Intelligent defect detection and repair method and system for vacuum isothermal forgings based on multi-modal sensing data

Abstract

The application relates to the technical field of intelligent manufacturing, in particular to an intelligent defect detection and repair method and system for vacuum isothermal forgings based on multi-modal sensing data. The method comprises the following steps: acquiring image data, thermal distribution data and stress monitoring data of a vacuum isothermal forging through a visual sensor, a thermal sensor and a stress sensor respectively, pre-processing the above data, and generating a multi-modal data set aligned in time and space; performing fusion processing on the multi-modal data set, extracting multi-dimensional features of defects through a deep learning network, and generating a defect feature sequence; optimizing a repair path planning model based on the defect feature sequence and combining stress distribution features, and generating a repair task sequence; and inputting the repair task sequence into an intelligent repair device to perform a repair operation. The application combines multi-modal sensing technology and a deep learning algorithm, realizes high-precision detection, classification and repair of defects of vacuum isothermal forgings, significantly improves production efficiency and product quality, and reduces the rate of defective products and repair costs.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing technology, and in particular to an intelligent defect detection and repair method and system for vacuum isothermal forgings based on multimodal sensing data. Background Technology

[0002] Vacuum isothermal forgings are widely used in high-end manufacturing fields such as aerospace, aviation, and gas turbines. Their production process places extremely high demands on the superplastic deformation of the billet, control of process parameters, and control of microstructure and properties. However, due to the complex process conditions and the extreme difficulty of operation under high temperatures, forgings are prone to complex defects such as cracks, uneven microstructure, and uneven heat distribution. If these defects are not detected and effectively repaired in a timely manner, they will seriously affect the quality and service performance of the forgings.

[0003] Existing defect detection and repair technologies mainly rely on single-modal sensors (such as visual ultrasound), which have low detection range and accuracy, making them difficult to adapt to the complex and diverse types of defects in forgings. Furthermore, traditional repair methods lack intelligent means for determining defect location and type, typically requiring decisions based on human experience, resulting in low repair efficiency and poor accuracy.

[0004] With the development of multimodal sensing technology and artificial intelligence algorithms, the application of deep learning algorithms based on multimodal sensing data fusion in complex industrial scenarios is gradually becoming a trend. Multimodal data can comprehensively utilize information from vision, thermal sensing, and stress sensing to characterize defects from multiple dimensions, while deep learning algorithms can achieve accurate defect classification and automatic optimization of repair strategies through data learning. However, there is currently a lack of a complete system that can integrate multimodal data acquisition, deep learning processing, and intelligent repair operations for defect detection and repair of vacuum isothermal forgings. Therefore, proposing a system that can efficiently and accurately complete defect detection and intelligent repair is key to solving the above problems. Summary of the Invention

[0005] This invention provides an intelligent defect detection and repair method and system for vacuum isothermal forgings based on multimodal sensing data, in order to solve the problem of how to comprehensively analyze and process the complex defects of vacuum isothermal forgings through multimodal sensing technology and deep learning algorithms, and achieve high-precision detection, classification and intelligent repair of defects.

[0006] To address the aforementioned technical problems, this invention provides an intelligent defect detection and repair method for vacuum isothermal forgings based on multimodal sensing data, comprising:

[0007] Image data of vacuum isothermal forgings are acquired by a vision sensor, heat distribution data are collected by a thermal sensor, and stress monitoring data are collected by a stress sensor. The image data, heat distribution data, and stress monitoring data are preprocessed to generate a spatiotemporally aligned multimodal dataset.

[0008] The multimodal dataset is fused, and multidimensional features of defects are extracted through a deep learning network to generate a defect feature sequence of defect location, category, and severity.

[0009] Based on the defect feature sequence and stress distribution characteristics, the repair path planning model is optimized to generate a repair task sequence of repair methods, repair parameters and path optimization schemes.

[0010] The repair task sequence is input into the intelligent repair equipment to perform repair operations on the vacuum isothermal forging. The repair status is monitored in real time by sensors, a repair quality report is generated, and the repair path planning model is fed back for dynamic adjustment.

[0011] Furthermore, the step of acquiring image data through a visual sensor includes:

[0012] The image data is subjected to noise filtering and feature enhancement to generate preprocessed image data.

[0013] Furthermore, the step of acquiring heat distribution data through a thermal sensor includes:

[0014] The heat distribution data is denoised and normalized to generate preprocessed heat distribution data.

