Rapid leakage diagnosis method for large die casting vacuum system

Abstract

The invention discloses a quick leakage diagnosis method for a large die casting vacuum system. Detection is synchronously completed in a pre-vacuumizing stage. The method comprises the following steps: firstly, synchronously acquiring multi-source data such as pressure, temperature, airflow and deformation, preprocessing, extracting and fusing multi-physical field features, and carrying out preliminary identification and coarse positioning; accurate positioning of a leakage point is realized through controllable excitation verification, and a leakage aperture is calculated based on an adaptive quantitative model. According to the method, the positioning precision and the quantification precision are remarkably improved through multi-stage innovation, the robustness of the system under the complex working condition is high, the rejection rate and the maintenance cost of castings can be effectively reduced, and prominent economic benefits are achieved.

Description

Technical Field

[0001] This invention belongs to the field of fault diagnosis technology for large die-casting production equipment, specifically a rapid leak diagnosis method for vacuum systems of large die-casting parts. Background Technology

[0002] The production of large die-cast parts (such as automotive engine blocks and aerospace structural components) relies on vacuum systems to maintain a negative pressure environment in the mold cavity to prevent internal defects such as porosity and shrinkage cavities during the molten metal filling process, thus ensuring the density and mechanical properties of the castings. Leaks in the vacuum system (especially minute leaks) can directly disrupt the stability of the negative pressure in the cavity, leading to a significant increase in the scrap rate of castings. At the same time, the sealing structure of large die-casting molds is complex (including multiple sealing units such as parting surfaces, slider mating surfaces, and ejector pin holes), and the vacuum system has many branches and large chamber volumes, making the location and quantification of leakage sources extremely difficult. Existing leak detection technologies have the following key drawbacks: Traditional methods mostly rely on post-production detection (such as penetration testing and hydrostatic testing), which cannot provide real-time diagnosis during the pre-vacuuming stage (the critical window period before molten metal injection) within the production cycle, resulting in the formation of defective castings and wasting materials and time. Existing technologies rely solely on pressure sensors to monitor changes in vacuum levels, which cannot capture multi-physical field coupling disturbances such as temperature field, flow field, and structural displacement field caused by leaks. They are easily affected by environmental interference, leading to misjudgments, and it is difficult to locate specific leak sealing units. Existing leakage orifice calculation models are mostly based on general fluid dynamics formulas, which are not adapted to the chamber structure, sealing form and on-site conditions (such as gas medium characteristics and chamber volume differences) of large die-casting vacuum systems. The quantification error is large and cannot provide accurate basis for maintenance decisions. Manual judgment or single-point inspection methods are time-consuming (usually several minutes to tens of minutes), which cannot match the short cycle requirements of large-scale die casting production (the pre-vacuuming stage usually only takes a few seconds to tens of seconds). Therefore, there is an urgent need for a diagnostic method that can respond quickly during the pre-vacuuming stage, perform multi-dimensional coupled detection, accurately locate the leak location, and quantify the degree of leakage, in order to overcome the shortcomings of existing technologies. Summary of the Invention

[0003] The purpose of this invention is to provide a rapid leak diagnosis method for vacuum systems of large die-cast parts, in order to solve the problems in the prior art mentioned in the background art, such as mismatch between detection timing and production cycle, insufficient matching between positioning accuracy and sealing unit, and incompatibility between anti-interference ability and field environment.

[0004] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: A rapid leak diagnosis method for a vacuum system used in large die castings, performed during the pre-vacuuming stage after mold closing and before molten metal injection, includes the following steps: Step S1, Multi-source data synchronous acquisition: Based on the vacuum system flow field and mold sealing structure, synchronously acquire pressure time series data of key nodes in the system, temperature distribution data of mold sealing path, airflow rate data of vacuum pump inlet, and micro-displacement deformation data of mold core sealing part; Step S2, Data Preprocessing: Time base alignment and noise suppression are performed on the collected heterogeneous data to form a unified time series dataset with spatiotemporal consistency; Step S3, Feature Extraction and Fusion: Extract the dynamic pressure decay characteristics, temperature spatial distribution gradient, airflow disturbance spectrum features and sealing surface micro-deformation parameters from the unified time series dataset, and construct a primary diagnostic feature vector through weighted fusion. Step S4, preliminary identification of leakage patterns: The preliminary diagnostic feature vector is input into a pre-trained die-casting-specific leakage pattern classifier to obtain preliminary leakage area determination results. The classifier adopts a hybrid architecture integrating decision tree and support vector machine, and is trained based on experimental samples of typical leakage scenarios. Step S5, coarse location of the leakage area: Based on the spatial topology diagram of the mold sealing unit, the preliminary leakage area determination result is mapped to the specific physical sealing unit; Step S6, Local pressure verification: Apply a controllable positive pressure pulse to the initially located suspected leak area and monitor the pressure response of the area and adjacent areas. Verify the leak location based on the propagation delay and amplitude attenuation law of the pressure wave. Step S7, Leakage degree quantification: Based on the fluid dynamics throttling principle, calculate the equivalent parameter of the leakage orifice diameter according to the pressure response data; Step S8, Diagnostic report generation: The equivalent parameters of the leakage aperture are rated according to the preset leakage level threshold system, and a structured diagnostic report containing the leakage location, level and treatment suggestions is generated and output.

[0005] According to the above technical solution, step S1, multi-source data synchronous acquisition includes: A high-precision absolute pressure sensor array was deployed based on the flow field simulation results of the vacuum system, covering key nodes of the main vacuum pipeline, branch pipelines and mold sealing chamber. The sampling frequency of the sensor was used to capture the dynamic response of leakage and ensure the integrity of the pressure time series data. A distributed fiber optic temperature sensing network is deployed along the mold sealing path (parting surface, slider mating surface, ejector pin hole) to continuously monitor the temperature distribution along the sealing path. Its spatial resolution is suitable for spatial location of leakage on the sealing surface. A thermal mass flow meter is installed in the stable region of the vacuum pump inlet flow field to collect airflow rate data in real time and capture airflow disturbance signals caused by leakage. Capacitive micro-displacement sensors are embedded in the mold slider locking mechanism, ejector pin guide sleeve and parting surface positioning pin to monitor the relative displacement deformation of the sealing surface under the action of mold closing force. The resolution is suitable for detecting micro-gap leakage. All sensors achieve nanosecond-level time synchronization through a hard-wired trigger synchronization mechanism, ensuring the spatiotemporal consistency of multi-source data.

[0006] According to the above technical solution, in step S2, data preprocessing specifically includes: Using the electrical signal of the die-casting machine closing the mold as the global time reference, the time stamp synchronization algorithm is used to uniformly calibrate the time of all sensor data to eliminate the time delay deviation of data acquisition. To address high-frequency noise in pressure time series data and airflow velocity data, an adaptive threshold denoising algorithm based on the Daubechies wavelet basis is adopted. Noise is filtered out through wavelet decomposition and threshold reconstruction. The threshold is dynamically adjusted based on the noise standard deviation and the signal energy ratio. To address the slow variation characteristics and spike interference of temperature distribution data, a moving window mid-range filtering algorithm based on data stationarity analysis is adopted, with the window length adaptively determined by the autocorrelation coefficient of the temperature data. To address the mechanical vibration interference in micro-displacement deformation data, a Kalman filter model is established. Random jitter is eliminated through state estimation and observation update iteration. The process noise and observation noise covariance of the filter model are optimized and determined based on offline calibration data.

