Injection product quality prediction method and system based on XGBoost-LSTM integrated model

Abstract

This invention discloses a method and system for predicting the quality of injection molded products based on an XGBoost-LSTM ensemble model. The method includes the following steps: S1, designing the gating and cooling systems according to the requirements of the injection molded product, determining key injection molding process parameters, and performing mold flow analysis to simulate actual production; S2, collecting Moldflow simulation analysis data; S3, preprocessing the simulation dataset and dividing it into training and test sets; S4, training the XGBoost-LSTM ensemble model using the training set; S5, verifying the prediction accuracy of the XGBoost-LSTM ensemble model using the test set. The system includes a mold flow analysis module, a simulation data acquisition module, a data preprocessing module, a model construction module, a model training module, and a model testing module. This invention accurately predicts product quality based on set injection molding parameters, reducing trial-and-error costs and improving production efficiency.

Description

Technical Field

[0001] This invention relates to the field of injection molding production technology, and in particular to a method and system for predicting the quality of injection molded products based on an XGBoost-LSTM integrated model. Background Technology

[0002] In the injection molding industry, quality control of injection molded parts is of paramount importance and is one of the core technologies for enterprises to control production costs and enhance market competitiveness. Currently, traditional quality control methods for injection molded products mainly rely on offline testing of various quality indicators after production. This not only incurs huge costs in terms of time and money but also fails to guarantee comprehensive and rapid inspection of product quality, hindering the improvement of the enterprise's overall production level and economic efficiency.

[0003] In recent years, with the rapid development of artificial intelligence technologies such as machine learning and deep learning, data-driven quality prediction methods have emerged and are gradually being applied to various industrial sectors. Currently, machine learning algorithms based on support vector machines, response surface analysis, artificial neural networks, and random forests are all being used in quality prediction. However, problems still exist, such as the underutilization of time-series information in industrial data and the unreliability of the accuracy of individual prediction models, which bring difficulties and challenges to the stable and efficient production of enterprises. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a method and system for predicting the quality of injection molded products based on the XGBoost-LSTM integrated model, which can quickly establish a black box model of the injection molding production process and accurately predict the product quality according to the set injection molding parameters, thereby effectively reducing the trial and error costs of enterprises and improving production efficiency.

[0005] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows:

[0006] A method for predicting the quality of injection molded products based on an XGBoost-LSTM ensemble model includes the following steps:

[0007] S1. Based on the requirements of injection molded products, use Moldflow software to design the gating system and cooling system, and determine the key injection molding process parameters based on orthogonal experimental design, and conduct mold flow analysis to simulate actual production;

[0008] S2. Collect Moldflow simulation analysis data to obtain the simulation dataset;

[0009] S3. Preprocess the simulation dataset and divide the preprocessed simulation dataset into training set and test set;

[0010] S4. Construct an XGBoost-LSTM ensemble model. Train the XGBoost-LSTM ensemble model using the training set to obtain a regression model for predicting the quality of injection molded products.

[0011] S5. Verify the prediction accuracy of the EXGBoost-LSTM ensemble model using the test set.

[0012] As a preferred embodiment, the key injection molding process parameters include injection time, holding time, cooling time, injection pressure, holding pressure, material temperature, nozzle temperature, and mold temperature.

[0013] As a preferred embodiment, the simulation dataset includes process parameter settings, product quality data, and real-time mold flow status data. The product quality data includes mold filling time, weld line, flow front temperature, volume shrinkage, and warpage. The real-time mold flow status data includes pressure at the injection point, clamping force, volume, and temperature.

[0014] As a preferred embodiment, the preprocessing of the simulation dataset includes the following steps:

[0015] S31. Remove outliers from product quality data;

[0016] S32. Normalize the simulation dataset;

[0017] S33. Perform data segmentation and reconstruction on the real-time model flow status data.

[0018] As a preferred approach, the key to segmenting and reconstructing the real-time model flow state data lies in: selecting T time steps as intervals, using a sliding window method to divide the sequence segments, and recombining them into new feature inputs.

[0019] As a preferred embodiment, the XGBoost-LSTM ensemble model includes an LSTM sub-model, an XGBoost sub-model, and a meta-model.