[0015] Furthermore, the step of acquiring stress monitoring data through a stress sensor includes:

[0016] The stress monitoring data is time-aligned and feature-extracted to generate stress gradient features consistent with the image data and thermal distribution data.

[0017] Furthermore, the fusion processing step of the multimodal dataset includes:

[0018] The image data is convolved to extract image defect features, the thermal distribution data is subjected to gradient analysis to extract temperature anomaly features, and the stress monitoring data is subjected to feature extraction to generate a multimodal feature sequence of defects.

[0019] Furthermore, the step of the deep learning network extracting multidimensional features includes:

[0020] The weights of each modal feature are calculated using an attention mechanism, and the image features, thermal distribution features, and stress features are weighted and fused to generate the defect location, category, and severity.

[0021] Furthermore, the optimization steps of the repair path planning model include:

[0022] Based on the defect feature sequence, a preliminary repair method and repair parameters are generated. The repair path is optimized by combining the stress distribution characteristics, and a repair task sequence is generated.

[0023] Furthermore, the repair task sequence includes:

[0024] The repair method, repair parameters, and path optimization scheme, wherein the path optimization scheme includes the spatial location of the repair points and the repair sequence.

[0025] Furthermore, the steps for executing and providing feedback on the repair task include:

[0026] Repair operations are performed using intelligent repair equipment according to the repair task sequence, real-time data on the repair status is collected, repair monitoring data is generated and fed back to the path planning model, and repair parameters and paths are dynamically adjusted.

[0027] Furthermore, a smart defect detection and repair system for vacuum isothermal forgings based on multimodal sensing data includes:

[0028] The data acquisition module is used to acquire the image data, thermal distribution data, and stress monitoring data;

[0029] The data processing and fusion module is used to preprocess and fuse the data to generate a multimodal feature sequence of defects;

[0030] A repair task generation module is used to generate a repair task sequence based on the defect feature sequence;

[0031] The intelligent repair module is used to execute the repair task sequence and monitor the repair status in real time through sensors, and feed back the real-time repair monitoring data to the repair task generation module.

[0032] The key innovations of this invention include:

[0033] (1) Combining multimodal data fusion with deep learning: This invention is the first to align visual, thermal and stress data in time and space, extract multidimensional features through a deep learning network, and analyze the properties and distribution of complex defects through multimodal fusion, thus solving the limitations of traditional single-modal detection.

[0034] (2) Dynamic repair task sequence generation and optimization: The repair path planning is dynamically adjusted by using the defect feature sequence and stress distribution characteristics to generate a repair task sequence including repair methods, repair parameters and optimization paths, thus realizing an intelligent repair process.

[0035] (3) Real-time feedback and closed-loop optimization mechanism: By monitoring the repair status in real time and feeding the repair monitoring data back to the path planning model for dynamic adjustment, a closed-loop system of detection, repair and verification is formed, which significantly improves the accuracy and adaptability of repair.

[0036] (4) Efficient application of mathematical models: Complex mathematical calculations are introduced into the detection and repair process, including convolutional neural network feature extraction, gradient analysis, dynamic path planning optimization, etc., which ensures the efficiency of data processing and model optimization and lays the foundation for large-scale industrial applications.

[0037] The following are its main beneficial effects:

[0038] (1) Improved detection accuracy: By spatiotemporal alignment and fusion processing of data from three sensors—visual, thermal, and stress—the surface and internal defect features of forgings are comprehensively analyzed, solving the problem of low detection accuracy in single-sensor modes. Deep learning networks further extract and classify multidimensional features, significantly improving the accuracy and comprehensiveness of defect detection.

[0039] (2) Optimize repair efficiency: Based on the defect feature sequence and combined with stress distribution features, an optimized repair path planning model is generated. By dynamically adjusting the repair method and parameters, the accuracy and efficiency of the repair process are improved and redundant repair steps are reduced.

[0040] (3) Enhance system adaptability: During the repair process, the repair status is monitored in real time by sensors and dynamically fed back to the path planning model to adjust the repair parameters and path, so as to ensure that the repair process can adapt to complex defect types and different working environments.

[0041] (4) Reduce production costs: Through intelligent defect detection and repair processes, manual intervention is reduced, the defect rate is lowered, and the use of repair resources is optimized, which significantly reduces production and repair costs.

