Spectrum / acoustic emission fusion monitoring method and device for fusion defects of laser powder bed
By using a fusion monitoring method combining spectral and acoustic emission signals, the problem of accuracy in defect monitoring during laser additive manufacturing was solved. This enabled the acquisition of multi-dimensional state information and defect identification during the laser additive manufacturing process, thereby improving part quality and optimizing the manufacturing process.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- HARBIN INST OF TECH
- Filing Date
- 2025-12-10
- Publication Date
- 2026-04-21
AI Technical Summary
Existing technologies are insufficient to comprehensively and accurately monitor and diagnose various defects in the laser additive manufacturing process, resulting in poor part quality and performance, which restricts its application in key areas.
A laser powder bed melting defect spectral/acoustic emission fusion monitoring method is adopted. By simultaneously acquiring acoustic emission signals and atomic emission spectral signals, feature extraction and fusion decision are performed using a one-dimensional convolutional neural network and a multi-scale spatial interactive fusion convolutional neural network. Defect monitoring is then performed in conjunction with DS evidence theory.
It enables the acquisition of multi-dimensional and multi-level state information of the laser additive manufacturing process, improves the accuracy of defect identification and part quality, and supports the comprehensive optimization of the manufacturing process.
Smart Images
Figure CN121899113A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to a spectral / acoustic emission fusion monitoring method and device, belonging to the field of metal laser additive manufacturing technology. Background Technology
[0002] Additive manufacturing (AM) is a technology that uses materials to create solid parts based on a three-dimensional digital model. Also known as 3D printing, it represents a major breakthrough in manufacturing technology. Metal laser additive manufacturing, as one of the key technologies in additive manufacturing, offers advantages such as short processing cycles, high material utilization, and excellent overall mechanical properties of the formed parts. Laser powder bed fused deposition (LPBF) is one of the mainstream technologies in metal laser additive manufacturing. Its working principle involves using a high-energy-density laser beam to melt metal powder along a planned path, depositing it layer by layer to form the final shape.
[0003] However, during laser additive manufacturing, various defects inevitably arise due to complex thermophysical processes, such as the interaction between high-energy lasers and materials, rapid melting and solidification, the generation and release of thermal stress, highly uneven temperature distribution, and high cooling rates. Common defect types include porosity, cracks, lack of fusion, and spheroidization. The presence of these defects severely affects the quality, performance, and reliability of parts, hindering the widespread application of laser additive manufacturing technology in key areas. Therefore, real-time monitoring and effective control of defects during laser additive manufacturing are crucial for improving the quality and performance of laser additive manufactured parts, and are also an important prerequisite for promoting the further development and application of laser additive manufacturing technology.
[0004] Currently, mainstream single-monitoring methods, whether based on acoustic, optical, or thermal signals, are insufficient for comprehensively and accurately diagnosing and predicting various types of defects and state characteristics. Similarly, control methods based on single monitoring methods struggle to achieve comprehensive optimization of the manufacturing process. Spectroscopic monitoring provides rich information about material elemental composition and energy level transitions, reflecting the microscopic physicochemical changes of materials under laser irradiation; while temporal acoustic spectral monitoring can analyze the propagation characteristics of stress waves within the material to reveal internal structural changes and defects. Integrating these two different types of monitoring data—spectral and temporal acoustic spectral data—allows for comprehensive acquisition of state information during the laser additive manufacturing process from multiple dimensions and levels. Summary of the Invention
[0005] To address the problem that existing technologies, whether based on acoustic, optical, or thermal signals, are unable to comprehensively and accurately diagnose and predict various types of defects and their state characteristics, this invention proposes a laser powder bed melting defect spectral / acoustic emission fusion monitoring method and device.
