Intelligent monitoring method and system for production of forgings

Abstract

The application relates to the technical field of intelligent monitoring, and discloses an intelligent monitoring method and system for forging production. The application comprises the following steps: collecting temperature data of a wind power flange forging and performing polar coordinate conversion to obtain polar coordinate temperature field data; inputting the polar coordinate temperature field data into a ring thermal field gradient decomposition network to perform feature extraction and obtain a temperature field feature vector; performing phase transition critical temperature sensitive enhancement on the temperature field feature vector to obtain a phase transition sensitive enhanced feature vector; performing latent space reconstruction and interpolation on the phase transition sensitive enhanced feature vector based on a heat conduction partial differential equation constraint to obtain full-field temperature distribution data; and generating temperature field anomaly scores and anomaly type probability distribution information according to the full-field temperature distribution data. The application effectively solves the monitoring problem of normal surface temperature but dangerous temperature gradient in the depth direction, and improves the production qualification rate of the wind power flange forging.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent monitoring, and in particular to an intelligent monitoring method and system for forging production. BACKGROUND

[0002] The temperature field monitoring technology in the traditional wind power flange forging heat treatment process mainly relies on a dispersed thermocouple array and a simple interpolation algorithm for temperature data acquisition and processing. This method estimates the temperature distribution between sparse measurement points by arranging a limited number of temperature sensors on the surface and inside the forging, using linear interpolation or spline interpolation and other mathematical methods, and determines whether the heat treatment process is normal based on a pre-set temperature threshold range. The existing monitoring system usually uses a Cartesian coordinate system for data processing, and uses a traditional convolutional neural network or a simple statistical analysis method for temperature anomaly detection.

[0003] However, the existing technology has obvious limitations in monitoring temperature field anomalies in large ring-shaped heterogeneous structures. First, when the flange diameter reaches 4-7 meters, the data sparsity problem caused by the limited number of temperature measurement points is more prominent, and it is difficult to effectively capture the heat flow dynamic characteristics and phase change relationship inside the forging; second, the traditional monitoring method has insufficient sensitivity to small anomalies in the critical phase change temperature range of 590-720℃, and it is difficult to timely detect temperature anomalies that may cause uneven internal structure, excessive residual stress or even micro-cracks in the forging; finally, the existing system lacks effective monitoring capability for temperature gradient changes in the depth direction, and cannot identify temperature standing wave phenomena that have normal surface temperature but dangerous temperature gradient in the depth direction.

[0004] How to construct a more suitable temperature field monitoring and data processing method for the ring-shaped geometric characteristics of the wind power flange; how to achieve sensitive detection and enhancement of small anomalies in the phase change critical temperature range under the condition of data sparsity; how to combine physical constraints with deep learning to improve the accuracy and physical consistency of temperature field reconstruction; how to establish a multi-scale anomaly feature extraction mechanism to identify hidden temperature anomaly patterns including "temperature standing wave". The existence of these technical problems makes it difficult for traditional monitoring methods to timely detect potential defects, often discovering problems through quality inspection after heat treatment is completed, resulting in increased scrap rate of high-value forgings and reduced production efficiency. SUMMARY

[0005] The present application provides an intelligent monitoring method and system for forging production, which effectively solves the monitoring problem of dangerous temperature gradient in the depth direction when the surface temperature is normal, and improves the production qualification rate of wind power flange forgings.

[0006] In a first aspect, the present application provides an intelligent monitoring method for forging production, which comprises:

[0007] temperature data collection and polar coordinate conversion are performed on the wind power flange forgings to obtain polar coordinate temperature field data;

[0008] The polar coordinate temperature field data are input into a ring thermal field gradient decomposition network for feature extraction to obtain a temperature field feature vector;

[0009] The temperature field feature vector is subjected to phase transition critical temperature sensitive enhancement to obtain a phase transition sensitive enhanced feature vector;

[0010] Based on the constraint of the heat conduction partial differential equation, the phase transition sensitive enhanced feature vector is subjected to latent space reconstruction and interpolation to obtain full-field temperature distribution data;

[0011] The temperature field anomaly score and anomaly type probability distribution information are generated according to the full-field temperature distribution data.

[0012] Optionally, in the first implementation manner of the first aspect of the present application, the temperature data collection and polar coordinate conversion performed on the wind power flange forgings to obtain polar coordinate temperature field data comprise:

[0013] The wind power flange forgings are divided into three regions including an outer ring, a middle ring and an inner ring along the radial direction, temperature measuring points are uniformly arranged in the circumferential direction of each region, and K-type thermocouples are installed at the temperature measuring points to obtain a multi-layer temperature measuring point structure;

[0014] Real-time data collection is performed on the K-type thermocouples in the multi-layer temperature measuring point structure to obtain original temperature data, and noise elimination and smoothing processing are performed on the original temperature data to obtain filtered temperature data;

[0015] The filtered temperature data are converted from a Cartesian coordinate system to a polar coordinate system to obtain a temperature field matrix represented in polar coordinates;

[0016] Double cubic spline interpolation and standardization processing are performed on the temperature field matrix represented in polar coordinates to obtain polar coordinate temperature field data.

[0017] Optionally, in the second implementation manner of the first aspect of the present application, the inputting of the polar coordinate temperature field data into a ring thermal field gradient decomposition network for feature extraction to obtain a temperature field feature vector comprises:

[0018] The polar coordinate temperature field data are input into an input layer of a ring thermal field gradient decomposition network for feature encoding to obtain an input feature vector;

[0019] The input feature vector is sequentially processed through four polar coordinate convolution modules connected in series in the ring thermal field gradient decomposition network to obtain a multi-scale feature map, wherein each polar coordinate convolution module comprises two polar coordinate convolution layers, one batch normalization layer and one LeakyReLU activation function.

[0020] inputting the multi-scale feature map into a ring-shaped self-attention module in the ring-shaped thermal field gradient decomposition network for processing to obtain an enhanced feature map;

[0021] fusing the enhanced feature map with a feature map of a corresponding layer in the multi-scale feature map to obtain a fused multi-scale temperature gradient feature;

[0022] inputting the fused multi-scale temperature gradient feature into a global polar coordinate pooling layer in the ring-shaped thermal field gradient decomposition network for processing to obtain a temperature field feature vector.

[0023] Optionally, in a third implementation manner of the first aspect of the present application, the phase transition sensitive enhancement on the temperature field feature vector to obtain a phase transition sensitive enhanced feature vector comprises:

[0024] converting the temperature field feature vector to a temperature domain space to obtain reconstructed temperature field data, and constructing a temperature sensitivity weight function based on the reconstructed temperature field data and a phase transition temperature interval of the wind power flange forge piece;

[0025] calculating the closeness of temperature values of each point in the reconstructed temperature field data to a phase transition critical temperature based on the temperature sensitivity weight function to generate a sensitivity map, and performing enhancement processing on the sensitivity map to obtain an enhanced sensitivity map;

[0026] performing multi-scale temperature gradient calculation and fusion on the reconstructed temperature field data through polar coordinate gradient operators with different kernel sizes in a multi-scale temperature gradient pyramid to obtain a comprehensive gradient feature map;

[0027] performing pixel-level multiplication operation on the enhanced sensitivity map and the comprehensive gradient feature map to obtain a phase transition sensitive gradient map, and fusing the phase transition sensitive gradient map and the enhanced sensitivity map to obtain a phase transition sensitive enhanced feature vector.