[0042] (5) Improve product quality: The system verifies the repair effect in real time during the repair process to ensure that the structural integrity and performance indicators of the repaired forgings meet the requirements and satisfy the strict quality standards of high-end manufacturing. Attached Figure Description

[0043] Figure 1 This is a flowchart illustrating the intelligent defect detection and repair method for vacuum isothermal forgings based on multimodal sensor data, provided in an embodiment of this application.

[0044] Figure 2 The diagram shows the structure of the intelligent defect detection and repair system for vacuum isothermal forgings based on multimodal sensing data, which is provided in the embodiments of this application. Detailed Implementation

[0045] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein in the specification of the application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application; the terms "comprising" and "having," and any variations thereof, in the specification, claims, and foregoing drawings of this application, are intended to cover non-exclusivity. The terms "first," "second," etc., in the specification, claims, and foregoing drawings of this application are used to distinguish different objects, not to describe a particular order.

[0046] In this document, the term "embodiment" means that a specific feature or structural characteristic described in connection with an embodiment may be present in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0047] Example 1: Refer to Figure 1 This is a flowchart illustrating the intelligent defect detection and repair method for vacuum isothermal forgings based on multimodal sensing data provided in this embodiment of the invention. The process may include at least steps S100-S400:

[0048] S100. Image data of the vacuum isothermal forging is acquired through a vision sensor, heat distribution data is collected through a thermal sensor, and stress monitoring data is collected through a stress sensor. The image data, heat distribution data, and stress monitoring data are preprocessed to generate a spatiotemporally aligned multimodal dataset.

[0049] S200. The multimodal dataset is fused and processed, and multidimensional features of defects are extracted through a deep learning network to generate a defect feature sequence of defect location, category, and severity.

[0050] S300. Based on the defect feature sequence and stress distribution characteristics, optimize the repair path planning model to generate a repair task sequence of repair methods, repair parameters and path optimization schemes.

[0051] S400. Input the repair task sequence into the intelligent repair equipment to perform repair operations on the vacuum isothermal forging, and monitor the repair status in real time through sensors, generate a repair quality report and feed it back to the repair path planning model for dynamic adjustment.

[0052] Step S100 includes at least steps S110-S130:

[0053] S110. Acquire image data through a vision sensor and perform preprocessing.

[0054] First, image data I(x,y,t) of the vacuum isothermal forging is acquired using a vision sensor, where x and y are the two-dimensional coordinates of the image, and t is the time series. This image data contains information about defects on the forging surface (cracks, scratches, deformation, etc.), but may also contain noise interference.

[0055] Furthermore, noise filtering is performed by applying a two-dimensional Gaussian filter G(x,y;σ) to I(x,y,t). The filtering formula is as follows:

[0056]

[0057] Where σ is the standard deviation of the filter, used to control the smoothing degree. The resulting I'(x,y,t) is the image data after removing high-frequency noise.

[0058] Furthermore, feature enhancement processing is performed. The Laplacian operator is used to enhance the features of high-gradient regions in the image; the formula is as follows:

[0059]

[0060] Where I''(x,y,t) is the enhanced image data used to extract crack edge and surface defect features.

[0061] The preprocessed image data I''(x,y,t) is obtained and used as the input for subsequent multimodal data fusion.

[0062] S120. Collect heat distribution data through a thermal sensor and perform preprocessing.

[0063] First, thermal distribution data is acquired. Specifically, thermal distribution data T(x,y,t) of the vacuum isothermal forging is acquired using a thermal sensor, where T represents the temperature value at the pixel location, x and y are spatial coordinates, and t is a time series. The abnormal temperature distribution caused by the forging process in the thermal distribution data can reflect potential defects inside the forging.

[0064] Further, noise reduction is performed. Time series smoothing is applied to T(x,y,t) using a moving average filter, with the following formula:

[0065]

[0066] Where n is the size of the time window. The filtered T'(x,y,t) eliminates high-frequency noise in the temperature sampling.

[0067] Furthermore, normalization is performed. T'(x,y,t) is normalized to bring the temperature values ​​to the range [0,1].

[0068]

[0069] in, and These represent the maximum and minimum values ​​of the heat distribution data, respectively.

[0070] The normalized thermal distribution data T''(x,y,t) is obtained and used as the input for subsequent multimodal data fusion.

[0071] S130. Stress monitoring data is collected and preprocessed using a stress sensor.

