Automatic Classification of Laser Peening Quality Using Acoustic Signals

Laser Shock Peening increases the fatigue life of metallic components by introducing beneficial compressive residual stresses. To achieve the desired effect, each individual laser pulse must be delivered correctly. Laser Shock Peening quality is typically verified by destructive and time-consuming residual stress measurements or by subjective operator judgement, which is non-objective and unsuitable for continuous in-line control. We propose a simple, low-cost and robust method based on the analysis of the acoustic response that automatically classifies individual laser pulses as defect-free or defective. We show that the acoustic response captured by a low-cost microphone carries sufficiently informative signatures to reliably distinguish correct from incorrect impacts and enables quality control at the level of single pulses. The method provides a non-destructive and objective route to real-time monitoring of Laser Shock Peening, with the potential to increase process reliability and support industrial deployment of this technology without the need for subsequent destructive measurements.

💡 Research Summary

The paper presents a low‑cost, real‑time quality‑monitoring solution for Laser Shock Peening (LSP) based solely on acoustic emissions captured with a standard USB condenser microphone. Experiments were conducted on AA2024 aluminium plates (100 × 100 × 5 mm) using an Nd:YAG laser (1064 nm, 1 Hz, 1.5 J). Two clearly distinct process conditions were created: “PROCESS OK” with a water layer on the surface (99 pulses) and “PROCESS NOT OK” without water (99 pulses). The acoustic signal was sampled at 44 kHz, amplitude‑based detection (mean + 3σ) was used to segment individual pulses, and a 0.4 s window (‑0.1 s to +0.3 s around the peak) was extracted for each event.

Signal preprocessing involved conversion to absolute amplitude, a 5 ms moving‑average smoothing, and normalization. Feature extraction combined global descriptors (RMS amplitude, spectral centroid, fraction of energy above 6 kHz) with time‑resolved descriptors calculated over five post‑peak intervals (energy ratios, high‑frequency ratios) and a count of secondary envelope peaks between 300 ms and 1 s. In total, 15 features were generated, capturing overall pulse energy, sharpness, and cavitation‑related high‑frequency content.

A Random Forest classifier (300 trees, class‑balanced weighting) was trained on a 60 % training set, validated on 20 % of the data, and finally tested on the remaining 20 % unseen samples. The model achieved 100 % accuracy on both validation and test sets, indicating that the selected acoustic features fully separate the two process states under the experimental conditions. Feature‑importance analysis revealed that the global RMS (total acoustic energy) was the strongest discriminator—defective pulses (no water) exhibited higher RMS due to reduced damping. Conversely, a higher spectral centroid and a larger high‑frequency energy ratio in the early post‑peak interval were characteristic of OK pulses, reflecting the presence of cavitation‑induced high‑frequency components.

The authors discuss the practical implications: the method requires only a cheap microphone and straightforward signal‑processing pipelines, making it suitable for integration into existing LSP control systems as an alarm that flags abnormal pulses in real time. Limitations include the narrow experimental scope (single alloy, geometry, microphone position) and the binary classification based on a pronounced process difference (water vs. no water). Future work should explore broader material sets, finer process deviations (e.g., variations in laser energy, coating thickness), and robustness against varying acoustic environments.

In conclusion, the study demonstrates that acoustic emissions in the audible band contain sufficient information to automatically classify individual LSP pulses with perfect accuracy under the tested conditions. By eliminating the need for destructive residual‑stress measurements, this approach promises to improve process reliability, reduce inspection costs, and accelerate the industrial adoption of Laser Shock Peening.