Online monitoring method and system for exhaust state of sprue bushing

Abstract

The invention relates to the technical field of intelligent manufacturing, in particular to an online monitoring method and system for the exhaust state of a sprue bushing, and the method comprises the steps: deploying a sensor array to synchronously collect multiple types of signals, comprehensively obtaining physical information in an exhaust channel, and providing data for subsequent analysis; multi-modal fusion features are extracted by performing time-frequency conjoint analysis on collected signals, key information of the exhaust process can be dug deeply from multiple dimensions, and accurate judgment of the exhaust state is assisted; the characteristics of the virtual flow field are calculated by constructing a digital twinborn model, so that a real exhaust scene can be simulated, and the gas flow condition can be intuitively presented; the characterization capability of the features on the exhaust state is enhanced by generating a cooperative discrimination vector; the abnormal state category is identified by means of the pre-training network, so that the problem can be quickly and accurately positioned, and countermeasures can be taken in time; the exhaust problem can be intelligently solved by self-adaptively adjusting parameters according to abnormal categories, the production process is optimized, and cost is reduced and efficiency is improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing technology, and in particular to a method and system for online monitoring of the venting status of a gating bushing. Background Technology

[0002] In injection molding, the sprue bushing is a key component of the mold gating system, and its built-in venting channels are crucial for expelling gases from the mold cavity. If venting is inadequate, residual gases may be compressed under high-pressure melt and potentially combust, or form surface defects such as gas marks and shrinkage cavities, severely impacting product quality. Traditional techniques primarily enhance passive venting capabilities by optimizing the size, layout, and connection method of the venting grooves in the sprue bushing to the mold cavity.

[0003] However, existing technologies have significant limitations in actual dynamic injection molding processes: they lack real-time sensing methods for the internal state of the venting channel, and can only infer whether venting is sufficient through indirect signals from product defects or the end of the cycle, failing to provide process early warning; they do not consider the real-time impact of dynamic changes in injection molding process parameters on gas flow behavior, resulting in a disconnect between venting efficiency and process status; when anomalies such as local blockage, gas stagnation, or backflow occur, existing systems cannot autonomously identify the specific anomaly type and make targeted process adjustments, still relying on manual intervention and trial and error, leading to long debugging cycles, significant material waste, and difficulty in meeting the intelligent control requirements of high-cycle, high-consistency production. Therefore, this invention proposes an online monitoring method and system for the venting status of the gating bushing. Summary of the Invention

[0004] The purpose of this invention is to solve the problems in the background art, and to propose an online monitoring method and system for the venting status of the gating bushing.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: The first aspect of this invention provides a method for online monitoring of the venting status of a gate bushing, comprising: S1. Deploy a multi-physics sensor array at key locations in the venting channel of the gate bushing to simultaneously collect pressure fluctuation signals, gas temperature gradient signals, and acoustic emission signals within the venting channel during the injection molding process. S2. Perform time-frequency joint analysis on the collected multi-physics field signals to extract the multi-modal fusion features of the exhaust process. The pressure fluctuation signal, gas temperature gradient signal and acoustic emission signal constitute the multi-physics field signal. The multi-modal fusion features include pressure peak frequency, temperature change phase difference and acoustic signal chaotic features. S3. Construct a physical-driven digital twin model of the venting state of the gating bushing and calculate the virtual flow field characteristics of the gas flow in the venting channel in real time. S4. Heterogeneous feature fusion of multimodal fusion features and virtual flow field features is performed to generate a collaborative discriminant vector for exhaust state; S5. Based on the collaborative discriminant vector of exhaust state, abnormal state categories are identified through a pre-trained temporal graph convolutional network; S6. If an abnormal state is identified, the local cooling parameters or injection speed curve near the exhaust channel will be adaptively adjusted according to the abnormal state category.

[0006] Furthermore, a multiphysics sensor array is deployed at key locations in the venting channel of the gating bushing to simultaneously acquire pressure fluctuation signals, gas temperature gradient signals, and acoustic emission signals within the venting channel during injection molding, including: Miniature pressure sensors, thermocouple arrays, and broadband acoustic emission sensors are respectively arranged at three key locations: the inlet, middle, and outlet of the exhaust channel, thus forming a multi-physics sensor array. With a sampling rate higher than the screw motion frequency of the injection molding machine, the data acquisition of each sensor in the multi-physics field sensor array is synchronously triggered to obtain the original synchronization signal set; The original synchronization signal set is preprocessed by bandpass filtering and power frequency noise suppression to generate time-aligned pressure fluctuation signal, gas temperature gradient signal and acoustic emission signal.

[0007] Furthermore, time-frequency joint analysis is performed on the acquired multi-physics field signals to extract multi-modal fusion features of the exhaust process, including: Based on the generated pressure fluctuation signal, wavelet packet decomposition is performed on it to extract the energy and zero-crossing rate of each subband, and the frequency of pressure peak occurrence per unit time is calculated to obtain the pressure peak frequency. Based on the generated gas temperature gradient signal, phase synchronization analysis is performed on the signals from different thermocouples to calculate the time difference between different sensors for the temperature change event and convert it into a phase difference, thereby obtaining the phase difference of the temperature change. Based on the generated acoustic emission signal, a recursive quantitative analysis is performed on it, and the length distribution entropy value of the diagonal structure in its recursive graph is calculated as a chaotic characteristic of the acoustic signal. The calculated pressure peak frequency, temperature change phase difference, and chaotic characteristics of the acoustic signal are combined to form a multimodal fusion feature.

[0008] Furthermore, a physical-driven digital twin model of the venting state of the gating bushing is constructed, and the virtual flow field characteristics of the gas flow within the venting channel are calculated in real time, including: Based on the three-dimensional geometric model of the gating bushing and the thermophysical property parameters of its material, a simplified physical model based on computational fluid dynamics and gas state equations is established as the core of the physics-driven digital twin model. The injection speed, melt temperature, and holding pressure of the injection molding machine are acquired in real time and input as dynamic boundary conditions into the simplified physical model. Within a simplified physical model, the transient flow equations of gas in the exhaust channel are solved using the finite volume method to simulate the virtual flow field. From the simulated virtual flow field, velocity nonuniformity, vorticity intensity, and local pressure gradient are extracted as virtual flow field features.