[0072] First, stress monitoring data is acquired. Stress monitoring data S(x,y,t) of the vacuum isothermal forging is collected using a stress sensor, where S represents the stress value at the pixel location, x and y are spatial coordinates, and t is the time series. This stress monitoring data reflects the stress distribution within the forging and can be used to identify potential structural defects.

[0073] Furthermore, the time series of S(x,y,t) are aligned with those of I''(x,y,t) and T''(x,y,t) to ensure that the three-modal data are at the same time points. It has consistency:

[0074]

[0075] Furthermore, spatial gradient analysis is performed on S(x,y,t) to calculate the local variation characteristics in the stress distribution, as shown in the formula:

[0076]

[0077] Wherein, ∇S(x,y,t) is the gradient vector of the stress distribution, used to locate the stress concentration region.

[0078] Furthermore, a spatiotemporally aligned multimodal dataset consistent with image data I''(x,y,t) and thermal distribution data T''(x,y,t) is generated.

[0079]

[0080] The features of thermal distribution and stress distribution are preprocessed in a unified manner and then input into the multimodal dataset D(x,y,t), providing a unified foundation for feature extraction in subsequent deep learning.

[0081] Step S200 includes at least steps S210-S230:

[0082] S210. Perform deep feature extraction on the image data in the multimodal dataset to generate an image defect feature sequence.

[0083] First, input preprocessed image data I''(x,y,t) from S100. Feature extraction is performed on I''(x,y,t) using a convolutional neural network (CNN). The formula for the convolution kernel is as follows:

[0084]

[0085] in, The convolutional kernel represents the feature values; k is the kernel size. is the kernel weight; b is the bias value.

[0086] Furthermore, the convolutional feature map is input into a max pooling layer to extract key feature points, resulting in an image defect feature matrix:

[0087]

[0088] in, This is a sequence of image defect features used to represent crack morphology and edge characteristics.

[0089] Connection instructions: It will be used as input to the subsequent S230 module and fused with other modal features.

[0090] S220. Extract features from the thermal distribution data in the multimodal dataset and generate a thermal distribution feature sequence based on the temperature anomaly region.

[0091] First, input the normalized heat distribution data T''(x,y,t) from S100. Apply the temperature gradient calculation to T''(x,y,t), expressed by the formula:

[0092]

[0093] in, It represents the local gradient of temperature change and is used to locate areas of temperature anomalies.

[0094] Furthermore, by setting a temperature threshold Extracting outlier regions from thermal distribution data:

[0095]

[0096] in, This is the set of coordinates for the abnormal region.

[0097] Furthermore, based on the distribution characteristics of the abnormal regions, a heat distribution feature sequence is generated:

[0098]

[0099] in, It represents the local anomaly features and gradient information of heat distribution.

[0100] Connection instructions: Will be compared with image feature sequences and stress characteristic sequence Integrate in S230.

[0101] S230. Perform multimodal fusion processing on the image defect feature sequence, thermal distribution feature sequence, and stress feature sequence to generate a defect feature sequence with defect location, category, and severity.

[0102] First, input the stress gradient feature sequence ∇S(x,y,t) from S130. Then, time-align the stress data with the image and thermal distribution data using the following formula:

[0103]

[0104] in, To ensure the consistency of the three-modal data over time after alignment.

[0105] Furthermore, FI,FT,∇S(x,y,t) are input into the deep learning fusion network, and a multi-layer attention mechanism is used to calculate the correlation weights between modalities:

[0106]

[0107] in, These are the fusion weights for image, thermal distribution, and stress features, respectively. , , This is the characteristic transformation matrix.

[0108] Furthermore, feature vectors representing the defect location, category, and severity are calculated based on the fusion weights:

[0109]

[0110] in, This represents the fused defect feature sequence, including location. ,category and severity .

[0111] Connection instructions: It will be directly used as input to the S300 module for repair path planning and optimization.

[0112] Step S300 includes at least steps S310-S330:

[0113] S310. Based on the defect category and location in the defect feature sequence, a preliminary repair method and repair parameters are generated.

[0114] First, input the defect feature sequence output by S230. ={ , , },

[0115] in, =( , , ) represents the spatial location of the defect; Defect category (e.g., cracks, holes, overheated areas); This indicates the severity of the defect.

[0116] Furthermore, according to and Choose an appropriate repair method The specific rules are as follows:

[0117]

[0118] in, and This is the threshold for the severity of the defect.

