Electric arc additive deposition process online monitoring and morphology prediction system based on deep learning

Through the online monitoring and morphology prediction system based on deep learning, the problems of inconsistency in molding and low degree of automation in arc additive manufacturing are solved, and high-precision control and morphology prediction of arc additive deposition process are achieved, which improves the Process accuracy and consistency.

CN119026461APending Publication Date: 2024-11-26XI AN JIAOTONG UNIV
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202411051151.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-01
Publication Date
2024-11-26

AI Technical Summary

Technical Problem

Arc additive manufacturing technology lacks a complete research system and standard specifications in process, process monitoring and quality evaluation, which makes the molding morphology difficult to be consistent with the design and the degree of automation is low, which limits the further development of the technology.

Method used

Using an online monitoring and morphology prediction system based on deep learning, the process parameters and welding machine voltage signals are obtained in real time through the data acquisition system, and a high-precision real-time morphology prediction model is constructed, combined with digital twin technology to perform three-dimensional demonstrations to achieve the process parameters. Automated precise control.

Benefits of technology

It realizes high-precision prediction of the solidification morphology of the molten pool during arc additive deposition, improves process accuracy and consistency, and supports real-time morphology prediction speed below 1m/min, which is far beyond the accuracy and flexibility of the existing technology. .

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119026461A_ABST
    Figure CN119026461A_ABST
Patent Text Reader

Abstract

According to the electric arc additive deposition process online monitoring and morphology prediction system based on deep learning, single-channel deposition is completed on an additive deposition platform, and technological parameters and welding machine voltage signals are obtained in real time through a data acquisition system; after the collected data is processed by the data preprocessing module, a data set of the real-time morphology prediction model is generated, the optimized and pre-trained real-time morphology prediction model is loaded into an upper computer, real-time prediction of the solidification morphology of the liquid molten pool is realized, and three-dimensional demonstration is performed through a digital twinning technology. The method is used for online predicting and controlling the solidification morphology of the liquid molten pool. The online monitoring technology is utilized, and real-time process parameters and voltage signals reflecting transient solidification information of the molten pool are efficiently recorded; meanwhile, a high-precision real-time morphology prediction model is constructed by means of a deep learning technology; and through fusion, it is ensured that the solidification morphology of the molten pool can be accurately predicted in the deposition process, and therefore automatic and accurate control over technological parameters is achieved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of metal additive manufacturing, and in particular to an online monitoring and morphology prediction system for an arc additive deposition process based on deep learning. Background Art

[0002] Metal additive manufacturing technology is an important driving force of the Fourth Industrial Revolution, especially suitable for high-end equipment manufacturing fields such as aerospace. In dealing with the difficulties in preparing medium and large complex components, arc additive manufacturing stands out from many metal additive technologies. This technology is an additive manufacturing method that uses arc as a heat source to melt metal wire and deposit layer by layer to form metal components. It has the advantages of efficient deposition, flexibility and low cost.

[0003] At present, arc additive manufacturing has not yet formed a complete research system and standard specifications in terms of process, process monitoring and quality assessment. The process is highly dependent on manual intervention, has a low degree of automation, and the high transient state during the solidification process is difficult to control, resulting in the formation morphology of metal components being difficult to be consistent with the designed three-dimensional model. These shortcomings limit the further development of the technology. The consistency of forming dimensional accuracy is considered to be a key challenge restricting the engineering application and rapid industrialization of arc additive technology. The key to solving this problem lies in building a high-precision prediction model from process parameters to the geometric morphology of deposited parts; in addition, the industry lacks a standardized online monitoring platform for arc additive forming processes, which is difficult to meet the needs of academic research and industrial applications.

[0004] In recent years, with the development of digital thread, interdisciplinary research and information fusion technology, the combination of online monitoring technology of multi-sensor information fusion and deep learning technology has provided new possibilities for solving the above problems. These technological advances are expected to realize the automation and intelligence of arc additive manufacturing process, improve process accuracy and consistency, and lay the foundation for industry standardization and wide application. Summary of the invention

[0005] In order to overcome the shortcomings of the above-mentioned prior art, the purpose of the present invention is to provide an online monitoring and morphology prediction system for arc additive deposition process based on deep learning, which utilizes online monitoring technology to efficiently record real-time process parameters and voltage signals reflecting the transient solidification information of the molten pool, and uses them as an important basis for subsequent analysis; at the same time, a high-precision real-time morphology prediction model is constructed with the help of deep learning technology; through fusion, it is ensured that the solidification morphology of the molten pool can be accurately predicted during the deposition process, thereby realizing automatic and precise control of process parameters.