[0009] Furthermore, the virtual flow field characteristics include velocity nonuniformity, vorticity intensity, and local pressure gradient, and their calculation methods include: In the simplified physical model, the exhaust channel is discretized into several control volumes; Based on the solved transient flow equations, the gas velocity vector, pressure scalar, and vorticity field distribution data for each control volume are obtained. Based on the gas velocity vectors of all control volumes, calculate the normalized value of their velocity standard deviation as the velocity non-uniformity. The maximum vorticity modulus value is extracted from the entire vorticity field distribution data and then dimensionlessized to serve as the vorticity intensity. Calculate the pressure difference between all adjacent control volumes along the axis of the exhaust channel, and take the maximum absolute value as the local pressure gradient.

[0010] Furthermore, the multimodal fusion features and virtual flow field features are heterogeneously fused to generate a collaborative discriminant vector for the exhaust state, including: The acquired multimodal fusion features are mapped to a high-dimensional feature space through a kernel function to obtain the mapped features; Simultaneously, the virtual flow field characteristics are encoded into structured vectors based on their inherent physical constraints; An attention mechanism is used to dynamically assign weights to the mapped features and structured vectors, where the weights are dynamically calculated based on the real-time correlation between various features and the current injection molding process conditions. The two sets of feature vectors after weight allocation are concatenated, and then dimensionality reduction and deep fusion are performed through a single-layer fully connected neural network for feature fusion to generate a collaborative discriminant vector for exhaust state.

[0011] Furthermore, a pre-trained temporal graph convolutional network is used to identify exhaust gas anomaly state categories, including: Obtain the collaborative discrimination vectors of the exhaust state and construct a collaborative discrimination vector sequence in chronological order; The co-discriminative vector sequence is constructed as a time graph structure, where the co-discriminative vector at each sampling time is treated as a node, and the edges between nodes are determined by both temporal adjacency and cosine similarity between co-discriminative vectors. The constructed temporal graph structure is input into a pre-trained temporal graph convolutional network, which performs multi-layer convolutional aggregation along the temporal and feature dimensions. The output layer of the temporal graph convolutional network is connected to a Softmax classifier to calculate the probability that the current exhaust state belongs to a preset category of normal exhaust, partial stagnation, complete blockage, or gas backflow. The exhaust state category corresponding to the maximum value in the obtained probability distribution is taken as the final abnormal state identification result.

[0012] Furthermore, the Softmax function is used to calculate the exhaust state probability, and the category corresponding to the maximum exhaust state probability is taken as the abnormal state identification result, including: The output layer of the temporal graph convolutional network generates a four-dimensional original score vector, which corresponds to four exhaust state categories: normal exhaust, partial stagnation, complete blockage, and gas backflow. The original four-dimensional score vector is input into the Softmax function to calculate the normalized probability value for each state category; Compare the probability values ​​of the four state categories and determine the highest probability value and its corresponding state category; If the state category corresponding to the maximum probability value is normal exhaust, and the maximum probability value is greater than the preset first confidence threshold, then the current exhaust state is determined to be normal. If the state category corresponding to the maximum probability value is any one of the state categories of partial stagnation, complete blockage, or gas backflow, and the corresponding maximum probability value is greater than the preset second confidence threshold, then the current exhaust state is determined to be an abnormal state corresponding to that category; wherein, the second confidence threshold is set to be higher than the first confidence threshold; If the maximum probability value is not greater than the corresponding confidence threshold, an uncertain state signal is output, and a manual review process is triggered.

[0013] Furthermore, the local cooling parameters or injection speed curve near the exhaust channel are adaptively adjusted according to the abnormal state category, including: Receive the identified abnormal state category results; If local stagnation is identified, a control command is generated to adjust the mapping relationship according to the pre-stored process parameters and increase the local cooling water flow rate set value of the corresponding exhaust channel area. If a complete blockage is detected, a control command is generated to adjust the mapping relationship according to the pre-stored process parameters and increase the injection speed setting. If gas backflow is detected, a control command is generated to activate the injection molding machine's exhaust auxiliary device and adjust the holding pressure curve; The generated control commands are issued and executed at the start of the next injection cycle, and sensor data is collected in real time after execution.

[0014] A second aspect of the present invention provides an online monitoring system for the venting status of a gating bushing, comprising: Multiphysics signal acquisition module: Deploy a multiphysics sensor array at key locations in the venting channel of the gate bushing to simultaneously acquire pressure fluctuation signals, gas temperature gradient signals, and acoustic emission signals within the venting channel during the injection molding process; Multimodal feature extraction module: Performs time-frequency joint analysis on the acquired multi-physics field signals to extract multimodal fusion features of the exhaust process, including pressure peak frequency, temperature change phase difference, and chaotic features of acoustic signals; Virtual flow field calculation module: Constructs a physical-driven digital twin model of the venting state of the gating bushing and calculates the virtual flow field characteristics of the gas flow in the venting channel in real time; Heterogeneous feature fusion module: Heterogeneously fuses multimodal fusion features with virtual flow field features to generate a collaborative discriminant vector for exhaust state; Abnormal state identification module: Based on the collaborative discriminant vector of exhaust state, the abnormal state category is identified through a pre-trained temporal graph convolutional network; Process parameter adjustment module: If an abnormal state is identified, the local cooling parameters or injection speed curve near the exhaust channel will be adaptively adjusted according to the abnormal state category.

[0015] Compared with existing technologies, the advantages of the present invention in providing an online monitoring method and system for the venting status of a gating bushing are as follows: 1) By deploying sensor arrays at key locations, multiple types of signals are collected synchronously to comprehensively acquire physical information within the exhaust channel, providing a rich and accurate data foundation for subsequent analysis and ensuring that various subtle changes in the exhaust process can be captured; by performing time-frequency joint analysis on the collected multiple types of signals, multi-modal fusion features are extracted to characterize the exhaust process from different dimensions, effectively mining the key information hidden in the signals and providing strong feature support for accurately judging the exhaust status. 2) By constructing a physical-driven digital twin model and calculating virtual flow field characteristics, the real exhaust process can be simulated, the gas flow situation can be reflected intuitively, and potential problems can be predicted in advance, providing a virtual reference for state assessment. By fusing multimodal fusion features with virtual flow field features, a collaborative discrimination vector is generated, which integrates the advantages of features from different sources, enhances the feature expression ability, and improves the accuracy and reliability of exhaust state discrimination. 3) By using a pre-trained network to identify abnormal state categories, it can quickly and accurately classify and promptly detect problems, providing a clear direction for subsequent processing and ensuring the stable operation of the injection molding process; by adaptively adjusting parameters according to the abnormality category, it can achieve intelligent control, effectively solve the venting problem, improve product quality and production efficiency, and reduce scrap rate and production costs. Attached Figure Description

[0016] Figure 1 This is a flowchart of an online monitoring method for the venting status of a gating bushing proposed in this invention.

