Injection molded article warpage defect prediction method and system based on physical information neural network

Abstract

This invention relates to the fields of injection molding quality control and artificial intelligence, specifically to a method and system for predicting warpage defects in injection molded products based on a physical information neural network. The method includes: acquiring the basic parameters of the injection molded product and a physical information neural network; collecting measured data and randomly selecting physical sampling points; substituting the measured data and physical sampling points into the network using a double loss function training method; the measured data includes the basic parameters and corresponding measured warpage displacement values; the physical sampling points output corresponding predicted temperature fields, predicted internal stress fields, and predicted warpage displacement fields through an output layer; inputting the basic parameters of the product to be predicted into the trained network, and outputting the warpage defect prediction results. This invention balances data-driven accuracy with the reliability of physical mechanisms, solving the technical problems of pure data-driven models relying on a large number of defect samples and the low efficiency of traditional physical simulation.

Description

Technical Field

[0001] This invention relates to the fields of injection molding quality control, artificial intelligence and physical simulation, and specifically to a method and system for predicting warpage defects in injection molded products based on a physical information neural network. Background Technology

[0002] Warpage defects are one of the most common quality problems in injection molded products. They manifest as unexpected out-of-plane deformation after demolding, directly affecting the product's assembly accuracy and performance. Existing warpage defect prediction methods are mainly divided into two categories:

[0003] One type is the purely data-driven artificial intelligence method, which trains neural network models through a large number of measured defect samples, relying on the completeness and diversity of the data. However, in actual engineering, the pass rate of injection molded products is extremely high, and it is difficult to obtain warpage defect samples, resulting in insufficient training data for the model, low prediction accuracy, poor generalization, and inability to adapt to products of different materials or structures.

[0004] Another type is the traditional physical simulation method (such as Moldflow software), which uses physical equations such as heat conduction and stress-strain to solve numerically. It does not rely on defect samples, but it suffers from low solution efficiency and cumbersome parameter adjustment, making it difficult to meet the needs of real-time prediction in production sites.

[0005] Physical Information Neural Networks (PINNs), as a novel AI model that integrates physical mechanisms and data-driven approaches, embed physical equations as constraints into the network training, ensuring prediction accuracy and consistency with the underlying mechanisms even with limited data. However, there is currently no technical solution to combine PINNs with the prediction of warpage defects in injection-molded products. There is an urgent need to construct PINN models adapted to the characteristics of injection molding to overcome the dual bottlenecks of sample dependence and low efficiency in existing methods. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention provides a method and system for predicting warpage defects in injection molded products based on a physical information neural network. The aim is to achieve accurate and efficient prediction of warpage defects through a physical equation constraint combined with a data-driven architecture.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] On one hand, embodiments of the present invention provide a method for predicting warpage defects in injection molded products based on a physical information neural network, the method comprising the following steps:

[0009] S100, Obtain the basic parameters and physical information neural network of the injection molded product. The physical information neural network includes an input layer, a network body, and an output layer. The basic parameters include process parameters, material parameters, and structural parameters.

[0010] S200: Collect measured data and randomly select physical sampling points. Substitute the measured data and the physical sampling points into a double loss function to train the physical information neural network, thereby obtaining the trained physical information neural network. The measured data includes basic parameters and corresponding measured warpage displacement values. The physical sampling points are output through the output layer, which outputs the corresponding predicted temperature field, predicted internal stress field, and predicted warpage displacement field.

[0011] S300: Input the basic parameters of the injection molded product to be predicted into the trained physical information neural network, and output the warpage defect prediction result.

[0012] Optionally, in S200, the step of collecting the measured data and randomly selecting physical sampling points, substituting the measured data and the physical sampling points into a double loss function to train the physical information neural network, and obtaining the trained physical information neural network, includes:

[0013] S210, collect multiple sets of measured data, input the measured data into the physical information neural network to obtain the predicted warping displacement, and calculate the data loss based on the predicted warping displacement and the measured warping displacement; the measured data includes process parameters, material parameters, structural parameters and corresponding measured warping displacement values.

