Nirmano Pulse

The Problem

Enterprise vibration monitoring systems cost thousands per sensor point. For a 50-machine shop floor, that is a six-figure deployment before you see a single data point. SMB manufacturers — the companies that would benefit most from machine health visibility — cannot justify that spend.

The result: most small and mid-size factories run machines until something breaks. Maintenance is reactive. OEE is a number someone estimates in a spreadsheet, if they track it at all. The data that could prevent downtime, predict failures, and optimize throughput sits uncollected because the hardware to capture it was designed for enterprise budgets.

What We Built

Two hardware variants that make vibration monitoring economically viable at any scale.

OEE Puck: LIS2DH12 accelerometer sampling at 100Hz. 12-month battery life on a CR2032. Detects machine state — running, idle, off, anomaly — and counts cycles. Enough to calculate OEE accurately on every machine in a facility. Full shop floor coverage costs less than a single enterprise sensor.

PdM Puck: IIS3DWB accelerometer sampling at 26.7kHz. Full vibration spectrum analysis. Detects bearing faults, imbalance, looseness, and misalignment. Predictive maintenance capability at a fraction of enterprise pricing.

Firmware: Rust on nRF52840 using the Embassy async runtime for power-efficient operation. On-device DSP — FFT, RMS calculation, z-score anomaly detection — runs locally. No cloud dependency for basic state detection. BLE 5.0 for gateway connectivity, designed for mesh deployment across factory floors.

The core DSP and state machine are working. Full product — gateway, dashboard, fleet management — is in active development. Pulse is currently in early access.

Architecture & Approach

Pulse started as a cost engineering problem. We worked backward from the question: what is the minimum viable hardware that delivers real operational value to a 50-machine shop?

1. **Design:** Defined two distinct use cases (OEE tracking vs. predictive maintenance) and designed separate hardware for each rather than one overbuilt device.

2. **Build:** Chose Rust for firmware — memory safety without garbage collection overhead matters when you are running DSP on a microcontroller with limited resources.

3. **Test:** Validated state machine accuracy against manual machine observations. Tuned anomaly detection thresholds on real industrial vibration data.

4. **Deploy:** BLE mesh architecture means adding sensors is physical installation only — no network infrastructure changes.

5. **Retain:** Device firmware updates over BLE. Detection models improve as fleet data grows.

What This Proves

Building AI-enabled systems for manufacturing requires understanding the full stack — from bare metal firmware to cloud analytics. Pulse is not a software company's attempt at hardware. It is purpose-built by someone who understands signal processing, embedded constraints, factory floor conditions, and the economic reality of SMB manufacturing.

The OEE Puck exists because we asked the right cost engineering question, not because we stripped features from an enterprise product.

When we say we understand production systems, this is what we mean.