[0017] Figure 2 This is a block diagram of an online monitoring system for the venting status of a gating bushing proposed in this invention. Detailed Implementation

[0018] 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.

[0019] Please see Figure 1 This invention provides a method for online monitoring of the venting status of a gate bushing, comprising: S1. Deploy a multi-physics sensor array at key locations in the venting channel of the gate bushing to simultaneously collect pressure fluctuation signals, gas temperature gradient signals, and acoustic emission signals within the venting channel during the injection molding process. S2. Perform time-frequency joint analysis on the collected multi-physics field signals to extract the multi-modal fusion features of the exhaust process. The pressure fluctuation signal, gas temperature gradient signal and acoustic emission signal constitute the multi-physics field signal. The multi-modal fusion features include pressure peak frequency, temperature change phase difference and acoustic signal chaotic features. S3. Construct a physical-driven digital twin model of the venting state of the gating bushing and calculate the virtual flow field characteristics of the gas flow in the venting channel in real time. S4. Heterogeneous feature fusion of multimodal fusion features and virtual flow field features is performed to generate a collaborative discriminant vector for exhaust state; S5. Based on the collaborative discrimination vector of exhaust state, the abnormal state categories of exhaust obstruction, gas retention or poor exhaust are identified by a pre-trained temporal graph convolutional network. S6. If an abnormal state is identified, the local cooling parameters or injection speed curve near the exhaust channel will be adaptively adjusted according to the abnormal state category.

[0020] It should be further explained that, in the specific implementation process, a multiphysics sensor array is deployed at key locations in the venting channel of the gating bushing to simultaneously collect pressure fluctuation signals, gas temperature gradient signals, and acoustic emission signals within the venting channel during injection molding, including: Miniature pressure sensors, thermocouple arrays, and broadband acoustic emission sensors are respectively arranged at three key locations: the inlet, middle, and outlet of the exhaust channel, thus forming a multi-physics sensor array. With a sampling rate higher than the screw motion frequency of the injection molding machine, the data acquisition of each sensor in the multi-physics field sensor array is synchronously triggered to obtain the original synchronization signal set; The original synchronization signal set is preprocessed by bandpass filtering and power frequency noise suppression to generate time-aligned pressure fluctuation signal, gas temperature gradient signal and acoustic emission signal. Specifically, the multiphysics sensor array is selected and arranged accordingly; the miniature pressure sensor is an absolute pressure sensor with a dynamic response frequency of not less than 10kHz and a range covering -0.1 to 1.0MPa, to accurately capture transient pressure fluctuations in the exhaust channel caused by gas compression and release; the thermocouple array consists of at least three K-type miniature armored thermocouples, arranged in a non-equidistant manner along the axis of the exhaust channel, for example, densely arranged in the expected eddy or stagnant areas, to obtain a spatial distribution that can reflect temperature gradient changes; the operating frequency band of the broadband acoustic emission sensor should cover 20kHz to 1MHz to simultaneously capture broadband noise generated by gas turbulence and high-frequency acoustic emission events generated by bubble collapse; during installation, all sensors are tightly coupled to the exhaust channel wall through a special sealing clamp to ensure signal transmission efficiency and prevent melt leakage; Using a built-in high-stability clock source as the master clock, a parallel trigger bus sends a unified trigger pulse to the signal conditioning modules attached to all sensors. The signal conditioning module is a dedicated circuit integrated into the sensor probe or installed nearby. Its functions include amplifying, filtering (pre-filtering), and electrically isolating the weak analog signal output by the sensor, conditioning it to a standard voltage range suitable for analog-to-digital converter sampling. The sampling rate is set to more than 5 times the maximum feed frequency of the injection molding machine screw. For example, if the maximum screw movement frequency is 50Hz, the sampling rate should be set to at least 250Hz to satisfy the Nyquist sampling theorem and preserve signal details. At the start of acquisition, each sensor synchronously starts analog-to-digital conversion, converting the analog signal into a digital signal, and transmitting the raw data stream with a unified timestamp to the central processing unit to form a set of raw synchronization signals that are strictly aligned in the time dimension. The original synchronization signal set is preprocessed; in the bandpass filtering stage, for the pressure fluctuation signal, the passband is set to 0.1Hz to twice the screw motion frequency to retain low-frequency fluctuations related to the injection cycle and filter out ultra-low frequency drift; for the acoustic emission signal, the passband is set to 50kHz to 800kHz to focus on high-frequency components related to material impact and bubble dynamics; power frequency noise suppression adopts an adaptive notch filter algorithm to detect and eliminate 50Hz / 60Hz and its harmonic interference from the power grid in real time; after preprocessing, three clean signals with fully aligned time axes and improved signal-to-noise ratio are generated, namely the pressure fluctuation signal, the gas temperature gradient signal, and the acoustic emission signal.

