EMBEDDED C DSP CORE • FREESTANDING ANSI C99 • DETERMINISTIC 312B SRAM
ARM CORTEX-M0+/M4/M7/M33 • ESP32-S3 • HOST SITL
FREESTANDING ANSI C99 • SOFTWARE ALERT LATCH • REPLAY RESULTS AVAILABLE

Embedded Vibration Monitoring for Industrial Motors & Pumps

Compute vibration-derived features and compare them with a startup baseline in embedded C. The library uses a 312-byte context, caller-provided sensor batches, and no network service.

01 / ARCHITECTURE

Embedded Software Library

Runs in host firmware that supplies tri-axial sample batches. Sensor drivers, acquisition timing, and any external hardware are outside the library.

02 / LOCAL REACTION

Software Alert Latch

Sets a software alert after six consecutive threshold-exceeding evaluations. At 500 Hz with one 32-sample batch every 64 ms, six calls span 384 ms from the first qualifying call to the sixth. The host decides any response.

03 / ISOLATION

100% Air-Gapped & Offline

Deterministic fixed-point arithmetic with zero Wi-Fi/cellular reliance, zero IT firewall disputes, and zero recurring cloud subscriptions.

SYSTEM SIGNAL CHAIN ARCHITECTURE SENSOR INPUT TO HOST RESPONSE
HOST-PROVIDED INTEGER SAMPLES • 0 DYNAMIC HEAP BUFFERS
STAGE 01 INGESTION
Transducer & SPI FIFO
The caller supplies exactly 32 signed samples for each axis. Sensor drivers, range conversion, and sample timing are implemented by the host.
SPI / I2C FIFO Zero-Copy Buffer
STAGE 02 GATING
Startup Baseline Heuristics
Baseline collection includes a 20-window stabilization period and a high-energy limiter. The library receives no RPM signal and does not identify transient sources.
Fixed Input Floor Baseline Heuristics
STAGE 03 DSP CORE
Fixed-Point Harmonics
Computes demeaned 3-axis RMS, crest factor, and rolling Goertzel metrics, then compares them with the startup baseline.
312B SRAM ~24k Cyc / ~4 µs Host
STAGE 04 PROTECTION
Software Alert & Host Integration
A software alert latches after six consecutive threshold-exceeding windows. Host firmware chooses any Modbus, alarm, or shutdown response. The 0–100 condition index is not a percentage of wear or remaining life.
Software Alert Latch Modbus RTU/TCP
RAM Footprint
312Bytes
0 BSS context / 312 B instance struct
Execution Latency
4.17µs
4.17 µs host avg / ~100 µs est. @ 240 MHz Cortex-M (~24k cycles)
Flash Code Size
8.4KB+
8.4 KB host / 16.2 KB Cortex-M4 (-O3; 6.5 KB -Os)
Dynamic Memory
0malloc
Zero heap / freestanding ANSI C99

Electrical, Memory & Real-Time Execution Profile

Resource and execution parameters across target toolchains: Flash & static stack measured with LLVM/Clang; execution timing measured on arm64 host (~4.17 µs avg) and estimated for bare-metal ARM Cortex-M (~24k–26k cycles).

Parameter Description Symbol Min Typ Max Unit Test Conditions & Verification Method
SRAM Allocation Footprint RAM_inst — 312 312 Bytes Caller-allocated context struct. 0 static BSS (ac_engine.o).
Dynamic Heap Allocation RAM_dyn 0 0 0 Bytes Zero dynamic heap (malloc = 0; llvm-nm -u = 0 undefined symbols).
Flash ROM Footprint (Thumb-2) ROM_flash 8.4 16.2 16.6 KB Static archive .text (-O3): 16.2 KB M4/M33; 16.6 KB M0+/M7 (6.5 / 6.8 KB -Os).
Estimated Cycle Latency (Cortex-M) t_est_M — ~100.0 < 240.0 µs Cycle-budget estimate (~24k cycles: ~100 µs @ 240 MHz; ~200 µs @ 120 MHz).
Core CPU Load (@ 64 ms cadence) LOAD_cpu 0.006 — < 0.35 % 0.0062% measured on arm64 host; < 0.35% estimated on 168 MHz Cortex-M4.
Defect Confirmation Trip Latency t_trip 384 544 1,088 ms Six consecutive threshold-exceeding windows are required before this software alert latch is set.
Real-Data Benchmark Coverage FAR 0.00 — — % 0.00% CritFAR on OpenBLDC M10/M30 & UORED healthy; 193 .acbin + dyno streams.
📄 Official AC Engine 5.0 Technical Dossier 2-PAGE SPEC
Memory & static stack breakdown tables across GCC/Clang targets, C99 header integration specs, and commercial OEM evaluation options.
Request 2-Page Technical Dossier →

Results Reported by the Included Replay Harnesses

The figures below are repository-reported outputs from local dataset replays and are not field accuracy, safety qualification, or guaranteed product performance.