[0014] S220: Randomly select time points at the end of the filling stage, the holding pressure stage, and the cooling stage; randomly select spatial points at locations with uneven wall thickness, near the gate, and at the edge region to generate space-time physical sampling points; input the physical sampling points into the physical information neural network to obtain the predicted temperature field, the predicted internal stress field, and the predicted warpage displacement field; calculate the physical loss based on the predicted temperature field, the predicted internal stress field, and the predicted warpage displacement field.

[0015] S230, the total loss is obtained by weighted summation of the data loss and the physical loss, and the network parameters are updated by minimizing the total loss through an optimization algorithm. This process is repeated iteratively until the training termination condition is met, and the trained physical information neural network is obtained.

[0016] Optionally, in S210, the calculation of data loss based on the predicted warping displacement and the measured value of the warping displacement includes:

[0017] The data loss is obtained by squaring the difference between the predicted warping displacement and the measured warping displacement, and then averaging the squared errors of all measured samples.

[0018] Optionally, in S220, the calculation of physical loss based on the predicted temperature field, the predicted internal stress field, and the predicted warping displacement field includes:

[0019] S221, Substitute the partial derivative of the predicted temperature field with respect to time and the Laplace operator of the predicted temperature field into the three-dimensional unsteady heat conduction equation to calculate the residual loss of the heat conduction equation.

[0020] S222, by substituting the predicted stress increment tensor, the predicted total strain increment tensor, the predicted viscous strain increment, and the predicted volume shrinkage strain increment into the thermo-elastic-plastic stress-strain equation, the residual loss of the stress-strain equation is calculated.

[0021] S223, Substitute the predicted warping displacement vector, the thermal stress load vector calculated based on the predicted temperature field, and the residual stress load vector calculated based on the predicted stress field into the structural deformation equilibrium equation to calculate the residual loss of the structural deformation equation.

[0022] S224, sum the mean square errors of the residual losses of the heat conduction equation, the stress-strain equation, and the structural deformation equation to obtain the physical loss.

[0023] Optionally, in S221, substituting the partial derivative of the predicted temperature field with respect to time and the Laplace operator of the predicted temperature field into the three-dimensional unsteady heat conduction equation to calculate the residual loss of the heat conduction equation includes:

[0024] The partial derivative of the predicted temperature field with respect to time is calculated to obtain the rate of temperature change. The Laplace operator of the predicted temperature field is calculated to obtain the temperature diffusion term. The rate of temperature change and the temperature diffusion term are substituted into the three-dimensional unsteady heat conduction equation to obtain the residual of the heat conduction equation. The residual of the heat conduction equation is squared and then the mean is calculated to obtain the residual loss of the heat conduction equation.

[0025] Optionally, in S222, substituting the predicted stress increment tensor, the predicted total strain increment tensor, the predicted viscous strain increment, and the predicted volumetric shrinkage strain increment into the thermoelastic-plastic stress-strain equation to calculate the residual loss of the stress-strain equation includes:

[0026] Calculate the first deviation between the predicted total strain increment tensor and the elastic strain component of the predicted stress increment tensor; calculate the second deviation between the predicted total strain increment tensor and the thermal expansion strain component; calculate the third deviation between the predicted total strain increment tensor and the predicted viscous strain increment; calculate the fourth deviation between the predicted total strain increment tensor and the predicted volumetric shrinkage strain increment; sum the first deviation, the second deviation, the third deviation, and the fourth deviation to obtain the stress-strain equation residual; square the stress-strain equation residual and calculate the mean value to obtain the stress-strain equation residual loss.

[0027] Optionally, in S223, substituting the predicted warping displacement vector, the thermal stress load vector calculated based on the predicted temperature field, and the residual stress load vector calculated based on the predicted internal stress field into the structural deformation equilibrium equation to calculate the residual loss of the structural deformation equation includes:

[0028] Calculate the first product of the product stiffness matrix and the predicted warping displacement vector, substitute the predicted temperature field into the thermal stress calculation model to obtain the thermal stress load vector, substitute the predicted internal stress field into the residual stress calculation model to obtain the residual stress load vector, sum the thermal stress load vector and the residual stress load vector to obtain the total load vector, calculate the second deviation between the first product and the total load vector to obtain the structural deformation equation residual, square the structural deformation equation residual and calculate the mean value to obtain the structural deformation equation residual loss.