[0006] In order to achieve the above object, the technical solution adopted by the present invention is:

[0007] An online monitoring and morphology prediction system for arc additive deposition process based on deep learning completes single-pass deposition on an additive deposition platform, and obtains process parameters and welding machine voltage signals in real time through a data acquisition system; the collected data is processed by a data preprocessing module to generate a data set of a real-time morphology prediction model, and the optimized and pre-trained real-time morphology prediction model is loaded into a host computer to achieve real-time prediction of the solidification morphology of the liquid molten pool, and a three-dimensional demonstration is performed through digital twin technology for online prediction and control of the solidification morphology of the liquid molten pool. The system is easy to update, and further data development and update can be carried out by replacing the real-time morphology prediction model.

[0008] An online monitoring and morphology prediction system for arc additive deposition process based on deep learning, comprising:

[0009] Additive deposition platform: It is divided into two parts. One part is the welding robot 1, the wire feeder 2 and the welding machine 3 forming the fuse system, and the other part is the three-axis motion platform 4 and the motion control cabinet 5 forming the relative motion system. The two parts work together to realize arc additive deposition; the welding robot 1 is used to fix the position and posture of the welding gun; the wire feeder 2 provides the original metal wire for forming; the welding machine 3 provides the heat source of the fuse and is connected to the host computer 6 to display the wire feeding speed of the wire feeder in real time; the three-axis motion platform 4 is connected to the host computer 6, and is used to control the welding gun to deposit at a fixed posture and a specified speed according to the input command; the motion control cabinet 5 drives the three-axis motor through the motion control card to execute the given motion command;

[0010] Data acquisition system: used to collect deposition process parameters, welding machine voltage data and deposited part three-dimensional morphology; process parameters include wire feeding speed and welding speed, which are respectively connected to the host computer 6 through the welding machine 3 and the three-axis motion platform 4 through the communication protocol to transmit information in real time; the high-frequency voltage acquisition system 7 is composed of a Ni acquisition board, a voltage transmitter and an isolator, which is used to collect the voltage data of the welding machine 3 and transmit it to the host computer 6 through a high-frequency signal; the three-dimensional morphology data of the deposited part is scanned and acquired by the first laser profile scanner 8 and the second laser profile scanner 9;

[0011] Data preprocessing module: used to perform data cleaning and voltage fluctuation feature extraction on the collected process parameters and high-frequency voltage signals to generate an input set of the morphology prediction model; at the same time, the point cloud data collected by the first laser profile scanner 8 and the second laser profile scanner 9 are registered and the cross-section point cloud is extracted to determine the consistent point cloud data volume of each cross section, and multiple sets of two-dimensional coordinates representing the point cloud are used as the output set of the real-time morphology prediction model;

[0012] Real-time morphology prediction model: Through orthogonal experiments, variables such as learning rate, number of neural network layers, number of neurons in each layer, and amount of data extracted per training round are optimized. The optimization method is to use five-fold cross-validation to evaluate the lowest overall prediction error of the test set.

[0013] Host computer 6: connected to the welding machine 3, the three-axis motion platform 4, and the high-frequency voltage acquisition system 7, the host computer 6 is implanted with a real-time morphology prediction model trained by a neural network, which is used to predict the cross-sectional morphology of the molten pool solidification weld according to the real-time process parameters and voltage fluctuations, and connect the cross-sectional morphologies along the deposition path to achieve three-dimensional digital twin geometric model mapping of the real-time deposition morphology.

[0014] Compared with the prior art, the present invention has the following beneficial effects:

[0015] (A) The present invention constructs an online monitoring and morphology prediction system for arc additive deposition process based on deep learning, and provides an online collection method for process parameters and high-frequency voltage during the deposition process, which can be effectively applied to arc additive platforms.