[0028] Optionally, in a fourth implementation manner of the first aspect of the present application, the latent space reconstruction and interpolation on the phase transition sensitive enhanced feature vector based on the heat conduction partial differential equation constraint to obtain full-field temperature distribution data comprises:

[0029] inputting the phase transition sensitive enhanced feature vector into an encoder in a latent space reconstruction algorithm for dimension reduction processing to obtain a low-dimensional latent space variable;

[0030] performing upsampling and reconstruction processing on the low-dimensional latent space variable through a decoder in the latent space reconstruction algorithm to obtain first temperature field data;

[0031] performing iterative calculation of minimizing reconstruction error and physical constraint error on the first temperature field data by applying heat conduction partial differential equation constraint to obtain second temperature field data;

[0032] performing space filling and sparse area completion processing on the second temperature field data to obtain full-field temperature distribution data.

[0033] Optionally, in a fifth implementation manner of the first aspect of the present application, the performing iterative calculation of minimizing reconstruction error and physical constraint error on the first temperature field data by applying heat conduction partial differential equation constraint to obtain second temperature field data comprises:

[0034] calculating temperature time derivative and spatial derivative based on the first temperature field data, and generating temperature field partial derivative data based on the temperature time derivative and the spatial derivative;

[0035] performing processing on the material composition vector and the local temperature information of the wind power flange forging by inputting the heat diffusion coefficient network to obtain the heat diffusion coefficient in the form of spatial variation;

[0036] performing processing on the first temperature field data by inputting the heat source function network to obtain the internal heat source function;

[0037] applying heat conduction partial differential equation constraint to construct a physical constraint loss function according to the temperature field partial derivative data, the heat diffusion coefficient and the internal heat source function;

[0038] combining the physical constraint loss function with a reconstruction loss function and a regularization term into a total loss function, and performing minimum iterative optimization on the total loss function by gradient descent method to obtain second temperature field data.

[0039] Optionally, in a sixth implementation manner of the first aspect of the present application, the generating temperature field anomaly score and anomaly type probability distribution information based on the full-field temperature distribution data comprises:

[0040] performing feature calculation at macro, meso and micro scales on the full-field temperature distribution data, calculating global temperature standard deviation and radial temperature gradient at the macro scale, calculating local temperature mean, standard deviation and skewness at the meso scale, and calculating temperature change rate and temperature acceleration at the micro scale to obtain a multi-scale temperature feature set;

[0041] extracting a temperature profile along the depth direction from the full-field temperature distribution data, and calculating depth direction temperature gradient and depth gradient variance based on the temperature profile to identify temperature standing wave phenomenon that the surface temperature is normal but there is dangerous temperature gradient in the depth direction to obtain depth anomaly feature;

[0042] calculating temperature field anomaly score for each spatial point based on the multi-scale temperature feature set and the depth anomaly feature.

[0043] The area where the temperature field anomaly score exceeds the anomaly warning threshold is segmented to obtain segmented areas, and the target feature vector of each segmented area is calculated. By comparing and analyzing the target feature vector with the historical defect database, temperature field anomaly type probability distribution information including anomaly type and probability distribution is generated.

[0044] In a second aspect, the present invention provides an intelligent monitoring system for forging production, the intelligent monitoring system for forging production comprising:

[0045] The acquisition module is used to collect temperature data of wind power flange forgings and perform polar coordinate conversion to obtain polar coordinate temperature field data;

[0046] An extraction module is used to input the polar coordinate temperature field data into a ring thermal field gradient decomposition network for feature extraction to obtain a temperature field feature vector;

[0047] An enhancement module, configured to perform phase change critical temperature sensitivity enhancement on the temperature field characteristic vector to obtain a phase change sensitivity enhancement characteristic vector;

[0048] A reconstruction module is used to perform latent space reconstruction and interpolation on the phase change sensitive enhancement eigenvector based on the constraints of the heat conduction partial differential equation to obtain full-field temperature distribution data;

[0049] A generation module is used to generate temperature field anomaly scores and anomaly type probability distribution information based on the full-field temperature distribution data.

[0050] The technical scheme provided by the application, through polar coordinate conversion and polar coordinate convolution layer design, the application is specially optimized for the wind power flange annular geometry, effectively solves the defects of the traditional convolution network in processing the temperature gradient continuity of the wind power flange inner and outer ring, so that the temperature field monitoring is more in line with the characteristics of the annular structure. Through the phase change critical temperature sensitive enhancement module, the application focuses on monitoring and enhancing the small abnormalities of the wind power flange in the key phase change temperature interval of 590-720 DEG C, so that the system can more sensitively capture the temperature anomalies in the phase change process. By using the latent space reconstruction algorithm constrained by the heat conduction partial differential equation, the application organically combines deep learning with thermodynamic physical laws, ensures that the reconstructed temperature field conforms to the heat conduction physical law, and improves the accuracy of temperature prediction in the sparse data area. Through the abnormal feature extraction of macro, meso and micro three scales, the application can comprehensively analyze the temperature field anomaly, especially the identification ability of the "temperature standing wave" phenomenon is greatly improved, effectively solving the monitoring problem of the dangerous temperature gradient in the depth direction while the surface temperature is normal. The application can identify the potential defect area in the early stage, provide sufficient intervention time for the operator, thereby effectively preventing the defect from forming and improving the production qualified rate of the wind power flange forging. It can accurately locate the temperature anomaly area, especially the small temperature anomaly monitoring of the key structure area such as the bolt hole is more accurate, and provides accurate spatial positioning information for defect prevention. BRIEF DESCRIPTION OF DRAWINGS

[0051] In order to more clearly illustrate the technical scheme of the embodiments of the application, the following will briefly introduce the drawings needed to be used in the embodiment description. Obviously, the drawings in the following description are some embodiments of the application, and those skilled in the art can obtain other drawings based on these drawings without creative labor.

[0052] Figure 1 An embodiment schematic diagram of the intelligent monitoring method of the forging production in the embodiment of the application;

[0053] Figure 2 An embodiment schematic diagram of the intelligent monitoring system of the forging production in the embodiment of the application. DETAILED DESCRIPTION

[0054] An embodiment of the present invention provides an intelligent monitoring method and system for forging production. The terms "first", "second", "third", "fourth", etc. (if any) in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way are interchangeable where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or apparatus that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to these processes, methods, products or apparatus.

[0055] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1 An embodiment of the intelligent monitoring method for forging production in the embodiment of the present invention includes:

[0056] Step S101: collecting temperature data of a wind turbine flange forging and performing polar coordinate conversion to obtain polar coordinate temperature field data;

[0057] It is understandable that the execution subject of the present invention can be an intelligent monitoring system for forging production, or a terminal or a server, which is not limited here. The embodiment of the present invention is described by taking a server as the execution subject as an example.

[0058] Specifically, a temperature monitoring structure was constructed based on a spatial layout. The forging was radially divided into three regions: the outer ring, the middle ring, and the inner ring. Temperature measurement points were evenly distributed around each region and stacked in layers along the thickness direction to form a three-dimensional temperature measurement point structure. K-type thermocouples were used as sensing elements, and raw temperature data was collected in real time at a sampling frequency of 5 Hz. The Savitzky-Golay filter algorithm was used to suppress noise and smooth the collected data, eliminating high-frequency disturbances while preserving the temperature trend. Considering the axisymmetric nature of the forging, the data was converted from a Cartesian coordinate system to a polar coordinate system, expressed in the form of T(r,θ,z,t), to better align the data structure with the flange geometry. Bicubic spline interpolation was used for spatial interpolation, increasing the angular resolution to 1° and the radial resolution to 10 mm. Finally, the temperature field data was normalized and mapped to the range 0 to 1, resulting in polar coordinate temperature field data that can be directly input into the deep learning model.

