Online monitoring and process compensation system and method for residual stress and deformation of die casting

Abstract

The invention relates to the technical field of die casting intelligent manufacturing, and discloses an online monitoring and process compensation method and system for residual stress and deformation of a die casting. The method comprises the following steps: embedding a distributed temperature-stress composite sensor array in a mold cavity, and synchronously acquiring temperature and stress signals; after the signal is purified, a space-time correlation matrix is constructed to quantify a thermal-mechanical coupling relation; identifying a stress distribution mode through a support vector machine model, and positioning a fluctuation abnormal region; analyzing a defect mechanism based on mutual information and Granger causality test, and calculating a pore formation probability in combination with fluid dynamics simulation data; matching the pore high-risk location with a historical crack correlation model to generate a crack prediction index; and performing inversion optimization on the mold filling speed and pressure parameters by using the potential quality hazard evaluation function. Real-time monitoring of residual stress and deformation in the die-casting process, defect dynamic traceability and process online optimization are achieved, and air hole and crack defects are effectively restrained.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing and metal forming technology, specifically to a method and system for online monitoring and process compensation of residual stress and deformation in die-cast parts. Background Technology

[0002] Die casting is a highly efficient metal forming process in manufacturing, widely used in fields such as automobiles and aerospace where high-performance components are required. The quality of the die casting directly affects the safety and reliability of the final product. During die casting, molten metal rapidly fills the mold cavity under high pressure and undergoes complex thermo-mechanical coupling, leading to residual stress and microscopic defects such as porosity, shrinkage cavities, or cracks within the casting, which in turn affect dimensional stability and mechanical properties.

[0003] Monitoring methods for die-cast part quality often rely on offline inspection techniques (such as X-ray inspection and ultrasonic testing) or localized sensing analysis based on limited measurement points. These methods primarily reflect the state information after forming, rarely covering the dynamic evolution characteristics throughout the entire forming process. These methods have limitations in capturing the spatiotemporal coupling relationship between temperature and stress fields during die casting, and their ability to trace the source of defects is weak. Especially in high-cycle, continuous production scenarios, the mapping relationship between process parameters and internal states is complex, and existing monitoring strategies still have room for improvement in the accuracy and response speed of identifying abnormal signals in key areas.

[0004] The internal stress distribution of die-cast parts is influenced by multiple factors, including material solidification characteristics, mold heat conduction conditions, and filling flow behavior, exhibiting high non-uniformity and time-varying characteristics. Stress concentration or abnormal fluctuations may occur in certain localized areas, potentially correlated with subsequent porosity formation or crack initiation. However, traditional analysis methods typically treat temperature and stress data independently, lacking a systematic model of their co-evolutionary dynamics, resulting in a time lag between defect prediction and process intervention. For example, a sudden change in local cooling rate during the filling stage may induce stress anomalies and promote gas retention; however, without simultaneously analyzing the temperature-stress coupling characteristics of this region, it is difficult to accurately determine the high-risk location of porosity. Although some studies have attempted to introduce numerical simulation to assist in defect prediction, simulation models often rely on ideal boundary conditions and have long computation cycles, making it difficult to integrate with real-time production line data for closed-loop control. Summary of the Invention

[0005] The purpose of this invention is to provide an online monitoring and process compensation method and system for residual stress and deformation of die castings. It solves the technical problems of traditional die casting quality monitoring relying on offline detection or local sensing, difficulty in capturing the spatiotemporal coupling characteristics of temperature field and stress field, and time lag in defect prediction and process intervention, which makes it difficult to trace the source and control porosity and crack-like defects in high-cycle continuous die casting production in real time.

[0006] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: To address the aforementioned technical problems, this invention provides an online monitoring and process compensation method for residual stress and deformation of die-cast parts. This method has advantages such as strong real-time performance, high accuracy of coupled modeling, and outstanding defect tracing capabilities. It is particularly suitable for dynamic prediction and closed-loop process control of porosity and crack-like defects in high-cycle continuous die-casting production.

[0007] This invention provides a method for online monitoring and process compensation of residual stress and deformation in die-cast parts, comprising the following steps: A distributed temperature-stress composite sensor array is embedded in the inner wall of the die-casting machine mold cavity and key structural areas to simultaneously collect real-time temperature field data and stress signal data at each measuring point during the die-casting process. The sensor array consists of miniature thermocouples and fiber optic strain sensors. The thermocouples are arranged in layers at 5–10 mm intervals along the depth direction of the mold, and the fiber optic strain sensors are arranged circumferentially along the cavity contour at 20–30 mm intervals. They are fixed in the mold steel substrate by high-temperature resistant ceramic encapsulation. The signal leads are arranged along the inner wall of the cooling channel (without obstructing the flow of cooling medium) and led out to the external data acquisition unit through the reserved wire holes in the mold.

[0008] Wavelet threshold denoising and sliding window midpoint filtering are applied to the collected raw temperature field data and stress signal data respectively to obtain purified temperature field data and purified stress signal data. The wavelet threshold denoising uses the db4 wavelet basis function for three-level decomposition, and the threshold is adaptively selected according to the SURE criterion. The window length of the sliding window midpoint filtering is set to 3 times the sampling period to eliminate instantaneous impact noise.

[0009] A spatiotemporal correlation matrix was constructed based on purification temperature field data and purification stress signal data. , where t represents the time step (unit: seconds, s). Representing spatial coordinates (unit: millimeters, mm), matrix elements , Let i be the sequence of temperature measurement points (unit: degrees Celsius, ℃). For the j-th stress measurement point sequence, (This represents the Pearson correlation coefficient).