[0016] (B) The present invention develops for the first time a real-time morphology prediction model for arc additive molten pool solidification based on deep learning. By loading the pre-trained model into the host computer and using the data collected by the data acquisition system and processed by the pre-processing module (including process parameters and continuous voltage data reflecting molten pool fluctuations), the solidification morphology of the wire melting point can be predicted in real time, thereby achieving high-precision prediction of the full characteristics of the solidified weld cross-section.

[0017] (C) The present invention proposes and applies for the first time a high-precision weld bead feature extraction method, which achieves high-precision extraction of weld bead size information through dual laser scanning and point cloud registration, far exceeding the accuracy of current destructive tests.

[0018] (D) The prediction speed of the real-time morphology prediction model of the present invention supports real-time morphology prediction at welding speeds lower than 1 m / min. When developing and updating data, only the real-time morphology prediction model needs to be adjusted, thereby enhancing the compatibility and scalability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a framework diagram of the present invention.

[0020] Figure 2 Schematic diagram of the additive deposition platform, data acquisition system, host computer, etc. of the present invention.

[0021] Figure 3 This is a flowchart of data set screening and deep learning model training and verification according to an embodiment of the present invention. DETAILED DESCRIPTION

[0022] The technical solution of the present invention is described in detail below in conjunction with the accompanying drawings and embodiments. It is obvious that the described embodiments are only a part of the present invention but not all embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present invention.

[0023] Reference Figure 1 , an online monitoring and morphology prediction system for arc additive deposition process based on deep learning, including an additive deposition platform, a data acquisition system, a data preprocessing module, a real-time prediction model and a host computer;

[0024] Additive Deposition Platform: Reference Figure 2 , divided into two parts: one part is the welding robot 1, the wire feeder 2 and the welding machine 3 forming the fuse system, and the other part is the three-axis motion platform 4 and the motion control cabinet 5 forming the relative motion system, and the two parts work together to realize arc additive deposition; the welding robot 1 is used to fix the position and posture of the welding gun; the wire feeder 2 provides the original metal wire for forming; the welding machine 3 provides the heat source of the fuse, and is connected to the host computer 6 to display the wire feeding speed of the wire feeder in real time; the three-axis motion platform 4 is connected to the host computer 6, and is used to control the welding gun to deposit at a fixed posture and a specified speed according to the input command; the motion control cabinet 5 drives the three-axis motor through the motion control card to execute the given motion command;

[0025] Data acquisition system: used to collect deposition process parameters, welding machine voltage data and deposited part three-dimensional morphology; process parameters include wire feeding speed and welding speed, which are connected to the host computer 6 through the welding machine 3 and the three-axis motion platform 4 to transmit information in real time; Figure 2 The high-frequency voltage acquisition system 7 is composed of a Ni acquisition board, a voltage transmitter and an isolator, and is used to collect the voltage data of the welding machine 3 and transmit it to the host computer 6 through a high-frequency signal; the three-dimensional morphology data of the deposited part is scanned and acquired by the first laser profile scanner 8 and the second laser profile scanner 9;

[0026] The data acquisition system of this embodiment is composed of three parts. The first part is to connect the wire feeder 2 and the three-axis motion platform 4 with the host computer 6 through a network cable to obtain the real-time wire feeding speed and welding speed; the second part is to collect real-time voltage data at a frequency of 100000Hz with the help of the built high-frequency voltage acquisition system 7; finally, the first laser profile scanner 8 and the second laser profile scanner 9 are used to scan and realize the high-precision spatial point cloud data acquisition of the solidified morphology of the weld;

[0027] Data preprocessing module: used to perform data cleaning and voltage fluctuation feature extraction on the collected process parameters and high-frequency voltage signals to generate an input set of the morphology prediction model; at the same time, the point cloud data collected by the first laser profile scanner 8 and the second laser profile scanner 9 are registered and the cross-section point cloud is extracted to determine the consistent point cloud data volume of each cross section, and multiple sets of two-dimensional coordinates representing the point cloud are used as the output set of the real-time morphology prediction model;

