Search

CN-122020521-A - Intelligent detection system and method for multiple physical characteristics of molten pool based on radiation optical signals

CN122020521ACN 122020521 ACN122020521 ACN 122020521ACN-122020521-A

Abstract

The invention discloses a molten pool multi-physical characteristic intelligent detection system and method based on a radiation light signal, which are constructed and can be used for detecting the molten pool multi-physical characteristic in a high-speed, long-time, synchronous and online manner, acquiring a large amount of data by printing through a plurality of parameter processes, calculating correlation coefficients between different physical characteristics and technological parameters of the molten pool and coupling relations between different physical characteristics, distributing different weights for each output characteristic in a loss function of a neural network model, establishing a one-dimensional convolutional neural network model, and outputting predicted values of the average temperature, area, width, depth and height of the molten pool in real time by taking the online acquisition photoelectric signal and the technological parameters as inputs.

Inventors

  • LIU ZHANWEI
  • ZHAO JINCHAO

Assignees

  • 北京理工大学

Dates

Publication Date
20260512
Application Date
20260114

Claims (10)

  1. 1. The intelligent detection system for the multiple physical characteristics of the molten pool based on the radiation optical signals is characterized by comprising a photoelectric signal acquisition module, a molten pool characteristic detection module and a signal processing module; The photoelectric signal acquisition module is used for acquiring radiation optical signals of two different wave bands of a molten pool, wherein the two different wave bands are 650 nm and 780 nm; The high-speed camera is used for acquiring molten pool images in the additive manufacturing process on line in real time, and further acquiring the molten pool area in the printing process through a gray segmentation algorithm; the laser scanner is used for performing off-line detection on the printing structural member after printing is finished, and obtaining a point cloud three-dimensional image of the structural member; The signal processing module is used for: 1) Receiving radiation light signals of two different wave bands of a molten pool, and calculating to obtain the average temperature of the molten pool; 2) Acquiring a printing process molten pool area through a gray segmentation algorithm according to an additive manufacturing process molten pool image acquired by a high-speed camera; 3) Calculating the three-dimensional shape of the surface of the printing part according to the point cloud three-dimensional image of the structural part, and further calculating melting height data and melting width data of a molten pool; 4) Time alignment is carried out on printer process parameters, radiation light signals, molten pool area, molten pool melting height, melting width data and melting depth data at each time point; 5) A one-dimensional convolutional neural network model is built and trained, wherein the technological parameters of a dual-band radiation optical signal and a laser printer at each time point are taken as model input, and real physical characteristic data at the corresponding time point are taken as output, so that the one-dimensional convolutional neural network model is trained, and the online prediction of the multidimensional physical characteristics of a molten pool is realized; wherein the real physical characteristic data comprises a bath mean temperature, a bath area, a bath width, a bath depth and a bath height.
  2. 2. The intelligent detection system for the multiple physical characteristics of the molten pool based on the radiation optical signals, as set forth in claim 1, is characterized in that the photoelectric signal acquisition module is fixed on a mechanical arm of a metal additive manufacturing laser generator, and has an included angle of 45 degrees with the horizontal direction of a working platform, so as to detect the radiation optical signals of the molten pool in the additive manufacturing process on line.
  3. 3. The intelligent detection system for the molten pool multiple physical characteristics based on the radiation optical signals, as claimed in claim 2, is characterized in that the photoelectric signal acquisition module comprises a first photoelectric detector, a second photoelectric detector, a data acquisition card and a molten pool scaling spectral filter light path, wherein the molten pool scaling spectral filter light path consists of a 650 nm narrow-band light filter, a 780 nm narrow-band light filter and a semi-reflective semi-transparent mirror; The molten pool image is firstly reduced to be completely received by the sensing target surface of the photoelectric detector by the molten pool zooming and light splitting filter light path, the reduced molten pool image is input to the half-reflecting half-lens, the transmission image enters the first photoelectric detector through the 650 nm narrow-band filter, the reflection image enters the second photoelectric detector through the 780 nm narrow-band filter, the signal processing module receives photoelectric signals received by the two photoelectric detectors, and the average temperature of the molten pool is calculated by applying the dual-band radiation optical signals by combining the colorimetric temperature measurement principle.