[0010] The spatiotemporal correlation matrix is ​​input into a pre-trained support vector machine classification model to extract high-dimensional feature vectors and output internal stress distribution pattern category labels. The support vector machine classification model uses a radial basis kernel function, and the training samples are derived from the corresponding sensor datasets of historical qualified parts and typical defective parts.

[0011] If the stress value corresponding to a local peak in the internal stress distribution pattern exceeds a preset threshold... If the spatial location is marked as an abnormal fluctuation region, time series data are extracted from all temperature measurement points corresponding to that region to form an abnormal temperature change sequence; the preset threshold The value is determined based on the 60% to 80% range of the material's yield strength combined with historical statistics of the mold's service life.

[0012] The mutual information and Granger causality tests were performed on the abnormal temperature change sequence and the stress signal data of the corresponding fluctuating abnormal area to calculate the temporal dependence strength between them. Then, the K-means clustering algorithm was used to group multiple batches of abnormal samples, with each group corresponding to a potential defect type, to form a defect mechanism classification result. The feature vector of the K-means clustering includes four dimensions: temperature decrease rate (unit: Kelvin per second, K / s), stress rise slope (unit: megapascal per millisecond, MPa / ms), peak duration (unit: millisecond, ms), and mutual information value. The number of clusters K is determined by the elbow rule. From the defect mechanism classification results, groups highly correlated with porosity formation were selected. Then, the corresponding fluid dynamics simulation data of the molding process for these groups was used, combined with local cooling rate and gas retention index, to calculate the probability distribution of porosity formation. (The range of values ​​is) The gas retention index is defined as the reciprocal of the ratio of the local solidification front advance velocity (unit: mm / s) to the melt flow velocity (unit: mm / s). Spatial matching is performed between the high-risk areas of pore formation and a pre-established correlation model for crack occurrence. If the probability distribution... The integral value in the high-risk area has a cosine similarity to the crack occurrence frequency at the corresponding location in the historical crack database that is higher than the threshold. (= value range is [0,1]), then the crack generation prediction index; The crack occurrence correlation model is a three-dimensional mesh mapping table that records the historical number of crack occurrences in each spatial unit under different process conditions; the threshold Set to 0.75; The crack occurrence prediction index is input into the quality hazard assessment function. ,in: The stomatal probability (within the range [0,1]), i.e. The maximum value or weighted average value in the high-risk area; The stress gradient (unit: megapascal per millimeter, MPa / mm) is calculated by dividing the stress difference between adjacent stress measurement points in the fluctuating anomaly region by the spatial distance. The volume shrinkage rate (with a value range of [0,1]) reflects the severity of solidification shrinkage. , , Let the weight coefficients be (, satisfying) The filling speed v (unit: meters per second, m / s) and pressure setpoint p (unit: megapascals, MPa) are obtained by optimizing the historical process parameter space through a multi-objective genetic algorithm; based on the output value of F(Q) (the value range is [0,1]), the optimized filling speed v (unit: meters per second, m / s) and pressure setpoint p (unit: megapascals, MPa) are retrieved. The optimized filling speed v and pressure setpoint p are written into the die-casting machine PLC control system to replace the current process parameters. In the next production cycle, the updated control instructions are executed, and a new round of temperature field and stress signal data are collected simultaneously to verify the defect suppression effect and serve as input data for the next iteration.

[0013] The installation method of the distributed temperature-stress composite sensor array is as follows: during the mold processing stage, a blind hole with a diameter of 1.5mm and a depth of 30mm is reserved, the packaged sensor is pressed into the hole, and the hole opening is sealed with high temperature metal glue to ensure that the thermal expansion coefficient matches the mold steel substrate and avoid loosening or signal drift due to thermal cycling.

[0014] The input to the support vector machine classification model is the top 10 principal components of the spatiotemporal correlation matrix after dimensionality reduction by principal component analysis, with a cumulative variance contribution rate of no less than 92%.

[0015] The fluid dynamics simulation data of the filling process was obtained by coupling the VOF (Volume of Fluid) model with the solidification heat transfer model. The simulation boundary conditions included the actual gate location, injection velocity curve and initial mold temperature.

[0016] The volume shrinkage rate V is calculated from the temperature range corresponding to the minimum point of the second derivative of the local cooling curve in the purification temperature field data, reflecting the severity of solidification shrinkage.

[0017] The PLC control system receives the v and p parameters sent by the host computer via the Modbus TCP protocol and completes the parameter loading before the next mold closing action is triggered, ensuring that process adjustments are synchronized with the production cycle.

[0018] Compared with the prior art, the present invention has the following beneficial effects: Existing technologies mostly monitor temperature or stress in isolation and at single points, failing to depict the inherent correlations within the strongly coupled process of die casting. This invention, through a distributed temperature-stress composite sensor array and its specific arrangement within the mold, achieves native, synchronous, and precise capture of the three-dimensional temperature and stress fields during die casting. This provides a unique and reliable data foundation for revealing the intrinsic mechanism of thermo-mechanical coupling, solving the technical problem of insufficient dimensionality and inability to reflect the essential nature of physical field interactions in traditional monitoring data.

[0019] This invention quantifies the temporal causal relationship between temperature and stress evolution under online conditions by constructing a spatiotemporal correlation matrix and integrating mutual information and causal verification analysis. Furthermore, it precisely correlates specific physical signals with defect formation mechanisms through clustering of anomaly patterns, thereby elevating quality analysis from "detecting anomalies" to "understanding the causes of anomalies." Simultaneously, this invention fuses real-time sensor data with pre-stored fluid dynamics simulation data based on physical mechanisms to calculate the probability of defect formation online and spatially match it with a historical failure database. This endows offline simulation models with the ability to perform online real-time calculations and predictions, enabling defect prediction to possess both physical realism and operational adaptability.