[0028] Real-time morphology prediction model: It is built based on deep learning technology and digital twin technology, optimizes model parameters through orthogonal experiments, and uses the average error loss of five-fold cross validation as the evaluation standard; the optimal model uses a learning rate of 0.001, a single sampling of 16, a hidden layer of 3 layers, and the number of neurons in each layer is 64, 128 and 256 respectively; the input of the neural network includes the fluctuating voltage value, process parameters and voltage fluctuation characteristic values ​​at the same time interval, which are determined to be 75 values ​​as a set of inputs after data preprocessing; the output is multiple sets of two-dimensional point cloud coordinates extracted from the weld cross section, which are determined to be 350 sets after data preprocessing, as shown in Figure 3 As shown;

[0029] Host computer 6: connected to welding machine 3, three-axis motion platform 4, and high-frequency voltage acquisition system 7, the host computer 6 is implanted with a real-time morphology prediction model trained by a neural network, which is used to predict the cross-sectional morphology of the solidified weld bead in the molten pool according to the real-time process parameters and voltage fluctuations, and connect each cross-sectional morphology along the deposition path to achieve three-dimensional digital twin geometric model mapping of the real-time deposition morphology. The embedded human-computer interactive display interface is used to display the motion posture of each axis, real-time process parameters, real-time voltage, and the three-dimensional model and morphology characterization parameters predicted by the real-time prediction model.

[0030] A method for building an online monitoring and morphology prediction system for arc additive deposition process based on deep learning, comprising the following steps:

[0031] Step 1: Arrange the additive deposition platform, data acquisition system and host computer 6;

[0032] Step 2: Initialize the additive deposition platform and data acquisition system, set the initial wire feeding speed and other process parameters on the teaching pendant of the welding machine 3, put the wire on the wire feeder 2, set the unidirectional motion instruction on the three-axis motion platform 4, connect the high-frequency voltage acquisition system 7, start the first laser profile scanner 8 and the second laser profile scanner 9, and start the host computer 6 software;

[0033] Step 3: Deposit a single channel and collect deposition process information. Invar alloy wire is used as the original deposition material. The wire feeding speed (m / min) is 2, 4, 6, 8, and 10, respectively. The welding speed (m / min) is 0.1, (0.1, 0.2), (0.1, 0.2, 0.3, 0.4), (0.1, 0.2, 0.3, 0.4, 0.5, 0.6), (0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7), respectively. The above parameters are formable parameters determined by actual collection. High-frequency voltage signals, wire feeding speeds, and welding speeds are collected in real time during the deposition process. After deposition, the single channel position is maintained and two laser profile scanners are used to obtain complete point cloud data.

[0034] Step 4: Data preprocessing stage: The collected voltage signals, wire feeding speed and welding speed are packaged into online data sets and aligned with the point cloud data in time; after cleaning and algorithm matching, a complete online data set of single material formable process parameters and its corresponding 350 sets of two-dimensional coordinates are obtained, such as Figure 3 As shown on the right;

[0035] Step 5: Build a real-time morphology prediction model, optimize model parameters and pre-train;

[0036] 5.1) Build Figure 3 The neural network learning framework shown;

[0037] 5.2) According to the design, Table 1 contains 63 models, each with 9 sets of parameter settings; each model has undergone 5-fold cross validation, with 100 rounds of training for each fold, and the best model parameters are selected by calculating the average error of cross validation; the final optimization parameters are a learning rate of 0.001, a single sampling number of 16, a hidden layer structure of 3 layers, and the number of neurons in each layer is 64, 128, and 256 respectively;

[0038] 5.3) Using the online dataset defined in step 4 as input, a single section corresponding to a single dataset is used as the target set for pre-training;

[0039] 5.4) The fitting error is 0.129 after verification on the test set. The error calculation is based on the following formula, where y i is a set of 2D coordinates of the cross-section point cloud;

[0040]

[0041] Table 1

[0042]

[0043] Step 6: The host computer loads the real-time morphology prediction model pre-trained in step 5 to improve the digital twin module;

[0044] Step 7: Deposition and verification, deposit a new single channel for prediction; the deposition speed required by conventional experiments can be met, and real-time prediction can be achieved when the welding speed is lower than 1m / min. During deposition, the collected real-time parameters are preprocessed and input into the real-time morphology prediction model, and the output results are reflected to the host computer;

[0045] Step 8: Data processing, the three-dimensional data generated by real-time prediction can be stored for subsequent multi-scale analysis and as a reference for further optimization of the real-time morphology prediction model.

