Machine Health Indicator (STM32L476)

This project implements a Machine Health Indicator on an STM32L476RG Nucleo board. It monitors machine health by analyzing acoustic patterns from a microphone using Digital Signal Processing (DSP) techniques.

System Architecture

The application follows a periodic monitoring cycle:

1. Data Acquisition: Samples are collected from a microphone.
2. Signal Processing: Time-domain data is converted to the frequency domain.
3. Anomaly Detection: The frequency spectrum is analyzed for abnormal patterns.
4. Reporting: Results are sent via UART and indicated via on-board LEDs.
5. Power Management: The system enters a low-power sleep mode between checks.

Key Components

1. Data Acquisition (DFSDM + DMA)

The system uses the Digital Filter for Sigma Delta Modulators (DFSDM) peripheral to interface with a digital microphone (acting as a vibration sensor).

• DFSDM Configuration: Configured with a Sinc4 filter and an oversampling ratio to achieve a sampling frequency ($f_s$) of 16.45 kHz.
• DMA: Direct Memory Access is used to transfer samples from the DFSDM filter to a buffer in RAM without CPU intervention.
• Synchronization: The application polls a volatile bool flag (mic_dma_finished) which is set in the HAL_DFSDM_FilterRegConvCpltCallback when the buffer is full.

2. Signal Processing

Once a buffer of 2048 samples is collected (representing ~125 ms of data), the system performs frequency analysis.

• Top-level Entry: detect_anomaly acts as the main entry point.
• Algorithm Comparison: The system toggles between two FFT implementations for performance benchmarking:
  • decompose_cmsis: Uses the optimized ARM CMSIS-DSP RFFT library.
  • decompose_kiss: Uses the KISS FFT library.
• Magnitude Calculation: The complex FFT output is converted to a magnitude spectrum.
• DC Removal: The 0Hz component is cleared to remove any DC offset.

3. Anomaly Detection

The detect_anomaly function in Core/Src/machine_health.c implements a Band Energy Ratio strategy:

• It calculates the energy in a Normal Band (0-500 Hz) and a Problem Band (1-8 kHz).
• An anomaly is flagged if the ratio of Problem Band energy to Normal Band energy exceeds a threshold (0.5).
• It also tracks and prints the peak frequency and maximum magnitude.

4. User Interface & Feedback

• UART Console: Detailed statistics including FFT cycles, peak frequency, and power metrics are printed to UART2 (115200 baud).
• LED Matrix: An 8x8 LED matrix (interfaced via SPI2) provides visual feedback:
  • Normal: Brief green flash.
  • Anomaly: Persistent red indication.

5. Power Management

To maximize energy efficiency, the system uses a dedicated power management module (Core/Src/power_manager.c):

• Uninterrupted Sleep: Instead of waking up every 1ms via SysTick, the system suspends the SysTick interrupt (HAL_SuspendTick()) before entering sleep.
• RTC Wakeup: The Real-Time Clock (RTC) Wakeup Timer is used to fire an interrupt after the desired sleep duration (5 seconds).
• Power Metrics: The system estimates energy consumption and average power based on CPU cycles spent in active vs. sleep modes.
• Wakeup Source: The CPU enters Sleep Mode via HAL_PWR_EnterSLEEPMode and remains in a low-power state until the RTC Wakeup interrupt occurs.

Build and Run

Prerequisites

• ARM GCC Toolchain (arm-none-eabi-gcc)
• CMake and Ninja
• STM32L476RG Nucleo Board

Build Commands

Using CMake presets:

cmake --preset Debug
cmake --build build/Debug

Flashing

Flash the resulting .elf or .bin file using OpenOCD, STM32CubeProgrammer, or by dragging it onto the Nucleo's USB drive.