[0021] It should be further explained that, in the specific implementation process, time-frequency joint analysis is performed on the collected multi-physics field signals to extract the multi-modal fusion features of the exhaust process, including: Based on the generated pressure fluctuation signal, wavelet packet decomposition is performed on it to extract the energy and zero-crossing rate of each subband, and the frequency of pressure peak occurrence per unit time is calculated to obtain the pressure peak frequency. Based on the generated gas temperature gradient signal, phase synchronization analysis is performed on the signals from different thermocouples to calculate the time difference between different sensors for the temperature change event and convert it into a phase difference, thereby obtaining the phase difference of the temperature change. Based on the generated acoustic emission signal, a recursive quantitative analysis is performed on it, and the length distribution entropy value of the diagonal structure in its recursive graph is calculated as a chaotic characteristic of the acoustic signal. The calculated pressure peak frequency, temperature change phase difference, and chaotic characteristics of the acoustic signal are combined to form a multimodal fusion feature. Specifically, the analysis of pressure fluctuation signals includes: decomposing the pressure fluctuation signal using a multi-order wavelet packet decomposition algorithm; selecting the db4 wavelet (i.e., the Daubechies series wavelet function with tight support, orthogonality, and a vanishing moment of 4) as the basis function, performing a 5-level complete decomposition to obtain 32 terminal sub-bands; calculating the proportion of energy of each terminal sub-band to the total signal energy, and counting the number of times the signal waveform of each sub-band crosses the zero level, i.e., the zero-crossing rate; and detecting pressure spikes on the original pressure fluctuation signal using a joint criterion based on amplitude and slope: when a... If the pressure value of a data point exceeds the average value of the five points before and after it within the sliding window by more than a preset peak detection amplitude (for example, the peak detection amplitude is set to 15% of the average value of the sliding window, which is used to determine whether pressure fluctuation constitutes a pressure peak event), and the sign of the difference value before and after that point changes from positive to negative, then it is determined to be a pressure peak event. The total number of peak events detected within a unit of time (i.e., a complete injection cycle) is counted and divided by the time length to obtain the pressure peak frequency. This feature reflects the severity of gas overcoming resistance during the exhaust process. The analysis of gas temperature gradient signals includes: identifying temperature abrupt change events from preprocessed multi-channel temperature signals, i.e., segments where the first derivative of temperature with time exceeds a preset temperature abrupt change detection threshold (e.g., this threshold is set to 1 degree Celsius per second, used to determine whether the rate of temperature change constitutes the minimum rate of change standard for a temperature abrupt change event); for the same temperature abrupt change event (such as the passage of a cold air mass), recording the time points at which it is detected on the thermocouples at the inlet, middle, and outlet of the channel; calculating the time difference between any two thermocouples (such as the inlet and outlet) detecting the same temperature abrupt change event; dividing this time difference by the duration of the temperature abrupt change event, and then multiplying by 360 degrees to convert the time delay into a phase difference; obtaining the temperature abrupt change phase difference by calculating the phase difference of all temperature abrupt change events within the analysis window and taking the statistical mean, which characterizes the propagation speed of temperature disturbances in the exhaust channel and indirectly reflects the smoothness of airflow; The analysis of acoustic emission signals includes: employing a recursive quantitative analysis method to reconstruct the phase space of the acoustic emission signals, transforming them into trajectories in a high-dimensional phase space by selecting appropriate time delays and embedding dimensions; calculating a recursion graph, which is a two-dimensional square matrix whose elements represent whether two points on the phase space trajectory are adjacent to each other within a preset recursive judgment distance threshold, where the recursive judgment distance threshold is set to 10% of the phase space diameter, used to determine whether two phase space points are close enough to be marked as recursion points in the recursion graph by the maximum Euclidean distance; analyzing the line segments formed by recursion points appearing continuously along the diagonal direction in the recursion graph (i.e., diagonal structure); calculating the distribution of the lengths of all line segments, and calculating the information entropy of this distribution, i.e., the length distribution entropy value, as a chaotic characteristic of the acoustic signal. This entropy value quantifies the degree of determinism of the acoustic emission dynamic system; among them, ordered periodic signals produce lower entropy, while chaotic or random signals produce higher entropy, thus distinguishing between steady airflow and complex acoustic events such as turbulence and bubble bursting; Finally, the calculated pressure peak frequency, temperature abrupt change phase difference, and acoustic signal chaotic features are normalized and then spliced ​​together to form a comprehensive feature vector, namely multimodal fusion features, which jointly describe the transient behavior of the exhaust process from three physical dimensions: pressure, temperature, and acoustics.

[0022] It should be further explained that, in the specific implementation process, a physical-driven digital twin model of the venting state of the gating bushing is constructed, and the virtual flow field characteristics of the gas flow within the venting channel are calculated in real time, including: Based on the three-dimensional geometric model of the gating bushing and the thermophysical property parameters of its material, a simplified physical model based on computational fluid dynamics and gas state equations is established as the core of the physics-driven digital twin model. The injection speed, melt temperature, and holding pressure of the injection molding machine are acquired in real time and input as dynamic boundary conditions into the simplified physical model. Within a simplified physical model, the transient flow equations of gas in the exhaust channel are solved using the finite volume method to simulate the virtual flow field. From the simulated virtual flow field, velocity nonuniformity, vorticity intensity, and local pressure gradient are extracted as virtual flow field features. The calculation method includes: discretizing the exhaust channel into several control volumes in the simplified physical model; obtaining the gas velocity vector, pressure scalar, and vorticity field distribution data for each control volume based on the solved transient flow equations; calculating the normalized value of the velocity standard deviation of all control volumes based on the gas velocity vectors, as the velocity nonuniformity; extracting the maximum vorticity modulus from the entire vorticity field distribution data and performing dimensionless processing, as the vorticity intensity; calculating the pressure difference between all adjacent control volumes along the exhaust channel axis, and taking the maximum absolute value as the local pressure gradient. Specifically, the core of constructing a physics-driven digital twin model is to simplify the physical model. The inputs include: a three-dimensional computer-aided design model of the gating bushing to define the geometric boundaries of the venting channel; and material properties such as thermal conductivity, specific heat capacity, and surface roughness of the mold steel and any residues that may adhere to it. A set of governing equations is established for the non-isothermal, compressible flow of gas within the microchannel. This set of governing equations integrates the conservation of mass and momentum based on the Navier-Stokes equations, constitutive relations based on the gas law (such as the ideal gas law), and considers wall slip effects. This model is a simplified model, meaning it reasonably simplifies the real three-dimensional turbulence model, for example, by using a reduced-order model from large eddy simulation, to meet real-time computation requirements while preserving the core physical mechanisms. The simplified physical model boundary conditions are dynamically input and solved. A real-time data interface reads the injection speed curve (varying over time), current melt temperature, and holding pressure setpoint from the injection molding machine controller. These parameters are mapped in real-time to the simplified physical model boundary conditions: injection speed determines the mass flow rate or pressure inlet condition at the channel inlet; melt temperature determines the thermal boundary conditions (isothermal or convective heat transfer) of the channel wall; holding pressure affects the gas back pressure. At each monitoring time step (e.g., 10 milliseconds), the latest boundary conditions are input into the simplified physical model. Within the simplified physical model, the computational domain (i.e., the exhaust channel) is spatially discretized using the finite volume method, dividing it into tens of thousands to hundreds of thousands of tiny control volumes (hexahedral meshes). For each control volume, the control equations are discretized and iteratively solved using a pressure-velocity coupling algorithm (e.g., SIMPLE or PISO) until convergence, thus obtaining the transient virtual flow field of gas flow within the exhaust channel at the current moment. The data includes the velocity vector, pressure scalar, and vorticity at the center point of each control volume. Finally, three virtual flow field features are extracted from the virtual flow field: Velocity nonuniformity extraction: Collect the velocity vector magnitudes (i.e., rate values) of all control volumes on the channel cross-section, and calculate the standard deviation of all rate values; divide the standard deviation by the average rate of the entire virtual flow field and normalize to obtain a dimensionless velocity nonuniformity index. A larger value indicates a more nonuniform velocity distribution in the virtual flow field, potentially indicating flow separation or blockage. Eddy power extraction: Calculate the eddy vector (curl of the velocity field) of each control volume and obtain its magnitude; iterate through all control volumes to find the maximum magnitude; divide the maximum eddy power by the ratio of the characteristic velocity to the characteristic length (e.g., the inlet horizontal velocity). The ratio of average velocity to the hydraulic diameter of the channel is dimensionless to obtain vorticity intensity, which reflects the intensity of rotational motion in the virtual flow field and is related to local vortices and energy dissipation. The local pressure gradient is extracted: along the axial direction of the exhaust channel, the static pressure difference between the center points of all adjacent control volumes is calculated sequentially, and the maximum absolute value of the static pressure difference is obtained. This maximum value is the local pressure gradient. This feature directly characterizes the section with the most severe pressure loss during the flow process and is a key indicator for judging local blockage or abrupt changes in the flow cross section. The three features of velocity nonuniformity, vorticity intensity and local pressure gradient together constitute a quantitative indicator describing the macroscopic state of the virtual flow field, namely the virtual flow field characteristics.

