Monitoring and Event Workflows#
Goal-oriented recipes for analyzing stored slow monitoring records, correlating transient event catalogs with multi-channel detector time series, and processing long continuous observations with bounded memory.
Note
These tutorials cover offline and post-run diagnostic workflows on recorded data files. They are designed for reproducible batch analysis and do not implement an online, 24/7 real-time telemetry daemon.
Analysis Pathways#
Separate slow diurnal/linear baseline drift from short pulse transients and persistent state steps.
Refine discrete catalog event times against witness sensors and align multi-channel time windows.
Stream long time-series records in bounded blocks, maintaining global PSD continuity and atomic checkpoints.
Workflow Comparison#
Workflow |
Primary Input |
Key Output |
Required Dependencies |
Missing Data / Boundary Handling |
|---|---|---|---|---|
Trend & State Analysis |
Multi-channel 1-minute slow records |
Estimated baseline, pulse & step candidate tables |
|
Valid contiguous runs, edge exclusion flags |
Catalog & Waveform Alignment |
Event timestamp catalog + time series |
Refined event times, channel-aligned arrays |
|
Half-open intervals, partial / gap / no_peak status |
Chunked Streaming |
Multi-channel file on disk (HDF5) |
Streaming PSD, time-binned RMS, checkpoints |
|
Bounded memory buffer, resume consistency |
Recommended Progression#
T1 to T2 Sequential Pathway: In diagnostic investigations, start with Long-term Trends (T1) to isolate anomalous intervals, transient bursts, and operational state steps across slow monitor channels. The resulting candidate event timestamps feed directly into Event Alignment (T2), where discrete triggers are refined against high-bandwidth witness channels and converted to structured
SegmentTableobjects.Independent High-Throughput Pathway (T5): When analyzing multi-hour or multi-day datasets that exceed available workstation RAM, use Chunked Processing (T5) independently. It provides bounded-memory sequential streaming, stateful digital filtering (
sosfilt(..., zi=...)), cumulative PSD estimation, and atomic checkpointing without loading the full recording into memory.