Control unit for executing tasks related to components
Through the dual neural network architecture combining physical first principles and modeling methods of measuring data, the problem of insufficient modeling speed and accuracy in the existing technology is solved, and faster and more accurate system modeling is achieved.
Patent Information
- Application Number
- CN202510125560.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Priority Date
- 2024-01-31
- Filing Date
- 2025-01-27
- Publication Date
- 2025-08-01
AI Technical Summary
The prior art is difficult to effectively combine physical first principles and measurement data in the modeling process, resulting in insufficient modeling speed and accuracy.
Using a bine neural network architecture, one for approximate states and the other for approximating unknown parts of the equation, modeling in combination with physical first principles and measured data.
Faster and more accurate system modeling is achieved, improving the accuracy and efficiency of the model.
Smart Images
Figure CN120406203A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to a control unit for performing tasks related to components. Background Art
[0002] System modeling is one of the core areas of any engineering and scientific research. It involves finding the mathematical relationship between system inputs and outputs. Broadly speaking, there are two main ways to model physical phenomena. The first is to use first principles to find the constitutive equations of the system, and the second is to use measurement data to model the relationship between the inputs and outputs of the system. In addition, many recent advances have been made in combining physics / first principles with measurement data to model systems. The expectation of this data-enhanced modeling is to achieve faster and more accurate solutions than can be achieved by pure physics or pure data-based modeling alone.
[0003] A technical paper titled "Deep Hidden Physics Models: Deep Learning of Nonlinear Partial Differential Equations" discloses that the time derivative of the state is known, while the other spatial derivatives of the state and the constitutive relationship of the spatial derivatives are unknown. Therefore, the unknown part is considered to depend on the spatial derivatives up to a certain order. And neural networks have been used to approximate this part. Thus, the solution strategy consists of two neural networks, one as an approximation of the state and the other as an approximation of the unknown part of the control equation. Brief Description of the Drawings
[0004] Figure 1 Illustrates a control unit for performing tasks related to components according to an embodiment of the present invention; and
[0005] Figure 2 Illustrates a flowchart of a method for performing tasks related to components according to the present invention. Detailed Description
[0006] Figure 1 Illustrates a control unit for performing tasks related to components according to an embodiment of the present invention. The control unit 10 monitors and identifies at least one incomplete function related to the component 12 in the operating mode. The control unit 10 manually corrects the at least one incomplete function and stores the data of the at least one incomplete function. The control unit 10 constructs and trains the intelligent module 14 using the stored data. The control unit 10 uses the trained intelligent module 14 to perform tasks related to the component 12 in a real-time environment.
[0007] Explain in further detail the construction of the control unit 10 and the components related to the control unit 10. The control unit 10 is a logic circuit and software program, which is implemented as one or more microprocessors, microcomputers, microcontrollers, digital signal processors, central processing units, state machines, logic circuits, and / or any component that operates on signals based on operation instructions. The control unit 10 includes an intelligent module 14, which uses the acquired / pre-stored data for training and construction. The control unit 10 includes a memory 16 for storing the acquired data. The operation of each component 12 involves multiple tasks and multiple functions for performing each task. The tasks include completed functions and uncompleted functions. The at least one uncompleted function is part of the task related to the component 12.
[0008] The functions involve multiple physical parameters, including force, distance, and perimeter. For example, in the manufacture of a machine, the component 12 is a screw. And in the manufacturing method, the tasks involved in the screw will be pressing, clinching, and crimping. And the function involved in pressing the screw 12 is how much distance the screw must enter the peripheral cavity of the machine. Another function can be how much perimeter of the screw 12 is required to firmly fix the screw 12 to the existing cavity in the machine.
[0009] The intelligent module is selected from a group of intelligent modules, which includes an artificial intelligence (AI) module, a deep learning (DL) module, and a machine learning module (ML). These modules include neural networks built and developed according to requirements. The component is a screw, and the at least one task is a fastening task, and at least one function related to the component is the distance required to tighten / fasten the screw. The data corrected manually is called the ground truth data for training the intelligent module.