[0046] In summary, the present invention is an online monitoring and morphology prediction system for arc additive deposition process based on deep learning, which provides a method for collecting process parameters and high-frequency voltage during the deposition process. The system can be easily applied to arc additive platforms. At the same time, a real-time morphology prediction model for molten pool solidification for arc additive based on deep learning is developed for the first time. By loading the pre-trained model in the host computer and using the data acquired by the data acquisition system and processed by the pre-processing module (including process parameters and continuous voltage data reflecting molten pool fluctuations), the solidification morphology of the wire melting position can be predicted in real time, thereby achieving high-precision prediction of the full characteristics of the cross-section of the solidified weld.

Claims

1. An online monitoring and morphology prediction system for arc additive deposition process based on deep learning, characterized by: Single-pass deposition is completed on the additive deposition platform, and the process parameters and welding machine voltage signals are obtained in real time through the data acquisition system. The collected data is processed by the data preprocessing module to generate a data set of the real-time morphology prediction model of the molten pool solidification. The optimized and pre-trained real-time morphology prediction model is loaded into the host computer to realize real-time prediction of the solidification morphology of the liquid molten pool, and a three-dimensional demonstration is performed through the digital twin technology for online prediction and control of the solidification morphology of the liquid molten pool.

2. An online monitoring and morphology prediction system for arc additive deposition process based on deep learning, characterized in that: include: The additive deposition platform is divided into two parts, one part is a welding robot (1), a wire feeder (2) and a welding machine (3) forming a fuse system, and the other part is a three-axis motion platform (4) and a motion control cabinet (5) forming a relative motion system, and the two parts work together to achieve arc additive deposition; the welding robot (1) is used to fix the position and posture of the welding gun; the wire feeder (2) provides raw metal wire for forming; the welding machine (3) provides a heat source for the fuse and is connected to a host computer (6) to display the wire feeding speed of the wire feeder in real time; the three-axis motion platform (4) is connected to the host computer (6) and is used to control the welding gun to deposit at a fixed posture and a specified speed according to input instructions; the motion control cabinet (5) drives the three-axis motor through a motion control card to execute a given motion instruction; Data acquisition system: used to collect deposition process parameters, welding machine voltage data and deposited part three-dimensional morphology; process parameters include wire feeding speed and welding speed, which are respectively connected to the host computer (6) through the welding machine (3) and the three-axis motion platform (4) through a communication protocol to transmit information in real time; the high-frequency voltage acquisition system (7) is composed of a Ni acquisition board, a voltage transmitter and an isolator, and is used to collect the voltage data of the welding machine (3) and transmit it to the host computer (6) through a high-frequency signal; The three-dimensional shape data of the deposited part is scanned and acquired by a first laser profile scanner (8) and a second laser profile scanner (9); Data preprocessing module: used for performing data cleaning and voltage fluctuation feature extraction on the collected process parameters and high-frequency voltage signals to generate an input set of the morphology prediction model; at the same time, performing registration and cross-section point cloud extraction on the point cloud data collected by the first laser profile scanner (8) and the second laser profile scanner (9) to determine the consistent point cloud data volume of each cross section, and using multiple sets of two-dimensional coordinates representing the point clouds as output sets of the real-time morphology prediction model; Real-time morphology prediction model: Through orthogonal experiments, variables such as learning rate, number of neural network layers, number of neurons in each layer, and amount of data extracted per training round are optimized. The optimization method is to use five-fold cross-validation to evaluate the lowest overall prediction error of the test set.

3. The arc additive deposition process online monitoring and morphology prediction system based on deep learning according to claim 2, characterized in that: The host computer (6) is implanted with a real-time morphology prediction model trained by a neural network, which is used to predict the cross-sectional morphology of the molten pool solidification weld according to the real-time process parameters and voltage fluctuations, and connect each cross-sectional morphology along the deposition path to achieve three-dimensional digital twin geometric model mapping of the real-time deposition morphology.

Citation Information

Cited By

  • Additive manufacturing surface topography prediction method and system based on multi-source molten pool feature fusion

    CN120763684A