[0020] As a preferred embodiment, training the XGBoost-LSTM ensemble model using a training set includes the following steps:

[0021] S41. Input the real-time model flow state data in the training set into the LSTM sub-model to obtain the first predicted value;

[0022] S42. Input the injection molding process parameter data and product quality data from the training set into the XGBoost sub-model to obtain the second predicted value;

[0023] S43. The first predicted value obtained from step S41 and the second predicted value obtained from step S42 are combined to form a new training set and input into the meta-model. The meta-model uses a weighted combination of the predicted values ​​of each sub-model as the final predicted value.

[0024] As a preferred embodiment, the meta-model uses the following formula to calculate the weighted combination of the predicted values ​​from each sub-model:

[0025]

[0026] d lstm +d xgb =1

[0027] Where, d lstm The weights of the predicted values ​​of the LSTM sub-model, d xgb The weights of the predicted values ​​in the XGBoost sub-model. This is the final predicted value. The first predicted value, This is the second predicted value.

[0028] A quality prediction system for injection molded products based on an XGBoost-LSTM ensemble model, comprising:

[0029] The mold flow analysis module is used to design gating and cooling systems using Moldflow software, determine key injection molding process parameters based on orthogonal experimental design, and simulate actual production through mold flow analysis.

[0030] The simulation data acquisition module is used to collect Moldflow simulation analysis data to obtain a simulation dataset.

[0031] The data preprocessing module is used to preprocess the simulation dataset and divide the preprocessed simulation dataset into training and testing sets.

[0032] The model building module is used to build an XGBoost-LSTM ensemble model; wherein the XGBoost-LSTM ensemble model includes an LSTM sub-model, an XGBoost sub-model, and a meta-model;

[0033] The model training module is used to input the training set into the XGBoost-LSTM ensemble model for training, resulting in a regression model for predicting the quality of injection molded products; and

[0034] The model testing module is used to verify the prediction accuracy of the XGBoost-LSTM ensemble model through a test set.

[0035] The beneficial effects of this invention are as follows:

[0036] This invention performs Moldflow analysis under key injection molding process parameters, collects and preprocesses simulation analysis data, then constructs sub-models based on XGBoost and LSTM machine learning algorithms, and fully utilizes process parameter data and real-time mold flow state data for training. Finally, a meta-model is constructed using a weighted combination method to obtain accurate and reliable quality prediction results. For injection molding plants, this invention can quickly establish a black-box model of the injection molding production process, accurately predict product quality based on set injection molding parameters, effectively reduce trial-and-error costs in factory production, and improve production efficiency. Attached Figure Description

[0037] Figure 1 This is a flowchart illustrating the injection molding product quality prediction method based on the XGBoost-LSTM ensemble model of the present invention.

[0038] Figure 2 This is a flowchart illustrating the preprocessing method for simulation datasets according to the present invention.

[0039] Figure 3 This is a flowchart illustrating the training method of the XGBoost-LSTM ensemble model using a training set according to the present invention.

[0040] Figure 4 This is a schematic diagram of the injection molding product quality prediction system based on the XGBoost-LSTM integrated model of the present invention. Detailed Implementation

[0041] The working principle of the present invention will be further described in detail below with reference to the accompanying drawings.

[0042] like Figure 1 As shown, a method for predicting the quality of injection molded products based on an XGBoost-LSTM ensemble model includes the following steps:

[0043] S1. Based on the requirements of injection molded products, use Moldflow software to design the gating system and cooling system, and determine the key injection molding process parameters based on orthogonal experimental design, and conduct mold flow analysis to simulate actual production;

[0044] The key injection molding process parameters include injection time, holding time, cooling time, injection pressure, holding pressure, material temperature, nozzle temperature, and mold temperature.

[0045] S2. Collect Moldflow simulation analysis data to obtain the simulation dataset;

[0046] The simulation dataset includes process parameter settings, product quality data, and real-time mold flow status data. The product quality data includes mold filling time, weld lines, flow front temperature, volume shrinkage, and warpage. The real-time mold flow status data includes pressure at the injection point, clamping force, volume, and temperature.