[0010] Figure 2 A flowchart of a method for tasks related to a component is illustrated. In step S1, at least one uncompleted function related to the component 12 in the operating mode is identified, and the uncompleted function is monitored by the control unit 10. In step S2, the at least one uncompleted function is corrected manually, and the data of the at least one uncompleted function is stored. In step S3, the intelligent module 14 is constructed and trained using the stored data. In step S4, the trained intelligent module 14 is used to perform the tasks related to the component 12 in a real-time environment.
[0011] Explain the method in detail. The control unit 10 identifies at least one unfinished function of the task related to the component 12. Monitor the unfinished function and correct it in two ways. One way is to correct it, that is, manually modify the required physical parameters by any domain expert and store the corrected data in the memory 16 of the control unit 10. This corrected data is called real data. This data is used to construct and train a neural network in the intelligent module 14. As mentioned above, when the component 12 is in the operating mode, there will be multiple unfinished functions identified and monitored during the component manufacturing process. The second way is to identify and monitor the unfinished function when the component 12 is in the operating mode. Use the pre-loaded data and with the help of the trained intelligent module 14, perform the task by completing the function in a real-time environment.
[0012] Use an example to disclose the above method. The component 12 is a screw that needs to be tightened in a machine. The amount by which the screw 12 needs to be tightened, the size of the screw 12, and the force used to tighten the screw 12 are several functions related to the screw 12. If any of these functions are unfinished, that is, the physical parameters related to these functions, such as distance, perimeter, and length, and the force to be applied, are not calculated, then a domain expert is used to correct / calculate these functions during the calibration process.
[0013] By constructing and training the neural network of the intelligent module 14, the corrected data (such as the magnitude of the force to be applied to tighten the screw 12, etc.) is stored in the intelligent module 14 of the control unit 10. During the real-time operation scenario, the control unit 10, with the help of the intelligent module 14, corrects the unfinished functions of the task related to the component 12 by using the real time that has been stored in the control unit 10. It should be noted that the above-disclosed process can be applied to any other process and to any other component known in the prior art.
[0014] It should be understood that the embodiments explained in the above description are merely illustrative and do not limit the scope of the present invention. Many such embodiments and other modifications and changes to the embodiments explained in the description are envisioned. The scope of the present invention is limited only by the scope of the claims.
Claims
1. A control unit (10) for performing tasks related to a component (12), the control unit (10) being adapted to: - Monitor and identify at least one incomplete function related to the component (12) in an operating mode; - Manually correct the at least one incomplete function and store the data of the at least one incomplete function in a memory (16) of the control unit (10); - Build and train an intelligent module (14) using the stored data; - Perform the tasks related to the component (12) in a real-time environment by using the trained intelligent module (14).
2. The control unit (10) according to claim 1, wherein the at least one incomplete function is part of the tasks related to the component (12).
3. The control unit (10) according to claim 1, wherein the intelligent module (14) is selected from modules including an artificial intelligence (AI) module, a deep learning (DL) module, and a machine learning (ML) module.
4. The control unit (10) according to claim 1, wherein the at least one incomplete function relates to a plurality of physical parameters, including force, perimeter, and distance.
5. The control unit (10) according to claim 1, wherein the component (12) is a screw, and the at least one task is a fastening task, and the at least one function related to the component (12) is the distance by which the screw needs to be fastened.
6. The control unit (10) according to claim 1, wherein the trained intelligent module (14) recommends a correction of parameters during any one of the tasks including pressing, bending, and pleating during a manufacturing process, based on the data.
7. The control unit (10) according to claim 1, wherein the data corrected manually is referred to as real data for training the intelligent module (14).
8. A method for performing tasks related to a component (12), the method comprising: Monitoring and identifying, by a control unit (10), at least one incomplete function related to the component in an operating mode; Manually correcting the at least one incomplete function and storing the data of the at least one incomplete function; Building and training an intelligent module (14) using the stored data; Performing the tasks related to the component (12) in a real-time environment by using the trained intelligent module (14).