ITTIA Edge AI Motor Health Monitoring Demo

This demonstration showcases how motor-related data, including current, vibration, temperature, and speed, is acquired, processed, and managed deterministically at the edge and transformed into AI-ready features for on-device inference. You will see how MCUs, sensors, Edge AI frameworks, and production-grade data management software work together to detect anomalies, predict failures, and deliver reliable motor health intelligence in real time.

Use Case: Predictive Maintenance for Industrial Motor Control

Challenge:

An industrial equipment manufacturer needs to:

• Continuously collect vibration, temperature, and current sensor data
• Store weeks of time-series history locally for analysis
• Run on-device anomaly detection
• Transmit only critical insights to a supervisory MPU, minimizing bandwidth

System Architecture

Hardware:

• MCU: STM32H573I-DK or B-U585I-IOT02A
• Sensors: MEMS accelerometer (pre-recorded), (can also support temperature sensor, current sensor)
• Comms: UART (can also support Ethernet)
• Storage: QSPI external Flash (can also support MicroSD)

Software Stack:

• OS: FreeRTOS (can also support Eclipse ThreadX RTOS, Zephyr)
• Database: ITTIA DB Lite
• AI Model: STM32Cube.AI anomaly detection model (can also support NanoEdge AI Studio, LiteRT, TensorFlow Lite Micro)
• Visualization: ITTIA Analitica dashboard via ITTIA Data Connect (optional)

Implementation

Schema Design

A time-series data stream and table capture sensor readings:

CREATE TABLE vibration_raw_samples (
    instance_id                INTEGER NOT NULL,
    sensor_id                  INTEGER NOT NULL,
    motor_id                   INTEGER NOT NULL,
    sample_time                TIMESTAMP NOT NULL,
    acceleration_mps2          FLOAT32 NOT NULL,
    weighted_acceleration_mps2 FLOAT32 NOT NULL,
    inverse_variance           FLOAT32 NOT NULL,
    PRIMARY KEY (instance_id, sensor_id)
);


CREATE TABLE motor_health_samples (
    instance_id         INTEGER NOT NULL,
    motor_id            INTEGER NOT NULL,
    sample_time         TIMESTAMP NOT NULL,
    acoustic_mic_V      FLOAT NOT NULL,
    acceleration1_mps2  FLOAT NOT NULL,
    acceleration2_mps2  FLOAT NOT NULL,
    acceleration3_mps2  FLOAT NOT NULL,
    temperature_C       FLOAT NOT NULL,
    PRIMARY KEY (instance_id, motor_id)
);

CREATE TABLE event_actions (
    instance_id  INTEGER NOT NULL,
    event_id     INTEGER NOT NULL,
    inference_id INTEGER NOT NULL,
    event_time   TIMESTAMP NOT NULL,
    reason_code  VARCHAR(32) NOT NULL, -- or event_type: "threshold crossed", "persistent anomaly", etc.
    severity     VARCHAR(32) NOT NULL,
    action_type  VARCHAR(32) NOT NULL,
    PRIMARY KEY (instance_id, event_id)
);

Data Ingestion

C code with ITTIA DB Lite API:

#include "motor_health_database.h" // generated from schema

int main() {
   db_init_ex(DB_API_VER, NULL);
   open_motor_health_database("motor_health", NULL);
   // ...
}

void log_sensor_data(uint64_t ts_ms, float mic, float vib[3], float temp) {
    db_t db;
    db_connect(&db, "motor_health", "ingestion", NULL, NULL);
    const motor_health_samples_row_t row = {
        .instance_id = 1,
        .motor_id = 1,
        .sample_time = {},
        .acoustic_mic_V = mic,
        .acceleration1_mps2 = vib[0],
        .acceleration2_mps2 = vib[1],
        .acceleration3_mps2 = vib[2],
        .temperature_C = temp,
    };
    put_motor_health_samples(db, row);
    db_disconnect(db);
} 

Local AI Inference 

Every second, an MCU callback function:

1. Receives the last N samples from an ITTIA DB Lite streaming window query
2. Normalizes and feeds them to an AI model
3. Flags anomalies and stores events in the anomaly events table.

Data Sharing

When an anomaly is detected:

• Summary (timestamp, type, score) sent over UART/Ethernet to PC (also supports i.MX MPU running Linux/QNX/VxWorks/etc.)
• MPU visualizes aggregated events in ITTIA Analitica dashboard
• Benefit: No raw sensor stream leaves MCU → bandwidth saved, privacy improved

Key Advantages

1. Deterministic performance, guaranteed latency for real-time loops
2. Security, STM32H5 TrustZone + ITTIA DB Lite secure storage
3. On-device intelligence—no cloud dependency for critical decisions
4. Interoperability—data can be synced to MPU or cloud when needed
5. Scalable—same schema/codebase runs on larger STM32H7 or MPU with ITTIA DB