The Signal Quality Engine is now open source

The Signal Quality Engine (SQE) is a Python library that processes raw SCADA telemetry through a fixed DSP pipeline — filters, drift detection, spike detection, frequency analysis, and a composite quality score — and it is now available under the Apache 2.0 license.

Why this exists#

RTU and PLC simulators generate noisy, sometimes pathological data. Downstream logic — alarm evaluation, historian compression, operator displays — needs to know how much to trust each value. Writing one-off threshold checks per signal doesn't scale, and tuning them by hand across hundreds of points is brittle.

SQE replaces that with a deterministic pipeline that scores every signal on every scan against five dimensions: noise, drift, spikes, oscillation energy, and missing samples. The output is a single Signal Quality Index (0–100) per point, per scan, with component breakdowns and automatic incident lifecycle events.

What's in the box#

• Filter bank — EWMA, high-pass, and moving average, all running in parallel per signal
• Drift detector — sustained, transient, and monotonic drift with configurable thresholds
• Spike detector — sliding-window variance with adaptive σ thresholds
• Oscillation detector — correlation-based and FFT frequency analysis
• Innovation model — Kalman-inspired residual tracking for early degradation signals
• SQI composite — weighted across all detector outputs, with trend tracking and quality classification
• Incident engine — lifecycle events (started, updated, resolved) with hysteresis to avoid flapping
• CLI tool — analyze, simulate, replay, and evaluate from the command line
• Integration adapters — drop-in connectors for the PointCore simulator, PLC scan engine, and comms front-end

Determinism matters#

Every computation in SQE is deterministic. No background threads, no async, no random seeds that vary between runs. Given the same input scans and config, the output — processed signals, SQI values, incident IDs — is byte-for-byte identical across runs. This is verified by the replay harness and a diff-based determinism check in the test suite.

Scan timing is predictable: ~0.1–0.2ms per signal per scan on consumer hardware, with fixed memory allocation. For a 100-signal setup inside a 100ms PLC scan window, SQE uses roughly 10–20% of the budget.

Try it#

pip install -e .
pip install -e ".[all]"  # with plotting support

# Quick demo
sqe simulate --duration 500 --noise 1.5 --plot

# Run the test suite
pytest -q

The repo is at github.com/dimilafl/rtu. Contributions, issues, and questions are welcome.