[0007] According to the above technical solution, in step S3, feature extraction and fusion specifically involve: Pressure dynamic decay characteristics: Dynamic trend analysis is performed on the preprocessed pressure time series data, and characteristic parameters characterizing the pressure drop rate caused by leakage are obtained by first-order difference and exponential fitting. Temperature spatial distribution gradient: Based on temperature distribution data, a temperature field model of the sealing path is constructed. The extreme values ​​and distribution characteristics of the temperature gradient are obtained through spatial differentiation operations, reflecting the local temperature rise caused by gas friction or adiabatic compression near the leak point. Spectral characteristics of airflow disturbance: Spectral analysis of airflow velocity data is performed to extract the characteristic main frequency band energy proportion corresponding to leakage disturbance, and to distinguish between normal airflow fluctuations and abnormal disturbances caused by leakage; Microscopic deformation parameters of sealing surface: Statistical analysis of micro-displacement data during the mold closing stabilization stage is performed. The degree of dispersion is calculated to reflect the change in the gap of the sealing surface and to identify microscopic displacement anomalies caused by sealing failure. Based on the importance differences of multi-physics field features, the Relief-F algorithm is used to calculate the weight coefficients of each feature, and the primary diagnostic feature vector is constructed through the following feature weighted fusion model: in, This is the fused primary diagnostic feature vector. This represents the normalized dynamic pressure decay characteristics. This represents the normalized temperature spatial gradient characteristics. This represents the normalized spectral characteristics of airflow disturbance. The normalized microscopic deformation characteristics of the sealing surface; Let be the weight coefficients of the features, where As the pressure feature weight, For temperature feature weights, Flow feature weights The deformation feature weights are determined by training and optimizing using the Relief-F algorithm combined with leakage samples from die-casting scenarios.

[0008] According to the above technical solution, the training process of the die-casting-specific leakage mode classifier includes: A sample library covering typical leakage scenarios of die casting molds was constructed, including four types of failure modes: slider seal failure, ejector pin hole gap leakage, parting surface warping leakage, and exhaust valve poor sealing. Each type of mode covers full-condition samples with different leakage scales. Based on the orthogonal experimental design principle, under standardized die-casting process parameters, multiple sets of repeated data were collected for each leakage configuration to ensure the statistical significance of the samples. Pressure dynamic attenuation characteristics, temperature spatial distribution gradient, airflow disturbance spectrum characteristics, and sealing surface micro-deformation parameters were extracted from the collected samples. The Relief-F algorithm was used to calculate the weight coefficients of each feature and a primary diagnostic feature vector was constructed through a weighted fusion model. At the same time, the corresponding fault category labels were labeled. A hybrid architecture classifier is trained by combining 10-fold cross-validation and grid search: the decision tree module implements coarse-grained leakage region partitioning based on the Gini impurity minimization criterion, and the support vector machine module implements fine-grained leakage pattern discrimination using radial basis kernel function. The depth of the decision tree and the penalty parameters and kernel parameters of the support vector machine are optimized through cross-validation.

[0009] According to the above technical solution, the construction of the spatial topology graph is specifically as follows: Based on the computer-aided design model of the mold assembly, the boundary representation method is used to extract all geometric surfaces involved in vacuum sealing and establish the geometric topology model of the sealing unit. Each sealed geometric surface is assigned a unique hierarchical coding identifier, which is associated with its functional component type, sealing method, and spatial correspondence with the sensor. Based on the spatial adjacency relationship of geometric facets and the gas flow path, a sparse adjacency relationship matrix is ​​constructed. The matrix elements are the gas flow weights between adjacent facets, and the weight values ​​are calculated based on the gap size and flow resistance characteristics. By calibrating and aligning the mold clamping reference with the sensor installation coordinates, a mapping index between the topology diagram and the physical space is established, enabling precise mapping from logical judgment results to the physical sealing unit.

[0010] According to the above technical solution, step S6, the local pressure verification specifically includes: Based on the initially located leakage area, an electromagnetically controlled vacuum shut-off valve is used to isolate the area under test from other parts of the system, ensuring the relevance of the excitation signal. A miniature pneumatic actuator is used to inject dry compressed air into a chamber near the suspected leak point to generate a positive pressure pulse signal that meets the requirements of hydrodynamic excitation. The pulse parameters are optimized based on the chamber volume and leakage response characteristics. The pressure response curves of the excitation chamber and adjacent chambers are synchronously acquired using a high-precision pressure sensor array. The pressure response delay time (the time difference between the propagation of the pressure wave from the excitation chamber to the adjacent chamber) and the amplitude attenuation coefficient (the ratio of the peak pressure in the adjacent chamber to the peak pressure in the excitation chamber) are extracted. The accuracy of the leak area is verified by analyzing the spatiotemporal distribution of response delay, thus eliminating misjudgments caused by sensor drift or environmental interference.

[0011] According to the above technical solution, in step S7, the equivalent parameter of the leakage orifice diameter is calculated using the following formula: in, For the equivalent parameter of the leakage orifice diameter, The pressure response delay time extracted in step S6. The peak pressure difference between the adjacent chamber and the excitation chamber. To stimulate the initial pressure of the chamber, The density of the gas medium in the die-casting environment. The dynamic viscosity of the gas medium, To maximize the effective volume of the chamber, This is a proportionality coefficient related to the mold sealing structure and the shape of the leakage channel.

[0012] Based on the above technical solution, the proportionality coefficient is determined through the following calibration process: Multiple standard leakage orifice plates with different orifice diameters were selected, covering the expected leakage scale range of the die-casting vacuum system. The orifice diameters of the orifice plates were precisely calibrated and met the traceability requirements. Under the same environmental conditions as the die-casting production site, a local pressure verification procedure was performed on each standard orifice plate to collect key parameters such as pressure response delay time and peak pressure difference. Based on the collected standard data, a constrained nonlinear least squares fitting algorithm is used to establish the mapping relationship between the k value and the experimental parameters. The fitting objective function is: in, For the standard number of orifice plates, Let be the actual aperture of the i-th standard orifice plate. Let be the measured pressure response delay time corresponding to the i-th standard orifice plate. The measured peak pressure difference corresponds to the i-th standard orifice plate. Let be the measured initial pressure of the excitation chamber corresponding to the i-th standard orifice plate. The initial iteration value is used; the optimal k calibration value is obtained through iterative calculation.

[0013] According to the above technical solution, the leakage level threshold system specifically includes: Based on the influence of different leakage pore sizes on the internal density and mechanical properties of die castings, the critical thresholds for each leakage level were determined through orthogonal experiments and statistical analysis, covering the full scale range from microscopic sealing gaps to severe leakage. Leakage levels are classified into four levels: Level 1 leakage corresponds to a leakage orifice diameter equivalent parameter that is less than the minimum leakage threshold allowed by the process, which is determined based on the requirement that the quality pass rate of die-cast parts is higher than a predetermined value; Level 2 leakage corresponds to an orifice diameter parameter that is between the minimum leakage threshold and the critical threshold that does not affect the current batch production; Level 3 leakage corresponds to an orifice diameter parameter that is between the above-mentioned critical threshold and the risk threshold that requires shutdown for maintenance; Level 4 leakage corresponds to an orifice diameter parameter that is greater than the risk threshold. The handling recommendations for each level are formulated based on the die-casting production scheduling and quality control logic to achieve closed-loop linkage between leak diagnosis and production execution. Among them, level four leaks trigger an emergency stop command for the die-casting machine.

