CORE REV: 5.0.0-PROD • MISRA-C/C++ ALIGNED • ZERO FPU REQUIRED
ARM CORTEX-M0+/M4/M7/M33 • ESP32 • HOST EVALUATION
PHYSICAL DYNAMOMETER LABORATORY BENCHMARK

Measured Vibration Analysis on Damaged Bearings

Physical sensor measurements from Analog Devices ADXL355 tri-axial accelerometers mounted on high-speed industrial dynamometer motors (OpenBLDC dataset).

TIME & FREQUENCY DOMAIN CORRELATION

Physical Signal Analysis on Held-Out Motor 30

Raw 20-bit dynamic acceleration waveform correlated against narrowband harmonic resonance extraction and edge health decay latching.

Physical Motor Signal Analysis Plot
PANEL 01 • TOP

Raw Dynamic Acceleration (mg)

Captures 3-axis accelerometer counts under normal running torque followed by real mechanical impact shocks when sub-surface ball/raceway spalling initiates.

PANEL 02 • BOTTOM LEFT

Harmonic Resonance Spectrum (0 - 250 Hz)

Dual-spectrum overlay showing healthy motor baseline (green dashed) vs. bearing defect spall peaks at 1X shaft speed (8.3 Hz), BPFO (94 Hz), and 2X BPFO (188 Hz).

PANEL 03 • BOTTOM RIGHT

Autonomous Health Index & ISO 10816

Continuous edge health rating mapped against ISO 10816 Zones A, B, C, and D. Latches decisive safety trip in 1.28 seconds (20 windows) upon defect inception.

EMPIRICAL FEATURE SHIFT METRICS
Statistical Vibration Parameter Healthy Motor 30 Baseline Damaged Bearing (Spalled) Metric Shift Δ Alarm Threshold Status
Acceleration RMS 18.4 mg (0.0184 g) 29.0 mg (0.0290 g) +57.6% Shift Decisive energy increase Breached Dynamic Limit
Peak-to-Peak Transient Shock 39.2 mg 111.0 mg +183.2% Shift Raceway micro-impacts Breached Absolute Floor
Vibration Kurtosis (Spikiness) 2.65 (Gaussian Distribution) 4.92 (Heavy-Tailed Impacts) +85.6% Shift Confirms metal spalling Critical Inception Flagged
Crest Factor (Peak / RMS) 2.13 (Normal Operational Hum) 3.83 (Severe Shock Impulses) +79.8% Shift Raceway pit impact Incipient Warning Tripped
BPFO Kinematic Harmonic Power 0.42 mg² (Floor Baseline) 4.88 mg² (Harmonic Spike) +1,061% Surge Outer race resonance Resonant Filter Locked
Autonomous Health Index 100.0% (Zone A • Good) 11.3% (Zone D • Trip Shutdown) Latched in 1.28s (20 windows) Hardware Relay Fired
MULTI-DAY CONTINUOUS STABILITY QUALIFICATION

Zero False Alarms Across 320 Million Sensor Samples

Industrial equipment manufacturers cannot tolerate nuisance false alarms or watchdog timer resets. AC Engine was subjected to multi-day continuous endurance qualification (10,000,000 windows / 7.41 days).

Multi-Day Endurance Stability Plot
Endurance Campaign Sample Volume False Alarm Trips Numerical Stability Metric
48-Hour Continuous Stress Stream 86,400,000 Samples (2.7M Windows) 0 Trips 0.0000000% FAR Mean Health = 85.01% (Zero Drift)
7.41-Day Continuous Multi-Day Run 320,000,000 Samples (10M Windows) 0 Trips 0.0000000% FAR Mean Health = 85.00% • StdDev = 0.08%
Real Bearing Spall Detection Latency Held-Out Motor 30 Damage Run Latched in 1.28s (20 windows) 93.7% detection rate across fault run

Zero-Buffer Streaming Architecture: Constant 375 Bytes RAM

Conventional cloud vibration algorithms attempt to buffer 10 to 60 seconds of raw high-speed waveform data in memory, requiring expensive external SDRAM and continuous batch transfers. AC Engine operates as a true real-time streaming pipeline designed specifically for resource-constrained microcontrollers:

1. Fixed O(1) Memory Footprint

Processes each incoming 32-sample sensor FIFO chunk in real-time. The entire state machine resides in a static 375-byte allocation with zero heap fragmentation (malloc = 0).

2. Fixed-Point Register Pipeline

RMS, Crest Factor, Kurtosis, and Kinematic Harmonics are evaluated in deterministic 32-bit registers with saturation clamping, preventing arithmetic overflow without requiring an FPU.

3. Numerical Drift Immunity

Recursive baseline tracking eliminates historical memory queues while maintaining flat baseline stability (85.00% ± 0.08%) across 320 million samples without drift or stack growth.

Request Evaluation Package for Your Dyno →