Hybrid Method for Quality Diagnosis in Rotary Friction Welding Using Vibration Signals

• Published in: Soldagem & Inspeção Vol. 30
• Main Authors: Alcantara, Daniel Soares de, Balestrassi, Pedro Paulo, Castro, Carlos Alberto Carvalho, Silva, Eduardo Pereira da
• Format: Journal Article
• Language: English
• Published: Associação Brasileira de Soldagem 2025
• Subjects: METALLURGY & METALLURGICAL ENGINEERING; Signal processing; Process characteristics; Welding monitoring; Spectral analysis
• ISSN: 0104-9224; 1980-6973; 1980-6973
• DOI: 10.1590/0104-9224/si30.03

Abstract The identification of relevant information in vibration signals has been a subject of study for decades, leading to significant advancements. This field is increasingly recognized as essential for diagnosis, prognosis, and failure prediction, ensuring safety, estimating quality conditions, and monitoring processes across various domains. Researchers have developed numerous approaches to extract and classify features within vibration signals, aiming to identify patterns and useful process information. Many of these approaches have proven effective with vibration signals from various manufacturing processes, including welding, and mechanical, civil, and electrical engineering systems. However, few studies have focused on vibration signals in the Friction Welding process, specifically Rotary Friction Welding (RFW). During RFW, simultaneous application of rotation and compressive forces joins two surfaces, generating complex vibration waves that carry unique information about deformations between the parts' surfaces. The main challenge in diagnosing vibrations in RFW is that the vibration signal is often contaminated with noise due to uneven heating, causing fluctuations in roughness, friction coefficient, and variable ripples over time. This study aims to develop a hybrid signal analysis methodology to investigate the potential for quality diagnosis using vibration signals from the continuous drive friction welding (CDFW) process. The proposed method combines the Dickey-Fuller test (DF) for stationarity identification, Empirical Mode Decomposition (EMD) for frequency recognition, Principal Component Analysis (PCA) for compression, visualization, and data classification, and Short-Time Fourier Transform (STFT) for graphical frequency representation over time. Analyses confirmed that vibration signals are significantly affected by corresponding changes in process evolution. It was demonstrated that vibration signals can characterize CDFW quality. Experimental results also showed that applying EMD, PCA, and STFT to vibration signals successfully identified relevant characteristics of process evolution in the time-frequency domain.