[0006] The technical solution adopted by the present invention to solve the above problems is as follows: The steps of the laser powder bed melting defect spectral / acoustic emission fusion monitoring method of the present invention include: Step 1: Collect multi-source synchronous monitoring data during the laser powder bed melting additive manufacturing process. The multi-source synchronous monitoring data includes acoustic emission signals and atomic emission spectrum signals during the laser powder bed melting process. Both acoustic emission signals and atomic emission spectrum signals are timestamped. Step 2: Preprocess the atomic emission spectrum signal and extract the spectral feature vector; Step 3: Perform time-frequency synchronous compression transformation on the acoustic emission signal to obtain the synchronous compression time-frequency spectrum of the acoustic emission signal; Step 4: Input the spectral feature vector into a one-dimensional convolutional neural network to extract spectral features, and input the synchronously compressed spectrum of the acoustic emission signal into a multi-scale spatial interactive fusion convolutional neural network to extract acoustic emission features; Step 5: Perform feature-level fusion of the extracted spectral features and acoustic emission features to obtain the first-level fusion decision result; Step 6: Obtain the independent decision results for acoustic emission and the independent decision results for spectral emission based on the outputs of 1D-CNN and MSIF-CNN, respectively; Step 7: Using the first-level fusion decision result, the acoustic emission independent decision result, and the spectral independent decision result as three evidence sources, perform decision-level fusion using DS evidence theory to output the final defect monitoring result.
[0007] Furthermore, in step 1, the acoustic emission signal is acquired using the QASS acoustic emission acquisition device, which includes a non-contact sound acquisition module; and atomic emission spectra are acquired using an AVATES multi-channel spectrometer.
[0008] Furthermore, the preprocessing of the atomic emission spectral signal in step 2 is as follows: the acquired raw spectral signal is subjected to dark background removal and normalization, and plasma temperature, electron density, spectral intensity and its fluctuation characteristics, and the intensity ratio of characteristic spectral lines of specific elements are extracted as spectral feature vectors.
[0009] Further, the specific process of step 3 is as follows: select the wavelet basis and the number of decomposition levels, and perform multi-scale wavelet decomposition on a noisy acoustic emission signal; perform threshold quantization on the noisy coefficients, that is, select an appropriate threshold for multi-scale wavelet decomposition; perform threshold quantization on the noisy coefficients, that is, select an appropriate threshold, set the coefficients below the threshold to zero, and the points above the threshold to the difference between the point value and the threshold; perform wavelet reconstruction to obtain the denoised acoustic emission signal. The acquired acoustic emission signal is bandpass filtered to remove environmental noise, and then time-frequency synchronous compression transformation is performed to generate a two-dimensional acoustic emission signal synchronous compression time spectrum with extremely high resolution.
[0010] The laser powder bed melting defect spectral / acoustic emission fusion monitoring device of the present invention includes a spectral acquisition module, an acoustic emission acquisition module, and a signal processing and fusion analysis unit; The spectral acquisition module is used to acquire the emission spectrum of the molten pool. It includes a collimating lens, a fiber optic probe, and a high-speed spectrometer. The acoustic emission acquisition module is used to acquire airborne acoustic emission signals. It includes a non-contact air coupling generator sensor, a preamplifier, and a high-speed data acquisition card. The signal processing and fusion analysis unit is used to run signal preprocessing algorithms and trained hybrid fusion models, and it is composed of industrial control units.
[0011] Furthermore, both the fiber optic probe and the air-coupled acoustic emission sensor are fixed to the operating table inside the forming chamber via a magnetic clamping device, enabling non-contact monitoring.
[0012] The beneficial effects of this invention are: 1. This invention fuses two different types of monitoring data, namely spectral and acoustic time spectrum, to comprehensively acquire state information of the laser additive manufacturing process from multiple dimensions and levels; 2. In this invention, an acoustic emission sensor and a spectrometer probe are mounted on the operating table of a laser powder bed forming chamber using a clamping device. The sensor and probe are aligned with the printing area through adjustment to obtain a clear and complete printing signal. Different feature parameters are extracted and machine learning is performed using a linear discriminant analysis method to integrate different defect types, establish a defect database, and achieve automatic defect identification. The identified defect model is then integrated into the defect database to improve the accuracy of subsequent judgments. Attached Figure Description
[0013] Figure 1 This is a schematic diagram of a laser powder bed melting defect spectral / acoustic emission fusion monitoring device according to the present invention; Figure 2 This is a flowchart of the data fusion process described in this invention; Figure 1In the middle, 1-Argon cylinder; 2-Powder spreading scraper; 3-Powder feeding cylinder lifting platform; 4-Powder feeding cylinder; 5-Forming cylinder lifting platform; 6-Forming cylinder; 7-Substrate; 8-Forming part; 9-Strong magnetic universal crab claw chuck; 10-Spectrometer probe head; 11-Spectrometer fiber optic cable; 12-Scanning system; 13-Air-coupled acoustic emission sensor; 14-Acoustic emission data cable; 15-Preamplifier; 16-Laser; 17-Fiber optic cable; 18-PC terminal. Detailed Implementation
[0014] Example 1 The process of manufacturing metal parts layer by layer using laser powder bed fusion molding equipment is as follows: First, based on the three-dimensional digital model of the part to be formed, the parts are layered using slicing software. Process parameters are set, including layer thickness, scanning spacing, laser power, scanning speed and scanning strategy, and a path control data file is generated that can be executed by the forming equipment.
