Full length articleLaser cladding state recognition and crack defect diagnosis by acoustic emission signal and neural network
Introduction
With the continuous development, laser cladding technology possesses the many advantages in manufacturing of large and complex metal workpiece and in processes of metal surface strengthening. Laser cladding technology is a technical process to melt the substrate and metal powder by high-power laser beam to form a new layer of melt on the substrate. The pool relies on spontaneous cooling to form a cladding layer with high hardness, good thermal stability and certain metallurgical, mechanical or physical properties on the surface of the base material [1], thereby significantly improving the wear resistance, corrosion resistance, and resistance of the metal material fatigue and high temperature resistance [2], [3]. However, because the laser cladding process is a complex process with multi-physics integration, multiple process parameters interaction among laser power, laser scanning speed, powder feed rate, etc., which will result in an interactive effect on the quality of the cladding metal layer [4]. The coating is prone to severe cladding defects such as cracks, which greatly limits the wider promotion and development of laser cladding technology [5]. The optimization of the cladding process parameter combination and the feedback adjustments to the monitoring of the molten pool state of the cladding process can greatly improve the quality of the cladding layer and reduce the cladding defects such as cracks [6].
Therefore, the research on the identification of cladding state during laser cladding and the detection of typical defects in cladding layers have become research hotspots at home and abroad in recent years. In fact, the identification of cladding state and online defect detection in laser cladding have also been a challenge in the field of metal additive manufacturing, there are still few research reports on this aspect [7]. For the detection of cladding layer defects during the laser cladding process, Haythem Gaja [8] of the US Missouri University of Technology used cladding powders with different ratios to cladding to artificially control the occurrence of crack defects or pore defects, and construct the artificial neural network model to identify the acoustic emission signals collected during the process. The experiment can successfully identify these two types of defects, indicating that the acoustic emission signals have great potential for defect detection during the laser cladding process. Barua [9] et al. proposed the RGB value detection method of the laser metal deposition (LMD) high-temperature melt channel image. In large-spot LMD additive manufacturing process, the camera was used to shoot the red hot melt channel, and the RGB value of the melt channel image was analyzed online. If abnormal highlight areas were found, it is inferred that there are large crack defects under this area, but this method is limited to identify big cracks. For the related research on the quality evaluation and identification of the cladding layer, for example, Sashevchik et al. [10] adjusted the process parameters to produce pores with different densities, and defined different quality grades with different porosities. The sound signal in the cladding process is recorded and classified by machine learning algorithms. The experiment obtains a good accuracy but the method of diagnosis on quality level, does not provide any adjustment for laser cladding equipment information, to improve the quality of cladding layer to provide timely, effective contribution. Ye [11] changed the process parameters to make the parts in classes of “good”, “medium”, “poor” and other quality levels, and built a quality classification model, based on deep confidence network (DBN), to analyze the process acoustic sound signal, which experiment achieved a higher accuracy. However, only two process parameters, scanning speed and laser power, were changed to obtain several different cladding qualities, and the influence of other process parameters on cladding quality was not involved. Hossein Taheri et al. [12] detected acoustic emission signals from cladding substrates, and defined five states of the process: machine in inactive state, powder feeding only on, normal cladding state, low laser power, and low powder feeding rate. The clustering analysis of acoustic emission signals show that the data of each category has better clustering effect. His work demonstrates the application potential of AE in the on-line monitoring of laser cladding technology, but there are no results on the effect of more process parameters on the cracks in the cladding layer.
In order to monitor the parameters’ changing of laser cladding process online and detect the generation of typical defect cracks in cladding layer, in this research study, the acoustic emission (AE) signal feature-vectors are extracted, based on the characteristic of the maximum magnitude and power energy of the signal, to extract the high-energy signal samples and the low-energy signal samples, which containing the high-energy impact signals and the low-energy base signal respectively. Then, a neural network for crack detection is applied on the high-energy signal samples and detect whether cracks are generated from the additive process of laser cladding. A neural network for identifying cladding state is applied on the low-energy signal samples and identify the current cladding state. Finally, according to the feature-vector extracted from these two parts of the data, the current state of the process is recognized as one state of the following: the state of normal cladding with good cladding layer quality; the state of a certain process parameter is abnormal and possible cracks in cladding layer; or the state of a certain process parameter is abnormal and the detected cracks.