[0020] This invention defines a quality hazard assessment function that integrates multi-source quality characteristics, and optimizes core process parameters based on its output. Finally, it achieves real-time closed-loop control of the die-casting machine via an industrial bus. This forms a complete adaptive loop of "perception-analysis-decision-execution-verification," enabling process optimization to be based on continuously evolving quantitative quality feedback, thus achieving proactive prevention and suppression of defects. By reconstructing the entire technology chain, this invention elevates quality control in the die-casting process to a new stage driven by online, proactive, data- and mechanism-based approaches, significantly enhancing the stability of high-cycle continuous production and the consistency of product intrinsic quality. Attached Figure Description

[0021] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained from these drawings without creative effort.

[0022] Figure 1 This is a schematic diagram of the overall architecture of the system of the present invention.

[0023] Figure 2 This is an overall flowchart of the method described in this invention. Detailed Implementation

[0024] In the following description, only certain exemplary embodiments are briefly described. As those skilled in the art will recognize, the described embodiments can be modified in various ways without departing from the spirit or scope of the embodiments of the invention. Therefore, the drawings and description are considered to be exemplary in nature and not restrictive.

[0025] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0026] Example 1: This embodiment discloses an online monitoring and process compensation method for residual stress and deformation of die-cast parts, including the following steps: Step 1, Embedding and Acquisition: A distributed temperature-stress composite sensor array consisting of miniature thermocouples and fiber optic strain sensors is embedded in the inner wall of the die-casting mold cavity and key structural areas to simultaneously acquire real-time temperature field data and stress signal data during the die-casting process. Step 2, Signal purification: The original temperature field data and stress signal data collected in Step 1 are subjected to wavelet threshold denoising and sliding window midpoint filtering respectively to obtain purified temperature field data and stress signal data. Step 3, Spatiotemporal coupling modeling: Based on the purification data obtained in Step 2, calculate the Pearson correlation coefficient between each temperature measurement point sequence and each stress measurement point sequence, and construct a spatiotemporal correlation matrix; input the matrix into a pre-trained support vector machine classification model to identify and output the current internal stress distribution pattern; Step 4, Abnormal Area Identification: If there is a local peak value in the internal stress distribution pattern identified in Step 3 where the stress value exceeds the preset threshold, then the spatial location corresponding to the peak value is marked as an abnormal fluctuation area, and all temperature measurement point sequences corresponding to the area are extracted to form an abnormal temperature change sequence. Step 5, Defect Mechanism Analysis: For the abnormal temperature change sequence obtained in Step 4 and the stress signal data of the corresponding fluctuation abnormal area, mutual information calculation and Granger causality test are performed to analyze their temporal dependence; then, based on the multidimensional feature vector containing the temperature decrease rate, stress rise slope, peak duration and mutual information value, a clustering algorithm is used to group the historical and current abnormal samples to form the defect mechanism classification results. Step 6, Defect Risk Prediction: From the defect mechanism classification results in Step 5, categories related to porosity formation are selected; the fluid dynamics simulation data of the filling process corresponding to the category is called, and the local cooling rate calculated from the measured temperature field data and the gas retention index calculated from the simulation data are combined to calculate the probability distribution of porosity formation in the mold cavity, and high-risk areas for porosity formation are identified accordingly; then, the high-risk areas for porosity formation are spatially matched with the three-dimensional crack association model that records historical crack information. If the matching degree is higher than the set threshold, a crack occurrence prediction index is generated. Step 7, Process parameter optimization: Input the crack occurrence prediction index, porosity formation probability, stress gradient of the fluctuating abnormal area, and volume shrinkage rate obtained in step 6 into the quality hazard assessment function for calculation; Based on the output value of the function, the optimized filling speed setting value and pressure setting value are obtained by inversion solution. Step 8, Closed-loop control and iteration: The optimized filling speed and pressure setpoints from Step 7 are written into the PLC control system of the die-casting machine for process execution in the next die-casting production cycle; at the same time, Step 1 is returned to collect a new round of monitoring data to verify the effect and serve as input for the next iteration optimization, thus forming a closed loop.

[0027] Furthermore, in step 1, the miniature thermocouples are arranged in layers at intervals of 5 mm to 10 mm along the depth direction of the mold; the fiber optic strain sensors are arranged circumferentially along the cavity contour at intervals of 20 mm to 30 mm.

[0028] Furthermore, in step 2, the wavelet threshold denoising uses the db4 wavelet basis function for three-level decomposition, and the threshold is adaptively selected according to the SURE criterion; the window length of the sliding window mid-range filter is set to 3 times the data sampling period.

[0029] Furthermore, in step 3, before inputting the spatiotemporal correlation matrix into the support vector machine classification model, principal component analysis is performed to reduce its dimensionality, and the top principal components with a cumulative variance contribution rate of not less than 92% are selected as the input features of the model.

[0030] Furthermore, in step 6, the gas retention index is defined as the reciprocal of the ratio of the local solidification front advance velocity to the melt flow velocity.

[0031] Furthermore, in step 7, the quality hazard assessment function is a weighted sum function of porosity probability, stress gradient, and volume shrinkage rate, and its weight coefficients are obtained by optimizing within the historical process parameter space using a multi-objective genetic algorithm.