[0029] Optionally, in S300, the step of inputting the basic parameters of the injection-molded product to be predicted into the trained physical information neural network and outputting the warpage defect prediction result includes:

[0030] S310, the process parameters, material parameters and structural parameters of the injection molded product to be predicted are input into the trained physical information neural network model, and the temperature field distribution, internal stress field distribution and warping displacement are output through the output layer;

[0031] S320, determine the warping defect level based on the warping displacement, identify the uneven cooling area based on the temperature field distribution, and locate the stress concentration part based on the internal stress field distribution.

[0032] On the other hand, embodiments of the present invention provide a warpage defect prediction system for injection molded products based on a physical information neural network, including:

[0033] At least one processor;

[0034] At least one memory for storing at least one program;

[0035] When the at least one program is executed by the at least one processor, the at least one processor performs the method described above.

[0036] On the other hand, embodiments of the present invention provide a computer-readable storage medium storing a processor-executable program, which, when executed by a processor, is used to perform the above-described method.

[0037] The embodiments of the present invention have the following beneficial effects:

[0038] This invention embeds the three-dimensional unsteady heat conduction equation, thermo-elastic-plastic stress-strain equation, and structural deformation equilibrium equation as physical constraints into the neural network training process, constructing a dual loss function framework that combines data loss and physical loss. This makes the network output both closely match the measured data and conform to physical laws, requiring only a small amount of measured data to achieve accurate prediction of warping defects, and significantly reducing the need for training samples.

[0039] This invention employs a multi-task output design to simultaneously predict temperature field distribution, internal stress field distribution, and warping displacement. It not only enables quantitative prediction of defects but also identifies uneven cooling areas based on the temperature field and locates stress concentration points based on the internal stress field, providing a basis for process optimization.

[0040] This invention utilizes an end-to-end learning mechanism of a physical information neural network, which significantly improves the efficiency of the prediction process compared to traditional physical simulation methods, making it suitable for real-time prediction scenarios in production sites.

[0041] This invention balances data-driven accuracy with physical mechanism reliability, has high average prediction accuracy, is applicable to warpage prediction of injection molded products of different materials and structures, has strong generalization ability, and is highly practical for engineering applications. Attached Figure Description

[0042] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0043] Figure 1 This is a flowchart illustrating the method for predicting warpage defects in injection molded products based on a physical information neural network in an embodiment of the present invention.

[0044] Figure 2 This is a structural block diagram of the injection molding product warpage defect prediction system based on physical information neural network in an embodiment of the present invention. Detailed Implementation

[0045] The following will provide a clear and complete description of the concept, specific structure, and technical effects of the present invention in conjunction with embodiments and accompanying drawings, so as to fully understand the purpose, solution, and effects of the present invention. It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other.

[0046] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. In the following description, when referring to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the embodiments of this invention; they are merely examples of apparatuses and methods consistent with some aspects of the embodiments of this invention as detailed in the appended claims.

[0047] Unless otherwise defined, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used in this invention is for descriptive purposes only and is not intended to limit the invention.

[0048] refer to Figure 1 ,like Figure 1 The figure shown is an embodiment of the present invention providing a method for predicting warpage defects in injection molded products based on a physical information neural network. The method includes the following steps:

[0049] S100, Obtain the basic parameters and physical information neural network of the injection molded product. The physical information neural network includes an input layer, a network body, and an output layer. The basic parameters include process parameters, material parameters, and structural parameters.

[0050] S200: Collect measured data and randomly select physical sampling points. Substitute the measured data and the physical sampling points into a double loss function to train the physical information neural network, thereby obtaining the trained physical information neural network. The measured data includes basic parameters and corresponding measured warpage displacement values. The physical sampling points are output through the output layer, which outputs the corresponding predicted temperature field, predicted internal stress field, and predicted warpage displacement field.

[0051] S300: Input the basic parameters of the injection molded product to be predicted into the trained physical information neural network, and output the warpage defect prediction result.

