CN112149810A - Deep learning-based method for predicting and transferring quality of metal injection molding sintered product - Google Patents
Deep learning-based method for predicting and transferring quality of metal injection molding sintered product Download PDFInfo
- Publication number
- CN112149810A CN112149810A CN202011201541.2A CN202011201541A CN112149810A CN 112149810 A CN112149810 A CN 112149810A CN 202011201541 A CN202011201541 A CN 202011201541A CN 112149810 A CN112149810 A CN 112149810A
- Authority
- CN
- China
- Prior art keywords
- deep learning
- model
- injection molding
- sintered product
- metal injection
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
Abstract
The invention discloses a method for predicting and transferring the quality of a metal injection molding sintered product based on deep learning, which comprises the following specific steps: the method comprises the following steps of firstly, loading a deep learning model of a trained performance prediction data of a certain sintered product; the second step, freezing the weight parameters of the X layer neurons; reading in N times of test data of the new product; the fourth step, test data are randomly divided into K groups, and the K-1 group is used for training and verifying the model; step five, if the verification result is not good, adjusting the super-parameters of the model to improve; if the verification result is good, using the Kth group of data to perform model test; step six, if the test result does not meet the requirements, changing X, updating and freezing the neuron parameters of the X layer, and carrying out model training and verification again; and if the test meets the requirements, outputting the performance prediction result of the new product. The method can realize the prediction of the performance of the new product according to a small amount of data of the new product.
Description
Technical Field
The invention relates to the technical field of deep learning models, in particular to a method for predicting and transferring the quality of a metal injection molding sintered product based on deep learning.
Background
In the technical field of MIM metal injection molding, a deep learning model can be formed by selecting sintering process parameters, the model learns the past experience, and the quality of a sintered product can be better calculated according to the sintering process parameters. The conventional deep learning method usually needs test data of a large number of products to train, and a model trained by test data of a single product has poor generalization and migration capability on the performance of other parts, and the model trained by a certain product cannot effectively predict the performance of other products.
Disclosure of Invention
The technical problem to be solved by the invention is as follows: in order to solve the problems in the background art, the quality prediction migration method of the metal injection molding sintered product based on deep learning is provided, the performance of the new product can be predicted by training small-batch data again on the basis of the deep learning model completed by the original training, and the workload of tests and algorithm model training in the development of the new product is effectively reduced.
The technical scheme adopted by the invention for solving the technical problems is as follows: a deep learning-based method for predicting and transferring quality of metal injection molding sintered products specifically comprises the following steps:
the method comprises the following steps of firstly, loading a deep learning model of a trained performance prediction data of a certain sintered product;
the second step, freezing the weight parameters of the X layer neurons;
reading in N times of test data of the new product;
the fourth step, test data are randomly divided into K groups, and the K-1 group is used for training and verifying the model;
step five, if the verification result is not good, adjusting the super-parameters of the model to improve; if the verification result is good, using the Kth group of data to perform model test;
step six, if the test result does not meet the requirements, changing X, updating and freezing the neuron parameters of the X layer, and carrying out model training and verification again; and if the test result meets the requirement, outputting the performance prediction result of the new product.
More specifically, in the above technical solution, in the first step, the trained performance prediction data of a certain sintered product includes test control parameters, a position of a part, performance, and date information.
More specifically, in the above technical solution, the control parameters include sintering temperature, temperature increase and decrease speed, heat preservation time, partial pressure, and flow rate of the cooling water jacket.
More specifically, in the above technical solution, the position of the part includes the number of layers, the number of columns, and the number of rows.
More specifically, in the above technical solution, the number of layers, the number of columns, and the number of rows are recorded in a 0ne-hot coding format.
More specifically, in the above technical solution, in the fifth step, the hyper-parameter includes a learning rate and an iteration step number.
More specifically, in the above technical solution, the hyper-parameter further includes the number of layers of the neural network, the number of neurons, the type of the activation function, and the number of frozen weight layers.
More specifically, in the above technical solution, the neuron is composed of a linear model and a nonlinear model.
More specifically, in the above technical solution, the neural network is formed by stacking a plurality of neurons in a transverse direction and a plurality of neurons in a longitudinal direction.
More specifically, in the above technical solution, the neural network includes an input layer, a hidden layer, and an output layer.
The invention has the beneficial effects that: the invention provides a deep learning metal injection molding sintered product quality prediction migration method based on part characteristics, which is based on a deep learning model containing part characteristics, utilizes old product sintering test data accumulated in the past, can train an algorithm model with migration learning capability by freezing partial parameter weights in an original deep learning model according to the change of part geometric parameters in input characteristics under the condition of a small amount of new product test data, can avoid the condition that DOE design in the past depends on the experience of an engineer during the development task of the new product, quickly and effectively evaluates the influence of process parameters on yield, provides effective suggestions for the engineer, shortens DOE test verification time, improves efficiency, generally has the DOE time of 3-6 months, and uses the migration model to predict the time of 1-3 months.
