Activa 6G BS-VI Engine Audio Dataset: Healthy and Defective Samples

Engine condition monitoring traditionally relies oninvasive sensor diagnostics and periodic servicing. This paperpresents a comprehensive, non-invasive acoustic-based approachusing multiple advanced signal processing techniques to detectengine faults through frequency-domain, time-frequency-domain,and perceptual-domain analysis of real-time engine sound signals.Experimental validation is conducted on the Honda Activa 6GBS-VI compliant 109.51 cc air-cooled single-cylinder four-strokeengine. Beyond classical Fast Fourier Transform (FFT), weemploy Continuous Wavelet Transform (CWT), Mel-FrequencyCepstral Coefficients (MFCC), and Short-Time Fourier Trans-form (STFT)-based spectrogram analysis. Experimental valida-tion on twenty-seven audio samples (twenty-one healthy, sixdefective) from actual Activa 6G BS-VI engines reveals: (1) FFT-based spectral centroid classification achieves 100% accuracy;(2) spectral centroid elevation of 70% indicates high-frequencynoise dominance; (3) spectral flatness increases 124% in defectiveengines; (4) harmonic-to-noise ratio degrades 4.2 dB. Spectro-gram analysis reveals temporal fault evolution. The ensembleclassifier achieves 100% accuracy enabling specific fault typeidentification. This cost-effective, non-invasive framework enablesreal-time predictive maintenance