[0052] This invention provides a method and system for predicting warpage defects in injection molded products based on a physical information neural network. By embedding the heat conduction equation, stress-strain equation, and structural deformation equation as physical constraints into the neural network training, a dual loss function framework that integrates data-driven and physical mechanisms is constructed. It can achieve accurate prediction of warpage defects with only a small amount of measured data, solving the technical problems of pure data-driven models relying on a large number of defect samples and the low efficiency of traditional physical simulation. It significantly improves prediction accuracy and computational efficiency and is applicable to warpage prediction of injection molded products of different materials and structures.

[0053] The core of this embodiment lies in the design of the Physical Information Neural Network (PINN) architecture and the construction of the dual loss function. The Physical Information Neural Network consists of three parts: an input layer, the network body, and an output layer.

[0054] The input layer receives basic parameters, covering core process, material, and structural parameters that affect warpage. These include: injection temperature, injection pressure, holding pressure, holding time, injection speed, cooling time, mold temperature, melt flow rate, raw material moisture content, number of gates, venting groove width, and average wall thickness. All parameters are measurable and adjustable engineering parameters, requiring no additional complex testing equipment.

[0055] The main network uses a 3-5 layer fully connected network with 128-512 neurons in the hidden layers. The activation function is ReLU or Tanh. Batch normalization layers are set between adjacent hidden layers to improve training stability and avoid overfitting. If it is necessary to handle temporal process parameters (such as the change of holding pressure over time), it can be replaced with an improved CNN (adding a one-dimensional convolutional layer to capture temporal features).

[0056] The output layer adopts a multi-task output design, simultaneously outputting three key results: temperature field distribution (temperature values ​​at each node), internal stress field distribution (stress tensors at each node), and warpage displacement (maximum warpage displacement and displacement distribution in key areas), which not only enables defect prediction but also allows analysis of the root causes of warpage.

[0057] In some embodiments, in S200, the step of collecting the measured data and randomly selecting physical sampling points, substituting the measured data and the physical sampling points into a double loss function to train the physical information neural network, and obtaining the trained physical information neural network, includes:

[0058] S210, collect multiple sets of measured data, input the measured data into the physical information neural network to obtain the predicted warping displacement, and calculate the data loss based on the predicted warping displacement and the measured warping displacement; the measured data includes process parameters, material parameters, structural parameters and corresponding measured warping displacement values.

[0059] S220: Randomly select time points at the end of the filling stage, the holding pressure stage, and the cooling stage; randomly select spatial points at locations with uneven wall thickness, near the gate, and at the edge region to generate space-time physical sampling points; input the physical sampling points into the physical information neural network to obtain the predicted temperature field, the predicted internal stress field, and the predicted warpage displacement field; calculate the physical loss based on the predicted temperature field, the predicted internal stress field, and the predicted warpage displacement field.

[0060] S230, the total loss is obtained by weighted summation of the data loss and the physical loss, and the network parameters are updated by minimizing the total loss through an optimization algorithm. This process is repeated iteratively until the training termination condition is met, and the trained physical information neural network is obtained.

[0061] In some embodiments, S210, calculating the data loss based on the predicted warpage displacement and the measured warpage displacement includes:

[0062] The data loss is obtained by squaring the difference between the predicted warping displacement and the measured warping displacement, and then averaging the squared errors of all measured samples.

[0063] Specifically, the data loss is based on a small amount of measured data (50-100 sets), and the mean square error between the predicted warping displacement and the measured value is calculated to ensure the model's practical adaptability:

[0064] ;

[0065] in, The first prediction for PINN Group sample warping displacement. The warping displacement was measured by a 3D scanner. This represents the actual number of samples measured.

[0066] In some embodiments, S220, calculating the physical loss based on the predicted temperature field, the predicted internal stress field, and the predicted warping displacement field includes:

[0067] S221, Substitute the partial derivative of the predicted temperature field with respect to time and the Laplace operator of the predicted temperature field into the three-dimensional unsteady heat conduction equation to calculate the residual loss of the heat conduction equation.

