STM32-based Real-time MCSA System

Variable-speed induction motors in industrial deployments require continuous, non-invasive condition monitoring to detect stator inter-turn insulation degradation, broken rotor bars, and mechanical bearing faults before catastrophic failures occur. The prior research prototype relied on a multi-chip legacy architecture combining an ATmega328P for analog sampling, an ESP32 for network bridge transmission, and a Raspberry Pi for processing. This multi-chip arrangement suffered from synchronization jitter, inter-chip communication bottlenecks, potential telemetry packet dropouts, and elevated component failure points in high-EMI industrial environments.

Embedded Systems R&D Intern under Prof. Aurobindo Routray at IIT Kharagpur. Designed the unified ARM STM32 embedded architecture, configured real-time signal conditioning firmware, engineered high-speed telemetry framing, and validated physical PCB hardware interfaces under noisy inverter-driven motor testing conditions.

The system consolidates sampling, conditioning, and telemetry into a single high-performance ARM Cortex-M STM32 microcontroller. Phase currents are acquired through non-invasive current transducers, conditioned via an internal operational amplifier configured as a Programmable Gain Amplifier (PGA), digitized by timer-triggered ADC modules synchronized with DMA circular buffers, and transmitted via an interrupt-driven binary framing protocol over high-speed UART operating at 921,600 baud.

Implemented software-defined Automatic Gain Control (AGC) utilizing the STM32's built-in OPAMP PGA to maintain optimal dynamic signal resolution across variable motor load regimes without clipping or quantization loss. Configured hardware timer-triggered ADC conversions with zero CPU polling overhead, utilizing Direct Memory Access (DMA) to stream continuous phase-current time-series data into memory. Formulated a packed binary telemetry protocol featuring checksum verification and header synchronization operating at 921,600 baud for real-time spectral processing on host workstations.

Successfully consolidated the 3-chip legacy architecture into a robust, single-chip STM32 embedded platform. Demonstrated zero sample dropouts during continuous multi-hour acquisition runs, verified clean spectral resolution for motor fault sideband identification, and fulfilled the telemetry performance milestones for the GAIL & CIL sponsored project.