[0023] It should be further explained that, in the specific implementation process, the multimodal fusion features and virtual flow field features are heterogeneously fused to generate a collaborative discriminant vector for the exhaust state, including: The acquired multimodal fusion features are mapped to a high-dimensional feature space through a kernel function to obtain the mapped features; Simultaneously, the virtual flow field characteristics are encoded into structured vectors based on their inherent physical constraints; An attention mechanism is used to dynamically assign weights to the mapped features and structured vectors, where the weights are dynamically calculated based on the real-time correlation between various features and the current injection molding process conditions. The two sets of feature vectors after weight allocation are concatenated, and then dimensionality reduction and deep fusion are performed through a single-layer fully connected neural network for feature fusion to generate a collaborative discriminant vector for exhaust state. Specifically, the nonlinear transformation of multimodal fusion features includes: taking the obtained multimodal fusion feature vector (which may contain elements with different dimensions and numerical ranges) as input; using a Gaussian radial basis kernel function to map the original multimodal fusion feature vector to a high-dimensional feature space (e.g., Hilbert space); this mapping process is achieved by calculating the similarity between the input vector and a set of predefined kernel function center points, thereby transforming the nonlinear relationships that may exist in the original multimodal fusion feature vector into an approximately linearly separable form in the high-dimensional space, enhancing the expressive power of the features; The structured encoding of virtual flow field features includes encoding the extracted velocity nonuniformity, vorticity intensity, and local pressure gradient into structured vectors. For example, the first component of the encoded vector is the product of velocity nonuniformity and vorticity intensity, used to characterize turbulence intensity; the second component is the ratio of local pressure gradient to the reciprocal of the channel's equivalent diameter, used to characterize the local drag coefficient; and the third component is the weighted sum of the three features: velocity nonuniformity, vorticity intensity, and local pressure gradient, with the weights determined by the channel's geometry. The dynamic weight allocation using an attention mechanism involves: inputting the mapped high-dimensional multimodal feature vector and the encoded structured virtual flow field feature vector in parallel into an attention module. The core of this module is a neural network, whose additional input is the current real-time injection molding process parameters (such as injection speed and holding pressure). The neural network learns to calculate the attention weights of the two sets of feature vectors: for multimodal features, the weights are based on the correlation between real-time signal features and typical failure modes under the current process; the weights of virtual flow field features are based on the approximation of the current simulation conditions to the real physical world (which can be indirectly reflected by model errors calibrated using historical data). The weight calculation is normalized using the Softmax function to ensure that the sum of the two sets of weights is 1. The multimodal feature vector is multiplied by its corresponding weight, and the virtual flow field feature vector is multiplied by its corresponding weight to obtain the weighted feature representation. Feature concatenation and deep fusion to generate a co-discriminative vector includes: concatenating two sets of weighted feature vectors along their feature dimensions to form a joint feature vector; inputting this joint vector into a single-layer fully connected neural network for feature fusion, where the fully connected layer performs dimensionality reduction and deep fusion: through linear transformation and nonlinear activation functions (such as ReLU), it learns the complex cross-relationships between the two sets of features and extracts the most discriminative information, reducing the feature dimension to a preset lower dimension (such as 32 or 64 dimensions); the output of this fully connected layer is the final highly integrated exhaust state co-discriminative vector, representing the core state of the integrated physical sensor data and digital twin simulation information.