[0014] Compared with the prior art, the present invention has the following beneficial effects: This invention, through multi-level technological innovation, efficiently completes testing during the pre-vacuuming stage while ensuring complete synchronization between the diagnostic process and production cycle. This not only solves the problem of delayed testing in traditional methods but also achieves an order-of-magnitude improvement in positioning accuracy through a three-level positioning mechanism, enabling precise correlation to specific sealing units and significantly enhancing maintenance efficiency. Furthermore, the quantification model, through targeted modifications, is fully adapted to the characteristics of large-scale die-casting vacuum systems, exhibiting significantly higher accuracy than general-purpose models and providing a reliable basis for maintenance decisions. Simultaneously, the system maintains high diagnostic accuracy under complex operating conditions thanks to multi-source fusion anti-interference mechanisms and noise suppression algorithms, demonstrating excellent robustness. In summary, this invention effectively reduces casting scrap rates and maintenance costs, extends mold life, and thus significantly reduces overall production costs, demonstrating outstanding economic benefits and application prospects. Attached Figure Description

[0015] Figure 1 This is a flowchart of the diagnostic method of the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Example 1 like Figure 1 As shown, the technical solution of this invention is based on a five-level core logic chain consisting of multi-physics signal coupling capture, feature engineering optimization, spatial topology mapping, controllable excitation verification, and precise quantization calibration. Through technological innovation and synergistic linkage in each stage, it achieves fast, accurate, stable, and precise leak diagnosis. The specific technical solution is as follows: Multi-source data synchronous acquisition: The core technical logic of this step is based on the multi-physics coupling effect of leakage, to achieve comprehensive capture of characteristic signals and ensure spatiotemporal consistency.

[0018] During the pre-vacuuming stage, simulation analysis of the vacuum system's flow field and the mold sealing structure clarifies the propagation path of leaked gas and the sensitive areas of physical field disturbances. Four types of sensor arrays are then strategically deployed: a pressure sensor array captures the non-steady-state decay of negative pressure caused by the leak; a temperature sensor network captures local temperature changes induced by the leaking gas flow; a gas flow rate sensor captures flow field disturbances caused by the leak; and a micro-displacement sensor captures changes in the microscopic gaps of the sealing surface. These four types of sensors correspond to the four core physical dimensions of the leakage process: pressure, temperature, flow rate, and structure. Their signals are highly complementary and can collectively construct a complete physical profile of the leak. To eliminate spatiotemporal biases from multi-source data, a hard-wired trigger synchronization mechanism is employed, using the electrical signal indicating the die-casting machine's mold closing as the global synchronization benchmark. This ensures that the sampling start time deviation of all sensors is controlled within nanoseconds, achieving strict data alignment in the time dimension. Simultaneously, the calibration and association between sensor coordinates and mold sealing unit coordinates ensures precise data matching in the spatial dimension.

[0019] Data preprocessing: The core technical logic of this step is to achieve noise suppression and data standardization based on the noise characteristics and data heterogeneity of different physical signals.

[0020] The complex environment of a die-casting workshop leads to various noise components in the collected data: high-frequency noise in pressure and airflow rate data mainly comes from sensor circuit noise and vacuum pump airflow pulsation; spike interference in temperature data mainly comes from localized high-temperature radiation on the mold surface; and random jitter in micro-displacement data mainly comes from the mechanical vibration of the die-casting machine. Different filtering algorithms are employed to address the different sources and characteristics of noise. Adaptive wavelet threshold denoising algorithm: The pressure and airflow rate signals are decomposed into low-frequency approximate components (effective features) and high-frequency detail components (noise) by using the Daubechies wavelet basis. The threshold is dynamically adjusted based on the noise standard deviation and the signal energy ratio. The high-frequency detail components are thresholded and then the signal is reconstructed. This effectively suppresses high-frequency noise and completely preserves the abrupt features caused by leakage. The moving window mid-range filtering algorithm based on autocorrelation coefficient: determines the stationary interval of the data by calculating the autocorrelation coefficient of the temperature signal, and adaptively adjusts the filtering window length accordingly. A larger window length is used for the stationary interval to improve the filtering effect, and a smaller window length is used for the abrupt interval to avoid feature distortion. It effectively removes peak interference while retaining the true change of temperature gradient. The Kalman filter algorithm based on the uniform motion model establishes the state equation and observation equation of the micro-displacement signal. Through offline calibration and optimization of the process noise covariance and observation noise covariance, it realizes the iterative operation of state estimation and observation update, effectively suppresses random jitter caused by mechanical vibration, and extracts the true deformation trend of the sealing surface.

[0021] Meanwhile, using the closing position electrical signal as the time reference, all sensor data are uniformly calibrated through a timestamp synchronization algorithm to eliminate transmission delay and sampling deviation in the data acquisition process, forming a unified time series dataset that is spatiotemporally consistent and has sufficient noise suppression.

[0022] Feature Extraction and Fusion: The core technical logic of this step is to extract highly identifiable leakage features from multiphysics data and achieve feature complementarity and redundancy suppression through weighted fusion. The feature extraction process strictly adheres to the principles of clear physical meaning, sensitivity to leakage, and robustness to interference, specifically including four types of core features: To achieve optimal fusion of multiple features, the Relief-F algorithm is used to calculate the importance weights of each feature. The Relief-F algorithm iteratively calculates the correlation strength between features and leakage mode categories, assigning higher weights to features with high leakage mode identifiability and lower weights to redundant or interference-sensitive features, with the weight coefficients satisfying normalization conditions. A primary diagnostic feature vector is constructed using the following feature-weighted fusion model to achieve complementary enhancement of multi-physics features: Each feature has undergone max-min normalization to eliminate the impact of dimensional differences on the fusion effect and ensure the effectiveness of the weight coefficients.