[0032] This embodiment also discloses an online monitoring and process compensation system for residual stress and deformation of die-cast parts, including: The sensor data acquisition module is used to simultaneously acquire real-time temperature field data and stress signal data during the die casting process through a distributed temperature-stress composite sensor array embedded in the die casting mold cavity; The signal purification module, connected to the sensor data acquisition module, is used to perform wavelet threshold denoising and sliding window value filtering on the acquired raw temperature field data and stress signal data respectively, to obtain purified temperature field data and stress signal data. The spatiotemporal coupling modeling module is connected to the signal purification module. It is used to construct a spatiotemporal correlation matrix based on the purified data, and to process the matrix using a pre-trained support vector machine classification model to identify and output the internal stress distribution pattern. An anomaly identification module, connected to the spatiotemporal coupling modeling module, is used to detect local peak values ​​where the stress value exceeds a preset threshold in the identified internal stress distribution pattern, mark the corresponding spatial location as a fluctuating anomaly region, and extract the temperature measurement point sequence corresponding to the region to form an abnormal temperature change sequence. The defect mechanism analysis module, connected to the anomaly identification module, is used to perform mutual information calculation and Granger causality test on the abnormal temperature change sequence and the stress signal data of the corresponding fluctuation anomaly region, and to perform cluster analysis based on the feature vector containing the temperature decrease rate, stress rise slope, peak duration and mutual information value to generate defect mechanism classification results. The porosity risk location module is connected to the defect mechanism analysis module. It is used to filter out categories related to porosity formation from the defect mechanism classification results, and calculate the probability distribution of porosity formation and locate high-risk areas of porosity formation by combining the corresponding fluid dynamics simulation data of the filling process, local cooling rate and gas retention index. The crack prediction module is connected to the porosity risk location module. It is used to spatially match the high-risk areas of porosity formation with the pre-stored three-dimensional crack association model, and generate a crack occurrence prediction index when the matching degree is higher than a set threshold. The process optimization module connects the crack prediction module and the anomaly identification module. It is used to calculate and invert based on the crack occurrence prediction index, the probability of porosity formation, the stress gradient and volume shrinkage rate of the fluctuation abnormal area, and the quality hazard assessment function to output the optimized filling speed setting value and pressure setting value. The control execution module connects the process optimization module to the die-casting machine PLC control system. It is used to write the optimized filling speed setting value and pressure setting value into the PLC control system for execution in the next production cycle and to trigger a new round of data acquisition to start the iterative optimization process.

[0033] Furthermore, the sensor in the sensing data acquisition module is encapsulated in high-temperature resistant ceramic and installed in a blind hole reserved during mold processing.

[0034] Furthermore, the control execution module is connected to the die-casting machine PLC control system via the Modbus TCP communication protocol, and completes the loading of optimized parameters before the next mold closing action begins.

[0035] To enable those skilled in the art to fully understand and implement this invention, the specific implementation principle of this invention will be further explained in detail below with reference to a specific application scenario.

[0036] See Figure 1 and Figure 2 In the mold cavity and key structural areas of the die-casting mold, several blind holes with a diameter of 1.5 mm and a depth of 30 mm are pre-drilled during the mold processing stage. The blind holes are arranged in layers at intervals of 5–10 mm along the depth direction of the mold to embed miniature thermocouples.

[0037] Meanwhile, another set of blind holes is opened on the contour line of the mold cavity with a circumferential spacing of 20–30 mm for installing fiber optic strain sensors.

[0038] All sensors are encapsulated in high-temperature resistant ceramic. The outer diameter of the encapsulation body is interference-fitted with the inner diameter of the blind hole. The sensors are firmly embedded into the mold steel substrate by hydraulic or mechanical pressing, ensuring that the distance between the sensor's temperature measuring end and the cavity surface is ≤1mm (to guarantee temperature measurement accuracy). The orifice is sealed with high-temperature metal glue to ensure no gap between the encapsulation body and the mold steel substrate, and to match their coefficients of thermal expansion, preventing loosening or signal drift during continuous thermal cycling. The temperature measuring end of the miniature thermocouple faces the cavity surface. Its signal lead and the fiber optic strain sensor's pigtail are laid along a preset path inside the mold and led out through the mold cooling channel to the outside of the mold, connecting to the data acquisition unit to form a distributed temperature-stress composite sensor array.

[0039] The temperature-stress composite sensor array outputs real-time raw temperature field data and stress signal data during the die-casting process. The raw data is first input to a signal purification module. The temperature signal undergoes three-level wavelet decomposition using the db4 wavelet basis function. Thresholds for each level are adaptively calculated based on the Stein unbiased likelihood estimation (SURE) criterion. High-frequency noise components are then soft-thresholded before signal reconstruction. The stress signal is filtered using a sliding window mid-range filter, with the window length set to three times the current sampling period (e.g., 3ms when the sampling frequency is 1kHz) to remove outliers caused by instantaneous interference such as injection impact and demolding vibration. Finally, purified temperature field data and purified stress signal data are output.

[0040] The purified data is then fed into the spatiotemporal coupling modeling module. This module uses time step t as an index and spatial coordinates (x, y, z) corresponding to the physical locations of each sensor to construct a dimensional model. spatiotemporal correlation matrix Where N is the number of temperature measurement points, M is the number of stress measurement points, and the matrix elements are... , This represents the temperature data (in °C) of the i-th miniature thermocouple over a time series. This represents the strain-to-stress data (unit: MPa) of the j-th fiber Bragg grating strain sensor within the corresponding time period. The Pearson correlation coefficient formula was used for calculation. Subsequently, principal component analysis (PCA) was performed on the matrix, and the top 10 principal components with a cumulative variance contribution rate of not less than 92% were extracted as feature vectors and input into a pre-trained support vector machine classification model. The model uses radial basis function (RBF), and its training samples are derived from sensor datasets corresponding to known qualified parts and typical defective parts (such as those with porosity and cracks) in historical production. After annotation, supervised learning labels are formed, and the model output is a category label for the internal stress distribution pattern, such as "uniform distribution", "edge concentration", and "local mutation".

