h1

h2

h3

h4

h5
h6
http://join2-wiki.gsi.de/foswiki/pub/Main/Artwork/join2_logo100x88.png

Simulated Patient Monitoring and Ventilation Dataset for Sensor Failure Detection

; ;

Online
DOI: 10.18154/RWTH-2026-06634
URL: https://publications.rwth-aachen.de/record/1038619/files/Dataset.zip

Einrichtungen

  1. Institut und Lehrstuhl für Medizinische Informatik (922710)


Inhaltliche Beschreibung (Schlagwörter)
patient simulation (frei) ; clinical decision support systems (frei) ; medical device data (frei) ; physiological time series (frei) ; patient monitoring (frei) ; mechanical ventilation (frei)

Kurzfassung
This dataset comprises 160 JSON Lines (JSONL) recordings (~2.5 million timestamped events in total) of simulated multi-parameter patient monitoring and ventilation data, collected using a Gaumard HAL S5301 patient simulation phantom, a Dräger Oxylog 3000 Plus transport ventilator, and a Viatom PC60FW pulse oximeter. Data were captured for ten simulated patient profiles (p01–p10), each defined by a set of respiratory-mechanics targets, baseline vitals, adapted to a virtual diagnosis. For each patient, recordings were made under four ventilator configurations (two volume-controlled and two pressure-controlled settings), two experimental conditions (“clean” vs. a kinked pressure and flow measurement line referred to as “knick”, German for “kink”), and two repeated runs of approximately 10–11 minutes each, yielding the observed 10 × 4 × 2 × 2 = 160 files. Each record carries a device identifier, channel name, a class label (waveform samples, periodic measurements, device settings, alarms, or text messages), a value, and an ISO-8601 timestamp. Two simulated devices are represented: a pulse-oximetry device (channels for pulse rate, SpO2, and perfusion index) and a mechanical ventilator (waveform, measurement, settings, alarm, and message channels). Fault (“knick”) recordings carry a dedicated binary annotation channel marking the onset of the induced fault partway through the run and show a correspondingly higher rate and diversity of alarm events than matched “clean” recordings. The dataset underwent a manual quality check: one “clean” recording (p01-VC1-clean-0) had an incomplete and partially mislabeled annotation channel, which was reconstructed as constant given that the recording is free of the fault. Two recordings (p04-PC2-knick-0 and -1) have partial or entirely missing ventilator data due to a recording dropout; as the missing values cannot be reliably reconstructed, they are documented as a known limitation rather than imputed. A per-file lookup table mapping filenames to patient, condition, and ventilator-setting metadata, together with per-file record counts, timestamps, and data-quality notes, is provided in the settings workbook.

OpenAccess:
Download fulltext ZIP
(additional files)

Dokumenttyp
Dataset

Format
online

Sprache
English

Interne Identnummern
RWTH-2026-06634
Datensatz-ID: 1038619

Beteiligte Länder
Germany

 GO


CC0: Public Domain Dedication ; OpenAccess

QR Code for this record

The record appears in these collections:
Document types > Other document types
Research data > Datasets
Publication server / Open Access
Faculty of Medicine (Fac.10)
Public records
526500\-2
Publications database

 Record created 2026-07-09, last modified 2026-11-18


OpenAccess:
Download fulltext ZIP
(additional files)
Rate this document:

Rate this document:
1
2
3
 
(Not yet reviewed)