[0023] Pressure dynamic decay characteristics: Under leak-free conditions, the pressure decay of the vacuum system follows an exponential law. Leakage will cause the decay coefficient to increase and deviate from the ideal curve. By performing first-order difference operation on the preprocessed pressure signal, the time series of the pressure change rate is obtained. Then, nonlinear least squares fitting is used to obtain the characteristic parameters (decay coefficient, initial pressure deviation) of the exponential decay model. These parameters can quantitatively reflect the negative pressure loss rate caused by leakage. Temperature spatial distribution gradient: When the leaking gas flows through the defect channel, it will experience local temperature rise (adiabatic compression) or temperature drop (adiabatic expansion) due to shear and compression effects, forming a temperature gradient field centered on the leak point. A three-dimensional temperature field model of the sealing path is constructed based on temperature distribution data. The temperature gradient vector (∂T / ∂x, ∂T / ∂y, ∂T / ∂z) is solved through spatial differentiation. The maximum magnitude of the gradient vector is taken as a characteristic parameter. The spatial location of this parameter is highly correlated with the leak point, and its magnitude reflects the energy intensity of the leaking gas flow. Spectral characteristics of airflow disturbance: In the absence of leakage, the airflow rate at the vacuum pump outlet is relatively stable, and the spectral energy is mainly concentrated in the low-frequency band; leakage leads to an increase in airflow turbulence intensity, resulting in energy concentration in a specific dominant frequency band (5-50Hz). By converting the airflow rate time-domain signal into a frequency-domain signal using Fast Fourier Transform (FFT), calculating the power spectral density (PSD), and extracting the energy proportion of the characteristic dominant frequency band as a characteristic parameter, this parameter can effectively distinguish between normal airflow fluctuations and abnormal disturbances caused by leakage. Microscopic deformation parameters of the sealing surface: The microscopic gaps of the sealing surface are the main channels for leakage. The force of the leaking airflow will disrupt the force balance of the sealing surface, resulting in micrometer-level relative displacement of the sealing surface. Statistical analysis of the micro-displacement signals during the mold closing stabilization stage is performed, and the standard deviation of the displacement sequence is calculated as a characteristic parameter. This parameter can quantitatively reflect the uniformity of the sealing surface gap. The larger the gap and the worse the uniformity, the larger the standard deviation value.

[0024] Preliminary Leakage Pattern Identification: The core technology of this step is based on machine learning models to achieve rapid classification of leakage patterns and preliminary identification of leakage regions. A hybrid architecture combining decision trees and support vector machines (SVM) is used to build the classifier, fully leveraging the complementary advantages of both models. The classifier training process is based on a sample library designed using orthogonal experiments. This library covers all typical leakage scenarios: leakage type (slider seal failure, ejector pin hole gap leakage, parting surface warping leakage, vent valve poor sealing), leakage scale (0.05mm-1.5mm, covering micro-leakage to severe leakage), and process parameters (different vacuum target values, different clamping forces), ensuring the comprehensiveness and representativeness of the samples. A combination of 10-fold cross-validation and grid search is used to optimize the model parameters, avoiding overfitting and ensuring the classifier's generalization ability.

[0025] Decision Tree Module: The C4.5 algorithm is used to construct the decision tree. The optimal splitting feature is selected based on the information gain ratio, enabling rapid processing of high-dimensional feature data and achieving coarse classification of leakage areas. The input to the decision tree is the fused primary diagnostic feature vector, and the output is a preset leakage area category (e.g., slider mating surface, pin hole, parting surface, exhaust valve). Multiple rounds of branching decisions narrow down the leakage area. The SVM module uses the region categories output by the decision tree as constraints to construct an SVM model with a radial basis function (RBF) kernel, handling nonlinear classification boundaries that the decision tree cannot distinguish. The SVM maps feature vectors to a high-dimensional feature space through the kernel function, constructing an optimal classification hyperplane to achieve fine-grained identification of leakage patterns and outputting specific preliminary leakage areas (such as the mating surface of slider #3 and the interface of branch pipes #2-4).

[0026] Coarse Leakage Area Location: The core technology of this step is based on digital topology mapping, which enables precise correlation between the leak area and the physical sealing unit, from the logical classification result. The construction process of the spatial topology diagram integrates digital modeling of the mold and flow field simulation analysis, specifically including three core steps: Based on the above spatial topology diagram, the preliminary leakage area determination result (logical classification result) output by the classifier is mapped to the physical sealing unit: first, the corresponding sealing unit category is matched by the coded identifier, and then the spatial range of the leakage area is narrowed by combining the adjacency matrix and the sensor spatial mapping index, so as to achieve coarse positioning of the leakage area (positioning accuracy ±10mm).

[0027] Geometric topology modeling: Based on the mold CAD model, the boundary representation method (B-rep) is used to extract all geometric patches involved in vacuum sealing, clarify the geometric contour, dimensional parameters and spatial position of each sealing unit, and assign a unique hierarchical coding identifier to each sealing unit (such as "HF-03" representing the 3rd slider mating surface). The code contains key information such as the sealing unit type and the mold area to which it belongs. Adjacency Matrix Construction: Based on gas flow path analysis, a sparse adjacency matrix of sealing units is constructed. The matrix elements represent the gas flow weights between adjacent sealing units. The weight values ​​are calculated based on gap size, flow resistance characteristics, and distance. The smaller the flow resistance and the closer the distance, the higher the weight value, reflecting the ease with which leaked gas propagates between different sealing units. Sensor spatial mapping: A coordinate transformation algorithm is used to associate the sensor's installation coordinates with the geometric coordinates of the sealing unit, establishing a mapping index between the sensor's acquisition range and the sealing unit. Each sensor corresponds to one or more sealing units, and the signals acquired by the sensors can directly reflect the physical state changes of the corresponding sealing unit.

[0028] Local pressure verification: The core technology of this step is based on the controllable pressure pulse excitation and pressure wave propagation characteristics to achieve accurate verification of the leak location and eliminate false positives. This step is a secondary verification of the coarse location results, enhancing leak characteristics and improving location accuracy through active excitation. Area Isolation: Based on the coarse location results, an electromagnetically controlled vacuum shut-off valve closes the connection between the suspected leak area and other areas, achieving independent isolation of the leak area. The isolation operation must ensure reliable sealing to prevent the excitation gas from diffusing into non-target areas and affecting the verification results. Positive pressure pulse design: A miniature pneumatic actuator injects dry compressed air into the isolated chamber, generating a controllable positive pressure pulse signal. Pulse parameters (amplitude, pulse width) are optimized based on chamber volume and leakage response characteristics: the amplitude must ensure the pressure wave effectively propagates to the leak point and generates a detectable response signal, while avoiding excessive pressure that could damage the mold sealing surface; the pulse width must cover the complete process of pressure wave propagation, reflection, and attenuation to ensure that all characteristics of the pressure response are captured. Pressure Response Acquisition and Analysis: A high-precision pressure sensor array is used to simultaneously acquire the pressure response curves of the excitation chamber and adjacent chambers. Two core characteristic parameters are extracted: pressure response delay time (the time difference between the propagation of the pressure wave from the excitation chamber to the adjacent chamber) and amplitude attenuation coefficient (the ratio of the peak pressure in the adjacent chamber to the peak pressure in the excitation chamber). The presence of a leak shortens the pressure wave propagation path and intensifies attenuation; therefore, the response delay time is shorter than in the leak-free state, and the amplitude attenuation coefficient is greater. By comparing the measured parameters with the baseline parameters in the leak-free state, the accuracy of the leak location is verified, and misjudgments caused by sensor drift, environmental interference, and other factors are eliminated, ultimately achieving millimeter-level (±5mm) location of the leak.

[0029] Leakage Quantification: The core technical logic of this step is to achieve precise quantification of the leakage orifice diameter based on a fluid dynamics model adapted to the die-casting scenario. The leakage orifice diameter is a core indicator characterizing the degree of leakage. Its calculation model is based on the fluid dynamics throttling principle, combined with multi-dimensional corrections based on the scenario characteristics of large-scale die-casting vacuum systems. The formula is as follows: The innovation of this model lies in the introduction of three correction terms specifically for large-scale die-casting scenarios, which breaks through the ideal assumptions of the general model: The physical meaning of each parameter in the model is clear and can be obtained through actual measurements: The pressure response delay time extracted during localized pressurization verification. The peak pressure difference between the adjacent chamber and the excitation chamber. To stimulate the initial pressure of the chamber, The density of the gas medium in the die-casting environment (calculated in real time based on temperature and humidity sensor data). The effective volume of the excitation chamber is calculated precisely based on the mold CAD model.