[0041] When the output category is "local mutation" or a similar high-risk pattern, the anomaly detection module is activated. This module iterates through the stress values ​​corresponding to all spatial locations, and if the stress value at a certain location exceeds a preset threshold... (Unit: MPa) Then mark the location as an abnormal fluctuation area. The value is determined based on the yield strength of the die-casting material (such as ADC12 aluminum alloy), typically ranging from 60% to 80%, and dynamically corrected by combining the stress statistical distribution under similar working conditions in the mold's service history. After marking, the system extracts the corresponding temperature time series from all the micro thermocouples covered by the abnormal fluctuation area, combines them into an abnormal temperature change series, and uses it as input for subsequent analysis.

[0042] The defect mechanism analysis module receives the abnormal temperature change sequence and its corresponding stress signal data. First, it calculates the mutual information value between the two to quantify the nonlinear dependency. Then, it performs a Granger causality test to determine whether the temperature change statistically leads to the stress response. Subsequently, multiple batches of historical abnormal samples are clustered according to four-dimensional feature vectors: temperature decrease rate (unit: K / s, calculated from the first derivative of the purification temperature field data), stress rise slope (unit: MPa / ms, calculated from the first derivative of the purification stress signal data), and peak duration (unit: ms, calculated from the stress exceeding...). The time interval from the point of return to below the threshold, and the aforementioned mutual information value. The K-means clustering algorithm is executed here, and the number of clusters K is determined by the elbow method, that is, calculating the sum of squares within the cluster (WCSS) under different K values, and selecting the K value corresponding to the inflection point. The clustering results form several groups, each corresponding to a potential defect mechanism, such as "rapid cooling leading to shrinkage stress concentration" and "gas retention at the end of filling causing local weakening", etc.

[0043] The porosity risk localization module filters out groups highly correlated with porosity formation from the above clustering results (e.g., groups containing high gas retention index characteristics). For these groups, pre-stored fluid dynamics simulation data of the filling process is invoked. This simulation data is obtained by coupling a VOF (Volume of Fluid) multiphase flow model with a solidification heat transfer model, and the boundary conditions include the actual gate location, measured injection velocity curves, and initial mold temperature. In the simulation results, the gas retention index of each spatial unit is calculated (defined as the reciprocal of the ratio of the local solidification front advance velocity to the melt flow velocity); simultaneously, combined with the local cooling rate (unit: K / s, calculated from the purification temperature field data), it is substituted into empirical formulas or machine learning regression models to output the porosity formation probability distribution P(x,y,z) in three-dimensional space. Based on this, high-risk areas for porosity formation are identified in the geometric space of the mold cavity.

[0044] The crack prediction module spatially matches the high-risk areas for porosity formation mentioned above with a pre-established crack occurrence correlation model. This correlation model is a three-dimensional meshed mapping table with a mesh resolution consistent with the spatial distribution of the sensors. Each mesh cell records the number of times a crack occurred at that location in historical production under different combinations of process parameters. The system calculates... The spatial integral value of the high-risk area is used to calculate the cosine similarity with the crack occurrence frequency vector of the corresponding grid cell in the historical crack database. If the similarity is higher than a preset threshold... If a crack is detected, a crack risk is identified, and a crack occurrence prediction index is generated. This index includes spatial location, risk level, and related process conditions.

[0045] The process optimization module receives crack occurrence prediction indicators and calculates a quality hazard assessment letter in conjunction with other quality characteristics. Where P is the probability of porosity formation, taken from the maximum value or weighted average of P(x,y,z) in the high-risk area; D is the stress gradient (unit: MPa / mm), calculated by dividing the stress difference of adjacent fiber optic strain sensors in the fluctuating abnormal area by the spatial distance; V is the volume shrinkage rate, which is obtained by analyzing the second derivative of the local cooling curve in the purification temperature field data, locating the rapid cooling segment after the solidification platform ends, extracting the intensity of volume shrinkage corresponding to this temperature range, and converting it into a shrinkage rate value using empirical formulas or table lookup methods. Weighting coefficients The optimal filling speed v (in m / s) and pressure setpoint p (in MPa) are obtained by optimizing the historical process parameter space using a multi-objective genetic algorithm, with the goal of minimizing F(Q) while satisfying the production cycle constraint. Based on the output value of F(Q), the optimal filling speed v (in m / s) and pressure setpoint p (in MPa) are obtained through inversion. This inversion process can be performed using a response surface model, a neural network surrogate model, or a physics-based inverse simulation.

[0046] The control execution module sends the optimized v and p parameters to the die-casting machine's PLC control system via an industrial Ethernet interface using the Modbus TCP protocol. Upon receiving the parameters, the PLC caches them in the process parameter register and loads them before the next mold closing action, ensuring the new parameters take effect during the injection stage of the next die-casting cycle. Simultaneously, after the start of a new production cycle, the sensor data acquisition module continues to collect new temperature field and stress signal data. After processing by the signal purification module, this data is input as verification data to the spatiotemporal coupling modeling module, forming a closed-loop process: real-time sensor data input → abnormal area marking → defect mechanism classification → joint prediction of porosity and cracks → quality hazard assessment → process parameter optimization → PLC control execution → new cycle data feedback, achieving dynamic iterative optimization.