[0047] S3. Preprocess the simulation dataset and divide the preprocessed simulation dataset into training set and test set;

[0048] like Figure 2 As shown, the preprocessing of the simulation dataset includes the following steps:

[0049] S31. Remove outliers from product quality data;

[0050] S32. Normalize the simulation dataset;

[0051] The formula for normalization is as follows:

[0052] Suppose there are N samples, and for the m-th feature of each sample:

[0053]

[0054] Wherein, the x im x is the original value of the m-th feature of the i-th sample. im * Let x be the normalized value of the m-th feature of the i-th sample. im * It is distributed in the interval [0, 1].

[0055] S33. Perform data segmentation and reconstruction on the real-time model flow status data.

[0056] The key to segmenting and reconstructing real-time model flow state data lies in selecting T time steps as intervals, using a sliding window method to divide the sequence segments, and recombining them into new feature inputs; thus ensuring the temporal correlation of the dataset.

[0057] To better observe the model's prediction results, the preprocessed simulation dataset was divided into a training set and a test set in a ratio of 8:2. 80% of the dataset was used as the training set to train the model, and 20% of the dataset was used as the test set to validate the trained model.

[0058] S4. Construct an XGBoost-LSTM ensemble model. Train the XGBoost-LSTM ensemble model using the training set to obtain a regression model for predicting the quality of injection molded products.

[0059] The XGBoost-LSTM ensemble model includes an LSTM sub-model, an XGBoost sub-model, and a meta-model.

[0060] like Figure 3 As shown, training the XGBoost-LSTM ensemble model using the training set includes the following steps:

[0061] S41. Input the real-time model flow state data in the training set into the LSTM sub-model to obtain the first predicted value;

[0062] LSTM models are deliberately designed to avoid long-term dependency problems. Remembering long-term information is the default behavior of LSTMs in practice. The key to an LSTM model is the cell state, with a horizontal line running across the top of the graph. The cell state is analogous to a conveyor belt, running directly across the entire chain with only a few linear interactions. Information flowing along it remains unchanged, which is easy. LSTM models have the ability to remove or add information to the cell state through carefully designed structures called "gates." A gate is a method of selectively allowing information to pass through, containing a sigmoid neural network layer and a pointwise multiplication operation. The sigmoid layer outputs a value between 0 and 1, describing how much quantity can pass through each part. 0 means no quantity can pass, and 1 means any quantity can pass. The LSTM model has three gates to protect and control the cell state, and its structural principle is as follows: Figure 3 As shown. These three gates are called the forget gate, input gate, and output gate, respectively.

[0063] This invention trains an LSTM sub-model using the model flow state data of the training set as input features, and includes the following steps:

[0064] First, the forgetting gate determines the information that needs to be discarded; the forgetting gate is accessed by checking X. t and h t-1 The information outputs a vector between 0 and 1, where the 0-1 values ​​represent the cell state C. t-1 Which information is retained and how much is discarded? 0 indicates no retention, 1 indicates all information is retained. The calculation formula is as follows:

[0065]

[0066] Where, x t W represents the input feature of the current sample at step t. f For weights, h t Represents the hidden node state, σ is the activation function, and b f For bias

[0067] Second, the information to be added is determined by the input gate, and the state features are output by the output gate.

[0068] 1) Using h t-1 and X t New candidate cell information is obtained through a tanh layer. The calculation formula is as follows:

[0069] i t =σ(W i [h t-1 x t ]+b i )

[0070]

[0071] Among them, i t The input gate is σ, and tanh are the activation functions, h t Indicates the hidden node state, b i and b C For the bias, W i and W C All are weights.

[0072] 2) Update old cell information C t-1 It becomes new cellular information C t The update rule involves using the forget gate to select and forget a portion of the old cell information, and using the input gate to select and add candidate cell information. Part of it receives new cellular information C t The calculation formula is as follows:

[0073]

[0074] Among them, f t For the Gate of Oblivion, i t This is the input gate.

[0075] Third, after updating the cell state, it is necessary to adjust the input h. t-1 and X t To determine which state features of the output cell, the input needs to pass through a sigmoid layer called the output gate to obtain the decision condition. Then, the cell state is passed through a tanh layer to obtain a vector of values ​​between -1 and 1. This vector is multiplied by the decision condition obtained from the output gate to obtain the final output of the RNN unit. The calculation formula is as follows:

[0076] o t =σ(W o [h t-1 x t ]+b o )

[0077] h t =o t *tanh(Ct )

[0078] Among them, o t For output gate, W o As the weight, b o σ is the bias, and tanh are both excitation functions.