[0015] Then, the equipment starts the forming cycle: The powder feeding cylinder 4 rises to a layer thickness height under the drive of the powder feeding cylinder lifting platform 3; The powder spreading blade 2 spreads the metal powder in the powder supply chamber evenly onto the surface of the substrate 7 in the forming cavity in a horizontal direction (e.g., from left to right) to form a powder thin layer of uniform thickness. The forming cavity is pre-filled with a high-purity inert protective gas (such as argon) to prevent the metal powder from oxidizing at high temperatures and to ensure forming quality. Laser 16 emits a high-energy laser beam, which is deflected at high speed by scanning system 12 and selectively irradiates the powder layer along a preset path, causing the metal powder in the irradiated area to melt rapidly and form a molten pool; the molten pool solidifies rapidly after the laser is removed, achieving metallurgical bonding with the already formed lower layer.
[0016] After completing the processing of the current layer, the equipment automatically enters the next cycle; The aforementioned cycle of "powder spreading—scanning—melting—descent" is repeated layer by layer until the entire three-dimensional part is manufactured. Throughout the manufacturing process, the spectral acquisition module and the acoustic emission acquisition module work synchronously to acquire the atomic emission spectral signals of the molten pool region and the acoustic emission signals during the process in real time, providing the original data foundation for the multimodal fusion defect monitoring described in this invention.
[0017] Example 2 The steps of a laser powder bed melting defect spectral / acoustic emission fusion monitoring method include: Step 1: Receive multi-source synchronous monitoring data during the laser powder bed melting additive manufacturing process. The multi-source synchronous monitoring data includes acoustic emission signals and atomic emission spectrum signals during the laser powder bed melting process, and both the acoustic emission signals and the atomic emission spectrum signals during the laser powder bed melting process are timestamped. Acoustic emission signals are acquired using the QASS acoustic emission acquisition device; the acoustic emission signal acquisition device includes: a non-contact sound acquisition module; Acoustic emission (AE) refers to the phenomenon in which a material generates transient elastic waves that propagate outward when subjected to external forces (such as stress, temperature changes, phase transitions, etc.) due to the rapid release of internal energy (such as crack propagation, plastic deformation, friction, corrosion, etc.). Atomic emission spectra were acquired using an AVATES multichannel spectrometer. The spectral signals mainly originated from plasma radiation and thermal radiation generated during the interaction between the laser and the metal powder. These spectral signals carry rich information about the process status and are important physical bases for online monitoring, defect identification, and process control. Step 2: Perform fusion analysis on acoustic emission / spectral monitoring data based on timestamps to extract key features characterizing forming quality; Key features include acoustic emission peak amplitude, duration, energy, count, rise time, peak amplitude frequency, spectral line intensity of the spectral signal, plasma temperature, electron density, and spectral line intensity ratio; Among them, acoustic emission signal characteristics describe porosity and crack defects; Peak amplitude: The maximum voltage value measured in the time-domain waveform of the acoustic emission signal; Duration: The time (ms) from the first threshold reached to the last threshold dropped back. Energy: The integral of the absolute voltage value of a signal over time.
[0018] Count: The number of times an acoustic emission signal exceeds a preset threshold.
