency, and is suitable for different equipment operating conditions and environmental noise. As an option, the infrared thermal imager can also collect local high-resolution temperature distribution in combination with the specific regional monitoring requirements of specific parts of the equipment.

[0159] The data processing module:

[0160] The data processing module is responsible for denoising and boundary optimization of the temperature distribution data collected by the infrared thermal imager, ensuring the physical consistency and calculation accuracy of the input data.

[0161] Specifically, the data processing module can remove high-frequency noise in the infrared data and perform heat flow balance optimization on the boundary regions of the device surface, avoiding data errors caused by complex geometric boundaries. As a possible implementation, the module supports multi-dimensional noise reduction processing of time series data to improve the time continuity of the thermal distribution.

[0162] The model construction module:

[0163] The model construction module is used to establish a temperature distribution model of the device based on a nonlinear heat conduction equation and a high-order partial differential extended model. The module can combine the initial temperature data provided by the infrared thermal imager to accurately describe the dynamic temperature changes inside the device using the physical properties of the device, such as thermal conductivity and thermal diffusion characteristics.

[0164] In some embodiments, the model construction module also supports adaptive adjustment of dynamic heat conduction parameters, such as temperature-dependent thermal conductivity, to improve the adaptability of the model under complex operating conditions.

[0165] The optimization and inversion module:

[0166] The optimization and inversion module optimizes and solves the temperature distribution model based on the variational regularization method, and uses the optimized results to invert the defect heat source distribution inside the device.

[0167] The module can efficiently handle the ill-posed problem in the heat conduction model, enhance the smoothness of the heat diffusion region and the clarity of the boundary through regularization, and improve the accuracy of defect area inversion. In addition, the module supports dynamic adjustment of the defect heat source distribution to adapt to changes in device operating conditions.

[0168] The output module:

[0169] The output module is responsible for presenting the analysis results in a visual form and generating a defect detection report. The output content includes the geometric characteristics of the defect (such as volume, shape and boundary), the thermal intensity distribution (such as total thermal intensity and gradient), and the spatial position coordinates of the defect.

[0170] In some embodiments, the output module supports the generation of a three-dimensional thermal distribution map for intuitive display of the spatial distribution of defects. At the same time, the module can export the detection results in a structured data format, facilitating the connection with the device monitoring system and realizing the whole life cycle management of the device state.

[0171] Although embodiments of the present application have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A method for detecting defects in a device based on infrared recognition, characterized in that, The method comprises the following steps: Collecting infrared thermal imaging data of the device surface in the running state to obtain a sequence of infrared thermal distribution changing over time; Performing noise reduction processing on the infrared thermal imaging data to eliminate high-frequency noise, retain effective signals, and optimize the boundary region using heat flow balance conditions; Constructing a nonlinear heat conduction model to describe the dynamic changes in the internal temperature distribution of the device; Optimizing and analyzing the heat diffusion characteristics of the defect region based on a high-order partial differential extended model; Using a variational regularization method to optimize and solve the heat conduction model to inverse the heat source distribution of the defect region; According to the heat source distribution of the defect region, locating the defect position inside the device and extracting the geometric shape and thermal characteristics of the defect; The noise reduction processing step comprises: Performing two-dimensional wavelet decomposition on the collected thermal imaging data to separate the data into high-frequency and low-frequency components; Using a soft threshold filtering method to reduce noise for the high-frequency component to filter random noise signals; Retaining effective signals for the low-frequency component and reconstructing the processed thermal imaging data; The boundary region optimization using heat flow balance conditions comprises: For the complex geometric boundary of the device surface, setting the boundary condition as the heat flow balance condition in the normal direction of the device surface; Using the Neumann boundary condition based on the shape of the device surface to ensure that the heat flow of the device surface does not flow out of the boundary; The high-order partial differential extended model comprises: Introducing a fourth-order partial differential term in the heat conduction model to describe the heat diffusion characteristics of the boundary of the defect region through the high-order form of the Laplacian operator; Through the introduction of the high-order term, the heat boundary definition of the defect region is enhanced, and the noise interference in the boundary region is reduced; The variational regularization method comprises: Defining an objective function based on the residual of the heat conduction equation, and balancing the fitting accuracy and data smoothness of the heat distribution through a regularization term; The regularization term includes a second-order partial differential regularization term and a fourth-order partial differential regularization term; Solving the objective function by a numerical optimization method to optimize the temperature distribution of the device surface and inside.

2. The method of claim 1, wherein, The infrared thermal imaging data collection step comprises: Using a high-resolution infrared thermal imager to obtain the temperature distribution of the device surface; Collecting multiple frames of thermal imaging data at a fixed sampling frequency in the time dimension to form a thermal distribution time sequence; Recording the operating condition parameters of the device, including the ambient temperature, load state, and operating time. 3.The infrared recognition-based device defect detection method of claim 1, wherein, The nonlinear heat conduction model is: Establishing a nonlinear heat conduction model to describe the dynamic behavior of the temperature distribution changing over time according to the thermal conductivity characteristics of the device material; In the model, the thermal conductivity coefficient is related to the temperature, and the heat diffusion behavior is adjusted by the heat source characteristics of the device defect region. 4.The infrared recognition-based device defect detection method of claim 1, wherein, The step of inverting the heat source distribution of the defect region comprises: Using the optimized temperature distribution to calculate the heat source intensity distribution inside the device; According to the spatial characteristics of the heat source distribution, identifying the position of the defect region and extracting the geometric characteristics and thermal intensity of the defect. 5.The infrared recognition-based device defect detection method of claim 1, wherein, The defect detection result obtained by inverting the heat source distribution of the defect region is used to generate a three-dimensional thermal distribution map of the defect region, and further used to provide a defect position, geometric shape, and thermal characteristic report for reference in device operation and maintenance decision-making.

6. The system for detecting defects of an apparatus based on infrared recognition according to any one of claims 1 to 5, characterized in that, It comprises: An infrared thermal imager for collecting the temperature distribution of the device surface; The data processing module is configured to perform noise reduction and boundary optimization on the collected infrared thermal imaging data. The model construction module is configured to establish a temperature distribution model based on a nonlinear heat conduction equation and a high-order partial differential extended model. The optimization and inversion module is configured to optimize and solve the temperature distribution by using a variational regularization method, and to invert the defect heat source. The output module is configured to generate a defect detection result, including geometric characteristics, thermal intensity distribution, and position coordinates of the defect.

Images