[0068] S222, by substituting the predicted stress increment tensor, the predicted total strain increment tensor, the predicted viscous strain increment, and the predicted volume shrinkage strain increment into the thermo-elastic-plastic stress-strain equation, the residual loss of the stress-strain equation is calculated.

[0069] S223, Substitute the predicted warping displacement vector, the thermal stress load vector calculated based on the predicted temperature field, and the residual stress load vector calculated based on the predicted stress field into the structural deformation equilibrium equation to calculate the residual loss of the structural deformation equation.

[0070] S224, sum the mean square errors of the residual losses of the heat conduction equation, the stress-strain equation, and the structural deformation equation to obtain the physical loss.

[0071] Specifically, the physical loss transforms the three types of core physical equations into residual losses, and the constraint network output satisfies the thermal-structural coupling mechanism.

[0072] Calculation of residual loss of heat conduction equation: Calculate the partial derivative of the predicted temperature field with respect to time to obtain the rate of temperature change, calculate the Laplace operator of the predicted temperature field to obtain the temperature diffusion term, substitute the rate of temperature change and the temperature diffusion term into the three-dimensional unsteady heat conduction equation to obtain the residual of heat conduction equation, and calculate the mean value after squaring the residual of heat conduction equation to obtain the residual loss of heat conduction equation.

[0073] The three-dimensional unsteady heat conduction equation is:

[0074] ;

[0075] Where ρ is the material density, cp is the specific heat capacity at constant pressure, t is time, k is the thermal conductivity as a function of temperature, T is the temperature, ∇ is the Laplace operator, and Φ is the viscous dissipation term. The term represents volumetric heat release; the boundary conditions are convective heat transfer boundary conditions. Where h is the heat transfer coefficient at the melt-mold interface. Let n be the mold temperature and n be the boundary normal vector.

[0076] The residual of the heat conduction equation is:

[0077] ;

[0078] Stress-strain equation residual loss calculation: Calculate the first deviation between the predicted total strain increment tensor and the elastic strain component of the predicted stress increment tensor; calculate the second deviation between the predicted total strain increment tensor and the thermal expansion strain component; calculate the third deviation between the predicted total strain increment tensor and the predicted viscous strain increment; calculate the fourth deviation between the predicted total strain increment tensor and the predicted volumetric shrinkage strain increment; sum the first deviation, second deviation, third deviation, and fourth deviation to obtain the stress-strain equation residual; square the stress-strain equation residual and calculate the mean value to obtain the stress-strain equation residual loss.

[0079] The thermoelastic-plastic stress-strain equation (incremental form) is as follows:

[0080] ;

[0081] in, Let the total strain increment tensor be... Let E be the stress increment tensor, and E be the elastic modulus as a function of temperature. Poisson's ratio, dT is the coefficient of thermal expansion, and dT is the temperature increment. For the Kronecker function, For viscous strain increment, This represents the volumetric shrinkage strain increment.

[0082] The residuals of the stress-strain equation are:

[0083] ;

[0084] in, The internal stress field predicted by the physical information neural network model. , These represent the predicted viscous strain increment and contraction strain increment, respectively.

[0085] Calculation of residual loss in structural deformation equation: Calculate the first product of the product stiffness matrix and the predicted warping displacement vector; substitute the predicted temperature field into the thermal stress calculation model to obtain the thermal stress load vector; substitute the predicted internal stress field into the residual stress calculation model to obtain the residual stress load vector; sum the thermal stress load vector and the residual stress load vector to obtain the total load vector; calculate the second deviation between the first product and the total load vector to obtain the residual of structural deformation equation; square the residual of structural deformation equation and calculate the mean value to obtain the residual loss of structural deformation equation.

[0086] The structural deformation equilibrium equation (after finite element discretization) is as follows:

[0087] ;

[0088] Where [K] is the product stiffness matrix, and {u} is the warping displacement vector. The thermal stress load vector, This is the residual stress load vector.

[0089] The residual of the structural deformation equation is:

[0090] ;

[0091] in, This is the predicted warping displacement vector. , These are the thermal stress load and residual stress load calculated based on the predicted temperature field and stress field, respectively.