[0030] Gas dynamic viscosity Correction: In large die-casting systems, the flow state of leaked gas may be in the transitional region between laminar and turbulent flow, and the influence of viscous forces on the flow cannot be ignored. Therefore, gas dynamic viscosity is introduced. It can correct for flow loss caused by viscous forces and improve the calculation accuracy of the model in the transition flow region; chamber volume Correction: The chamber volume of a large die-casting vacuum system is much larger than that of a conventional leak detection scenario. The larger the chamber volume, the more significant the buffering effect on the pressure response and the smoother the pressure change rate. By using the cube root of the chamber volume as a scaling factor, the influence of volume on the pressure response is corrected, allowing the model to adapt to chamber structures of different volumes. The proportionality coefficient k is used to correct for the non-ideal characteristics of the leakage channel, including factors such as the channel cross-sectional shape (e.g., wedge, rectangle, annular), inner wall roughness, and channel length. These factors exhibit significant randomness and complexity in large-scale die-casting scenarios and cannot be accurately described by theoretical formulas; they must be determined through experimental calibration.

[0031] Diagnostic Report Generation: The core technical logic of this step is to establish a correlation system between leakage levels and treatment recommendations based on the quality requirements of the die-casting process, thereby enabling the engineering application of the diagnostic results. The construction of the leakage level threshold system is based on orthogonal experiments and statistical analysis. By systematically studying the influence of different leakage pore sizes on the internal density and mechanical properties of die-cast parts, the critical thresholds for each level are determined. Based on the above-mentioned level determination results, a structured diagnostic report is generated, which includes the leak location (specific sealing unit and coordinate range), leak level (level 1 to level 4), leak orifice diameter equivalent parameters (accurate to 0.01mm), and handling suggestions (maintenance timing, maintenance location, and maintenance method). The report is then pushed to the die-casting machine control system and production management platform through the industrial communication interface to achieve closed-loop linkage between leak diagnosis, production execution, and equipment maintenance.

[0032] Level 1 leakage (normal sealing): The equivalent parameter of the leakage orifice diameter is ≤ the minimum leakage threshold allowed by the process. This threshold is determined based on the requirement that the quality pass rate of the die casting is higher than a predetermined value (such as 99.5%). At this time, the leakage has no significant impact on the quality of the die casting and no maintenance operation is required. Level 2 leakage (minor leakage): The leakage orifice equivalent parameter is between the minimum leakage threshold and the critical threshold that does not affect the current batch production. At this time, the impact of leakage on die casting quality is within acceptable range, but maintenance is required after the current production batch is completed. Level 3 leakage (moderate leakage): The leakage orifice diameter equivalent parameter is between the above critical threshold and the risk threshold that requires shutdown for maintenance. At this time, the leakage has caused fluctuations in the die casting quality. It is necessary to stop the machine for maintenance immediately after the current mold production is completed to avoid batch defects. Level 4 Leakage (Severe Leakage): The equivalent parameter of the leakage pore size is greater than the risk threshold. At this time, the leakage will directly lead to serious defects in the die castings. It is necessary to immediately trigger the emergency stop command of the die casting machine, stop production and carry out emergency maintenance.

[0033] Example 2 This embodiment is a further refinement of Embodiment 1. The application scenario of this embodiment is a large engine block die-casting production line of an automotive parts company. The target die-cast part is a six-cylinder engine block (weight 50kg, maximum dimensions 800mm×500mm×300mm), with a mold cavity volume of 12L. The vacuum system adopts a central vacuum station + distributed pipeline design, including a main vacuum pipeline (50mm diameter), 4 branch pipelines (25mm diameter), and a mold sealing unit including a parting surface (3.2m circumference), 8 slider mating surfaces (each with an area of ​​0.08-0.12m²), 32 ejector pin holes (8mm diameter), and 4 vacuum channel interfaces. The production process requires a pre-vacuuming stage duration of 8 seconds, a target cavity vacuum degree of -0.095MPa (absolute pressure 5kPa), and an internal defect pass rate of ≥99% for the die-cast part.

[0034] Step 1, Multi-source data synchronous acquisition: Based on the vacuum system flow field simulation results (using ANSYS Fluent software for flow field simulation to analyze the propagation path and sensitive areas of leaked gas), optimize the deployment of four types of sensor arrays: Pressure sensor array: Eight high-precision absolute pressure sensors are deployed in the middle section of the main vacuum pipeline, at the ends of the four branch pipelines, and at the four corners of the mold cavity (upper left, upper right, lower left, and lower right). The sensor sampling frequency is set to 200Hz to ensure that dynamic pressure changes caused by leakage can be captured (the pressure change rate caused by leakage can reach 0.1MPa / s, and the 200Hz sampling frequency can satisfy the Nyquist sampling theorem and avoid signal aliasing). Distributed fiber optic temperature sensing network: A fiber optic temperature sensing network is deployed along the parting surface, the edges of the eight sliding block mating surfaces, and the central area. The spatial resolution of the sensing network is set to 5mm to ensure that it can capture local temperature changes near the leak point (the temperature change caused by the leak can reach 0.5-2℃, and the 5mm spatial resolution can accurately locate the spatial position of the temperature change). Airflow rate sensor: A thermal mass flow meter with a sampling frequency of 100Hz is installed in the stable flow field region at the vacuum pump inlet (1.5m away from the vacuum pump inlet) to capture airflow rate fluctuations caused by leakage; Micro-displacement sensors: Forty capacitive micro-displacement sensors are embedded at both ends of the eight slider locking mechanisms and at the top of the 32 ejector pin guide sleeves. The sensor resolution is set to 0.1μm to monitor the micro-displacement deformation of the sealing surface. Synchronization mechanism: The electrical signal (rising edge) of the die-casting machine closing the mold is used as the global synchronization trigger signal. The synchronous sampling of all sensors is achieved through a hard-wired synchronization module (synchronization error ≤10ns), ensuring the consistency of multi-source data in the time dimension.