  4. 4. The intelligent detection system for the multiple physical characteristics of the molten pool based on the radiation optical signals, which is characterized in that the photoelectric signal acquisition module further comprises a first convex lens and a second convex lens, and the first convex lens and the second convex lens are combined, so that when parallel light is input, the output is also parallel light.
  5. 5. The intelligent detection system for the multiple physical characteristics of the molten pool based on the radiation optical signals, which is disclosed in claim 1, is characterized in that the molten pool characteristic detection module comprises a high-speed camera which is mounted on an aluminum alloy structure frame fixedly connected with a printing laser head and is used for acquiring the image of the molten pool in the additive manufacturing process in real time on line.
  6. 6. The intelligent detection system for the multiple physical characteristics of the molten pool based on the radiation optical signals, which is disclosed in claim 1, is characterized in that a laser starting signal is led out from a printing laser head, and a photoelectric detector and a high-speed camera are synchronously triggered outwards, so that the photoelectric signal and the molten pool area data are synchronously corresponding in real time, and the time alignment of each signal is realized.
  7. 7. The intelligent detection system for the multiple physical characteristics of the molten pool based on the radiation optical signals, according to claim 1, wherein the process parameters of the adopted laser printer comprise laser power, scanning speed, spot size and powder ejection speed.
  8. 8. The intelligent detection system for molten pool multiple physical characteristics based on radiation optical signals according to claim 1, wherein, when determining the weight of the loss function, the system comprises the steps of statistically analyzing the variance of each output characteristic and/or the correlation coefficient matrix between every two characteristics, specifically comprising the following steps: Firstly, carrying out standardization processing on the real physical characteristic data; calculating the variance value of each output characteristic after standardization; Thirdly, calculating a pearson correlation coefficient matrix between the standardized output features; And fourthly, based on the variance value and the correlation coefficient matrix, the highest weight is allocated to the core physical characteristics with large variance value and in the center of the coupling relation network, and the lower weight is allocated to the physical characteristics with small variance value and weak coupling with other characteristics.
  9. 9. The intelligent detection system for the molten pool multiple physical characteristics based on the radiation optical signals is characterized in that the one-dimensional convolution neural network model structure sequentially comprises an image input layer, a first convolution module, a second convolution module, a full connection module and an output full connection layer, wherein the size of the image input layer is 6, the image input layer is adaptive to the number of input characteristics, the first convolution module comprises a first convolution layer, a batch normalization layer, an activation function layer and a maximum pooling layer, the second convolution module comprises a second convolution layer with more filters, a batch normalization layer, an activation function layer and a pooling layer, the flattening layer, the full connection module comprises at least two full connection layers with 128 and 64 neurons, each layer is connected with the activation function and a discarding layer in sequence, the number of the neurons is 5, the neurons are matched with the number of characteristics to be predicted, and the weighted regression output layer calculates weighted losses through a forward propagation function.
  10. 10. A method of detecting a molten bath multiphysics intelligent detection system according to claim 1, comprising: Firstly, on-line collecting dual-band radiation optical signals of a molten pool through the photoelectric signal collecting module; Secondly, acquiring real physical characteristic data of a molten pool through the real characteristic detection module of the molten pool; Thirdly, receiving and preprocessing the photoelectric signals, the process parameters and the real physical characteristic data through the signal processing module; Fourth, analyzing variances of all output features and a correlation coefficient matrix thereof based on the real physical feature data; Fifthly, distributing weights for all output features in the loss function according to the analysis result, and obtaining the loss function through weighted summation; Step six, constructing and training the one-dimensional convolutional neural network model, and optimizing by using a loss function; And seventhly, inputting the online acquired technological parameters and photoelectric signals into a trained network model, and outputting predicted values of the average temperature, area, width, depth and height of the molten pool in real time.