[0024] It should be further explained that, in the specific implementation process, a pre-trained temporal graph convolutional network is used to identify the categories of abnormal exhaust states, including: Obtain the collaborative discrimination vectors of the exhaust state and construct a collaborative discrimination vector sequence in chronological order; The co-discriminative vector sequence is constructed as a time graph structure, where the co-discriminative vector at each sampling time is treated as a node, and the edges between nodes are determined by both temporal adjacency and cosine similarity between co-discriminative vectors. The constructed temporal graph structure is input into a pre-trained temporal graph convolutional network, which performs multi-layer convolutional aggregation along the time dimension and the feature dimension to extract deep spatiotemporal pattern features. The output layer of the temporal graph convolutional network is connected to a Softmax classifier to calculate the probability that the current exhaust state belongs to one of the preset categories: normal exhaust, partial stagnation, complete blockage, or gas backflow. The exhaust state category corresponding to the maximum value in the obtained probability distribution is taken as the final abnormal state identification result. The process is as follows: the output layer of the temporal graph convolutional network generates a four-dimensional original score vector, corresponding to the four exhaust state categories: normal exhaust, partial stagnation, complete blockage, and gas backflow. The four-dimensional original score vector is input into the Softmax function to calculate the normalized probability value of each state category. The probability values ​​of the four state categories are compared to determine the maximum probability value and its corresponding state category. If the state category corresponding to the maximum probability value is normal exhaust, and the maximum probability value is greater than the preset first confidence threshold, then the current exhaust state is determined to be normal. If the state category corresponding to the maximum probability value is any of the following state categories: partial stagnation, complete blockage, or gas backflow, and the corresponding maximum probability value is greater than the preset second confidence threshold, then the current exhaust state is determined to be an abnormal state corresponding to that category. The second confidence threshold is set to be higher than the first confidence threshold. If the maximum probability value is not greater than the corresponding confidence threshold (i.e., it does not exceed the first confidence threshold in normal conditions, or it does not exceed the second confidence threshold in abnormal conditions), then an uncertain state signal is output, and a manual review process is triggered. Specifically, constructing the time graph structure includes: continuously acquiring a sequence of collaborative discriminant vectors within a time window (e.g., including the data from the two seconds prior to the current moment); treating each collaborative discriminant vector in the sequence as a node in the time graph structure; establishing edges between nodes using two preset rules: one is the temporal adjacency rule, where each node establishes undirected edges with the nodes at its previous and next time points, forming a time chain; the other is the feature similarity rule, calculating the cosine similarity between any two collaborative discriminant vectors. If the similarity exceeds a preset similarity connection threshold (e.g., set to 0.85, used to determine whether the state features of two different times are sufficiently similar to establish a direct connection in the time graph structure), then an edge is established between them to capture associations between discontinuous time points but with similar states; ultimately forming a dynamically weighted time graph structure. A temporal graph convolutional network is set up and applied for feature learning. This pre-trained temporal graph convolutional network contains multiple temporal graph convolutional layers. The operation performed by each layer is as follows: for each node in the temporal graph structure, it aggregates its own features, the features of its neighboring nodes connected by temporal edges, and the features of its neighboring nodes connected by feature-similar edges. During aggregation, neighbors of different edge types are assigned different weights (learnable parameters). Through multi-layer stacking, the temporal graph convolutional network captures the evolution pattern of node states over time and the propagation characteristics of states in the temporal graph structure, thereby extracting deep spatiotemporal pattern features that describe the dynamic changes of exhaust states within the entire time window. At the end of the temporal graph convolutional network, global pooling (such as attention pooling) is used to summarize the features of the entire temporal graph structure into a fixed-dimensional global state description vector. State classification and confidence determination include: inputting the global state description vector into the classification fully connected layer, outputting a four-dimensional original score vector, with the four dimensions corresponding to normal exhaust C1, partial stagnation C2, complete blockage C3, and gas backflow C4 respectively; inputting the original score vector into the Softmax function to calculate the normalized probability values ​​P(C1), P(C2), P(C3), and P(C4) of the four exhaust state categories, whose sum is 1; finding the maximum probability value P0 and its corresponding state category C0; Two confidence thresholds are preset: a first confidence threshold Tl (e.g., 0.7) and a second confidence threshold Th (e.g., 0.9), and the first confidence threshold Th is greater than the second confidence threshold Tl. If C0 is normal exhaust C1 and the maximum probability value P0 is greater than the first confidence threshold Tl, then it is finally determined to be a normal state; If C0 is any of the abnormal state categories (C2, C3, C4) and the maximum probability value P0 is greater than the second confidence threshold Th, then it is finally determined to be an abnormal state. If none of the above conditions are met (i.e., C0 is normal exhaust C1 but the maximum probability value P0 is not greater than the first confidence threshold Tl, or C0 is any abnormal state category C2 / C3 / C4 but the maximum probability value P0 is not greater than the second confidence threshold Th), then the classification confidence is considered insufficient, and an uncertain state signal is output. The uncertain state signal will trigger a prominent prompt on the human-machine interface and package all the original data, features and preliminary classification results of the current time period for offline review and analysis by operators or advanced diagnostic systems.

[0025] It should be further explained that, in the specific implementation process, the local cooling parameters or injection speed curve near the exhaust channel are adaptively adjusted according to the type of abnormal state, including: Receive the identified abnormal state category results; If local stagnation is identified, a control command is generated to adjust the mapping relationship according to the pre-stored process parameters and increase the local cooling water flow rate set value of the corresponding exhaust channel area. If a complete blockage is detected, a control command is generated to adjust the mapping relationship according to the pre-stored process parameters and increase the injection speed setting. If gas backflow is detected, a control command is generated to activate the injection molding machine's exhaust auxiliary device and adjust the holding pressure curve; The generated control commands are issued and executed at the start of the next injection cycle, and sensor data after execution is collected in real time to monitor the adjustment effect; Specifically, it receives the final state identification result (normal, some kind of definite abnormality, or uncertain); when the final state identification result is a definite abnormal state, it enters the corresponding adaptive control mechanism and calls the pre-stored process parameter adjustment mapping relationship library, which is the core knowledge base that is established offline and optimized online. If local stagnation is identified: the control algorithm responds by adjusting the mapping relationship according to the pre-stored process parameters, finds the corresponding cooling water flow adjustment coefficient from the mapping relationship, and generates a control command that aims to increase the local cooling water flow setpoint of the corresponding exhaust channel area. That is, based on the original process setpoint, a higher flow setpoint is generated according to the found cooling water flow adjustment coefficient. This control command aims to improve flow by enhancing local cooling to promote gas contraction. It is sent to the control valve of the corresponding cooling circuit in real time through the fieldbus. If a complete blockage is identified: the control algorithm responds by adjusting the mapping relationship according to the pre-stored process parameters. It retrieves the corresponding injection speed adjustment coefficient and the preset flash prevention threshold (the highest process parameter (such as injection speed or mold cavity pressure) safety limit value used to prevent melt overflow from the mold parting surface and generate flash defects) from the process parameter adjustment mapping relationship. The generated control command aims to increase the injection speed setpoint. That is, at the corresponding stage of the original injection speed curve, a new speed setpoint is calculated according to the retrieved injection speed adjustment coefficient, ensuring that the new setpoint does not exceed the flash prevention threshold. This command aims to clear the blockage by moderately increasing the injection driving force, while strictly preventing flash generation. At the same time, the corrected speed curve is sent to the injection molding machine controller. If gas backflow is identified: the control algorithm responds by adjusting the mapping relationship according to the pre-stored process parameters. The generated control command consists of two parts: first, sending an opening pulse signal to the integrated micro exhaust valve or negative pressure suction device (i.e. exhaust auxiliary device) to force the backflow gas to be discharged; second, adjusting the holding pressure setting curve, for example, fine-tuning the holding pressure setting value of a specific stage according to the mapping relationship (such as appropriately reducing it) to balance the pressure and facilitate exhaust. All generated control commands (all adjustments to process settings) are set to take effect at the start of the next injection cycle to avoid interfering with the current molding cycle. After the control command is issued, the effect monitoring mode is immediately entered. In the following 1-3 injection cycles, sensor signals are continuously collected and analyzed, multimodal fusion features, virtual flow field features, and collaborative discrimination vectors are recalculated, and the output probability changes of the time-series convolutional network are observed. If the probability value of the corresponding abnormal state category continues to decrease and returns to the normal range, the adjustment is considered effective. Otherwise, the adjustment strategy is upgraded (such as increasing the adjustment range) or the state uncertainty signal is triggered again, indicating that manual process intervention or equipment inspection is required.