[0092] The total physical loss term is the sum of the mean square errors of all residuals:

[0093] ;

[0094] Where M is the number of physical sampling points (recommended value is 500-1000), and the sampling points cover the key time points and key spatial areas of the product throughout the entire injection molding process.

[0095] The total loss function is obtained by weighted summation of data loss and physical loss:

[0096] ;

[0097] in, The data loss weight (values ​​range from 1.0 to 2.0). This is the weight for physical loss (values ​​range from 0.3 to 0.8). For data loss, This is a physical loss.

[0098] The selection strategy for physical sampling points follows a dual-dimensional coverage principle of time and space. The time dimension covers key stages of the entire injection molding process: the final filling stage (the instant the melt fills the cavity), the holding pressure stage (the process of compensation and compaction), and the cooling stage (the process of solidification and stress relaxation). The spatial dimension covers key areas of the product: areas with uneven wall thickness (large temperature gradients), near the gate (concentration of residual flow stress), and edge areas (constraint boundary effects). Through a random sampling strategy, the uniform distribution of physical constraints in the spatiotemporal domain is ensured, avoiding physical losses and imbalances caused by sampling bias.

[0099] The specific training process is as follows:

[0100] Data acquisition: Collect 50-100 sets of measured data, including process parameters (injection temperature 180-240℃, injection pressure 50-150MPa, etc.), material parameters (density, elastic modulus, etc.), structural parameters (wall thickness, number of gates, etc.) and the measured value of the warpage displacement of the corresponding product (detected by 3D scanner, accuracy ±0.01mm).

[0101] Physical sampling point generation: Randomly select 10-20 time points at the end of the filling stage, the holding pressure stage, and the cooling stage, and select 50-100 spatial points at the uneven wall thickness of the product, near the gate, and at the edge area to form 500-1000 physical sampling points.

[0102] Training settings: Adam optimizer is used, learning rate is 0.001-0.0001, batch size is 32-64, and the maximum number of iterations is 100. The training termination condition is that the warping prediction accuracy on the validation set is ≥96%, or the loss function does not decrease for 10 consecutive iterations.

[0103] Model optimization: If the proportion of physical loss is too high, increase it appropriately. If the data is overfitted, increase... And increase the number of physical sampling points.

[0104] In some embodiments, S300, the step of inputting the basic parameters of the injection-molded product to be predicted into the trained physical information neural network and outputting the warpage defect prediction result includes:

[0105] S310, the process parameters, material parameters and structural parameters of the injection molded product to be predicted are input into the trained physical information neural network model, and the temperature field distribution, internal stress field distribution and warping displacement are output through the output layer;

[0106] S320, determine the warping defect level based on the warping displacement, identify the uneven cooling area based on the temperature field distribution, and locate the stress concentration part based on the internal stress field distribution.

[0107] The specific prediction process is as follows:

[0108] Prediction process: Input the material parameters, structural parameters and proposed process parameters of the product to be predicted into the trained PINN model. The model outputs the temperature field distribution, internal stress field distribution and maximum warping displacement within 0.1-1s.

[0109] Result determination: If the predicted maximum warping displacement is <0.2mm, it is determined as no obvious warping defect; if 0.2mm≤warping displacement<0.5mm, it is determined as slight warping; if the warping displacement is ≥0.5mm, it is determined as severe warping.

[0110] Root cause analysis: Based on the output temperature field distribution, areas of uneven cooling can be identified; based on the internal stress field distribution, stress concentration points can be located, providing a basis for process optimization (such as adjusting mold temperature and optimizing cooling water channels).

[0111] Experimental verification:

[0112] The method of this invention is based on the PyTorch framework, and the experimental environment is deployed on a server equipped with NVIDIA GPUs. During the training phase, the Adam optimizer is used with a learning rate of 0.001 and a batch size of 32. Training is iterated until the warp prediction accuracy on the validation set reaches above 96% or the loss function shows no significant decrease after 10 consecutive iterations.

[0113] The method was validated on a test dataset containing injection molded products of various materials and structures. The results show that the method of the present invention only requires 50-100 sets of measured data to complete effective training, which reduces the training sample requirement by more than 60% compared with the pure data-driven method; the average prediction accuracy reaches more than 96%, which meets the needs of practical engineering applications; and the time taken for a single prediction is 0.1-1 second, which is significantly better than the traditional physical simulation method.

