Аннотация
The transition toward Industry 4.0 and predictive maintenance architectures demands scalable, highly accurate, and robust sensor networks. Traditional vibration monitoring relies heavily on contact-based piezoelectric accelerometers or fragile, prohibitively expensive Laser Doppler Velocimeters (LDVs). While acoustic sensing via MEMS microphones offers a non-contact alternative, its efficacy is severely degraded by the broadband ambient noise typical of industrial environments. This paper presents the mathematical modeling, hardware design, and empirical validation of a highly cost-effective Self-Mixing Interferometry (SMI) sensor explicitly optimized for edge computing topologies within Industrial Internet of Things (IIoT) deployments. Utilizing a commercial ADL-65075TL (650 nm) laser diode and a custom-engineered Analog Front-End (AFE) built around the high-performance OPA2134 operational amplifier, the system extracts nanometer-scale microvibration signatures strictly through optical feedback. Overcoming severe analog constraints, the typically dual-supply AFE was engineered to operate on a strict single-supply 5 V logic architecture. By synthesizing a precision 2.5 V virtual ground, the system successfully isolates microampere-level interferometric fringes while actively preventing signal clipping during negative phase excursions. This conditioned analog signal is deterministically digitized at 3.7 kHz via the 12-bit Analog-to-Digital Converter (ADC) of an ESP32 edge microcontroller. The architecture was empirically validated in a real-time predictive maintenance scenario involving the structural resonance analysis of metal containers on a motorized conveyor belt. The localized Fast Fourier Transform (FFT) algorithm successfully resolved target resonance frequencies in under 18 ms, dynamically determining internal fluid levels without requiring cloud offloading. This research demonstrates that laboratory-grade interferometric sensing can be seamlessly deployed at the network edge, drastically reducing computational latency and financial overhead for structural health monitoring.