[0059] Step S102: inputting polar coordinate temperature field data into an annular thermal field gradient decomposition network for feature extraction to obtain a temperature field feature vector;

[0060] Specifically, the polar coordinate temperature field tensor is input into the input layer of the network for initial feature encoding processing. This input layer converts the five-dimensional input tensor (batch size, number of channels, radial dimension, angular dimension, and depth dimension) into a unified input feature vector form, enabling the subsequent convolution module to perform spatial structure perception and local pattern extraction. After completing the preliminary encoding, the temperature field data is sequentially processed through the main network structure consisting of four serially connected polar coordinate convolution modules. Each convolution module contains two polar coordinate convolution layers, a batch normalization layer, and a LeakyReLU activation function. The convolution layer uses kernel weights designed in polar coordinate format to adapt to the periodicity of the ring structure and ensure feature continuity between 0° and 360° in the angular dimension; the batch normalization layer is used to stabilize the distribution differences between batches and improve the convergence speed of network training; the LeakyReLU activation function introduces a non-zero slope when processing negative interval data to alleviate the gradient vanishing problem. In terms of channel count, the first polar convolution module has 32 output channels, and the output channels of each subsequent module are doubled to 64, 128, and 256, respectively, forming a multi-scale feature map structure. To improve the model's sensitivity to key regions of the temperature field and its ability to express long-range dependencies, the multi-scale feature map is fed into a circular self-attention module for global enhancement. This module, through three 1×1 convolutional layers, maps the query Q, key K, and value V matrices, respectively. An attention mechanism builds relationships between spatial locations, enabling the network to generate more significant responses to temperature anomalies. The enhanced feature map is fused with the original feature map from each layer of the multi-scale convolution module. Feature concatenation is performed using skip connections, effectively combining low-level details with high-level abstract semantics, improving the model's ability to resolve complex temperature gradient fields at multiple scales. Finally, the fused multi-scale temperature gradient features are fed into a global polar pooling layer, which performs a weighted average of the three-dimensional spatial structure using polar coordinates, outputting a temperature field feature vector with the dimensions (batch size, 256, 1, 1, 1).

[0061] Step S103, performing phase change critical temperature sensitivity enhancement on the temperature field eigenvector to obtain a phase change sensitivity enhancement eigenvector;

[0062] Specifically, the temperature field feature vector is mapped and reduced to the temperature domain space through a fully connected neural network layer, restoring its spatial structure and numerical physical meaning, and outputting reconstructed temperature field data consistent with the original input spatial resolution. Based on the reconstructed temperature field data and the typical phase transition temperature interval (590℃, 650℃, 720℃) of the wind power flange steel material, a temperature sensitivity weight function in the form of a Gaussian distribution is constructed, which calculates the exponential distance between the temperature value of each spatial point and the phase transition critical temperature, and introduces a width control parameter σ to adjust the distribution range of the response curve. The response of each spatial position in the reconstructed temperature field data is calculated to obtain the original sensitivity map with a value range of 0 to 1, and the larger the value, the closer the temperature of the point to the phase transition critical temperature. To enhance the model's ability to distinguish subtle fluctuations near key points, a hyperbolic tangent activation function is used for nonlinear amplification processing of the sensitivity map, and the slope parameter β is adjusted to control the enhancement amplitude. At the same time, a multi-scale temperature gradient pyramid structure is constructed, and four different kernel sizes (3×3, 5×5, 7×7, 9×9) of polar coordinate gradient operators are applied to the reconstructed temperature field in turn to extract gradient information at different scales. These gradient results are normalized after calculation and input into the feature pyramid fusion network to form a comprehensive gradient feature map through upsampling and concatenation operations, covering temperature distribution patterns from micro-local disturbance to macro-gradient anomaly. The enhanced sensitivity map and the comprehensive gradient feature map are subjected to pixel-level multiplication operation to generate a phase transition sensitive gradient map, realizing the dual coupling of temperature change intensity and critical response weight. Finally, the phase transition sensitive gradient map and the enhanced sensitivity map are fused in the channel to obtain the final phase transition sensitive enhanced feature vector through channel weighting strategy.

[0063] Step S104, based on the heat conduction partial differential equation constraint, the phase transition sensitive enhanced feature vector is reconstructed and interpolated in the latent space to obtain the full-field temperature distribution data;

[0064] Specifically, the phase transition sensitive enhanced feature vector is input into the encoder module of the latent space reconstruction algorithm for dimension compression processing. The encoder adopts a multi-layer polar coordinate convolution structure, and extracts deep semantic information layer by layer through down-sampling operation with a step of 2, compresses the data dimension while preserving the spatial structure features, maps the input high-dimensional features to a low-dimensional latent space with a dimension of 128, and the latent variable retains important features such as temperature gradient, sensitivity enhancement and structure boundary. The low-dimensional latent space variable is input into the decoder module, and the inverse expansion is performed through the multi-layer up-sampling operation of the symmetric structure. The decoder uses transpose polar coordinate convolution combined with instance normalization and PReLU activation function to gradually reconstruct the spatial distribution features and restore the first temperature field data consistent with the original spatial resolution.

[0065] The first temperature field data is subjected to iterative calculation of minimizing reconstruction error and physical constraint error by applying heat conduction partial differential equation constraints. The optimization process is based on the heat conduction control equation, which simulates the diffusion evolution behavior of the temperature field by finite difference approximation of forward difference and central difference. The model introduces a spatially varying form of the thermal diffusion coefficient function, which is inferred by a perception network based on local temperature and material chemical composition parameters (C, Si, Mn, Cr, Mo, Ni) to reflect the changes in thermal diffusion rate in different regions. The internal heat source term is modeled by a neural network based on the U-Net architecture, which outputs the spatial heat source distribution based on the historical temperature sequence and the current temperature state. By alternately updating the reconstruction result and the equation constraint target, the model iteratively obtains the second temperature field data that meets the fitting degree of the observation data and the physical law of heat conduction.

[0066] Based on the first temperature field data, the partial derivative is calculated by calculating the temperature time derivative by forward difference and the second derivative of the three-dimensional grid in space based on the central difference format, to form the temperature field partial derivative data. The material composition vector and local temperature information are input into the thermal diffusion coefficient network, which uses a multilayer perception machine structure to output a spatially varying form of the thermal diffusion coefficient. The first temperature field data is input into the heat source function network, which uses the U-Net architecture to fuse the current temperature state and the historical temperature evolution sequence to output an internal heat source function with spatio-temporal correlation characteristics. The temperature partial derivative data, thermal diffusion coefficient and internal heat source function are introduced into the heat conduction partial differential equation to construct a physical constraint loss function, which is combined with the reconstruction loss function and the L1 regularization term to form the total loss function. Gradient descent method is used for minimizing iterative optimization, so that the model converges to the direction of improving reconstruction accuracy and enhancing physical consistency at the same time, and finally obtains the second temperature field data. The model performs spatial filling and sparse region completion processing based on the second temperature field data, combining the latent space interpolation capability and the spatial continuity learning mechanism, and finally outputs the full-field temperature distribution data.