[0035] Step 2, Data Preprocessing: A unified timestamp (timestamp accuracy ±10ns) is added to the data collected by each sensor using a timestamp synchronization algorithm. The time zero point (t=0) is taken as the trigger time of the closing electrical signal. All data are time-aligned to eliminate the time inconsistency caused by sensor transmission delay (usually 1-5ms) and sampling trigger deviation. An adaptive thresholding denoising algorithm based on the Daubechies 4 wavelet basis is used to perform three-level wavelet decomposition on the pressure and airflow velocity data, obtaining approximate components and three detail components. The threshold is calculated using the formula... (N is the number of sampling points, σ ​​is the noise standard deviation, estimated by calculating the median absolute deviation of the detail components). After soft thresholding of the detail components (setting the coefficients with absolute values ​​less than λ to zero and subtracting λ from the coefficients with absolute values ​​greater than λ), the signal is reconstructed, which effectively suppresses sensor circuit noise and vacuum pump airflow pulsation. A moving window mid-range filtering algorithm based on autocorrelation coefficient is adopted to calculate the autocorrelation coefficient of temperature data (lag order 10). When the autocorrelation coefficient is ≥0.8, the data is determined to be in the stationary range, and the window length is set to 7 sampling periods (35ms). When the autocorrelation coefficient is ≥0.8, the data is determined to be in the abrupt change range, and the window length is set to 3 sampling periods (15ms), effectively eliminating the spike interference caused by high temperature radiation on the mold surface. Establish the Kalman filter equation based on the uniform motion model, and the state equation is as follows: (A is the state transition matrix, A=[1,Δt;0,1], Δt is the sampling period of 5ms), the observation equation is (z k =Hx k +v k , where H is the observation matrix, H=[1,0]). The process noise covariance Q=diag([1e-8,1e-8]) and the observation noise covariance R=1e-6 are determined based on offline calibration data (collecting 100 sets of micro-displacement data under leak-free conditions and calculating the noise standard deviation). The random jitter is effectively suppressed through 5 iterations.

[0036] Step 3, Feature Extraction and Fusion: Perform first-order difference operations on the preprocessed pressure data to obtain the pressure change rate sequence, and then use an exponential function. (P0 is the initial pressure deviation, α is the attenuation coefficient,) Nonlinear least squares fitting was performed on the environmental pressure to obtain the attenuation coefficient α = 0.025s⁻¹, and after normalization, F P =0.72; Based on temperature distribution data, a three-dimensional temperature field model of the sealing path is constructed using the triangulation method. The temperature gradient vector is calculated using the central difference method, and the maximum magnitude of the gradient vector ∇T is obtained. max=0.8℃ / mm, normalized F T =0.65; Spectral characteristics of airflow disturbance: FFT transformation was performed on the airflow velocity data (1024 sampling points), the power spectral density was calculated, and the energy proportion of the 5-15Hz main frequency band, η=0.38, was extracted. After normalization, F... V =0.59; The standard deviation of the micro-displacement data during the stable phase of mold closing (t=3-6 seconds) is calculated to be σ=0.3μm. After normalization, F D =0.61; The weight coefficients are obtained by training using the Relief-F algorithm. =0.35、 =0.25、 =0.2、 =0.2, fused feature vector =0.66.

[0037] Step 4, Preliminary Leakage Mode Identification: Sample Library Construction: An orthogonal experimental design was used to construct the sample library. Experimental factors included leakage type (4 levels), leakage scale (5 levels), and process parameters (2 levels: vacuum -0.09MPa, -0.095MPa). 30 samples were collected for each experimental combination, totaling 4×5×2×30=1200 samples. Sample data included raw data from multiple sensor sources, preprocessed data, feature vectors, and corresponding fault category labels. The model parameters were optimized using 10-fold cross-validation and grid search. The decision tree depth was optimized to 8 layers, and the information gain ratio threshold was set to 0.1. The penalty parameter C=10 for the SVM and the radial basis function kernel parameter γ=0.1. After training, the overall accuracy of the classifier reached 96.8%, with a 98.2% accuracy rate for identifying slider mating surface leaks and a 97.5% accuracy rate for identifying pinhole leaks. fuse feature vectors =0.66 Input classifier, decision tree module output leakage region category is slider mating surface class, SVM module output preliminary leakage region is slider mating surface No. 3.

[0038] Step 5, coarse location of the leakage area: Based on the mold CAD model (drawn in SolidWorks2023), the geometric patch of the mating surface of the No. 3 slider is extracted using the boundary representation method. This patch is rectangular (300mm×400mm), assigned the code HF-03, and associated with its functional component (left slider), sealing type (face seal) and the corresponding 3 pressure sensors, 4 temperature sensors and 4 micro-displacement sensors. The gas flow weights of the mating surface of slider #3 with the adjacent cavity surface, the mating surface of slider #2, and the parting surface are 0.75, 0.3, and 0.5, respectively. The weight values ​​are calculated based on the gap size (the design gap between the mating surface of slider #3 and the cavity surface is 0.02mm, with low flow resistance and high weight). Through coordinate transformation (converting the sensor mounting coordinates from the machine tool coordinate system to the mold coordinate system), it was determined that the acquisition range of the three pressure sensors corresponding to the mating surface of the third slider covers the entire area of ​​the mating surface, the four temperature sensors are located at the four corners of the mating surface, and the four micro-displacement sensors are located at the edge of the mating surface. Based on the above mapping relationship, the preliminary identification results are mapped to the mating surface of the third slider of the physical sealing unit, with a coarse positioning range of the entire area of ​​the mating surface (300mm × 400mm) and a positioning error of ±10mm.

[0039] Step 6, Local pressure verification: Close the communication channel between the chamber of the No. 3 slider mating surface and other areas by electromagnetically controlling the vacuum shut-off valve. The response time of the shut-off valve is ≤50ms to ensure that the isolation operation is completed within the pre-vacuuming stage. The miniature pneumatic actuator injects dry compressed air, with the positive pressure pulse amplitude set to 50 kPa (absolute pressure 55 kPa) and the pulse width set to 2 seconds. The amplitude was chosen because 50 kPa generates a sufficiently strong pressure wave while remaining below the pressure resistance limit of the mold sealing surface (100 kPa), thus preventing damage to the sealing surface. The pulse width was chosen because the propagation and attenuation time of the pressure wave within the chamber is approximately 1.5 seconds, and a 2-second pulse width allows for complete capture of the pressure response process. A pressure sensor with a sampling frequency of 1kHz was used to acquire pressure response curves of the excitation chamber (the chamber containing the mating surface of the third slider) and the adjacent chamber (the left side region of the cavity). The pressure response delay time was extracted. =0.04s (time difference of pressure wave propagation from the excitation chamber to the adjacent chamber), amplitude attenuation coefficient =0.62 (ratio of peak pressure in the adjacent chamber (31 kPa) to peak pressure in the excitation chamber (50 kPa). Compared with the reference parameters under leak-free conditions ( =0.06s, amplitude attenuation coefficient=0.45) Compared with the previous time, the delay time is shortened and the attenuation coefficient is increased, which verifies that the leakage location is in the middle area of ​​the mating surface of the No. 3 slider (coordinates X=520mm, Y=380mm, Z=120mm), with a positioning error of ±3mm.

[0040] Step 7, Leakage Quantification: The gas medium is air. Based on on-site environmental sensor data (temperature 25℃, humidity 50%), the gas medium density ρ = 1.2 kg / m³ and dynamic viscosity μ = 1.8 × 10⁻⁻⁻⁶ are calculated. 5 Pa·s; Effective volume of the excitation chamber V = 0.8L = 8 × 10⁻ 4m³ (accurately calculated based on the mold CAD model); initial pressure of the excitation chamber =50 kPa (absolute pressure); peak pressure difference between adjacent chambers and the excitation chamber. =31kPa; proportionality coefficient k=0.85 (determined through calibration experiments); Substituting the above parameters into the leakage orifice diameter calculation model, the calculation process is as follows: Step 1, calculate the molecular part: =0.02232 Pa²·s²; The second step is to calculate the denominator: ≈5568Pa・kg / m³; Third step, calculate the ratio within the square root: 0.02232 / 5568 ≈ 4.01 × 10⁻ 6 m²; The fourth step is to calculate the leakage orifice diameter: d≈0.00028m=0.28mm.