[0114] refer to Figure 2 This invention also provides a warpage defect prediction system for injection molded products based on a physical information neural network, comprising:

[0115] At least one processor;

[0116] At least one memory for storing at least one program;

[0117] When the at least one program is executed by the at least one processor, the at least one processor performs the method described above.

[0118] The content of the above method embodiments is applicable to this embodiment. The specific functions implemented in this embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments. Therefore, they will not be repeated here.

[0119] This invention also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the method described above. This electronic device can be any smart terminal, including tablet computers, in-vehicle computers, etc.

[0120] It is understood that the content of the above method embodiments is applicable to this device embodiment. The specific functions implemented by this device embodiment are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0121] This invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method.

[0122] It is understood that the content of the above method embodiments is applicable to this storage medium embodiment. The specific functions implemented in this storage medium embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0123] This invention also provides a computer program product, including a computer program or computer instructions, which are stored in a memory. A processor of a computer device reads the computer program or computer instructions from the memory and executes the computer program or computer instructions, causing the computer device to perform the above-described method.

[0124] It is understood that the content of the above method embodiments is applicable to the embodiments of this program product. The specific functions implemented by the embodiments of this program product are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0125] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0126] It will be understood by those skilled in the art that all or some of the steps and systems in the methods disclosed above can be implemented as software, firmware, hardware, and suitable combinations thereof. Some or all of the physical components can be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit. Such software can be distributed on a computer-readable medium, which can include computer storage media (or non-transitory media) and communication media (or transient media). As is known to those skilled in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and is accessible to a computer. Furthermore, as is known to those skilled in the art, communication media typically include computer-readable instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.

[0127] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

Claims

1. A method for predicting warpage defects of an injection molded article based on a physical information neural network, characterized by, The method includes the following steps: S100, Obtain the basic parameters and physical information neural network of the injection molded product. The physical information neural network includes an input layer, a network body, and an output layer. The basic parameters include process parameters, material parameters, and structural parameters. S200: Collect measured data and randomly select physical sampling points. Substitute the measured data and the physical sampling points into a double loss function to train the physical information neural network, thereby obtaining the trained physical information neural network. The measured data includes basic parameters and corresponding measured warpage displacement values. The physical sampling points are output through the output layer, which outputs the corresponding predicted temperature field, predicted internal stress field, and predicted warpage displacement field. S300: Input the basic parameters of the injection molded product to be predicted into the trained physical information neural network, and output the warpage defect prediction result.

2. The method according to claim 1, characterized in that, In S200, the step of collecting the measured data and randomly selecting physical sampling points, substituting the measured data and the physical sampling points into a double loss function to train the physical information neural network, and obtaining the trained physical information neural network, includes: S210, collect multiple sets of measured data, input the measured data into the physical information neural network to obtain the predicted warping displacement, and calculate the data loss based on the predicted warping displacement and the measured warping displacement; the measured data includes process parameters, material parameters, structural parameters and corresponding measured warping displacement values. S220: Randomly select time points at the end of the filling stage, the holding pressure stage, and the cooling stage; randomly select spatial points at locations with uneven wall thickness, near the gate, and at the edge region to generate space-time physical sampling points; input the physical sampling points into the physical information neural network to obtain the predicted temperature field, the predicted internal stress field, and the predicted warpage displacement field; calculate the physical loss based on the predicted temperature field, the predicted internal stress field, and the predicted warpage displacement field. S230, the total loss is obtained by weighted summation of the data loss and the physical loss, and the network parameters are updated by minimizing the total loss through an optimization algorithm. This process is repeated iteratively until the training termination condition is met, and the trained physical information neural network is obtained.

3. The method according to claim 2, characterized in that, In S210, the calculation of data loss based on the predicted warping displacement and the measured value of the warping displacement includes: The data loss is obtained by squaring the difference between the predicted warping displacement and the measured warping displacement, and then averaging the squared errors of all measured samples.