[0067] Step S105, generating temperature field anomaly score and anomaly type probability distribution information according to full-field temperature distribution data.

[0068] Specifically, the full-field temperature distribution data is subjected to multi-scale spatial analysis to reveal temperature anomaly characteristics at different levels. On the macro scale, the overall temperature standard deviation is calculated to reflect the uniformity of the heat field, and the radial temperature gradient is calculated to capture the spatial offset trend of large-scale heat distribution; on the mesoscale, the local area temperature is scanned by the sliding window method to calculate the temperature mean, standard deviation and skewness of each local area to capture regional thermal anomalies such as local overheating or uneven cooling; on the micro scale, the temperature change rate and temperature acceleration of each spatial point are analyzed to identify abnormal temperature fluctuations or short-term drastic changes, forming a multi-scale temperature feature set.

[0069] To identify the "temperature standing wave" hidden danger (normal surface temperature but sharp temperature change in the depth direction), temperature profiles are extracted from the full-field temperature distribution data at multiple radial and angular positions to analyze the temperature variation in the thickness direction, and the temperature gradient in the depth direction and its variance are calculated. When a large temperature change is found below the stable surface temperature, it is identified as a depth anomaly feature.

[0070] Based on the multi-scale temperature feature set and the depth anomaly feature, a multivariate anomaly evaluation model is used to calculate the temperature field anomaly score for each spatial point. The higher the score, the more abnormal the temperature behavior. For points with an anomaly score higher than the warning threshold, a density clustering algorithm is applied to segment the abnormal regions. Each abnormal region is encoded as a target feature vector, including parameters such as average anomaly score, maximum anomaly value, area range, maximum temperature deviation, temperature gradient amplitude, and depth gradient statistics. Through similarity calculation and multi-dimensional matching with a historical defect database, each abnormal region is assigned a specific anomaly type label, and a probability distribution is generated based on the matching confidence, forming temperature field anomaly type probability distribution information.

[0071] In a specific embodiment, the process of step S101 can specifically include the following steps:

[0072] The wind power flange forging is divided into three regions including an outer ring, a middle ring and an inner ring along a radial direction, temperature measuring points are uniformly arranged in a circumferential direction of each region, and K-type thermocouples are installed at the temperature measuring points to obtain a multi-layer temperature measuring point structure;

[0073] Real-time data acquisition is performed on the K-type thermocouples in the multi-layer temperature measuring point structure to obtain original temperature data, and noise elimination and smoothing processing are performed on the original temperature data to obtain filtered temperature data;

[0074] The filtered temperature data are converted from a Cartesian coordinate system to a polar coordinate system to obtain a temperature field matrix represented in polar coordinates;

[0075] The temperature field matrix represented in polar coordinates is subjected to bicubic spline interpolation and standardization processing to obtain polar coordinate temperature field data.

[0076] Specifically, based on the circular ring axial symmetry characteristic of the wind power flange forging, the radial partition method is used to divide it into three thermal characteristic regions of the outer ring, the middle ring and the inner ring. The temperature measuring points are uniformly arranged in the circumferential direction of each ring at equal angles to ensure consistent angular data acquisition density, and the temperature measuring points are laid out layer by layer in the thickness direction at a fixed interval to form a three-dimensional multi-layer temperature measuring point system in the radial, angular and depth directions. K-type thermocouples are installed at each node as temperature sensing units. The K-type thermocouples have good high-temperature adaptability, stability and anti-interference ability, and are suitable for accurate temperature measurement of different positions, depths and heating stages of large-diameter, heavy-mass and high-thermal-capacity forgings such as wind power flanges.

[0077] All K-type thermocouples are connected to a high-frequency data acquisition system, and the sampling frequency is set to 5 Hz to realize the tracking of temperature changes with a resolution of seconds. Due to the influence of thermal radiation disturbance, electromagnetic interference, wire thermal inertia and thermocouple contact error in the measurement environment, the original temperature data contains high-frequency noise and local jitter, so the Savitzky-Golay filtering algorithm is used for smoothing processing, and the short-period fluctuations are eliminated by polynomial fitting within the local sliding window while the overall trend is preserved.

[0078] Considering the axisymmetric characteristics of the wind power flange, the filtered three-dimensional temperature measurement data is converted from the Cartesian coordinate system (x, y, z) to the polar coordinate system (r, θ, z), where r is the radial position, θ is the angular position, and z is the thickness direction. The polar coordinate temperature field matrix is interpolated by bicubic spline interpolation, and the continuous change relationship between data points in the local plane is reconstructed by bicubic polynomial fitting, which improves the angular temperature distribution accuracy to 1° and the radial accuracy to 10 mm. Finally, the interpolated high-resolution polar coordinate temperature field matrix is standardized, and the temperature value is transformed to the interval of 0 to 1 by linear mapping, using the maximum and minimum values of the historical sample data as the normalization boundary to improve the training stability and model generalization ability, and outputting the polar coordinate temperature field data.

[0079] In a specific embodiment, the process of performing step S102 can specifically include the following steps:

[0080] The polar coordinate temperature field data is input into the input layer of the annular thermal field gradient decomposition network for feature encoding to obtain an input feature vector;

[0081] The input feature vector is sequentially processed by four polar coordinate convolution modules connected in series in the annular thermal field gradient decomposition network to obtain a multi-scale feature map, wherein each polar coordinate convolution module includes two polar coordinate convolution layers, a batch normalization layer and a LeakyReLU activation function;

[0082] The multi-scale feature map is input into the annular self-attention module in the annular thermal field gradient decomposition network for processing to obtain an enhanced feature map;

[0083] The enhanced feature map is fused with the feature map of the corresponding layer in the multi-scale feature map to obtain a fused multi-scale temperature gradient feature;

[0084] The fused multi-scale temperature gradient feature is input into the global polar coordinate pooling layer in the annular thermal field gradient decomposition network for processing to obtain a temperature field feature vector.

[0085] Specifically, the polar coordinate temperature field data is input into the annular thermal field gradient decomposition network as a five-dimensional tensor (batch_size, 1, Hr, Hq, Hz), where 1 represents a single-channel temperature value, Hr, Hq, and Hz represent the spatial dimensions in the radial, angular, and depth directions, respectively. The input layer performs low-level feature extraction on the original temperature field through polar coordinate format convolution operation, and completes the encoding of the input feature vector from pure temperature value to directional gradient and local response capability.

[0086] The input feature vector is sequentially processed by four polar coordinate convolution modules connected in series, each module including two polar coordinate convolution layers, a batch normalization layer, and a LeakyReLU activation function. The polar coordinate convolution operation adopts a weight convolution kernel consistent with the geometric characteristics of the polar coordinate, sets an annular padding strategy in the angular dimension to preserve the periodic continuity in the q direction, and uses a reflection padding in the radial dimension to avoid boundary gradient loss. The output channel number of the first convolution module is 32, and then each layer is doubled in turn to form a step-by-step enhancement structure of 64, 128, and 256, generating a set of multi-scale feature maps.