[0119] Furthermore, the repair parameters are calculated. Specifically, for each repair method... Calculate its corresponding parameter set Such as laser power Heat treatment temperature The calculation formula is:

[0120]

[0121] Where f is the parameter generation function.

[0122] Connection instructions: Preliminary repair method and repair parameters It will be used as input to S320 for path optimization.

[0123] S320. Combining stress distribution characteristics and defect location, optimize the repair path planning model and generate an optimized repair path scheme.

[0124] First, input the stress gradient characteristics ∇S(x,y,t) of S130 and the defect location. Calculate the stress distribution around the defect. :

[0125]

[0126] in, This represents the local volume where the defect is located.

[0127] Furthermore, based on the repair method and stress distribution Establish a repair path planning model:

[0128]

[0129] in, To repair path length; The stress value is the area traversed by the path. , This represents the path weight coefficient.

[0130] Furthermore, an optimized repair path scheme is generated: a dynamic programming algorithm is used to solve the repair path planning model to generate an optimized path scheme. The repair order and path points.

[0131] Connection Explanation: Repair Path Optimization Solution Repair methods Repair parameters It will be used as an input to S330.

[0132] S330. Verify the repair path planning model, and integrate the repair method, repair parameters and path optimization scheme to generate a repair task sequence.

[0133] First, use simulation tests to evaluate the path optimization scheme. Verification was conducted to calculate the feasibility of the path.

[0134]

[0135] in, The cost of repairing the areas traversed by the path; This refers to stress changes after repair. , To verify the weighting coefficients.

[0136] Furthermore, the repair method Repair parameters and path optimization scheme Integrate and generate a repair task sequence :

[0137]

[0138] Connection Notes: Repair Task Sequence It will be used as input to the S400 module for repair device execution.

[0139] Step S400 includes at least steps S410-S430:

[0140] S410. Input the repair task sequence into the intelligent repair device and execute the repair operation.

[0141] First, input the repair task sequence output by S330. ,in, Repair methods (such as laser cladding, mechanical compression, and local heat treatment); To repair parameters (such as laser power) Heat treatment temperature wait); The repair path optimization scheme includes path points and repair order.

[0142] Furthermore, the repair parameters will be... The commands are parsed into device instructions, and specific repair operations are executed through the intelligent repair equipment. For laser repair, the laser equipment power is controlled. and scanning speed The relationship is expressed as:

[0143]

[0144] in, This refers to the laser energy per unit area.

[0145] Furthermore, based on the repair path The drive repair equipment moves along the set path points and to the defect location. Repairs were completed point by point.

[0146] Connection instructions: After the repair operation is completed, the repair status is collected by the sensor and used as the input of S420.

[0147] S420: The repair status is monitored in real time through sensors, and the stress and heat distribution data of the repair area are processed to generate real-time repair monitoring data.

[0148] First, image data of the repaired area is acquired using a visual sensor. Heat distribution data is collected through thermal sensors. Stress distribution data is collected through stress sensors. .

[0149] Furthermore, the collected image data Perform a difference analysis with the unrestored data I''(x,y,t) to calculate the visual changes in the restored area:

[0150]

[0151] For heat distribution data By comparing the data T''(x,y,t) before the repair, the temperature change after the thermal anomaly repair can be located:

[0152]

[0153] Stress distribution data Gradient change analysis was performed with respect to the gradient ∇S(x,y,t) before repair:

[0154] ΔS(x,y,t)=∇ (x,y,t)−∇S(x,y,t)

[0155] Furthermore, by integrating visual, thermal distribution, and stress change data, real-time monitoring data for repair is generated. :

[0156]

[0157] Connection Notes: Real-time monitoring data As input to the S430, it is used to adjust the repair parameters and paths.

[0158] S430. Input the real-time monitoring data of the repair into the repair path planning model, dynamically adjust the repair method and parameters, and generate the final repair quality report.

[0159] First, real-time monitoring data Input the repair path planning model and adjust the repair parameters. and path Dynamic adjustments are made, and the adjustment formula is as follows:

[0160]

[0161]

[0162] in, These are the adjusted repair parameters; The adjusted repair path; , This is for adjusting the coefficient.

[0163] Furthermore, based on the adjusted repair results, the residual severity of the defect after repair is calculated. Repair completion rate:

[0164]

[0165] like If the error is less than or equal to ϵ, the repair is considered successful, where ϵ is the allowable error.

[0166] Furthermore, the adjusted repair parameters ,path Repair quality report Integrate to complete the final repair loop.