[0041] Step 8, proportional coefficient k calibration: Select 6 standard leakage orifice plates with orifice diameters of 0.05mm, 0.1mm, 0.3mm, 0.8mm, 1.2mm and 1.5mm respectively. The orifice plates are calibrated by the National Institute of Metrology, and the calibration error is ≤±0.005mm. Under the same environmental conditions as the production site (temperature 25℃, humidity 50%, atmospheric pressure 101kPa), each standard orifice plate was installed on the standard mounting base on the mating surface of the No. 3 slider, and a local pressure verification procedure was performed to collect parameters such as , , and for each group of experiments. A constrained nonlinear least squares fitting algorithm is used, with the constraint k>0, and the fitting objective function is: in, Let be the actual aperture of the i-th standard orifice plate. These are the measured parameters from the corresponding experiment. The initial iteration value ( =1.0). Iterative calculations were performed using the Levenberg-Marquardt iterative algorithm, with the convergence condition being that the change in the sum of squared residuals ≤1e-6. The optimal k value was finally obtained as 0.85, with a fitting error ≤±2%.

[0042] Step 9, Leakage Level Determination and Report Generation: The leakage level threshold system in this embodiment is determined based on orthogonal experiments: Level 1 leakage ≤ 0.1mm, Level 2 leakage 0.1-0.3mm, Level 3 leakage 0.3-0.8mm, Level 4 leakage > 0.8mm. The calculated leakage orifice diameter d = 0.28mm, which belongs to Level 2 leakage (minor leakage); Generate a structured diagnostic report containing the following core information: Leakage location: Middle of the mating surface of slider No. 3 (coordinates X=520±3mm, Y=380±3mm, Z=120±3mm); Leakage level: Level 2 (minor leak); Leakage orifice diameter: 0.28 mm (error ±0.02 mm); Recommended action: After the current production batch is completed, disassemble slide number 3, replace the slide seal (material: fluororubber, hardness 70HA), and readjust the slide clearance to 0.02mm ± 0.005mm; The report is pushed to the die-casting machine control system (Siemens S7-1500) and production management platform via the PROFINET industrial Ethernet interface, enabling real-time display of diagnostic results and automatic creation of maintenance tasks.

[0043] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0044] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A rapid leak diagnosis method for vacuum systems of large die-cast parts, characterized in that: This process is performed during the pre-vacuuming stage after the die-casting machine closes the mold and before the molten metal is injected, and includes the following steps: Step S1, Multi-source data synchronous acquisition: Based on the vacuum system flow field and mold sealing structure, synchronously acquire pressure time series data of key nodes in the system, temperature distribution data of mold sealing path, airflow rate data of vacuum pump inlet, and micro-displacement deformation data of mold core sealing part; Step S2, Data Preprocessing: Time base alignment and noise suppression are performed on the collected heterogeneous data to form a unified time series dataset with spatiotemporal consistency; Step S3, Feature Extraction and Fusion: Extract the dynamic pressure decay characteristics, temperature spatial distribution gradient, airflow disturbance spectrum features and sealing surface micro-deformation parameters from the unified time series dataset, and construct a primary diagnostic feature vector through weighted fusion. Step S4, preliminary identification of leakage patterns: The preliminary diagnostic feature vector is input into a pre-trained die-casting-specific leakage pattern classifier to obtain preliminary leakage area determination results. The classifier adopts a hybrid architecture integrating decision tree and support vector machine, and is trained based on experimental samples of typical leakage scenarios. Step S5, coarse location of the leakage area: Based on the spatial topology diagram of the mold sealing unit, the preliminary leakage area determination result is mapped to the specific physical sealing unit; Step S6, Local pressure verification: Apply a controllable positive pressure pulse to the initially located suspected leak area and monitor the pressure response of the area and adjacent areas. Verify the leak location based on the propagation delay and amplitude attenuation law of the pressure wave. Step S7, Leakage degree quantification: Based on the fluid dynamics throttling principle, calculate the equivalent parameter of the leakage orifice diameter according to the pressure response data; Step S8, Diagnostic report generation: The equivalent parameters of the leakage aperture are rated according to the preset leakage level threshold system, and a structured diagnostic report containing the leakage location, level and treatment suggestions is generated and output.

2. The rapid leak diagnosis method for a vacuum system of large die-cast parts according to claim 1, characterized in that: In step S1, the synchronous acquisition of multi-source data includes: A high-precision absolute pressure sensor array was deployed based on the flow field simulation results of the vacuum system, covering key nodes of the main vacuum pipeline, branch pipelines and mold sealing chamber. The sampling frequency of the sensor was used to capture the dynamic response of leakage and ensure the integrity of the pressure time series data. A distributed fiber optic temperature sensing network is deployed along the mold sealing path to continuously monitor the temperature distribution on the sealing path. Its spatial resolution is suitable for spatial location of leakage on the sealing surface. A thermal mass flow meter is installed in the stable region of the vacuum pump inlet flow field to collect airflow rate data in real time and capture airflow disturbance signals caused by leakage. Capacitive micro-displacement sensors are embedded in the mold slider locking mechanism, ejector pin guide sleeve and parting surface positioning pin to monitor the relative displacement deformation of the sealing surface under the action of mold closing force. The resolution is suitable for detecting micro-gap leakage. All sensors achieve nanosecond-level time synchronization through a hard-wired trigger synchronization mechanism, ensuring the spatiotemporal consistency of multi-source data.

3. The rapid leak diagnosis method for a vacuum system of large die-cast parts according to claim 2, characterized in that: In step S2, data preprocessing specifically includes: Using the electrical signal of the die-casting machine closing the mold as the global time reference, the time stamp synchronization algorithm is used to uniformly calibrate the time of all sensor data to eliminate the time delay deviation of data acquisition. To address high-frequency noise in pressure time series data and airflow velocity data, an adaptive threshold denoising algorithm based on the Daubechies wavelet basis is adopted. Noise is filtered out through wavelet decomposition and threshold reconstruction. The threshold is dynamically adjusted based on the noise standard deviation and the signal energy ratio. To address the slow variation characteristics and spike interference of temperature distribution data, a moving window mid-range filtering algorithm based on data stationarity analysis is adopted, with the window length adaptively determined by the autocorrelation coefficient of the temperature data. To address the mechanical vibration interference in micro-displacement deformation data, a Kalman filter model is established. Random jitter is eliminated through state estimation and observation update iteration. The process noise and observation noise covariance of the filter model are optimized and determined based on offline calibration data.