[0087] To strengthen the model's ability to identify local thermal anomalies and model long-range dependencies, the multi-scale feature maps are input into the annular self-attention module. This module generates query Q, key K, and value V matrices through three independent 1x1 convolution layers, performs core attention mechanism calculation: matrix point multiplication captures similarity relationship, softmax normalization obtains attention weight, and multiplication with value matrix obtains enhanced feature map. Due to the use of polar coordinate feature structure, the attention mechanism naturally preserves the angular periodicity and radial distribution regularity, enabling the enhanced feature map to accurately focus on the local area of abnormal thermal gradient.

[0088] The network is designed with a multi-scale skip connection fusion mechanism to fuse the enhanced feature map with the corresponding hierarchical feature map output by the four convolution modules, perform a splicing operation in the channel dimension, and combine a channel attention modulation strategy to allocate fusion weights, thereby constructing a fused multi-scale temperature gradient feature set with global vision and local recognition accuracy. Finally, the fused feature set is input into the global polar coordinate pooling layer, which uses a polar coordinate weight mechanism to weight and compress the entire spatial domain feature tensor, outputting a temperature field feature vector with a dimension of (batch_size, 256, 1, 1, 1).

[0089] In a specific embodiment, the process of performing step S103 can specifically include the following steps:

[0090] The temperature field feature vector is converted to the temperature domain space to obtain reconstructed temperature field data, and a temperature sensitivity weight function is constructed based on the reconstructed temperature field data and the phase transition temperature interval of the wind power flange forge piece;

[0091] Based on the temperature sensitivity weight function, the proximity between the temperature value of each point in the reconstructed temperature field data and the critical temperature of the phase change is calculated to generate a sensitivity map, and the sensitivity map is enhanced to obtain an enhanced sensitivity map;

[0092] Multi-scale temperature gradient calculation and fusion are performed on the reconstructed temperature field data through polar coordinate gradient operators with different kernel sizes in the multi-scale temperature gradient pyramid to obtain a comprehensive gradient feature map.

[0093] The enhanced sensitivity map is multiplied pixel-wise with the comprehensive gradient feature map to obtain a phase change sensitive gradient map, and the phase change sensitive gradient map is fused with the enhanced sensitivity map to obtain a phase change sensitive enhanced feature vector.

[0094] Specifically, the temperature field feature vector is remapped back to the actual temperature domain space, thus recovering its physical meaning and constructing a feature expression form that can directly perceive the critical state of the material. The mapping process relies on a fully connected mapping network to transform the temperature field feature vector through multiple neural layers into a spatial data tensor of the same dimension as the original polar coordinate temperature field, obtaining reconstructed temperature field data with physical meaning. In order to enable the model to focus on the phase transition temperature interval specific to the wind power flange heat treatment process, i.e., the key temperature points corresponding to the structural evolution stages such as austenitizing, ferrite-pearlite transformation, etc., such as 590 degrees Celsius, 650 degrees Celsius, and 720 degrees Celsius, a set of temperature sensitivity weight functions centered on the phase transition critical temperature is constructed. The function is modeled in the form of a Gaussian distribution, which sets a temperature center point and a standard deviation width parameter σ, so that the closer any spatial point is to the critical temperature, the higher its sensitivity value will be, and when the temperature deviates from the center point, its weight will rapidly decrease. The numerical calculation result of the temperature sensitivity weight function falls between 0 and 1, forming a smooth response curve with a peak response to the critical temperature and a rapid decay to the non-critical interval. The system applies this function to each spatial location of the reconstructed temperature field data, calculating the proximity of its temperature value to each phase transition critical temperature point by point, forming a sensitivity map that reflects which regions in the entire heat treatment domain are most likely to be at the critical edge of structural phase transition. Although the basic sensitivity map has numerical response capability to the critical temperature, its numerical changes may still be inhibited by the temperature gradient smoothing effect, resulting in insufficient sensitivity to small fluctuations near the critical point, so further enhancement processing is performed. A nonlinear activation strategy is used to amplify the sensitivity map, and the hyperbolic tangent function is chosen as the main nonlinear enhancement means, and by adjusting the slope parameter of the function, the response of the sensitive area is amplified, and the enhanced sensitivity map is obtained. In this map, spatial points close to the phase transition critical temperature will have a significantly amplified response, thus forming a higher recognition in numerical terms, enabling the model to more focusedly identify small temperature fluctuations related to phase transition in subsequent processing. At the same time, in order to enhance the model's ability to analyze temperature change patterns, a multi-scale temperature gradient pyramid structure is introduced, which captures gradient information of temperature distribution changes at different scales through the structure. The pyramid model sets up polar coordinate gradient operators with different kernel sizes to calculate the gradient of the reconstructed temperature field data using convolution kernels of 3x3, 5x5, 7x7, and 9x9. Each polar coordinate gradient operator estimates the directional gradient of the temperature change in the local neighborhood based on its kernel center, while preserving the angular periodicity and radial boundary symmetry. The gradient maps output by each layer reflect the temperature change trend at different scales, with small-scale kernels perceiving local disturbances and large-scale kernels capturing global temperature difference distributions.The gradient map features from different scales are input into a feature fusion module, which combines multiple layer outputs into a comprehensive gradient feature map by upsampling and channel-wise weighting strategy. The map reflects the micro-local and macro-global heat distribution state in space and has high structural expression ability. The enhanced sensitivity map and the comprehensive gradient feature map are subjected to pixel-level multiplication operation to construct a phase change sensitive gradient map. The operation essentially couples the spatial temperature change rate and the response ability to the phase change critical temperature, so that the spatial points with strong gradient change and in the critical temperature interval obtain higher response values in the map, while the non-critical regions will not be highlighted even if there is temperature change, thereby focusing the model's attention on the locations with the highest evolution risk. In order to integrate the response features with high physical sensitivity back into the main feature channel, the phase change sensitive gradient map and the enhanced sensitivity map are combined through a channel fusion mechanism. The fusion process is realized by channel weighting superposition, that is, the global contour information of the sensitivity map responding to the critical temperature is retained, and the directional information of the gradient map on the change trend is injected, to form the final output of the phase change sensitive enhanced feature vector.

[0095] In a specific embodiment, the process of performing step S104 can specifically include the following steps:

[0096] The phase change sensitive enhanced feature vector is input into an encoder in the latent space reconstruction algorithm for dimension reduction processing to obtain a low-dimensional latent space variable;

[0097] The low-dimensional latent space variable is subjected to upsampling and reconstruction processing by a decoder in the latent space reconstruction algorithm to obtain first temperature field data;

[0098] The first temperature field data is subjected to iterative calculation of minimizing reconstruction error and physical constraint error by applying a heat conduction partial differential equation constraint to obtain second temperature field data;

[0099] The second temperature field data is subjected to spatial filling and sparse region completion processing to obtain full-field temperature distribution data.