[0167] Connection Notes: Repair Quality Report This will serve as the basis for verifying the system's closed loop and will be fed back to the preceding modules to optimize the model.

[0168] The key innovations of this invention include:

[0169] (1) Combining multimodal data fusion with deep learning: This invention is the first to align visual, thermal and stress data in time and space, extract multidimensional features through a deep learning network, and analyze the properties and distribution of complex defects through multimodal fusion, thus solving the limitations of traditional single-modal detection.

[0170] (2) Dynamic repair task sequence generation and optimization: The repair path planning is dynamically adjusted by using the defect feature sequence and stress distribution characteristics to generate a repair task sequence including repair methods, repair parameters and optimization paths, thus realizing an intelligent repair process.

[0171] (3) Real-time feedback and closed-loop optimization mechanism: By monitoring the repair status in real time and feeding the repair monitoring data back to the path planning model for dynamic adjustment, a closed-loop system of detection, repair and verification is formed, which significantly improves the accuracy and adaptability of repair.

[0172] (4) Efficient application of mathematical models: Complex mathematical calculations are introduced into the detection and repair process, including convolutional neural network feature extraction, gradient analysis, dynamic path planning optimization, etc., which ensures the efficiency of data processing and model optimization and lays the foundation for large-scale industrial applications.

[0173] The following are its main beneficial effects:

[0174] (1) Improved detection accuracy: By spatiotemporal alignment and fusion processing of data from three sensors—visual, thermal, and stress—the surface and internal defect features of forgings are comprehensively analyzed, solving the problem of low detection accuracy in single-sensor modes. Deep learning networks further extract and classify multidimensional features, significantly improving the accuracy and comprehensiveness of defect detection.

[0175] (2) Optimize repair efficiency: Based on the defect feature sequence and combined with stress distribution features, an optimized repair path planning model is generated. By dynamically adjusting the repair method and parameters, the accuracy and efficiency of the repair process are improved and redundant repair steps are reduced.

[0176] (3) Enhance system adaptability: During the repair process, the repair status is monitored in real time by sensors and dynamically fed back to the path planning model to adjust the repair parameters and path, so as to ensure that the repair process can adapt to complex defect types and different working environments.

[0177] (4) Reduce production costs: Through intelligent defect detection and repair processes, manual intervention is reduced, the defect rate is lowered, and the use of repair resources is optimized, which significantly reduces production and repair costs.

[0178] (5) Improve product quality: The system verifies the repair effect in real time during the repair process to ensure that the structural integrity and performance indicators of the repaired forgings meet the requirements and satisfy the strict quality standards of high-end manufacturing.

[0179] Example 2: Figure 2 A structural block diagram of an intelligent defect detection and repair system for vacuum isothermal forgings based on multimodal sensor data, according to an embodiment of the present invention, is shown. Figure 2 As shown, the system may include:

[0180] The data acquisition module 10 acquires image data I(x,y,t) of the vacuum isothermal forging through a vision sensor and collects defect information (such as cracks, scratches and deformation) on the surface of the forging.

[0181] Thermal distribution data T(x,y,t) of forgings are collected by thermal sensors to capture abnormal temperature distributions in forgings and reflect internal defects.

[0182] By collecting stress distribution data S(x,y,t) using stress sensors, the stress concentration of forgings can be monitored, and potential structural defects can be located.

[0183] The data processing and fusion module 20 preprocesses the collected data, including noise filtering and feature enhancement of image data, denoising and normalization of thermal distribution data, and time alignment and gradient calculation of stress distribution data.

[0184] Preprocessed multimodal data A unified input is used to perform deep learning fusion and extract multidimensional features of defects.

[0185] Output defect feature sequence ={ , , },in For the location of the defect, As a defect category, This indicates the severity of the defect.

[0186] Repair task generation module 30, based on defect feature sequence Based on the stress distribution characteristics ∇S(x,y,t), an appropriate repair method is selected. (Laser cladding, mechanical compression, local heat treatment).

[0187] Generate repair parameters Such as laser power Heat treatment temperature and optimized repair path .

[0188] Output repair task sequence This serves as input for subsequent repair operations.

[0189] The intelligent repair and feedback module 40 will sequence the repair tasks. Enter the intelligent repair device and execute the specific repair operation according to the path RdR_{d}.

[0190] The repair status is monitored in real time using sensors, and image data of the repaired product is collected. Heat distribution data Stress data .

