Creep age forming method and device based on digital twin technology

Patent metadata

Field	Value
Publication number	US-2026044650-A1
Application number	US-202519294689-A
Country	US
Kind code	A1
Filing date	Aug 8, 2025
Priority date	Aug 8, 2024
Publication date	Feb 12, 2026
Grant date	—

Abstract

A creep age forming method and device based on digital twin technology provided by the present disclosure. The method acquires sensor data of the aluminum alloy panel during a creep age forming process and inputs the sensor data into the first prediction model and obtains the predicted springback amount and yield strength output by the first prediction model. First difference values between the predicted springback amount and yield strength with the target springback amount and yield strength are determined. Based on the first difference values, first process parameters are determined for updating the process of the aluminum alloy panel by using the second prediction model. The first process parameters are sent to the autoclave, and the autoclave is controlled to perform a process operation with the first process parameters. The present disclosure can adjust the process based on real-time sensor data to obtain components with precise forming and target performance.

First claim

What is claimed is:

1 . A creep age forming method based on digital twin technology, comprising:
acquiring sensor data of an aluminum alloy panel during a creep age forming process; inputting the sensor data into a first prediction model and obtaining a predicted springback amount and yield strength output by the first prediction model, wherein, the first prediction model is obtained by using a first training method on an initial network model, and the initial network model is an LSTM algorithm model; inputs of the first prediction model comprise time, temperature, heating rate, and pressure; outputs of the first prediction model comprise the predicted springback amount and yield strength of the aluminum alloy panel; comparing the predicted springback amount and yield strength with a target springback amount and yield strength, and determining first difference values between the predicted springback amount and yield strength and the target springback amount and yield strength; determining, based on the first difference values, first process parameters for updating a process of the aluminum alloy panel by using a second prediction model; sending the first process parameters to an autoclave, and controlling the autoclave to perform process operations with the first process parameters; wherein, the second prediction model is obtained by using a second training method on the initial network model; inputs of the second prediction model comprise process updating time, time, temperature, heating rate, and pressure; outputs of the second prediction model comprise the predicted springback amount and yield strength of the aluminum alloy panel; wherein, the step of obtaining the first prediction model comprises: conducting creep age forming tests on a specimen-level aluminum alloy and obtaining a sample data set; constructing a material constitutive equation using the sample data set; performing a finite element simulation on a creep age forming process of a component-level aluminum alloy panel, based on the material constitutive equation, and obtaining a corresponding relationship between strain amount and displacement of a feature point on the aluminum alloy panel at different times; determining, based on the corresponding relationship, initial process parameters of the aluminum alloy panel and the springback amount and yield strength at mold-fitting time during the finite element simulation process; taking the initial process parameters and the springback amount and yield strength at the mold-fitting time as first training samples and training the initial network model to obtain the first prediction model; wherein, the step of obtaining the second prediction model comprises: during the finite element simulation process, defining a time when room temperature loading is completed as a starting time, and a time when the aluminum alloy panel is completely fitted to the mold as an ending time; determining a plurality of process update times between the starting time and the ending time according to a preset process update frequency to obtain a process update time sequence; acquiring finite element simulation process data corresponding to each process update time in the process update time sequence; training the initial network model using the process update time sequence and the finite element simulation process data corresponding to each process update time in the process update time sequence as second training data, and obtaining the second prediction model.

2 . The method according to claim 1 , wherein the material constitutive equation comprises:

ε1
=

f
⁡
(

σ
,
T
,
t
,
γ

)

=

f
⁡
(
σ
)

⁢

f
⁡
(
T
)

⁢

f
⁡
(
γ
)

⁢

f
⁡
(
t
)

;

Formula
⁢

(
1
)

wherein, ε 1 is a creep strain; σ is an applied creep stress; T is a creep temperature; γ is a creep heating rate; t is a creep time, and f(σ), f(T), f(γ), f(t) are a creep stress function, a creep temperature function, a creep heating rate function and a creep time function, respectively.

3 . The method according to claim 1 , wherein the step of obtaining the sample data set comprises:
conducting creep age forming tests on a specimen-level aluminum alloy and obtaining an initial sample data set; determining an optimal setting range for a K-value in a K-nearest neighbor algorithm; setting the K-value in the K-nearest neighbor algorithm based on the optimal setting range, and expanding the initial sample data set based on the K-nearest neighbor algorithm with the K-value setting to generate a new sample data set.

4 . The method according to claim 1 , wherein the corresponding relationship between strain amount and displacement of feature point on the aluminum alloy panel at different times comprises:

ε2
=

L
-

L
⁢
0

L
⁢
0

,

formula
⁢

(
2
)

wherein, ε 2 is the strain amount of the feature point on the alum

Assignees

• Univ Central South

Inventors

• Zhan Lihua
• Li Shuangbo
• Yang Youliang

Classifications

• G06N3/0442

  characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU] · CPC title

• G06F30/27 Primary

  using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model · CPC title

• G06F30/23 Primary

  using finite element methods [FEM] or finite difference methods [FDM] · CPC title

• G06N3/0985

  Hyperparameter optimisation; Meta-learning; Learning-to-learn · CPC title

• G06N3/084

  Backpropagation, e.g. using gradient descent · CPC title