[0100] Specifically, the phase-sensitive enhanced feature vector is input into the encoder part of the latent space structure, which adopts a multi-level polar coordinate convolutional neural network. Through successive convolution, normalization and nonlinear activation operations, the spatial and channel dimensions of the input features are gradually compressed, and their gradient characteristics, spatial periodicity and phase-sensitive information are preserved in the process. The high-dimensional tensor is mapped to a low-dimensional latent space variable. The dimension of the latent space is set to between 128 and 256, and the compressed tensor contains the spatial layout compression structure of the temperature field, as well as the material thermal property characteristics, temperature gradient information and phase transition zone response, thereby realizing the effective transition from physical perception to structural abstraction. In the decoder part, a set of polar coordinate transpose convolutional network structures symmetric to the encoder is used to upsample the low-dimensional latent space variable and reconstruct the temperature field. Through transpose convolution operations, the compressed information is gradually restored to the spatial scale of the original temperature field, during which instance normalization and PReLU activation functions are used to ensure the preservation of gradient continuity and temperature change smoothness while amplifying spatial information. The decoding process not only restores the dimension, but also maps and restores the physical state, because the features carried in the latent space encoding have been endowed with the abstract rules of temperature evolution and material behavior by network training. The first temperature field data generated in this way restores the complete three-dimensional spatial structure in form and effectively predicts the real temperature evolution process in overall trend, but since the prediction process mainly relies on the feature reconstruction ability and has not been forced to follow the partial differential equation model of heat conduction, there are local fluctuations that violate physical laws in the details, especially near the heat diffusion boundary, heat source disturbance point or sparse temperature measurement area. To improve the physical consistency of the reconstructed data, the first temperature field data is input into the physical constraint optimization module based on the heat conduction partial differential equation as the initial solution, and a joint loss function is constructed, which considers the reconstruction error term, the physical partial derivative residual term and the latent space regularization term. The physical constraint part is obtained by discretizing and approximating the heat conduction master equation, calculating the first-order derivative of the temperature field with respect to time and the second-order derivatives with respect to space (radial, angular and depth), and combining the material thermal diffusion coefficient function and the internal heat source term to construct the residual term. The thermal diffusion coefficient is modeled in a spatially varying manner, which is dynamically output by a perception machine network according to the local temperature and material composition parameters; while the internal heat source function is input into the U-Net structure by the historical temperature sequence and the current temperature state to generate, which is used to describe the non-uniform energy input process in the actual heat treatment process. The above partial derivatives are realized by finite difference approximation, so that the model directly optimizes the partial differential equation residual in the training framework. In the optimization process, the total loss function is taken as the target, and the model parameters are updated iteratively by gradient descent method, so that the generated second temperature field data fits the original observation data as much as possible while meeting the physical constraints of the heat conduction equation.The essence of this stage is to find the temperature field evolution trajectory most consistent with the observation in the solution space of the physical law by guiding the model, and the output result exhibits higher physical rationality in local continuity, heat conduction propagation direction, and abnormal point diffusion path, thereby overcoming the limitations of purely data-driven models in non-uniform materials, complex boundaries, and transient dynamic response. The second temperature field data is processed by spatial filling and sparse area completion. The heat diffusion law and neighborhood relationship learned in the latent space by the heat conduction constraint model are comprehensively utilized, and a structure-preserving interpolation method is used to fill in the missing areas. The specific way includes propagating and extending the gradient direction of the adjacent known areas, and smoothing fitting the reconstructed boundary through spatial consistency rules, so that the interpolation result is not only numerically continuous, but also conforms to the natural boundary propagation trend in heat diffusion behavior, and finally the full-field temperature distribution data is obtained.

[0101] In a specific embodiment, the process of performing step applying heat conduction partial differential equation constraint to perform iterative calculation of minimizing reconstruction error and physical constraint error on the first temperature field data to obtain the second temperature field data can specifically include the following steps:

[0102] Based on the first temperature field data, calculate the temperature time derivative and spatial derivative, and generate temperature field partial derivative data based on the temperature time derivative and spatial derivative;

[0103] The material composition vector and local temperature information of the wind power flange forging are input into the heat diffusion coefficient network for processing to obtain the spatial variation form of the heat diffusion coefficient;

[0104] The first temperature field data is input into the heat source function network for processing to obtain the internal heat source function;

[0105] Applying heat conduction partial differential equation constraint, constructing physical constraint loss function according to temperature field partial derivative data, heat diffusion coefficient and internal heat source function;

[0106] Combining the physical constraint loss function with the reconstruction loss function and the regularization term into the total loss function, minimizing and iteratively optimizing the total loss function by gradient descent method to obtain the second temperature field data.

[0107] Specifically, the key partial derivative terms, i.e., the first-order derivative of temperature with respect to time and the second-order derivatives with respect to each spatial dimension, are calculated based on the first temperature field data. The time derivative part is used to reflect the transient change characteristics of the temperature field at different times, and its value is approximated by forward difference, i.e., the instantaneous change rate is obtained by dividing the difference between the temperature values of adjacent time frames by the time interval. The spatial derivative part is numerically approximated by central difference, in which the second-order derivatives in the radial, angular, and thickness directions are used to describe the heat diffusion behavior in space. These spatial derivatives can depict the heat conduction direction, heat flux variation, and temperature gradient change trend in the local region, thereby forming a set of temperature field partial derivative data. The microscopic composition factors of the wind power flange material are introduced because the thermal diffusion capacity of the material not only varies with space but also significantly depends on its composition and temperature state. The vector reflecting the chemical composition of the material (such as the mass percentage of elements C, Si, Mn, Cr, Mo, Ni, etc.) and the temperature value of each spatial point in the first temperature field are input into the pre-trained thermal diffusion coefficient network. This network is based on a multi-layer perceptron structure, and its output is a spatially varying thermal diffusion coefficient, i.e., the thermal diffusion coefficient dynamically changes with the spatial coordinates, thereby truly reflecting the non-uniformity of thermal diffusion caused by composition unevenness, different heat treatment histories, and microstructure differences in complex forgings. This output not only improves the fitting ability of the thermal diffusion field distribution but also enables the model to have stronger explanatory ability for material behavior at the physical level. After determining the thermal diffusion coefficient to characterize the regulatory effect of internal energy input on thermal behavior, a heat source function is constructed, which represents the heat generation per unit volume per unit time. Since this heat source term is related to not only time variation but also historical temperature state, current process parameters, and furnace thermal environment, a deep neural network with U-Net structure is used as the heat source function prediction model. The input includes the first temperature field distribution at the current time point and its historical temperature sequence at several previous time steps, and the output is a spatially distributed heat source intensity tensor. This model preserves the spatial distribution information through down-sampling encoding and up-sampling decoding structure, and extracts multi-scale features through jump connection, so that the output heat source function not only has spatio-temporal consistency but also can reflect the non-steady-state heat source changes caused by local temperature difference, thermal radiation interference, or phase change exothermicity. Based on the main partial differential equation of heat conduction, a residual expression is constructed, and a physical constraint loss function is defined. In actual modeling, the combination of the above partial derivatives, thermal diffusion coefficient, and internal heat source term is considered as the expected physical relationship, which is compared with the actual derivative characteristics shown by the first temperature field. The difference is the deviation of the current model output from the heat conduction law, which is integrated in space to form the loss function. This physical constraint loss function measures the consistency deviation between the reconstructed temperature field and the actual thermal diffusion process. If this value is smaller, it means that the current temperature evolution is more consistent with the physical control law; if the value is larger, it means that the model has serious unreasonable areas that violate the thermodynamic equilibrium or heat diffusion process.Meanwhile, to ensure the model still maintains high fitting accuracy to the observation data beyond the physical constraints, a data reconstruction loss is introduced into the total loss function, which measures the numerical deviation between the first temperature field and the actual observation value; meanwhile, a latent space regularization term is introduced to impose regularization on the latent space variable output by the encoder, preventing the latent variable distribution from being too dispersed, improving the convergence stability and generalization ability of the model. The three are combined to form the total loss function. After constructing the complete loss function, an optimization method based on the gradient descent strategy is used to train all the learnable parameters of the neural network through backpropagation, and the network weights are updated iteratively, so that the model gradually tends to an optimal state that takes into account both physical laws and observation data in each iteration. In this process, the model corrects the local disturbance in the first temperature field that violates the heat diffusion law and optimizes the heat gradient distribution in the entire spatial domain, so that the final output of the second temperature field data exhibits high consistency and physical interpretability in terms of time evolution, spatial propagation, and boundary continuity.