[0047] During operation, all modules are deployed on a host industrial computer and connected to the local control system of the die-casting machine via an isolation gateway to ensure data security and real-time performance. The signal leads of the distributed temperature-stress composite sensor array are led out through the mold cooling channel and connected to a high-precision synchronous data acquisition card with a sampling frequency of no less than 1kHz, ensuring complete capture of critical stages such as injection, holding pressure, and cooling. The mold steel matrix is ​​made of H13 hot-work mold steel, whose thermal conductivity matches the sensor encapsulation ceramic material, avoiding excessive interface thermal resistance that could affect temperature measurement accuracy. The fiber optic strain sensor has a center wavelength around 1550nm, and strain information is demodulated in real time using a wavelength demodulator, achieving a strain resolution of [missing information]. Temperature cross-sensitivity is eliminated through reference grating compensation.

[0048] In practical applications, taking a die-cast automotive engine bracket as an example, the critical area of ​​the mold cavity is equipped with 12 miniature thermocouples (three layers, four in each layer) and 18 fiber optic strain sensors (evenly distributed along the contour). During the 35th continuous production run, the anomaly detection module detected a stress value of 280 MPa in a region near the end of the gate, exceeding... (Set to 260MPa), marked as an abnormal fluctuation area.

[0049] Clustering results from the defect mechanism analysis module showed that the anomaly belonged to the gas retention plus rapid cooling group. The porosity risk location module, using the filling simulation data corresponding to the product, calculated the gas retention index for this area to be 1.8, the cooling rate to be 85 K / s, and the porosity formation probability P=0.68. The crack prediction module, comparing with the historical database, found that the cosine similarity between the crack occurrence frequency vector at this location under similar operating conditions and the current P distribution was 0.78, which is higher than... Therefore, crack prediction indices are generated. The process optimization module calculates... The inversion revealed that v needed to be reduced from 3.2 m / s to 2.9 m / s, and p from 85 MPa to 92 MPa. The control execution module issued the parameters, and after the 36th iteration of the new parameters, monitoring data showed that the peak stress in the region decreased to 240 MPa, and the porosity probability decreased to 0.32, verifying the effectiveness of the compensation.

[0050] Specific examples are as follows: When deploying the system described in this invention on a die-casting production line for an automotive engine bracket, blind holes are drilled in the critical areas of the mold cavity and miniature thermocouples and fiber optic strain sensors are embedded therein. The miniature thermocouples are arranged in three layers along the depth direction of the mold, with four in each layer, for a total of 12. The fiber optic strain sensors are arranged along the contour of the cavity at a circumferential spacing of approximately 25 mm, for a total of 18. All sensors are encapsulated in high-temperature resistant ceramic and then pressed into the H13 mold steel substrate. The orifices are sealed with high-temperature metal glue. The signal leads are led out through the cooling channel and connected to a data acquisition card with a synchronous sampling frequency of 1 kHz to ensure that the complete thermo-mechanical response during the injection, holding, and cooling stages is fully captured.

[0051] During the 35th die-casting cycle, the sensor data acquisition module outputs raw temperature and stress signals in real time. The signal purification module performs db4 wavelet three-level decomposition and SURE threshold denoising on both types of signals, as well as 3ms sliding window mid-range filtering, effectively eliminating transient interference caused by injection impact and obtaining purified temperature field data and purified stress signal data with high signal-to-noise ratio. The spatiotemporal coupling modeling module is based on the physical coordinates of each sensor. With time step t, construct 3D spatiotemporal correlation matrix The Pearson correlation coefficient between each pair of temperature-stress measurement points was calculated, and principal component analysis was performed on the matrix. The top 10 principal components with a cumulative variance contribution rate of 93.5% were extracted and input into a pre-trained support vector machine classification model. Due to the significant characteristics of gas stagnation and rapid cooling defect samples in the historical data, the model outputs local mutation mode labels.

[0052] The anomaly detection module was then activated, traversing all stress measurement points. It was found that a fiber optic strain sensor located near the end of the gate measured a stress of 280 MPa, exceeding the dynamic threshold set based on the yield strength of ADC12 aluminum alloy (approximately 325 MPa). Therefore, the spatial location was marked as an abnormal fluctuation area, and the temperature sequences of four miniature thermocouples within its coverage area were extracted to form an abnormal temperature change sequence.

[0053] The defect mechanism analysis module calculated the mutual information between the sequence and the corresponding stress signal, obtaining a value of 0.87, indicating a strong nonlinear dependence. The Granger causality test p-value was less than 0.01, confirming that the temperature change was the Granger cause of the stress response. Subsequently, the temperature drop rate (85 K / s), stress rise slope (120 MPa / ms), peak duration (18 ms), and mutual information value (0.87) of this anomaly were used as a four-dimensional feature vector and input into a K-means clustering model with K=5 determined by the elbow rule. The cluster centers were matched with historical gas retention and rapid cooling groups to complete the defect mechanism classification.

[0054] The porosity risk location module calls upon the fluid dynamics simulation data of the filling process corresponding to this product. This data is generated by coupling a VOF multiphase flow and solidification heat transfer model. The boundary conditions include the actual gate location, the measured injection velocity of 3.2 m / s, and the initial mold temperature of 220℃. In the simulation domain, the gas retention index of this abnormal fluctuation region is calculated to be 1.8 (defined as the reciprocal of the ratio of the solidification front advance velocity to the melt flow velocity). Combined with the measured cooling rate of 85 K / s, and substituted into the pre-trained XGBoost regression model, the probability of porosity formation at this point is output as P=0.68. This location is then marked as a high-risk area for porosity in the three-dimensional geometric model of the mold cavity.