[0026] Please see Figure 2 This invention provides an online monitoring system for the venting status of a gate bushing, comprising: Multiphysics signal acquisition module: Deploy a multiphysics sensor array at key locations in the venting channel of the gate bushing to simultaneously acquire pressure fluctuation signals, gas temperature gradient signals, and acoustic emission signals within the venting channel during the injection molding process; Multimodal feature extraction module: Performs time-frequency joint analysis on the acquired multi-physics field signals to extract multimodal fusion features of the exhaust process, including pressure peak frequency, temperature change phase difference, and chaotic features of acoustic signals; Virtual flow field calculation module: Constructs a physical-driven digital twin model of the venting state of the gating bushing and calculates the virtual flow field characteristics of the gas flow in the venting channel in real time; Heterogeneous feature fusion module: Heterogeneously fuses multimodal fusion features with virtual flow field features to generate a collaborative discriminant vector for exhaust state; Abnormal state identification module: Based on the collaborative discriminant vector of exhaust state, the abnormal state category is identified through a pre-trained temporal graph convolutional network; Process parameter adjustment module: If an abnormal state is identified, the local cooling parameters or injection speed curve near the exhaust channel will be adaptively adjusted according to the abnormal state category.

[0027] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. The focus of each embodiment is on its differences from other embodiments. In particular, the apparatus embodiments are described simply because they are fundamentally based on the method embodiments; relevant details can be found in the descriptions of the method embodiments.

[0028] For ease of description, the above devices are described separately by function as various units. Of course, in implementing this application, the functions of each unit can be implemented in one or more software and / or hardware.

[0029] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0030] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0031] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0032] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0033] Secondly: The accompanying drawings of the embodiments disclosed in this invention only involve the structures involved in the embodiments disclosed in this invention. Other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of this invention can be combined with each other. In conclusion, the above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for online monitoring of the venting status of a gating bushing, characterized in that: S1. Deploy a multi-physics sensor array at key locations in the venting channel of the gate bushing to simultaneously collect pressure fluctuation signals, gas temperature gradient signals, and acoustic emission signals within the venting channel during the injection molding process. S2. Perform time-frequency joint analysis on the collected multi-physics field signals to extract the multi-modal fusion features of the exhaust process. The pressure fluctuation signal, gas temperature gradient signal and acoustic emission signal constitute the multi-physics field signal. The multi-modal fusion features include pressure peak frequency, temperature change phase difference and acoustic signal chaotic features. S3. Construct a physical-driven digital twin model of the venting state of the gating bushing and calculate the virtual flow field characteristics of the gas flow in the venting channel in real time. S4. Heterogeneous feature fusion of multimodal fusion features and virtual flow field features is performed to generate a collaborative discriminant vector for exhaust state; S5. Based on the collaborative discriminant vector of exhaust state, abnormal state categories are identified through a pre-trained temporal graph convolutional network; S6. If an abnormal state is identified, the local cooling parameters or injection speed curve near the exhaust channel will be adaptively adjusted according to the abnormal state category.

2. The method for online monitoring of the venting status of a gating bushing according to claim 1, characterized in that, A multiphysics sensor array is deployed at key locations in the venting channel of the gating bushing to simultaneously acquire pressure fluctuation signals, gas temperature gradient signals, and acoustic emission signals within the venting channel during injection molding, including: Miniature pressure sensors, thermocouple arrays, and broadband acoustic emission sensors are respectively arranged at three key locations: the inlet, middle, and outlet of the exhaust channel, thus forming a multi-physics sensor array. With a sampling rate higher than the screw motion frequency of the injection molding machine, the data acquisition of each sensor in the multi-physics field sensor array is synchronously triggered to obtain the original synchronization signal set; The original synchronization signal set is preprocessed by bandpass filtering and power frequency noise suppression to generate time-aligned pressure fluctuation signal, gas temperature gradient signal and acoustic emission signal.

3. The method for online monitoring of the venting status of a gating bushing according to claim 1, characterized in that, Time-frequency joint analysis was performed on the acquired multi-physics field signals to extract multi-modal fusion features of the exhaust process, including: Based on the generated pressure fluctuation signal, wavelet packet decomposition is performed on it to extract the energy and zero-crossing rate of each subband, and the frequency of pressure peak occurrence per unit time is calculated to obtain the pressure peak frequency. Based on the generated gas temperature gradient signal, phase synchronization analysis is performed on the signals from different thermocouples to calculate the time difference between different sensors for the temperature change event and convert it into a phase difference, thereby obtaining the phase difference of the temperature change. Based on the generated acoustic emission signal, a recursive quantitative analysis is performed on it, and the length distribution entropy value of the diagonal structure in its recursive graph is calculated as a chaotic characteristic of the acoustic signal. The calculated pressure peak frequency, temperature change phase difference, and chaotic characteristics of the acoustic signal are combined to form a multimodal fusion feature.

