Author(s)

Jiayu Wang ¹, Jiali Gao ¹, Pengfei Qiu ¹

Affiliation(s)

¹ University of Shanghai for Science and Technology, Shanghai, China

Corresponding Author

Jiali Gao

ABSTRACT

Precise control of defocus distance is key to ensuring the cladding quality in laser powder feeding additive manufacturing. To achieve online real-time monitoring and intelligent identification of the defocus state, this study proposes an intelligent identification framework that integrates principal component analysis (PCA) with deep learning. Three models—multilayer perceptron (MLP), autoencoder (AE), and convolutional neural network (CNN)—were systematically constructed and compared. The results show that, through its inherent convolution and pooling operations, the PCA-CNN model can adaptively capture the gradual changes in the two-dimensional spatial morphology of the molten pool caused by minor variations in defocus distance, achieving an overall accuracy of 99.28% on an independent test set. This performance is significantly superior to that of PCA-MLP (95.34%) and PCA-AE (93.19%). The proposed PCA-CNN intelligent identification framework enables rapid and accurate determination of the defocus state during the cladding process, providing a reliable technical foundation for subsequent real-time compensation of process parameters and closed-loop quality control.

KEYWORDS

Laser additive manufacturing, defocus distance, principal component analysis, online monitoring, convolutional neural network

CITE THIS PAPER

Jiayu Wang, Jiali Gao, Pengfei Qiu. Research on Intelligent Identification of Defocus Distance in Laser Powder Feeding Additive Manufacturing Based on PCA-CNN. Journal of Engineering Mechanics and Machinery (2026). Vol. 11, No. 1, 193-204. DOI: http://dx.doi.org/10.23977/jemm.2026.110118.

REFERENCES

[1] Zhou Y, Jiang D, Al-Akailah A, et al. Understanding the formation of laser-induced melt pools with both wire and powder feeding in directed energy deposition[J]. Additive Manufacturing, 2024, 89: 104312.
[2] Kittel J, Wendt F, Hoelters S, et al. Approach for advanced working distance monitoring and control capability in laser metal deposition processing for additive manufacturing[J]. Journal of Laser Applications, 2023, 35.
[3] Gong X, You W, Li X, et al. Modeling the influence of injection parameters on powder efficiency in laser cladding[J]. Welding in the World, 2020, 64(8): 1437-1448.
[4] Shi T, Lu B, Shen T, et al. Closed-loop control of variable width deposition in laser metal deposition[J]. The International Journal of Advanced Manufacturing Technology, 2018, 97(9): 4167-4178.
[5] Chabot A, Rauch M, Hascoet J. Novel control model of Contact-Tip-to-Work Distance (CTWD) for sound monitoring of arc-based DED processes based on spectral analysis[J]. The International Journal of Advanced Manufacturing Technology, 2021, 116: 1-10.
[6] Stehmar C, Gipperich M, Kogel-Hollacher M, et al. Inline Optical Coherence Tomography for Multidirectional Process Monitoring in a Coaxial LMD-w Process[J]. Applied Sciences, 2022, 12(5).
[7] Ye J, Bab-Hadiashar A, Hoseinnezhad R, et al. Predictions of in-situ melt pool geometric signatures via machine learning techniques for laser metal deposition[J]. International Journal of Computer Integrated Manufacturing, 2023, 36(9): 1345-1361.
[8] Hsu H W, Lo Y L, Lee M H. Vision-based inspection system for cladding height measurement in Direct Energy Deposition (DED)[J]. Additive Manufacturing, 2019, 27: 372-378.
[9] Da Silva A, Frostevarg J, Kaplan A F H. Melt pool monitoring and process optimisation of directed energy deposition via coaxial thermal imaging[J]. Journal of Manufacturing Processes, 2023, 107: 126-133.
[10] Costa A P de A, Choren R, Pereira D A de M, et al. Integrating multicriteria decision making and principal component analysis: a systematic literature review[J]. Cogent Engineering, 2024, 11(1).
[11] Khan T A, Tulsi J, Alam M, et al. Analysis and visualization of fraud detection patterns through data mining and classification using MLP and hybrid deep learning model[J]. Egyptian Informatics Journal, 2025, 32: 100829.
[12] Liu W, Gai M. PV-MLP: A lightweight patch-based multi-layer perceptron network with time-frequency domain fusion for accurate long-sequence photovoltaic power forecasting[J]. Renewable Energy, 2025, 251.
[13] Mysliwiec P, Kubit A, Szawara P. Optimization of 2024-T3 Aluminum Alloy Friction Stir Welding Using Random Forest, XGBoost, and MLP Machine Learning Techniques[J]. Materials, 2024, 17(7).

Sponsors, Associates, and Links

---

• Cybernetics and Mechatronics
  
  
• Digital Manufacturing and Process Management
  
  
• Ultra-Precision Machining Process
  
  
• Journal of Robotics and Biomimetics
  
  
• Prognostics, Diagnostics and Health Management
  
  
• Micro-Electro-Mechanical Systems
  
  
• Journal of Precision Instrument and Machinery
  
  
• Engineering and Solid Mechanics
  
  
• Fracture and Damage Mechanics
  
  
• Frontiers in Tribology
  
  
• Fluid and Power Machinery
  
  
• Chemical Process Equipment
  
  
• Journal of Assembly and Manufacturing
  
  
• Mechanical Vibration and Noise