[0055] The crack prediction module spatially aligns the high-risk area with a pre-established 3D meshed crack correlation model. Each mesh cell in the model stores the frequency of crack occurrence at that location under the same process conditions in the past 200 simulations. The system then calculates the probability of porosity formation at this point. Using the high-risk location as the center and combining the probability values ​​of its surrounding neighboring grid cells, the overall spatial integral value of the grid cell containing the high-risk location is calculated, yielding 0.65. This value is then compared with the historical crack frequency vector. The cosine similarity calculation yielded a similarity of 0.78, which is higher than the preset threshold. Therefore, a crack occurrence prediction index is generated, which includes spatial coordinates, a risk level of "high", and associated process conditions such as "high-speed filling and excessively fast end cooling".

[0056] The process optimization module receives this indicator and simultaneously calculates the quality hazard assessment function. Where P is taken as 0.68, and D is the stress gradient, calculated by dividing the stress difference (280–210 = 70 MPa) between adjacent fiber optic strain sensors in the fluctuating anomaly region by the spatial distance (25 mm), i.e., D = 2.8 MPa / mm. After normalization based on the historical maximum stress gradient, its value is 1.2; V is obtained by analyzing the temperature range (580℃→520℃) corresponding to the minimum point of the second derivative of the cooling curve in this region, and the volume shrinkage rate is found to be 0.15 from the table. Weighting coefficient The result was obtained by optimizing the multi-objective genetic algorithm under the constraint that the cycle time is ≤45s. .

[0057] Based on this value, a neural network surrogate model trained with historical process parameters as input and F(Q) as output is invoked to inversely determine the optimal filling speed. Pressure setpoint .

[0058] The multi-objective genetic algorithm uses process parameters and corresponding quality assessment results from historical production data as the training set. Its main objective is to minimize the quality hazard assessment function F(Q), while also taking into account constraints such as production cycle time. It performs multi-generation iterative optimization to find a set of optimal weight coefficients α, β, and γ for online evaluation.

[0059] The control execution module sends v and p to the die-casting machine PLC control system via Industrial Ethernet using the Modbus TCP protocol. Before the mold closing signal is triggered in the 36th cycle, the PLC writes the new parameters into the injection control register. After the new cycle begins, the sensor data acquisition module collects data again. After being processed by the signal purification module, the data is input into the spatiotemporal coupling modeling module. Verification shows that the peak stress in this area has decreased to 240 MPa, the porosity probability P has decreased to 0.32, and F(Q) has decreased to 0.41. This proves that the process compensation effectively suppresses the defect initiation trend and achieves dynamic iterative optimization from anomaly detection to parameter adaptive adjustment.

[0060] This invention, through a distributed temperature-stress composite sensor array and a multi-level data fusion analysis architecture, can transform traditional offline detection into online proactive intervention. The achievement of its technical effect depends on the high reliability of the sensor embedded in the mold steel matrix, the quantitative characterization of thermo-mechanical co-evolution by the spatiotemporal coupling matrix, and the defect prediction-process inversion closed loop based on the fusion of physical mechanisms and data-driven approaches. Thus, it can achieve accurate source tracing and real-time suppression of porosity and crack-like defects without interrupting continuous production.

[0061] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0062] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. It should be noted that any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for online monitoring and process compensation of residual stress and deformation in die-castings, characterized in that, Includes the following steps: Step 1, Embedding and Acquisition: A distributed temperature-stress composite sensor array consisting of miniature thermocouples and fiber optic strain sensors is embedded in the inner wall of the die-casting mold cavity and key structural areas to simultaneously acquire real-time temperature field data and stress signal data during the die-casting process. Step 2, Signal purification: The original temperature field data and stress signal data collected in Step 1 are subjected to wavelet threshold denoising and sliding window midpoint filtering respectively to obtain purified temperature field data and stress signal data. Step 3, Spatiotemporal coupling modeling: Based on the purification data obtained in Step 2, calculate the Pearson correlation coefficient between each temperature measurement point sequence and each stress measurement point sequence, and construct a spatiotemporal correlation matrix; input the matrix into a pre-trained support vector machine classification model to identify and output the current internal stress distribution pattern; Step 4, Abnormal Area Identification: If there is a local peak value in the internal stress distribution pattern identified in Step 3 where the stress value exceeds the preset threshold, then the spatial location corresponding to the peak value is marked as an abnormal fluctuation area, and all temperature measurement point sequences corresponding to the area are extracted to form an abnormal temperature change sequence. Step 5, Defect Mechanism Analysis: The mutual information calculation and Granger causality test are performed on the abnormal temperature change sequence obtained in Step 4 and the stress signal data of the corresponding fluctuation abnormal area to analyze their temporal dependence. Subsequently, based on a multidimensional feature vector containing temperature decrease rate, stress rise slope, peak duration and mutual information value, a clustering algorithm is used to group historical and current abnormal samples to form defect mechanism classification results. Step 6, Defect Risk Prediction: From the defect mechanism classification results in Step 5, select categories related to porosity formation; The fluid dynamics simulation data of the filling process corresponding to this category is called, and the local cooling rate calculated from the measured temperature field data and the gas retention index calculated from the simulation data are combined to calculate the probability distribution of porosity formation in the mold cavity, and high-risk areas for porosity formation are identified accordingly. Then, the high-risk areas for porosity formation are spatially matched with a three-dimensional crack association model that records historical crack information. If the matching degree is higher than a set threshold, a crack occurrence prediction index is generated. Step 7, Process parameter optimization: Input the crack occurrence prediction index, porosity formation probability, stress gradient of the fluctuating abnormal area, and volume shrinkage rate obtained in step 6 into the quality hazard assessment function for calculation; Based on the output value of the function, the optimized filling speed setting value and pressure setting value are obtained by inversion solution. Step 8, Closed-loop control and iteration: The optimized filling speed and pressure setpoints from Step 7 are written into the PLC control system of the die-casting machine for process execution in the next die-casting production cycle; at the same time, Step 1 is returned to collect a new round of monitoring data to verify the effect and serve as input for the next iteration optimization, thus forming a closed loop.