Drawings
In order to more clearly illustrate the embodiments or technical solutions in the prior art of the present invention, the drawings used in the description of the embodiments or prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments described in the present application, and other drawings can be obtained by those skilled in the art without creative efforts.
Fig. 1 is a schematic diagram of the operation of the present invention.
Detailed Description
In order to make the technical problems, technical solutions and advantageous effects solved by the present invention more clearly apparent, the present invention is 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.
Referring to fig. 1, a method for predicting migration of a sintered product based on deep learning metal injection molding specifically comprises the following steps:
the method comprises the following steps of firstly, loading a deep learning model of the trained performance prediction data of a certain sintered product, wherein the deep learning model is based on a deep neural network and is trained through control parameters and performance parameters of other products.
The second step, freezing the weight parameter of the neuron in the X layer, wherein the setting method of the weight parameter is as follows: the deep neural network model changes the traceable parameter of the layer needing to be frozen into False through a keras interface.
And thirdly, reading in N times of test data of the new product, wherein the test data comprises sintering process parameters, product geometry, material data and performance data.
The fourth step, test data are randomly divided into K groups, the K-1 group is used for training and verifying a model, target data are performance data of the product, such as hardness, density, shrinkage rate, flatness and the like, and the mapping relation between input and the target data is learned through model training; the verified metrics include mean square error MSE of the predicted value and the true value, and R2_ score evaluation criteria of the sklern library.
Step five, if the verification result is not good, adjusting the super-parameters of the model to improve; and if the verification result is good, performing model test by using the Kth group of data.
Step six, if the test result does not meet the requirement after training, verifying and testing in sequence, changing X, updating the frozen X layer neuron parameters, and performing model training and verifying again; and if the test meets the requirement after training, verification and test in sequence, outputting the performance prediction result of the new product.
In the first step, the trained performance prediction data of a certain sintered product comprises test control parameters, part positions, performance and date information. The control parameters comprise sintering temperature, temperature rising and falling speed, heat preservation time, partial pressure, flow of the cooling water jacket and the like. The position of the part includes the number of layers, the number of columns, the number of rows, etc.
The number of layers, the number of columns and the number of rows are recorded in a 0ne-hot coding mode and in a non-numerical mode. The 0ne-hot encoding encodes N states using N-bit state registers.
In the fifth step, the hyper-parameter refers to the network structure of the model, such as the number of layers of the neural network, the number of neurons, the type of activation function, the number of iteration steps, the number of frozen weight layers, the learning rate, and the like.
Neurons consist of linear and non-linear models.
The linear model is constructed as follows: assuming a function of this linear model: y is wx + b, where x is a 1xn vector matrix, each vector value in the matrix represents a feature of the sample, w is a weight matrix of nx1 (the weight of the corresponding vector), and b is the bias term.
The function of the linear model is defined as follows:
the non-linear model is located behind a linear model, also called the excitation function or activation function.
The function of the nonlinear model is defined as follows:
the activation function is the value of tanh,
since the linear model y before the nonlinear model is wx + b, the following derivation can be made:
Z=wx+b
the nonlinear model function becomes:
the neural network is formed by transversely stacking a plurality of neurons and longitudinally stacking a plurality of neurons.
The neural network includes an input layer, a hidden layer, and an output layer.
The input layer is responsible for directly receiving input vectors, and data is not processed under normal conditions and the number of layers of the neural network is not counted.
The hidden layer is the most important part of the whole neural network, and can be one layer or N layers, and each neuron of the hidden layer processes data.
The output layer is used to output the value of the whole network process, which may be a sort vector value or a continuous value generated like a linear regression.
When a factory has a production task of a new product, the traditional DOE (design of materials) test method usually needs 11 to 27 sets of tests to determine process parameters, the test time is required to be 3 to 6 months after all the tests are completed, the yield of samples also depends on the experience of test design engineering seriously, and the risk is high.
It should be noted that: DOE (design of experiments) plays a very important role in the whole process of quality control, and is an important guarantee for improving the quality of products and the process flow. DOE step: firstly, screening main significant factors; secondly, finding out the best production condition combination; thirdly, the combination of the optimal production conditions is proved to have reproducibility.
Judging the success of the DOE first-stage experiment:
firstly, 1-4 significant factors appear in ANOVA analysis;
and secondly, the cumulative contribution rate of the significant factors is more than 75%.
