Spectral Features Analysis for Print Quality Prediction in Additive Manufacturing: An Acoustics-Based Approach

Area of Science:

• Materials Science and Engineering
• Manufacturing Technology
• Acoustics and Signal Processing

Background:

• Quality prediction in additive manufacturing (AM) is critical for high-risk sectors like aerospace and biomedicals.
• Acoustic sensing offers a non-destructive method for detecting variations in 3D printing processes.
• Fused Deposition Modeling (FDM) is a widely used AM technique where quality control is essential.

Purpose of the Study:

• To investigate the efficacy of acoustic data analysis for predicting 3D print quality in FDM.
• To extract and analyze time and frequency-domain acoustic features for quality classification.
• To implement and compare various machine learning algorithms for acoustic-based print quality prediction.

Main Methods:

• Collected acoustic data streams from FDM 3D-printed samples with varying layer thicknesses.
• Preprocessed audio samples using Harmonic-Percussive Source Separation (HPSS) and extracted spectral features using Librosa.
• Employed eight machine learning classifiers (including XGBoost, Random Forest, SVM) for print quality prediction based on acoustic features.

Main Results:

• Identified spectral flatness, spectral centroid, power spectral density, and RMS energy as key acoustic features.
• The Extreme Gradient Boosting (XGBoost) model achieved the highest prediction accuracy (91.3%), precision (88.8%), recall (92.9%), F1-score (90.8%), and AUC (96.3%).
• Demonstrated the potential of acoustic signatures for differentiating between print quality levels.

Conclusions:

• Acoustic-based analysis combined with machine learning provides a robust method for real-time quality prediction in FDM 3D printing.
• This approach lays the groundwork for acoustic-based quality control systems in additive manufacturing.
• The methodology can be extended to other AM techniques and diverse manufacturing quality assessment applications.