View Full Laboratory Benchmark Data →
OPENBLDC ADXL355 FLEET (REAL 1 kHz / 500 Hz)
3 Motors × Front/Rear Tri-Axial

Real MEMS FIFO telemetry on Motor 10, Motor 20, and Motor 30 nominal + notched bearing runs.

0.00% CritFAR (M10F / M30F+R)
512–704 ms 1st crit @ 1 kHz; M20 drift & M10R noted
FEMTO-ST PRONOSTIA (8 REAL RUNS)
4,000–5,000 N Overload Campaign

25.6 kHz decimated to 500 Hz .acbin streams across Bearings 1_1–3_2 run-to-failure.

8/8 Progression Tracked
B1_1 holds 0 crit D0–D6 & trips D7 to 8% health
UORED-VAFCLS (60 BEARINGS, CC BY 4.0)
Healthy / Developing / Fault Triplets

42 kHz decimated to 500 Hz across 20 bearing groups (Inner, Outer, Cage, Ball).

20/20 Healthy 0-Crit Held-Out
16/20 Dev & 14/20 Fault flagged; 2/20 cal pre-defect
UNIVERSITY OF OTTAWA TVS (36 RUNS)
Variable-Speed VFD Sweeps

200 kHz decimated to 500 Hz under Profiles A–D acceleration and deceleration.

0 Crit Profiles A–C Healthy
Inner 3/3 (A–D), Outer 8/9; Profile D sweep stress
CWRU & MFPT (36 .ACBIN @ 500 Hz)
2 HP Motor & Stepped Radial Loads

CWRU 0.007" faults (0–3 HP) + MFPT (3 baselines, 7 variable-load IR, 10 OR).

12/12 CWRU Faults; 5/7 MFPT IR
1–3 HP 0% CritFAR; 0 HP drift & 500 Hz OR limit
NASA IMS (8 CH) & SEU GEARBOX (45 CH)
Run-to-Failure & Parallel/Planetary

500 Hz decimated streams; Paderborn, 4TU & PHM09 explicitly excluded (size/license).

IMS 1st B3/B4 & SEU Chipped/Surf
EndH 8–15%; SEU 20_0 parallel H 92% EndH

Embedded Library and External Monitoring Systems

These architectures make different tradeoffs. Hardware, installation, connectivity, operating cost, and performance depend on the selected system and deployment; this repository does not establish a vendor comparison.

Architecture Attribute AC Engine (Embedded C Core) External Wireless Pucks (SKF / Banner / ifm)
Unit OEM BOM Cost Depends on host MCU, sensor, and integration costs; not estimated here. Varies by product, installation, and service plan.
Physical Integration Requires host software integration and a configured vibration sensor. Depends on the selected monitoring system.
Power Source & Lifespan Depends on the host equipment and sensor interface. Depends on the selected monitoring system.
Defect Trip & Protection Latency Six consecutive threshold-exceeding evaluation calls; host controls any response. Depends on product configuration and connectivity.
Cloud & Gateway Dependency The library itself has no networking code. Host firmware may add network services. Depends on the selected monitoring system.
Recurring Cloud & SaaS Costs Commercial terms are separate from this software repository. Depends on the vendor and service plan.
OEM Value Capture Depends on the host product and license. Depends on the selected monitoring system.

Confidential 4-Step Dyno Evaluation

Engineered for VP of Engineering and firmware teams to evaluate our pre-compiled static library on your test dynos with zero integration risk.

STEP 01

Mutual NDA Execution

Standard 2-page mutual agreement protecting your proprietary kinematics and our compiled static library.

STEP 02

Pre-Compiled SDK Delivery

We deliver AC_Engine_v5.0_Commercial_SDK.zip containing stripped .a binaries for your exact MCU architecture.

STEP 03

Dyno Bench Validation (60 Days)

Integrate the archive with a supported host toolchain, feed configured sensor batches, and compare results with your own test data. Integration time depends on the host system.

STEP 04

Production Design Win

Upon bench qualification, transition smoothly to tiered volume licensing or enterprise site agreement. Zero recurring cloud fees.

Start Commercial Evaluation →

Built by Systems Engineers

Bridging hands-on mechanical machinery experience with aerospace-grade software determinism.

Albert Cornelius - Founder & Embedded Architect

Albert Cornelius

FOUNDER & CTO
Embry-Riddle Aeronautical University | Embedded Systems & Flight Software Reliability

Specializing in deterministic edge DSP and autonomous vibration diagnostics for industrial rotating machinery. Developed AC Engine 5.0 to solve the false-alarm and memory bottleneck in OEM motor drives by bridging practical mechanical shop floor experience (heavy construction machinery and diesel wheel ends) with rigorous flight software discipline (NASA F Prime, freestanding ANSI C99).