Description

Intelligent detection system and method for multiple physical characteristics of molten pool based on radiation optical signals Technical Field The invention belongs to the technical field of infrared light force mechanics, and particularly relates to a molten pool multi-physical-characteristic intelligent detection system and method based on a radiation light signal. Background Metal additive manufacturing, which is one of the core technologies of a new industrial revolution, thoroughly changes the manufacturing range of traditional material reduction and isomorphism, and directly manufactures metal parts with complex geometric shapes based on three-dimensional model data by melting metal powder or wire layer by a high-energy beam (such as laser and electron beam). The stacking mode from bottom to top gives the unparalleled design freedom degree, can realize the processes which cannot be achieved by traditional manufacturing such as dot matrix structure, integrated molding, functional gradient material and the like, and has huge application potential in the fields such as aerospace, biomedical, high-end die, energy equipment and the like. However, metal additive manufacturing processes involve physicochemical phenomena under extreme conditions, transient high temperatures, extremely high temperature gradients, and rapid fusing processes. This unbalanced physical process makes the formed part extremely susceptible to various internal defects such as porosity, unfused, spheroidization, dimensional deviations, and residual stresses and cracks. These defects seriously threaten the mechanical properties, fatigue life and service reliability of the components, and restrict the further development and application of the technology in the field of high-end intelligent manufacturing. Therefore, the realization of on-line quality monitoring of metal additive manufacturing process is a key way to break through the bottleneck of the technology development. In metal additive manufacturing processes, the melt pool is a direct product of the interaction of the high energy beam with the metal material and is also the basic unit of the printing member, the characteristics of which directly determine the quality of the process. The temperature of the molten pool affects the fused structure and the crystal grain morphology, and the geometric characteristics of the molten pool such as width, length, depth and area are one of the most core physical characteristics in the metal additive manufacturing process, reflect the advantages and disadvantages of the process, are the most visual basis for judging whether a process window is reasonable or not, and the unstable size can directly cause various defects to affect the quality of the final part. Currently, monitoring techniques for molten pools rely primarily on a single sensor approach. For example, a high-speed camera or thermal infrared imager is used to obtain the geometric profile or temperature distribution of the molten pool. However, the methods have obvious limitations that the image-based method has large data volume and low processing speed, is difficult to meet the real-time requirement of a high-speed manufacturing process, is easy to be interfered by severe environments such as high temperature, smoke dust and the like on site, is difficult to synchronously acquire information of multiple physical fields by a single sensor, has strong nonlinear coupling relation between a temperature field and a geometric field of a molten pool, and cannot fully and accurately reflect the real complex state of the molten pool by analyzing a certain characteristic in isolation. Therefore, there is an urgent need in the art for an innovative method and system for comprehensively considering the multi-feature coupling relation of a molten pool and realizing high-speed, online, long-term and synchronous intelligent detection of multi-dimensional physical features of the molten pool. Disclosure of Invention In view of the above, the present invention is to provide a system and a method for intelligently detecting multiple physical characteristics of a molten pool based on a radiation optical signal. A molten pool multi-physical-characteristic intelligent detection system based on a radiation optical signal comprises a photoelectric signal acquisition module, a molten pool characteristic detection module and a signal processing module; The photoelectric signal acquisition module is used for acquiring radiation optical signals of two different wave bands of a molten pool, wherein the two different wave bands are 650 nm and 780 nm; The high-speed camera is used for acquiring molten pool images in the additive manufacturing process on line in real time, and further acquiring the molten pool area in the printing process through a gray segmentation algorithm; the laser scanner is used for performing off-line detection on the printing structural member after printing is finished, and obtaining