[0019] Rise time: The time interval between first reaching the threshold and the peak value; Peak amplitude frequency: The frequency corresponding to the maximum power spectral density in the acoustic emission signal; Spectral signals can characterize defects such as elemental content, porosity, collapse, spheroidization, and splashing. Spectral line intensity: the peak intensity of a spectral line of a specific element; Plasma temperature: reflects the thermodynamic state of the plasma and is calculated using the Boltzmann plot method; Boltzmann plot: For multiple spectral lines of the same element, plot ln( I ki λki / g k A ki ) vs. E k The slope of the graph is inversely proportional to temperature.
[0020] Electron density: Reflects the degree of ionization of the plasma, calculated using the Stark broadening method; Stark broadening method: Measure the full width at half maximum (FWHM) of an elemental spectral line and substitute it into an empirical formula.
[0021] Step 3: Evaluate the quality of the formed part based on key features and obtain the evaluation results; Step 4: While collecting data from multiple sources, cache the data and convert it to a specified file format before saving it to the storage module; Step 5: Output feedback information based on the evaluation results in Step 3.
[0022] An apparatus for realizing a spectral / acoustic emission fusion monitoring method for defects in laser powder bed melting includes a metal laser powder bed melting system, an acoustic emission and spectral signal acquisition system, and a signal processing system; Metal laser powder bed melting system is used to realize 3D printing of metal components, printing metal components with different defect types or levels; The signal acquisition system is connected to the metal laser powder bed melting system to acquire acoustic emission and spectral signals during the 3D printing process of metal components; The signal processing system is connected to the signal acquisition system to extract feature parameters to build a machine learning model. The machine learning model is used to identify and classify unknown signals in the printing process, thereby realizing online monitoring of internal defects in the metal laser powder bed melting system.
[0023] Working principle Multimodal signal synchronous acquisition: During the operation of the LPBF device, a spectrometer is used to acquire the emission spectrum signal of the molten pool region at the laser application point in real time, with a resolution of no less than 0.2 nm and a sampling frequency of no less than 20 Hz. Simultaneously, a non-contact air-coupled acoustic emission sensor is used to acquire the acoustic emission signal generated during the LPBF process in real time, with a resonant frequency of 40 kHz and an operating frequency range of 35-45 kHz. Signal preprocessing and feature extraction Spectral signal processing: The acquired raw spectral signals are processed by removing dark background and normalizing, and plasma temperature, electron density, spectral intensity and its fluctuation characteristics, and the intensity ratio of characteristic spectral lines of specific elements are extracted as spectral feature vectors.
[0024] Acoustic emission signal processing: By selecting an appropriate wavelet basis and decomposition level, a noisy acoustic emission signal is decomposed into a multi-scale wavelet decomposition. The noisy coefficients are then thresholded, i.e., an appropriate threshold is selected for multi-scale wavelet decomposition. The noisy coefficients are then thresholded, i.e., an appropriate threshold is selected, and coefficients below the threshold are set to zero, while coefficients above the threshold are set to the difference between the value of that point and the threshold. Wavelet reconstruction is then performed to obtain the denoised acoustic emission signal.
[0025] The acquired acoustic emission signal is bandpass filtered to remove environmental noise, and then time-frequency synchronous compression transformation is performed to generate a two-dimensional acoustic emission signal synchronous compression time spectrum with extremely high resolution.
[0026] The feature vectors output from the two branches are fused at the feature level. Specifically, the two feature vectors are concatenated to form a fused feature vector, which is then input into a fully connected layer for dimensionality reduction and nonlinear mapping to obtain the first-level fusion decision result.
[0027] Constructing a training dataset: LPBF experiments were conducted under different process parameters, and spectral and acoustic emission signals were collected simultaneously. Defects in the formed parts were then identified through subsequent CT scans or metallographic analysis to establish a "multimodal signal-defect type" labeled dataset.
[0028] The acquired raw spectral signals and acoustic emission signals are preprocessed separately. The raw spectral signals are subjected to wavelength calibration, dark background removal and normalization to obtain a one-dimensional sequence of spectral intensity changes over time. The acquired acoustic emission signals are bandpass filtered to remove noise and then subjected to time-frequency synchronous compression transformation to generate a two-dimensional synchronous compressed time-domain spectrum of acoustic emission signals with extremely high resolution.