[0079] Fourth, calculate the first predicted value. The calculation formula is as follows:

[0080]

[0081] in, W is the first predicted value. y As the weight, b y denoted as a bias, and tanh is an activation function.

[0082] S42. Input the injection molding process parameter data and product quality data from the training set into the XGBoost sub-model to obtain the second predicted value;

[0083] This invention uses injection molding process parameter data as input features, and corresponding label data as product quality data. The features and label data are input into an XGBoost sub-model. During training, multiple regression trees (tree nodes) are constructed, and each regression tree can be split into multiple leaf nodes, where each leaf node represents a process parameter feature value. The default splitting principle for leaf nodes is to select the feature with the highest information gain, i.e., a greedy criterion. Each leaf node in the regression tree has a weight value during splitting, and the next leaf node to split in each regression tree is determined by the weight of the input features. In each training round, the weights of the leaf nodes are adjusted according to the objective function. After multiple iterations, the optimal regression tree is formed. The prediction values ​​of all regression trees are summed to obtain the prediction result of the XGBoost sub-model. The specific calculation formula is as follows:

[0084]

[0085]

[0086] F m (x i ) = F m-1 (x i )+f m (x i )

[0087]

[0088] Where F is the function space containing K regression trees, f m (x i Let F be the tree model function in the current m-th iteration. m-1Let L be the function space fixed after the first m-1 iterations of training, L be the loss function, and Ω be the regularization term representing the model complexity, used to control overfitting; Ω is the value. This is the predicted value after training a single sample is completed; This is the predicted value after training with N samples, i.e., the second predicted value.

[0089] S43. The first predicted value obtained from step S41 and the second predicted value obtained from step S42 are combined to form a new training set and input into the meta-model. The meta-model uses a weighted combination of the predicted values ​​of each sub-model as the final predicted value.

[0090] The meta-model uses the following formula to calculate the weighted combination of the predicted values ​​from each sub-model:

[0091]

[0092] d lstm +d xgb =1

[0093] Where, d lstm The weights of the predicted values ​​of the LSTM sub-model, d xgb The weights of the predicted values ​​in the XGBoost sub-model. This is the final predicted value. The first predicted value, This is the second predicted value.

[0094] S5. Verify the prediction accuracy of the XGBoost-LSTM ensemble model using the test set.

[0095] By improving prediction accuracy, the relationship between key injection molding process parameters and injection molded product quality can be determined, allowing users to easily adjust these parameters. When the prediction accuracy reaches a threshold, the final prediction value calculated by the current XGBoost-LSTM integrated model is an estimate of the injection molding parameters. This enables accurate prediction of product quality based on set injection molding parameters, effectively reducing trial-and-error costs and improving production efficiency for enterprises.

[0096] like Figure 4 As shown, an injection molding product quality prediction system based on an XGBoost-LSTM ensemble model includes:

[0097] The mold flow analysis module is used to design gating and cooling systems using Moldflow software, determine key injection molding process parameters based on orthogonal experimental design, and simulate actual production through mold flow analysis.

[0098] The simulation data acquisition module is used to collect Moldflow simulation analysis data to obtain a simulation dataset.

[0099] The data preprocessing module is used to preprocess the simulation dataset and divide the preprocessed simulation dataset into training and testing sets.

[0100] The model building module is used to build an XGBoost-LSTM ensemble model; wherein the XGBoost-LSTM ensemble model includes an LSTM sub-model, an XGBoost sub-model, and a meta-model;

[0101] The model training module is used to input the training set into the XGBoost-LSTM ensemble model for training, resulting in a regression model for predicting the quality of injection molded products; and

[0102] The model testing module is used to verify the prediction accuracy of the XGBoost-LSTM ensemble model through a test set.

[0103] The beneficial effects of this invention are as follows:

[0104] This invention performs Moldflow analysis under key injection molding process parameters, collects and preprocesses simulation analysis data, then constructs sub-models based on XGBoost and LSTM machine learning algorithms, and fully utilizes process parameter data and real-time mold flow state data for training. Finally, a meta-model is constructed using a weighted combination method to obtain accurate and reliable quality prediction results. For injection molding plants, this invention can quickly establish a black-box model of the injection molding production process, accurately predict product quality based on set injection molding parameters, effectively reduce trial-and-error costs in factory production, and improve production efficiency.