Section snippets
Experimental conditions
The equipment system framework used in the experiment in this paper is shown in Fig. 1, which mainly includes coaxial powder feeding laser cladding system, cooling device and the acoustic emission signal acquisition system. During the experiment, the cooling device will be supplied with circulating cold water and act as a clamp to clamp the plate-type cladding substrate. The acoustic emission receiver is installed on one side of the substrate and connected to other acquisition equipment for AE
Acoustic emission detection technology
When the material or structure is subjected to external forces, the internal local stress concentration will result in an unstable stress distribution. When the strain energy under such a distribution accumulates to a certain extent, the unstable high energy state will be changed over to a stable low energy state. The state transition includes plastic strain, rapid phase transition, crack generation, fracture development, etc. In this transition, part of the strain energy is released in the
Laser cladding state recognition and crack defect diagnosis model (SRCD)
Based on the theories mentioned above, the work done in this research includes AE signal detection of laser cladding process and the state identification of predefined cladding states, through AE signal process and feature extraction. The functions of laser cladding state recognition and crack defect diagnosis model (SRCD) are presented in Fig. 5. Acoustic emission signals in laser cladding process are collected and uploaded to the PC, and then input into the networks after the data
Model training and result analysis
In order to intuitively verify the experimental effects of the proposed method, the model is verified by various given signals in the data set. The scrambled signal data samples with labels were fed into the module to obtain the recognition results of each signal sample. The recognition results are compared with the real tags, and the consistency of the recognition results with the real tags are obtained. The ratio of the correct number of samples to the total number of samples (Eq. (8)) is
Conclusion
The research in this article studied that changing several main process parameters has different effects on the cladding process and resultant layer quality. The acoustic emission signal during the cladding process is recorded online by the acoustic emission receiver on the substrate. The collected acoustic emission signals are classified to low-energy signals and high-energy signals according to the characteristic parameters. Then, a cladding state recognition network and a crack detection
CRediT authorship contribution statement
Kaiqiang Li: Conceptualization, Methodology, Validation, Formal analysis, Investigation, Data curation, Writing - original draft, Visualization. Tao Li: Supervision, Methodology, Investigation, Writing - review & editing. Min Ma: Validation, Investigation, Supervision. Dong Wang: Validation, Investigation, Supervision. Weiwei Deng: Resources. Huitian Lu: Methodology, Supervision.
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgments
This work was jointly supported by the National Natural Science Foundation of China (No. 51975099, No. 51875075) and the Key Research and Development Program of Ningxia Hui Autonomous Region of China (No.2018BDE02045)
References (29)
- et al.
A study of Ni-based WC composite coatings by laser induction hybrid rapid cladding with elliptical spot
Appl. Surf. Sci.
(2008) - et al.
In situ monitoring of selective laser melting using plume and spatter signatures by deep belief networks
ISA T
(2018) - et al.
Acoustic emission studies for characterization of fatigue crack growth in 316LN stainless steel and welds
Appl. Acoust.
(2017) - et al.
Analysis of defect generation in Ti–6Al–4V parts made using powder bed fusion additive manufacturing processes
Additive Manuf.
(2014) - Q. Li, Study on Morphology and Crack of Laser Cladding Layer of Ni60A Self-fluxing Alloy, Dalian University of...
Laser cladding
J. Laser Appl.
(1999)- et al.
Multi-objective optimization of coaxial powder feeding laser cladding based on NSGA-II
Chinese J. Laser
(2020) - H. Xu, Study on Shape Control Process of 316L Stainless Steel Solid Forming by Laser Cladding, Dalian University of...
- et al.
Defect detection in selective laser melting technology by acoustic signals with deep belief networks
INT J ADV MANUF TECH
(2018) - et al.
A defects detection method for laser metal deposition and thermal fields simulation
J. Xi’an University Technol.
(2019)