[0108] In a specific embodiment, the process of performing step S105 can specifically include the following steps:

[0109] The full-field temperature distribution data is calculated for macro, meso, and micro three scales, the global temperature standard deviation and the radial temperature gradient are calculated at the macro scale, the local temperature mean, standard deviation, and skewness are calculated at the meso scale, and the temperature rate of change and temperature acceleration are calculated at the micro scale to obtain a multi-scale temperature feature set;

[0110] The temperature profile along the depth direction is extracted from the full-field temperature distribution data, and the depth direction temperature gradient and depth gradient variance are calculated based on the temperature profile to identify the temperature standing wave phenomenon that the surface temperature is normal but there is a dangerous temperature gradient in the depth direction, and obtain a depth anomaly feature;

[0111] Based on the multi-scale temperature feature set and the depth anomaly feature, the temperature field anomaly score of each spatial point is calculated;

[0112] The regions with temperature field anomaly scores exceeding the anomaly warning threshold are segmented to obtain segmented regions, and the target feature vector of each segmented region is calculated, and the temperature field anomaly type probability distribution information containing the anomaly type and probability distribution is generated by comparing and analyzing the target feature vector with the historical defect database.

[0113] Specifically, on the macro scale, the global temperature standard deviation is calculated to measure the degree of thermal field dispersion of the entire forging, and the temperature gradient is calculated in the radial direction, and by extracting the temperature difference rate of change along the inner ring to the outer ring direction, the global gradient image reflecting the main diffusion path of heat flow and the steepness of the gradient is obtained. On the mesoscale, a sliding window mechanism (30° angle, 100mm radial, 50mm thickness) is introduced to extract local statistical features in different regions, including local temperature mean, standard deviation and skewness, to describe the heating strength, fluctuation amplitude and distribution symmetry of the local region. On the micro scale, the first and second order difference operations are performed on the temperature time series of each space point to calculate the temperature rate of change and temperature acceleration, which describes the short-time dynamic behavior in the heat treatment process.

[0114] Depth direction anomaly identification is performed according to the typical structural risk characteristics of the wind power flange. Starting from each angle θ and radial position r, the temperature profile curve is extracted along the thickness direction from the surface to the inner layer, and the temperature gradient and its variance in the depth direction are calculated as the depth gradient non-uniformity index. When the surface temperature is stable but the deep layer appears sharp temperature gradient change and there is severe fluctuation, it is marked as "temperature standing wave" phenomenon, indicating that there may be potential defects such as stress concentration, insufficient material organization transformation or blocked heat diffusion in the forging.

[0115] Based on the complete temperature feature set composed of macro, meso, micro and depth anomaly characteristics, a weighted comprehensive method is used to calculate the temperature field anomaly score of each space point. This process considers global statistical fluctuations, local structural disturbances, dynamic anomaly responses and depth temperature mutations, and assigns different weights to various features through a feature weighting model, and outputs a normalized anomaly score value. With the early warning threshold as the boundary, the points with anomaly scores higher than the threshold are clustered in space using a density clustering algorithm (such as DBSCAN), and a continuous anomaly region set is obtained.

[0116] Each anomaly region is structured and coded, and a target feature vector containing multiple attributes such as average anomaly score, maximum anomaly value, spatial area, maximum temperature deviation value, maximum temperature gradient, maximum depth gradient, anomaly distribution directionality and time dynamic characteristics is calculated. The vector is input into an anomaly type recognition system based on a historical defect database, and is matched with historical defect templates through K nearest neighbor matching, distance function fitting or atlas similarity calculation, and the matching strength is output to output the anomaly type classification result and probability distribution information.

[0117] The above describes the intelligent monitoring method for the production of forgings in the embodiments of the present application, and the intelligent monitoring system for the production of forgings in the embodiments of the present application is described below. Please refer to Figure 2 An embodiment of the intelligent monitoring system for the production of forgings in the embodiments of the present application includes:

[0118] The acquisition module 201 is configured to acquire temperature data of the wind power flange forge piece and perform polar coordinate conversion to obtain polar coordinate temperature field data.

[0119] The extraction module 202 is configured to input the polar coordinate temperature field data into a ring thermal field gradient decomposition network to perform feature extraction, and obtain a temperature field feature vector.

[0120] The enhancement module 203 is configured to perform phase transition critical temperature sensitive enhancement on the temperature field feature vector to obtain a phase transition sensitive enhanced feature vector.

[0121] The reconstruction module 204 is configured to perform latent space reconstruction and interpolation on the phase transition sensitive enhanced feature vector based on a heat conduction partial differential equation constraint to obtain full-field temperature distribution data.

[0122] The generation module 205 is configured to generate temperature field anomaly score and anomaly type probability distribution information according to the full-field temperature distribution data.

[0123] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described system, system and unit can refer to the corresponding processes in the foregoing method embodiments, which will not be described here.

[0124] The integrated unit, if realized in the form of a software function unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application or the whole or part of the technical solutions that essentially contribute to the prior art can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a smart monitoring device for a forge piece production (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0125] The above-described embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacements for some technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. An intelligent monitoring method for forging production, characterized in that: include: Perform temperature data collection and polar coordinate conversion on wind turbine flange forgings to obtain polar coordinate temperature field data; Inputting the polar coordinate temperature field data into the annular thermal field gradient decomposition network for feature extraction to obtain a temperature field feature vector, including: inputting the polar coordinate temperature field data into the input layer of the annular thermal field gradient decomposition network for feature encoding to obtain an input feature vector; processing the input feature vector in turn through four series-connected polar coordinate convolution modules in the annular thermal field gradient decomposition network to obtain a multi-scale feature map, wherein each polar coordinate convolution module includes two polar coordinate convolution layers, a batch normalization layer and a LeakyReLU activation function; inputting the multi-scale feature map into the annular self-attention module in the annular thermal field gradient decomposition network for processing to obtain an enhanced feature map; fusing the enhanced feature map with the feature map of the corresponding layer in the multi-scale feature map to obtain a fused multi-scale temperature gradient feature; inputting the fused multi-scale temperature gradient feature into the global polar coordinate pooling layer in the annular thermal field gradient decomposition network for processing to obtain a temperature field feature vector; Performing phase change critical temperature sensitivity enhancement on the temperature field feature vector to obtain a phase change sensitivity enhanced feature vector, including: converting the temperature field feature vector into a temperature domain space to obtain reconstructed temperature field data, and constructing a temperature sensitivity weight function based on the reconstructed temperature field data and the phase change temperature range of the wind power flange forging; calculating the proximity of the temperature value of each point in the reconstructed temperature field data to the phase change critical temperature based on the temperature sensitivity weight function to generate a sensitivity map, and performing enhancement processing on the sensitivity map to obtain an enhanced sensitivity map; performing multi-scale temperature gradient calculation and fusion on the reconstructed temperature field data by using polar coordinate gradient operators of different kernel sizes in a multi-scale temperature gradient pyramid to obtain a comprehensive gradient feature map; performing a pixel-level product operation on the enhanced sensitivity map and the comprehensive gradient feature map to obtain a phase change sensitive gradient map, and fusing the phase change sensitive gradient map with the enhanced sensitivity map to obtain a phase change sensitive enhanced feature vector; Based on the constraints of the heat conduction partial differential equation, the phase change sensitive enhancement eigenvector is reconstructed and interpolated in the latent space to obtain the full-field temperature distribution data; A temperature field anomaly score and anomaly type probability distribution information are generated according to the full-field temperature distribution data.