Judging the success of the DOE second stage experiment: no significant factors were present in the ANOVA analysis.
The invention loads the trained deep learning model of the original product, and when a new product is tested (3 groups to 5 groups) 1/6, the algorithm model parameters are circularly frozen, and the test data of the new product is used for repeating training, verification and test, thereby predicting the product performance at an earlier stage of the test.
The invention has the key point that a training method for the transfer learning of a sintered product performance prediction model is set up, and an optimal frozen layer and a hyper-parameter combination are searched through a cycle test by sequentially freezing the weights of different neuron layers in an original prediction model and adjusting hyper-parameters such as model learning speed, calculation step number and the like.
The invention provides a deep learning metal injection molding sintered product quality prediction migration method based on part characteristics, which is based on a deep learning model containing part characteristics, utilizes old product sintering test data accumulated in the past, can train an algorithm model with migration learning capability by freezing partial parameter weights in an original deep learning model according to the change of part geometric parameters in input characteristics under the condition of a small amount of new product test data, can avoid the condition that DOE design in the past depends on the experience of an engineer during the development task of the new product, quickly and effectively evaluates the influence of process parameters on yield, provides effective suggestions for the engineer, shortens DOE test verification time, improves efficiency, generally has the DOE time of 3-6 months, and uses the migration model to predict the time of 1-3 months.
The above description is only for the preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any person skilled in the art should be considered to be within the technical scope of the present invention, and the technical solutions and the inventive concepts thereof according to the present invention are equivalent to or changed within the technical scope of the present invention.
Claims (10)
1. A deep learning-based method for predicting and transferring quality of metal injection molding sintered products is characterized by comprising the following steps:
the method comprises the following steps of firstly, loading a deep learning model of a trained performance prediction data of a certain sintered product;
the second step, freezing the weight parameters of the X layer neurons;
reading in N times of test data of the new product;
the fourth step, test data are randomly divided into K groups, and the K-1 group is used for training and verifying the model;
step five, if the verification result is not good, adjusting the super-parameters of the model to improve; if the verification result is good, using the Kth group of data to perform model test;
step six, if the test result does not meet the requirements, changing X, updating and freezing the neuron parameters of the X layer, and carrying out model training and verification again; and if the test result meets the requirement, outputting the performance prediction result of the new product.
2. The deep learning metal injection molding sintered product quality prediction migration method according to claim 1, characterized in that: in the first step, the trained performance prediction data of a certain sintered product comprises test control parameters, part positions, performance and date information.
3. The deep learning metal injection molding sintered product quality prediction migration method according to claim 2, characterized in that: the control parameters comprise sintering temperature, temperature rising and falling speed, heat preservation time, partial pressure and cooling water jacket flow.
4. The deep learning metal injection molding sintered product quality prediction migration method according to claim 2, characterized in that: the part position comprises the number of layers, the number of columns and the number of rows.
5. The deep learning metal injection molding sintered product quality prediction migration method according to claim 4, wherein: and the number of layers, the number of columns and the number of rows are recorded in a 0ne-hot coding mode.
6. The deep learning metal injection molding sintered product quality prediction migration method according to claim 1, characterized in that: in the fifth step, the hyper-parameters include a learning rate and an iteration step number.
7. The deep learning-based metal injection molding sintered product quality prediction migration method according to claim 6, characterized in that: the hyper-parameters further comprise the number of layers of the neural network, the number of neurons, the type of the activation function and the number of frozen weight layers.
8. The deep learning metal injection molding sintered product quality prediction migration method according to claim 1, characterized in that: the neuron consists of a linear model and a nonlinear model.
9. The deep learning metal injection molding sintered product quality prediction migration method according to claim 7, characterized in that: the neural network is formed by transversely stacking a plurality of neurons and longitudinally stacking a plurality of neurons.
