Bridge TDMS Explorer

Prototype tools for inspecting Memorial Bridge-style TDMS sensor files.

What It Does

• Finds .tdms files in a folder.
• Ignores _version_ copies and Decimate files for analysis.
• Detects the available time range from filename timestamps.
• Parses active float64 TDMS channels, including repeated raw chunks.
• Reconstructs wall-clock timestamps from the filename plus the TDMS Time channel.
• Combines all normal files selected by the dashboard time range.
• Builds a sensor catalog with active and configured-but-inactive channels.
• Computes raw channel summaries, 1-minute trend features, event candidates, and sensor health flags.
• Discovers correlated channel groups from windowed sensor features.
• Classifies event candidates into traffic/vibration, boat collision / impact, and drawbridge-operation-like families.
• Reports behavior shifts only when three or more correlated channels agree.
• Decodes BDI sensor labels into approximate physical bridge locations and highlights supporting sensors for impact candidates.
• Provides a Streamlit dashboard for exploring files, signals, traffic-like events, trends, and health.

Run The Dashboard

Install the Python dependencies first:

python3 -m pip install -r requirements.txt

python3 -m streamlit run app.py

Use the same Python environment for both commands. For example, if Streamlit is installed in Anaconda, run /opt/anaconda3/bin/python -m pip install -r requirements.txt and /opt/anaconda3/bin/python -m streamlit run app.py.

On launch, use the sidebar folder browser or enter the local folder that contains your TDMS files. You can select either a folder with TDMS files directly inside it or a parent folder with daily TDMS subfolders.

On startup the app scans for new or changed TDMS files and ingests them into a local DuckDB/Parquet cache under cache/. The sidebar then shows the available cached time range. Use the Analysis Time Range start/end date and time inputs to choose the interval to query. When more than a week of data is available, the default launch range is the latest week.

TDMS timestamps are stored and queried as UTC in the cache. The dashboard range controls, chart axes, event timelines, and added *_eastern table/download columns show the same moments in Eastern bridge local time, including daylight saving time changes.

TDMS discovery is recursive, so a folder organized like tdms_files/ with subfolders such as tdms_files_2026-06-17_07-00-01/ and tdms_files_2026-06-18_07-00-01/ can be loaded from the parent folder.

Startup progress distinguishes scanning files, ingesting new files, updating the index, and ready state. Once ingestion is complete, the app queries only the selected time range and requested channels from the local cache instead of combining every selected TDMS file in memory.

Use the sidebar View selector to switch between pages. Only the selected view is computed, which keeps display-setting changes responsive on large TDMS selections.

Complete rosette strain-gage groups are detected from active SG-...-H, SG-...-V, and SG-...-D channels that share the same location, side, and rosette designation. The Files view lists these groups, and the raw signal, event detection, and trend views can focus on rosette groups instead of requiring the three component channels to be selected one by one. The anomaly model also treats complete rosettes as domain-confirmed groups, so all three orientations can confirm a rosette-specific shift without relying only on learned correlation. VGP_... and VGT_... channels are decoded as TTDS vertical guide post sensors and are not used as rosette-group evidence.

Anomaly Detection

The app includes an explainable first-pass anomaly workflow:

• Correlation Groups: finds channels that move together over the selected time range.
• Event Detection: detects rolling-RMS bursts and classifies traffic/vibration, impact candidates, and operation-like events.
• Anomaly Review: separates urgent boat collision / impact candidates from reportable operation or behavior shifts.

Behavior shifts are gated by group support. By default, a reported shift requires at least three channels in the same rosette or correlated group to show compatible abnormal movement. This helps avoid false alarms from a noisy or failed single sensor.

Sensor placement maps are decoded from the BDI installation-plan naming scheme. Locations 1-8 are on the south fixed span; locations 9-10 are on the south tower. Impact-candidate review highlights the sensors supporting those candidate events.

Run A CLI Summary

python3 analyze_tdms.py --cache

Build or update the scalable app cache:

python3 analyze_tdms.py --ingest

The legacy --cache command writes per-file Parquet files under cache/:

• raw samples
• sensor catalog
• channel summary
• 1-minute trend features
• sensor health flags

The scalable --ingest command writes:

• cache/bridge_index.duckdb
• partitioned raw samples under cache/samples/
• partitioned trend features under cache/features/
• per-file metadata under cache/metadata/

Notes

The parser is intentionally scoped to the TDMS layout in these bridge-monitoring files:

• channel data stored as float64
• group name AllNormal
• repeated raw data chunks
• normal files named like *_Normal_Data_AllNormal.tdms

If later files contain strings, timestamps, or other numeric TDMS raw types as active sample channels, the parser should be extended before interpreting those channels.