4. The method for online monitoring of the venting status of a gating bushing according to claim 1, characterized in that, A physical-driven digital twin model of the venting state of the gating bushing is constructed, and the virtual flow field characteristics of the gas flow within the venting channel are calculated in real time, including: Based on the three-dimensional geometric model of the gating bushing and the thermophysical property parameters of its material, a simplified physical model based on computational fluid dynamics and gas state equations is established as the core of the physics-driven digital twin model. The injection speed, melt temperature, and holding pressure of the injection molding machine are acquired in real time and input as dynamic boundary conditions into the simplified physical model. Within a simplified physical model, the transient flow equations of gas in the exhaust channel are solved using the finite volume method to simulate the virtual flow field. From the simulated virtual flow field, velocity nonuniformity, vorticity intensity, and local pressure gradient are extracted as virtual flow field features.

5. The method for online monitoring of the venting status of a gating bushing according to claim 4, characterized in that, Virtual flow field characteristics include velocity nonuniformity, vorticity intensity, and local pressure gradient, and their calculation methods include: In the simplified physical model, the exhaust channel is discretized into several control volumes; Based on the solved transient flow equations, the gas velocity vector, pressure scalar, and vorticity field distribution data for each control volume are obtained. Based on the gas velocity vectors of all control volumes, calculate the normalized value of their velocity standard deviation as the velocity non-uniformity. The maximum vorticity modulus value is extracted from the entire vorticity field distribution data and then dimensionlessized to serve as the vorticity intensity. Calculate the pressure difference between all adjacent control volumes along the axis of the exhaust channel, and take the maximum absolute value as the local pressure gradient.

6. The method for online monitoring of the venting status of a gating bushing according to claim 1, characterized in that, Multimodal fusion features are heterogeneously fused with virtual flow field features to generate a collaborative discriminant vector for exhaust state, including: The acquired multimodal fusion features are mapped to a high-dimensional feature space through a kernel function to obtain the mapped features; Simultaneously, the virtual flow field characteristics are encoded into structured vectors based on their inherent physical constraints; An attention mechanism is used to dynamically assign weights to the mapped features and structured vectors, where the weights are dynamically calculated based on the real-time correlation between various features and the current injection molding process conditions. The two sets of feature vectors after weight allocation are concatenated, and then dimensionality reduction and deep fusion are performed through a single-layer fully connected neural network for feature fusion to generate a collaborative discriminant vector for exhaust state.

7. The method for online monitoring of the venting status of a gating bushing according to claim 1, characterized in that, The exhaust gas anomaly state categories are identified using a pre-trained temporal graph convolutional network, including: Obtain the collaborative discrimination vectors of the exhaust state and construct a collaborative discrimination vector sequence in chronological order; The co-discriminative vector sequence is constructed as a time graph structure, where the co-discriminative vector at each sampling time is treated as a node, and the edges between nodes are determined by both temporal adjacency and cosine similarity between co-discriminative vectors. The constructed temporal graph structure is input into a pre-trained temporal graph convolutional network, which performs multi-layer convolutional aggregation along the temporal and feature dimensions. The output layer of the temporal graph convolutional network is connected to a Softmax classifier to calculate the probability that the current exhaust state belongs to a preset category of normal exhaust, partial stagnation, complete blockage, or gas backflow. The exhaust state category corresponding to the maximum value in the obtained probability distribution is taken as the final abnormal state identification result.

8. The method for online monitoring of the venting status of a gating bushing according to claim 7, characterized in that, The Softmax function is used to calculate the exhaust state probability, and the category corresponding to the maximum exhaust state probability is taken as the abnormal state identification result, including: The output layer of the temporal graph convolutional network generates a four-dimensional original score vector, which corresponds to four exhaust state categories: normal exhaust, partial stagnation, complete blockage, and gas backflow. The original four-dimensional score vector is input into the Softmax function to calculate the normalized probability value for each state category; Compare the probability values ​​of the four state categories and determine the highest probability value and its corresponding state category; If the state category corresponding to the maximum probability value is normal exhaust, and the maximum probability value is greater than the preset first confidence threshold, then the current exhaust state is determined to be normal. If the state category corresponding to the maximum probability value is any one of the state categories of partial stagnation, complete blockage, or gas backflow, and the corresponding maximum probability value is greater than the preset second confidence threshold, then the current exhaust state is determined to be an abnormal state corresponding to that category; wherein, the second confidence threshold is set to be higher than the first confidence threshold; If the maximum probability value is not greater than the corresponding confidence threshold, an uncertain state signal is output, and a manual review process is triggered.

9. The method for online monitoring of the venting status of a gating bushing according to claim 1, characterized in that, Adaptively adjust local cooling parameters or injection speed curves near the exhaust channel based on the type of abnormal condition, including: Receive the identified abnormal state category results; If local stagnation is identified, a control command is generated to adjust the mapping relationship according to the pre-stored process parameters and increase the local cooling water flow rate set value of the corresponding exhaust channel area. If a complete blockage is detected, a control command is generated to adjust the mapping relationship according to the pre-stored process parameters and increase the injection speed setting. If gas backflow is detected, a control command is generated to activate the injection molding machine's exhaust auxiliary device and adjust the holding pressure curve; The generated control commands are issued and executed at the start of the next injection cycle, and sensor data is collected in real time after execution.

10. An online monitoring system for the venting status of a gating bushing, characterized in that, The system, which is applied to the online monitoring method for the venting status of a gating bushing as described in any one of claims 1-9, comprises: Multiphysics signal acquisition module: Deploy a multiphysics sensor array at key locations in the venting channel of the gate bushing to simultaneously acquire pressure fluctuation signals, gas temperature gradient signals, and acoustic emission signals within the venting channel during the injection molding process; Multimodal feature extraction module: Performs time-frequency joint analysis on the acquired multi-physics field signals to extract multimodal fusion features of the exhaust process, including pressure peak frequency, temperature change phase difference, and chaotic features of acoustic signals; Virtual flow field calculation module: Constructs a physical-driven digital twin model of the venting state of the gating bushing and calculates the virtual flow field characteristics of the gas flow in the venting channel in real time; Heterogeneous feature fusion module: Heterogeneously fuses multimodal fusion features with virtual flow field features to generate a collaborative discriminant vector for exhaust state; Abnormal state identification module: Based on the collaborative discriminant vector of exhaust state, the abnormal state category is identified through a pre-trained temporal graph convolutional network; Process parameter adjustment module: If an abnormal state is identified, the local cooling parameters or injection speed curve near the exhaust channel will be adaptively adjusted according to the abnormal state category.

Images