2. The intelligent monitoring method for forging production according to claim 1, characterized in that: The temperature data of the wind turbine flange forging is collected and polar coordinate converted to obtain polar coordinate temperature field data, including: The wind turbine flange forging is radially divided into three regions including an outer ring, a middle ring, and an inner ring. Temperature measuring points are evenly arranged circumferentially in each region, and K-type thermocouples are installed at the temperature measuring points to obtain a multi-layer temperature measuring point structure. Performing real-time data acquisition on the K-type thermocouples in the multi-layer temperature measurement point structure to obtain raw temperature data, and performing noise elimination and smoothing processing on the raw temperature data to obtain filtered temperature data; Converting the filtered temperature data from a Cartesian coordinate system to a polar coordinate system to obtain a temperature field matrix represented by polar coordinates; Bicubic spline interpolation and standardization are performed on the temperature field matrix represented by polar coordinates to obtain polar coordinate temperature field data.

3. The intelligent monitoring method for forging production according to claim 1, characterized in that: Based on the constraints of the heat conduction partial differential equation, the phase change sensitive enhancement eigenvector is reconstructed and interpolated in the latent space to obtain the full-field temperature distribution data, including: Inputting the phase change sensitive enhanced feature vector into an encoder in a latent space reconstruction algorithm for dimensionality reduction processing to obtain a low-dimensional latent space variable; Upsampling and reconstructing the low-dimensional latent space variables by a decoder in the latent space reconstruction algorithm to obtain first temperature field data; Applying a heat conduction partial differential equation constraint to perform an iterative calculation of minimizing a reconstruction error and a physical constraint error on the first temperature field data to obtain second temperature field data; Perform space filling and sparse region completion processing on the second temperature field data to obtain full-field temperature distribution data.

4. The intelligent monitoring method for forging production according to claim 3, characterized in that: The applying of the heat conduction partial differential equation constraint to perform iterative calculation of minimizing reconstruction error and physical constraint error on the first temperature field data to obtain second temperature field data includes: Calculating a temperature time derivative and a space derivative based on the first temperature field data, and generating temperature field partial derivative data based on the temperature time derivative and the space derivative; The material composition vector and local temperature information of the wind turbine flange forging are input into the thermal diffusion coefficient network for processing to obtain the thermal diffusion coefficient in spatial variational form. Inputting the first temperature field data into a heat source function network for processing to obtain an internal heat source function; Applying heat conduction partial differential equation constraints to construct a physical constraint loss function according to the temperature field partial derivative data, the thermal diffusion coefficient and the internal heat source function; The physical constraint loss function, the reconstruction loss function and the regularization term are combined into a total loss function, and the total loss function is minimized and iteratively optimized using a gradient descent method to obtain second temperature field data.

5. The intelligent monitoring method for forging production according to claim 1, characterized in that: Generating the temperature field anomaly score and anomaly type probability distribution information according to the full-field temperature distribution data includes: Performing feature calculations on the full-field temperature distribution data at three scales, namely, macroscopic, mesoscopic, and microscopic, calculating the global temperature standard deviation and radial temperature gradient at the macroscopic scale, calculating the local temperature mean, standard deviation, and skewness at the mesoscopic scale, and calculating the temperature change rate and temperature acceleration at the microscopic scale, to obtain a multi-scale temperature feature set; Extracting a temperature profile along the depth direction of the full-field temperature distribution data, and calculating the depth direction temperature gradient and depth gradient variance based on the temperature profile, identifying the temperature standing wave phenomenon where the surface temperature is normal but a dangerous temperature gradient exists in the depth direction, and obtaining the depth anomaly feature; Calculating a temperature field anomaly score for each spatial point based on the multi-scale temperature feature set and the depth anomaly feature; The area where the temperature field anomaly score exceeds the anomaly warning threshold is segmented to obtain segmented areas, and the target feature vector of each segmented area is calculated. By comparing and analyzing the target feature vector with the historical defect database, temperature field anomaly type probability distribution information including anomaly type and probability distribution is generated.

6. An intelligent monitoring system for forging production, characterized in that: A method for implementing the intelligent monitoring method for forging production according to any one of claims 1 to 5, wherein the intelligent monitoring system for forging production comprises: The acquisition module is used to collect temperature data of wind power flange forgings and perform polar coordinate conversion to obtain polar coordinate temperature field data; An extraction module is used to input the polar coordinate temperature field data into a circular thermal field gradient decomposition network for feature extraction to obtain a temperature field feature vector, including: inputting the polar coordinate temperature field data into the input layer of the circular thermal field gradient decomposition network for feature encoding to obtain an input feature vector; processing the input feature vector in turn through four series-connected polar coordinate convolution modules in the circular thermal field gradient decomposition network to obtain a multi-scale feature map, wherein each polar coordinate convolution module includes two polar coordinate convolution layers, a batch normalization layer and a LeakyReLU activation function; inputting the multi-scale feature map into the circular self-attention module in the circular thermal field gradient decomposition network for processing to obtain an enhanced feature map; fusing the enhanced feature map with the feature map of the corresponding layer in the multi-scale feature map to obtain a fused multi-scale temperature gradient feature; inputting the fused multi-scale temperature gradient feature into the global polar coordinate pooling layer in the circular thermal field gradient decomposition network for processing to obtain a temperature field feature vector; An enhancement module is used to perform phase change critical temperature sensitivity enhancement on the temperature field feature vector to obtain a phase change sensitivity enhanced feature vector, including: converting the temperature field feature vector into a temperature domain space to obtain reconstructed temperature field data, and constructing a temperature sensitivity weight function based on the reconstructed temperature field data and the phase change temperature range of the wind power flange forging; calculating the proximity between the temperature value of each point in the reconstructed temperature field data and the phase change critical temperature based on the temperature sensitivity weight function to generate a sensitivity map, and enhancing the sensitivity map to obtain an enhanced sensitivity map; performing multi-scale temperature gradient calculation and fusion on the reconstructed temperature field data through polar coordinate gradient operators of different kernel sizes in a multi-scale temperature gradient pyramid to obtain a comprehensive gradient feature map; performing a pixel-level product operation on the enhanced sensitivity map and the comprehensive gradient feature map to obtain a phase change sensitivity gradient map, and fusing the phase change sensitivity gradient map with the enhanced sensitivity map to obtain a phase change sensitivity enhanced feature vector; A reconstruction module is used to perform latent space reconstruction and interpolation on the phase change sensitive enhancement eigenvector based on the constraints of the heat conduction partial differential equation to obtain full-field temperature distribution data; A generation module is used to generate temperature field anomaly scores and anomaly type probability distribution information based on the full-field temperature distribution data.

Images