[0191] The visual, thermal distribution, and stress changes of the repaired area are analyzed to generate real-time monitoring data. .

[0192] The repair parameters and paths are dynamically adjusted based on monitoring data, and a repair quality report is generated in the end. .

[0193] Beneficial effects of the embodiments:

[0194] (1) Efficient defect detection and repair closed loop: The system realizes a closed loop of the entire process from defect detection and classification to repair operation, which significantly improves the efficiency of detection and repair.

[0195] (2) Multimodal data fusion improves accuracy: Through multimodal fusion of visual, thermal and stress data, the system can fully capture the surface and internal defect features of forgings, thus improving the accuracy of defect detection.

[0196] (3) Intelligent repair task planning: Combining deep learning algorithms and dynamic path optimization models, the system can generate customized repair tasks for different defect categories and severity, ensuring the accuracy and adaptability of the repair process.

[0197] (4) Dynamic feedback mechanism enhances reliability: The real-time monitoring and feedback mechanism enables the system to dynamically adjust the repair parameters and path according to the repair effect, thereby improving the success rate of repair and the quality of forgings.

[0198] Obviously, the embodiments described above are only some embodiments of this application, not all embodiments. The accompanying drawings show preferred embodiments of this application, but do not limit the patent scope of this application. This application can be implemented in many different forms; conversely, the purpose of providing the embodiments is to make the disclosure of this application more thorough and comprehensive. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing specific embodiments and make equivalent substitutions for some of the technical features. Any equivalent structures made using the content of this application's specification and drawings, directly or indirectly applied to other related technical fields, are similarly within the scope of patent protection of this application.

Claims

1. A method for intelligent defect detection and repair of vacuum isothermal forgings based on multimodal sensing data, characterized in that, Includes the following steps: Image data of vacuum isothermal forgings are acquired by a vision sensor, heat distribution data are collected by a thermal sensor, and stress monitoring data are collected by a stress sensor. The image data, heat distribution data, and stress monitoring data are preprocessed to generate a spatiotemporally aligned multimodal dataset. The multimodal dataset is fused, and multidimensional features of defects are extracted through a deep learning network to generate a defect feature sequence of defect location, category, and severity. Based on the defect category and location in the defect feature sequence, a preliminary repair method and repair parameters are generated. Combining stress distribution characteristics and defect location, the repair path planning model is optimized to generate an optimized repair path scheme. Specifically, this includes: inputting stress gradient characteristics and defect location, calculating the stress distribution around the defect; establishing a repair path planning model based on the repair method and stress distribution; solving the repair path planning model using a dynamic programming algorithm to generate an optimized path scheme; validating the repair path planning model, and integrating the repair method, repair parameters, and optimized path scheme to generate a repair task sequence. The repair task sequence is input into the intelligent repair equipment to perform repair operations on the vacuum isothermal forging. The repair status is monitored in real time by sensors, a repair quality report is generated and fed back to the repair path planning model, and the repair parameters and repair path are dynamically adjusted.

2. The method according to claim 1, characterized in that, The steps for acquiring image data of vacuum isothermal forgings using a vision sensor include: The image data is subjected to noise filtering and feature enhancement to generate preprocessed image data.

3. The method according to claim 1, characterized in that, The steps for acquiring heat distribution data using a thermal sensor include: The heat distribution data is denoised and normalized to generate preprocessed heat distribution data.

4. The method according to claim 1, characterized in that, The steps for acquiring stress monitoring data using a stress sensor include: The stress monitoring data is time-aligned and feature-extracted to generate stress gradient features consistent with the image data and thermal distribution data.

5. The method according to claim 1, characterized in that, The process of fusing multimodal datasets includes: Image defect features are extracted by performing convolution operations on image data, temperature anomaly features are extracted by gradient analysis on thermal distribution data, and features are extracted from stress monitoring data to generate a multimodal feature sequence of defects.

6. The method according to claim 1, characterized in that, The steps for extracting multidimensional features of defects using deep learning networks include: By calculating the correlation weights between different modalities through a multi-layer attention mechanism, image defect features, thermal distribution features, and stress gradient features are weighted and fused to generate defect location, category, and severity.

7. The method according to claim 1, characterized in that, The repair task sequence includes: The repair method, repair parameters, and path optimization scheme, wherein the path optimization scheme includes the spatial location of the repair points and the repair order.

Images