[0105] The above description is merely a preferred embodiment of the present invention. Any minor modifications, equivalent changes, and alterations made to the above embodiments based on the technical solution of the present invention shall fall within the scope of the technical solution of the present invention.

Claims

1. A method for predicting the quality of injection molded products based on an XGBoost-LSTM ensemble model, characterized in that, Includes the following steps: S1. Based on the requirements of injection molded products, use Moldflow software to design the gating system and cooling system, and determine the key injection molding process parameters based on orthogonal experimental design, and conduct mold flow analysis to simulate actual production; S2. Collect Moldflow simulation analysis data to obtain the simulation dataset; S3. Preprocess the simulation dataset and divide the preprocessed simulation dataset into training set and test set; S4. Construct an XGBoost-LSTM ensemble model. Train the XGBoost-LSTM ensemble model using the training set to obtain a regression model for predicting the quality of injection molded products. S5. Verify the prediction accuracy of the XGBoost-LSTM ensemble model using the test set; In step S4, the XGBoost-LSTM ensemble model includes an LSTM sub-model, an XGBoost sub-model, and a meta-model. Training the XGBoost-LSTM ensemble model using a training set includes the following steps: S41, inputting real-time mold flow state data from the training set into the LSTM sub-model to obtain a first predicted value; S42, inputting injection molding process parameter data and product quality data from the training set into the XGBoost sub-model to obtain a second predicted value; S413, inputting a new training set composed of the first predicted value obtained in step S41 and the second predicted value obtained in step S42 into the meta-model. The meta-model uses a weighted combination of the predicted values ​​from each sub-model as the final predicted value. The calculation formula for the weighted combination of the predicted values ​​from each sub-model in the meta-model is as follows: ,in, The weights of the predicted values ​​of the LSTM sub-model The weights of the predicted values ​​in the XGBoost sub-model. This is the final predicted value. The first predicted value, This is the second predicted value.

2. The injection molding product quality prediction method based on the XGBoost-LSTM ensemble model according to claim 1, characterized in that: The key injection molding process parameters include injection time, holding time, cooling time, injection pressure, holding pressure, material temperature, nozzle temperature, and mold temperature.

3. The injection molding product quality prediction method based on the XGBoost-LSTM ensemble model according to claim 1, characterized in that: The simulation dataset includes process parameter settings, product quality data, and real-time mold flow status data. The product quality data includes mold filling time, weld line, flow front temperature, volume shrinkage, and warpage. The real-time mold flow status data includes pressure at the injection point, clamping force, volume, and temperature.

4. The injection molding product quality prediction method based on the XGBoost-LSTM ensemble model according to claim 3, characterized in that: The preprocessing of the simulation dataset includes the following steps: S31. Remove outliers from product quality data; S32. Normalize the simulation dataset; S33. Perform data segmentation and reconstruction on the real-time model flow status data.

5. The injection molding product quality prediction method based on the XGBoost-LSTM ensemble model according to claim 4, characterized in that: The key to segmenting and reconstructing real-time model flow state data lies in: selecting a time step as the interval, using a sliding window method to divide the sequence into segments, and recombining them into new feature inputs.

6. A system for predicting the quality of injection molded products based on an XGBoost-LSTM ensemble model, employing the method for predicting the quality of injection molded products based on an XGBoost-LSTM ensemble model as described in any one of claims 1-5, characterized in that: include: The mold flow analysis module is used to design gating and cooling systems using Moldflow software, determine key injection molding process parameters based on orthogonal experimental design, and simulate actual production through mold flow analysis. The simulation data acquisition module is used to collect Moldflow simulation analysis data to obtain a simulation dataset. The data preprocessing module is used to preprocess the simulation dataset and divide the preprocessed simulation dataset into training and testing sets. The model building module is used to build an XGBoost-LSTM ensemble model; wherein the XGBoost-LSTM ensemble model includes an LSTM sub-model, an XGBoost sub-model, and a meta-model; The model training module is used to input the training set into the XGBoost-LSTM ensemble model for training, resulting in a regression model for predicting the quality of injection molded products; and The model testing module is used to verify the prediction accuracy of the XGBoost-LSTM ensemble model through a test set.

Images