[0029] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent substitutions, and improvements made to the above embodiments without departing from the scope of the present invention, based on the technical essence of the present invention and within the spirit and principles of the present invention, shall still fall within the protection scope of the present invention.
Claims
1. A method for monitoring laser powder bed melting defects using a spectral / acoustic emission fusion method, characterized in that, The specific steps include: Step 1: Collect multi-source synchronous monitoring data during the laser powder bed melting additive manufacturing process. The multi-source synchronous monitoring data includes acoustic emission signals and atomic emission spectrum signals during the laser powder bed melting process. Both acoustic emission signals and atomic emission spectrum signals are timestamped. Step 2: Preprocess the atomic emission spectrum signal and extract the spectral feature vector; Step 3: Perform time-frequency synchronous compression transformation on the acoustic emission signal to obtain the synchronous compression time spectrum of the acoustic emission signal; Step 4: Input the spectral feature vector into a one-dimensional convolutional neural network to extract spectral features, and input the synchronously compressed spectrum of the acoustic emission signal into a multi-scale spatial interactive fusion convolutional neural network to extract acoustic emission features; Step 5: Perform feature-level fusion of the extracted spectral features and acoustic emission features to obtain the first-level fusion decision result; Step 6: Obtain the independent decision results for acoustic emission and the independent decision results for spectral emission based on the outputs of 1D-CNN and MSIF-CNN, respectively; Step 7: Using the first-level fusion decision result, the acoustic emission independent decision result, and the spectral independent decision result as three evidence sources, perform decision-level fusion using DS evidence theory to output the final defect monitoring result.
2. The laser powder bed melting defect spectral / acoustic emission fusion monitoring method according to claim 1, characterized in that, In step 1, the acoustic emission signal is acquired using the QASS acoustic emission acquisition device, which includes a non-contact sound acquisition module; and atomic emission spectra are acquired using an AVATES multi-channel spectrometer.
3. The laser powder bed melting defect spectral / acoustic emission fusion monitoring method according to claim 1, characterized in that, The preprocessing of atomic emission spectral signals in step 2 is as follows: the acquired raw spectral signals are subjected to dark background removal and normalization, and plasma temperature, electron density, spectral intensity and its fluctuation characteristics, and the intensity ratio of characteristic spectral lines of specific elements are extracted as spectral feature vectors.
4. The laser powder bed melting defect spectral / acoustic emission fusion monitoring method according to claim 1, characterized in that, Step 3 is as follows: Select the wavelet basis and the number of decomposition levels, and perform multi-scale wavelet decomposition on a noisy acoustic emission signal; perform threshold quantization on the noisy coefficients, that is, select an appropriate threshold for multi-scale wavelet decomposition; perform threshold quantization on the noisy coefficients, that is, select an appropriate threshold, set the coefficients below the threshold to zero, and the points above the threshold to the difference between the point value and the threshold; perform wavelet reconstruction to obtain the denoised acoustic emission signal. The acquired acoustic emission signal is bandpass filtered to remove environmental noise, and then time-frequency synchronous compression transformation is performed to generate a two-dimensional acoustic emission signal synchronous compression time spectrum with extremely high resolution.
5. A monitoring device for implementing the method as described in any one of claims 1 to 4, characterized in that, It includes a spectral acquisition module, an acoustic emission acquisition module, and a signal processing and fusion analysis unit; The spectral acquisition module is used to acquire the emission spectrum of the molten pool. It includes a collimating lens, a fiber optic probe, and a high-speed spectrometer. The acoustic emission acquisition module is used to acquire airborne acoustic emission signals. It includes a non-contact air coupling generator sensor, a preamplifier, and a high-speed data acquisition card. The signal processing and fusion analysis unit is used to run signal preprocessing algorithms and trained hybrid fusion models, and it is composed of industrial control units.
6. The laser powder bed melting defect spectral / acoustic emission fusion monitoring device according to claim 5, characterized in that, Both the fiber optic probe and the air-coupled acoustic emission sensor are fixed to the operating table inside the forming chamber by a magnetic clamping device, enabling non-contact monitoring.