4. The method according to claim 2, characterized in that, In S220, the calculation of physical loss based on the predicted temperature field, the predicted internal stress field, and the predicted warping displacement field includes: S221, Substitute the partial derivative of the predicted temperature field with respect to time and the Laplace operator of the predicted temperature field into the three-dimensional unsteady heat conduction equation to calculate the residual loss of the heat conduction equation. S222, by substituting the predicted stress increment tensor, the predicted total strain increment tensor, the predicted viscous strain increment, and the predicted volume shrinkage strain increment into the thermo-elastic-plastic stress-strain equation, the residual loss of the stress-strain equation is calculated. S223, Substitute the predicted warping displacement vector, the thermal stress load vector calculated based on the predicted temperature field, and the residual stress load vector calculated based on the predicted stress field into the structural deformation equilibrium equation to calculate the residual loss of the structural deformation equation. S224, sum the mean square errors of the residual losses of the heat conduction equation, the stress-strain equation, and the structural deformation equation to obtain the physical loss.

5. The method according to claim 4, characterized in that, In S221, the step of substituting the partial derivative of the predicted temperature field with respect to time and the Laplace operator of the predicted temperature field into the three-dimensional unsteady heat conduction equation to calculate the residual loss of the heat conduction equation includes: The partial derivative of the predicted temperature field with respect to time is calculated to obtain the rate of temperature change. The Laplace operator of the predicted temperature field is calculated to obtain the temperature diffusion term. The rate of temperature change and the temperature diffusion term are substituted into the three-dimensional unsteady heat conduction equation to obtain the residual of the heat conduction equation. The residual of the heat conduction equation is squared and then the mean is calculated to obtain the residual loss of the heat conduction equation.

6. The method according to claim 4, characterized in that, In S222, the step of substituting the predicted stress increment tensor, the predicted total strain increment tensor, the predicted viscous strain increment, and the predicted volumetric shrinkage strain increment into the thermoelastic-plastic stress-strain equation to calculate the residual loss of the stress-strain equation includes: Calculate the first deviation between the predicted total strain increment tensor and the elastic strain component of the predicted stress increment tensor; calculate the second deviation between the predicted total strain increment tensor and the thermal expansion strain component; calculate the third deviation between the predicted total strain increment tensor and the predicted viscous strain increment; calculate the fourth deviation between the predicted total strain increment tensor and the predicted volumetric shrinkage strain increment; sum the first deviation, the second deviation, the third deviation, and the fourth deviation to obtain the stress-strain equation residual; square the stress-strain equation residual and calculate the mean value to obtain the stress-strain equation residual loss.

7. The method according to claim 4, characterized in that, In S223, the step of substituting the predicted warping displacement vector, the thermal stress load vector calculated based on the predicted temperature field, and the residual stress load vector calculated based on the predicted internal stress field into the structural deformation equilibrium equation to calculate the residual loss of the structural deformation equation includes: Calculate the first product of the product stiffness matrix and the predicted warping displacement vector, substitute the predicted temperature field into the thermal stress calculation model to obtain the thermal stress load vector, substitute the predicted internal stress field into the residual stress calculation model to obtain the residual stress load vector, sum the thermal stress load vector and the residual stress load vector to obtain the total load vector, calculate the second deviation between the first product and the total load vector to obtain the structural deformation equation residual, square the structural deformation equation residual and calculate the mean value to obtain the structural deformation equation residual loss.

8. The method according to claim 1, characterized in that, In S300, the step of inputting the basic parameters of the injection-molded product to be predicted into the trained physical information neural network and outputting the warpage defect prediction result includes: S310, the process parameters, material parameters and structural parameters of the injection molded product to be predicted are input into the trained physical information neural network model, and the temperature field distribution, internal stress field distribution and warping displacement are output through the output layer; S320, determine the warping defect level based on the warping displacement, identify the uneven cooling area based on the temperature field distribution, and locate the stress concentration part based on the internal stress field distribution.

9. A warpage defect prediction system for injection molded products based on physical information neural networks, characterized in that, include: At least one processor; At least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor performs the method as described in any one of claims 1 to 8.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 8.

Images