4. The rapid leak diagnosis method for a vacuum system of large die-cast parts according to claim 3, characterized in that: In step S3, feature extraction and fusion specifically involve: Pressure dynamic decay characteristics: Dynamic trend analysis is performed on the preprocessed pressure time series data, and characteristic parameters characterizing the pressure drop rate caused by leakage are obtained by first-order difference and exponential fitting. Temperature spatial distribution gradient: Based on temperature distribution data, a temperature field model of the sealing path is constructed. The extreme values ​​and distribution characteristics of the temperature gradient are obtained through spatial differentiation operations, reflecting the local temperature rise caused by gas friction or adiabatic compression near the leak point. Spectral characteristics of airflow disturbance: Spectral analysis of airflow velocity data is performed to extract the characteristic main frequency band energy proportion corresponding to leakage disturbance, and to distinguish between normal airflow fluctuations and abnormal disturbances caused by leakage; Microscopic deformation parameters of sealing surface: Statistical analysis of micro-displacement data during the mold closing stabilization stage is performed. The degree of dispersion is calculated to reflect the change in the gap of the sealing surface and to identify microscopic displacement anomalies caused by sealing failure. Based on the importance differences of multi-physics field features, the Relief-F algorithm is used to calculate the weight coefficients of each feature, and the primary diagnostic feature vector is constructed through the following feature weighted fusion model: in, This is the fused primary diagnostic feature vector. This represents the normalized dynamic pressure decay characteristics. This represents the normalized temperature spatial gradient characteristics. This represents the normalized spectral characteristics of airflow disturbance. The normalized microscopic deformation characteristics of the sealing surface; The weight coefficients for the features are determined by training and optimization using the Relief-F algorithm combined with leakage samples from die-casting scenarios.

5. The rapid leak diagnosis method for a vacuum system of large die-cast parts according to claim 4, characterized in that: The training process for the die-casting-specific leakage pattern classifier includes: A sample library covering typical leakage scenarios of die casting molds was constructed, including four types of failure modes: slider seal failure, ejector pin hole gap leakage, parting surface warping leakage, and exhaust valve poor sealing. Each type of mode covers full-condition samples with different leakage scales. Based on the orthogonal experimental design principle, under standardized die-casting process parameters, multiple sets of repeated data were collected for each leakage configuration to ensure the statistical significance of the samples. Pressure dynamic attenuation characteristics, temperature spatial distribution gradient, airflow disturbance spectrum characteristics, and sealing surface micro-deformation parameters were extracted from the collected samples. The Relief-F algorithm was used to calculate the weight coefficients of each feature and a primary diagnostic feature vector was constructed through a weighted fusion model. At the same time, the corresponding fault category labels were labeled. A hybrid architecture classifier is trained by combining 10-fold cross-validation and grid search: the decision tree module implements coarse-grained leakage region partitioning based on the Gini impurity minimization criterion, and the support vector machine module implements fine-grained leakage pattern discrimination using radial basis kernel function. The depth of the decision tree and the penalty parameters and kernel parameters of the support vector machine are optimized through cross-validation.

6. The rapid leak diagnosis method for a vacuum system of large die-cast parts according to claim 5, characterized in that: The construction of the spatial topology graph is as follows: Based on the computer-aided design model of the mold assembly, the boundary representation method is used to extract all geometric surfaces involved in vacuum sealing and establish the geometric topology model of the sealing unit. Each sealed geometric surface is assigned a unique hierarchical coding identifier, which is associated with its functional component type, sealing method, and spatial correspondence with the sensor. Based on the spatial adjacency relationship of geometric facets and the gas flow path, a sparse adjacency relationship matrix is ​​constructed. The matrix elements are the gas flow weights between adjacent facets, and the weight values ​​are calculated based on the gap size and flow resistance characteristics. By calibrating and aligning the mold clamping reference with the sensor installation coordinates, a mapping index between the topology diagram and the physical space is established, enabling precise mapping from logical judgment results to the physical sealing unit.

7. The rapid leak diagnosis method for a vacuum system of large die-cast parts according to claim 1, characterized in that: Step S6, the local pressure verification specifically includes: Based on the initially located leakage area, an electromagnetically controlled vacuum shut-off valve is used to isolate the area under test from other parts of the system, ensuring the relevance of the excitation signal. A miniature pneumatic actuator is used to inject dry compressed air into a chamber near the suspected leak point to generate a positive pressure pulse signal that meets the requirements of hydrodynamic excitation. The pulse parameters are optimized based on the chamber volume and leakage response characteristics. A high-precision pressure sensor array was used to synchronously acquire the pressure response curves of the excitation chamber and adjacent chambers, and the pressure response delay time and amplitude attenuation coefficient were extracted. The accuracy of the leak area is verified by analyzing the spatiotemporal distribution of response delay, thus eliminating misjudgments caused by sensor drift or environmental interference.

8. A rapid leak diagnosis method for a vacuum system of large die-cast parts according to claim 7, characterized in that: In step S7, the equivalent parameter of the leakage orifice diameter is calculated using the following formula: in, For the equivalent parameter of the leakage orifice diameter, The pressure response delay time extracted in step S6. The peak pressure difference between the adjacent chamber and the excitation chamber. To stimulate the initial pressure of the chamber, The density of the gas medium in the die-casting environment. The dynamic viscosity of the gas medium, To maximize the effective volume of the chamber, This is a proportionality coefficient related to the mold sealing structure and the shape of the leakage channel.

9. A rapid leak diagnosis method for a vacuum system of large die-cast parts according to claim 8, characterized in that: The scaling factor is determined through the following calibration process: Multiple standard leakage orifice plates with different orifice diameters were selected, covering the expected leakage scale range of the die-casting vacuum system. The orifice diameters of the orifice plates were precisely calibrated and met the traceability requirements. Under the same environmental conditions as the die-casting production site, a local pressure verification procedure was performed on each standard orifice plate to collect key parameters such as pressure response delay time and peak pressure difference. Based on the collected standard data, a constrained nonlinear least squares fitting algorithm is used to establish the mapping relationship between the k value and the experimental parameters. The fitting objective function is: in, For the standard number of orifice plates, Let be the actual aperture of the i-th standard orifice plate. Let be the measured pressure response delay time corresponding to the i-th standard orifice plate. The measured peak pressure difference corresponds to the i-th standard orifice plate. Let be the measured initial pressure of the excitation chamber corresponding to the i-th standard orifice plate. The initial iteration value is used; the optimal k calibration value is obtained through iterative calculation.

10. A rapid leak diagnosis method for a vacuum system of large die-cast parts according to claim 1, characterized in that: The leakage level threshold system specifically includes: Based on the influence of different leakage pore sizes on the internal density and mechanical properties of die castings, the critical thresholds for each leakage level were determined through orthogonal experiments and statistical analysis, covering the full scale range from microscopic sealing gaps to severe leakage. Leakage levels are classified into four levels: Level 1 leakage corresponds to a leakage orifice diameter equivalent parameter that is less than the minimum leakage threshold allowed by the process, which is determined based on the requirement that the quality pass rate of die-cast parts is higher than a predetermined value; Level 2 leakage corresponds to an orifice diameter parameter that is between the minimum leakage threshold and the critical threshold that does not affect the current batch production; Level 3 leakage corresponds to an orifice diameter parameter that is between the above-mentioned critical threshold and the risk threshold that requires shutdown for maintenance; Level 4 leakage corresponds to an orifice diameter parameter that is greater than the risk threshold. The handling recommendations for each level are formulated based on the die-casting production scheduling and quality control logic to achieve closed-loop linkage between leak diagnosis and production execution. Among them, level four leaks trigger an emergency stop command for the die-casting machine.

Images