10. The deep learning metal injection molding sintered product quality prediction migration method according to claim 9, characterized in that: the neural network comprises an input layer, a hidden layer and an output layer.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011201541.2A CN112149810A (en) | 2020-11-02 | 2020-11-02 | Deep learning-based method for predicting and transferring quality of metal injection molding sintered product |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011201541.2A CN112149810A (en) | 2020-11-02 | 2020-11-02 | Deep learning-based method for predicting and transferring quality of metal injection molding sintered product |
Publications (1)
Publication Number | Publication Date |
---|---|
CN112149810A true CN112149810A (en) | 2020-12-29 |
Family
ID=73955152
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202011201541.2A Pending CN112149810A (en) | 2020-11-02 | 2020-11-02 | Deep learning-based method for predicting and transferring quality of metal injection molding sintered product |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112149810A (en) |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106802977A (en) * | 2016-12-14 | 2017-06-06 | 同济大学 | One kind is used for sintering performance index prediction and Quality evaluation method |
CN107609647A (en) * | 2017-10-16 | 2018-01-19 | 安徽工业大学 | One kind is based on BP neural network roll alloy mechanical property Forecasting Methodology |
CN109726824A (en) * | 2018-12-05 | 2019-05-07 | 中科恒运股份有限公司 | The transfer learning method and terminal device of training pattern |
CN109902861A (en) * | 2019-01-31 | 2019-06-18 | 南京航空航天大学 | A kind of order manufacturing schedule real-time predicting method based on the double-deck transfer learning |
CN110070217A (en) * | 2019-04-11 | 2019-07-30 | 武汉科技大学 | A kind of Forcasting Sinter Quality method of Kernel-based methods parameter |
CN110910969A (en) * | 2019-12-04 | 2020-03-24 | 云南锡业集团(控股)有限责任公司研发中心 | Tin-bismuth alloy performance prediction method based on transfer learning |
-
2020
- 2020-11-02 CN CN202011201541.2A patent/CN112149810A/en active Pending
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106802977A (en) * | 2016-12-14 | 2017-06-06 | 同济大学 | One kind is used for sintering performance index prediction and Quality evaluation method |
CN107609647A (en) * | 2017-10-16 | 2018-01-19 | 安徽工业大学 | One kind is based on BP neural network roll alloy mechanical property Forecasting Methodology |
CN109726824A (en) * | 2018-12-05 | 2019-05-07 | 中科恒运股份有限公司 | The transfer learning method and terminal device of training pattern |
CN109902861A (en) * | 2019-01-31 | 2019-06-18 | 南京航空航天大学 | A kind of order manufacturing schedule real-time predicting method based on the double-deck transfer learning |
CN110070217A (en) * | 2019-04-11 | 2019-07-30 | 武汉科技大学 | A kind of Forcasting Sinter Quality method of Kernel-based methods parameter |
CN110910969A (en) * | 2019-12-04 | 2020-03-24 | 云南锡业集团(控股)有限责任公司研发中心 | Tin-bismuth alloy performance prediction method based on transfer learning |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
KR102295805B1 (en) | Method for managing training data | |
CN109902862A (en) | A kind of time series forecasting system of time of fusion attention mechanism | |
CN112734002B (en) | Service life prediction method based on data layer and model layer joint transfer learning | |
CN112578089B (en) | Air pollutant concentration prediction method based on improved TCN | |
CN114912578A (en) | Training method and device of structure response prediction model and computer equipment | |
CN114418234A (en) | Power battery manufacturing capacity online prediction method based on reinforcement learning | |
CN110738363A (en) | photovoltaic power generation power prediction model and construction method and application thereof | |
CN116894180B (en) | Product manufacturing quality prediction method based on different composition attention network | |
CN112149810A (en) | Deep learning-based method for predicting and transferring quality of metal injection molding sintered product | |
CN116960962A (en) | Mid-long term area load prediction method for cross-area data fusion | |
Dube et al. | Machine Learning Approach to Predict Aerodynamic Performance of Underhood and Underbody Drag Enablers | |
CN114529040A (en) | On-line prediction method for assembly error of electromechanical product | |
CN115168602A (en) | Triple classification method based on improved concepts and examples | |
CN114692507A (en) | Counting data soft measurement modeling method based on stacking Poisson self-encoder network | |
CN114676887A (en) | River water quality prediction method based on graph convolution STG-LSTM | |
CN114943328A (en) | SARIMA-GRU time sequence prediction model based on BP neural network nonlinear combination | |
CN114912342A (en) | Packaging lead bonding process parameter optimization method based on multiple quality parameters | |
Wang et al. | Recurrent neural networks and its variants in remaining useful life prediction | |
CN110674883A (en) | Active learning method based on k nearest neighbor and probability selection | |
CN112330029A (en) | Fishing ground prediction calculation method based on multilayer convLSTM | |
CN114896864A (en) | Counting data soft measurement modeling method based on attention Poisson self-encoder network | |
Yang et al. | Remaining Useful Life Prediction Based on Stacked Sparse Autoencoder and Echo State Network | |
Aliabadian et al. | Optimizing the performance of ANFIS using the genetic algorithm to estimate the deformation modulus of rock mass | |
CN114282614B (en) | Medium-long runoff prediction method for optimizing CNN-GRU based on random forest and IFDA | |
CN114965120A (en) | Material mechanics parameter identification method based on artificial neural network and press-in test |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20201229 |