2. The method for online monitoring and process compensation of residual stress and deformation of die-cast parts according to claim 1, characterized in that, In step 1, the miniature thermocouples are arranged in layers at intervals of 5 mm to 10 mm along the depth direction of the mold; the fiber optic strain sensors are arranged circumferentially along the cavity contour at intervals of 20 mm to 30 mm.

3. The method for online monitoring and process compensation of residual stress and deformation of die-cast parts according to claim 1, characterized in that, In step 2, the wavelet threshold denoising uses the db4 wavelet basis function for three-level decomposition, and the threshold is adaptively selected according to the SURE criterion; the window length of the sliding window mid-range filter is set to 3 times the data sampling period.

4. The method for online monitoring and process compensation of residual stress and deformation of die-cast parts according to claim 1, characterized in that, In step 3, before inputting the spatiotemporal correlation matrix into the support vector machine classification model, principal component analysis is performed to reduce its dimensionality, and the top principal components with a cumulative variance contribution rate of not less than 92% are selected as the input features of the model.

5. The method for online monitoring and process compensation of residual stress and deformation of die-cast parts according to claim 1, characterized in that, In step 6, the gas retention index is defined as the reciprocal of the ratio of the local solidification front advance velocity to the melt flow velocity.

6. The method for online monitoring and process compensation of residual stress and deformation of die-cast parts according to claim 1, characterized in that, In step 7, the quality hazard assessment function is a weighted sum of porosity probability, stress gradient and volume shrinkage rate, and its weight coefficients are obtained by optimizing within the historical process parameter space using a multi-objective genetic algorithm.

7. An online monitoring and process compensation system for residual stress and deformation of die-cast parts, used to execute the online monitoring and process compensation method for residual stress and deformation of die-cast parts as described in any one of claims 1-6, characterized in that, include: The sensor data acquisition module is used to simultaneously acquire real-time temperature field data and stress signal data during the die casting process through a distributed temperature-stress composite sensor array embedded in the die casting mold cavity; The signal purification module, connected to the sensor data acquisition module, is used to perform wavelet threshold denoising and sliding window value filtering on the acquired raw temperature field data and stress signal data respectively, to obtain purified temperature field data and stress signal data. The spatiotemporal coupling modeling module is connected to the signal purification module. It is used to construct a spatiotemporal correlation matrix based on the purified data, and to process the matrix using a pre-trained support vector machine classification model to identify and output the internal stress distribution pattern. An anomaly identification module, connected to the spatiotemporal coupling modeling module, is used to detect local peak values ​​where the stress value exceeds a preset threshold in the identified internal stress distribution pattern, mark the corresponding spatial location as a fluctuating anomaly region, and extract the temperature measurement point sequence corresponding to the region to form an abnormal temperature change sequence. The defect mechanism analysis module, connected to the anomaly identification module, is used to perform mutual information calculation and Granger causality test on the abnormal temperature change sequence and the stress signal data of the corresponding fluctuation anomaly region, and to perform cluster analysis based on the feature vector containing the temperature decrease rate, stress rise slope, peak duration and mutual information value to generate defect mechanism classification results. The porosity risk location module is connected to the defect mechanism analysis module. It is used to filter out categories related to porosity formation from the defect mechanism classification results, and calculate the probability distribution of porosity formation and locate high-risk areas of porosity formation by combining the corresponding fluid dynamics simulation data of the filling process, local cooling rate and gas retention index. The crack prediction module is connected to the porosity risk location module. It is used to spatially match the high-risk areas of porosity formation with the pre-stored three-dimensional crack association model, and generate a crack occurrence prediction index when the matching degree is higher than a set threshold. The process optimization module connects the crack prediction module and the anomaly identification module. It is used to calculate and invert based on the crack occurrence prediction index, the probability of porosity formation, the stress gradient and volume shrinkage rate of the fluctuation abnormal area, and the quality hazard assessment function to output the optimized filling speed setting value and pressure setting value. The control execution module connects the process optimization module to the die-casting machine PLC control system. It is used to write the optimized filling speed setting value and pressure setting value into the PLC control system for execution in the next production cycle and to trigger a new round of data acquisition to start the iterative optimization process.

8. The online monitoring and process compensation system for residual stress and deformation of die-cast parts according to claim 7, characterized in that, The sensor in the sensing data acquisition module is encapsulated in high-temperature resistant ceramic and installed in a blind hole reserved during mold processing.

9. The online monitoring and process compensation system for residual stress and deformation of die-cast parts according to claim 7, characterized in that, The control execution module is connected to the die-casting machine PLC control system via the Modbus TCP communication protocol and loads the optimized parameters before the next mold closing action begins.

Images