Rolling bearings are essential components in rotating machinery. Harsh operating conditions frequently lead to the structural failure of bearings caused by mechanical vibration. While advanced vibration sensors capture high-fidelity data, the resulting signals are obscured by significant background noise, hindering the extraction of incipient fault features. To address the challenges of early-stage fault detection in industrial environments, a smart sensing framework that integrates an adaptive signal-processing layer directly into advanced vibration sensors was developed. This framework utilizes a kurtosis-guided adaptive wavelet thresholding method to transform raw, noisy sensor data into high-fidelity diagnostic information at the sensing node. By embedding intelligence and advanced sensors, the limitations caused by background noise can be addressed while preserving critical fault-related transients, facilitating autonomous edge-based diagnostics. By integrating the signal kurtosis and the decomposition level into the thresholding function, the algorithm adaptively suppresses noise while preserving the integrity of fault-related transients. To validate the approach, a simulation model was established with an inner ring fault frequency of 100 Hz and a carrier frequency of 2000 Hz, augmented with real-world noise extracted from the Case Western Reserve University data. The algorithm isolates the fault frequency and its corresponding harmonics at 200 and 300 Hz, whereas variational mode decomposition and standard wavelet thresholding do not accurately identify these harmonic markers in high-noise environments. This algorithm and the research results provide an efficient signal-processing framework essential for the development of smart sensors used for edge-based, early-stage fault diagnostics.