Temperature-modulated gas sensor signal

Temperature-modulated gas sensor signal

Abstract: Data is the conductance of a temperature-modulated gas sensor exposed to several concentrations of several gases. Both classification (gas type) and (selective) quantification are of interest.

Source:
Creator: Lab for Measurement Technology, Saarland University, 66123 Saarbrücken, Germany
Contact: m.bastuck@lmt.uni-saarland.de, info@lmt.uni-saarland.de

Dataset:
The data set was experimentally obtained from a semiconductor gas sensor (UST GGS 1330) with temperature cycled operation (TCO). The sensor temperature was linearly increased from 200 °C to 400 °C within 20 s, and back to 200 °C within another 20 s. This cycle is repeated during the whole measurement (~18 h). During the measurement, the sensor was exposed to four different gases (carbon monoxide, CO, ammonia, NH3, nitrogen dioxide, NO2, and methane, CH4) in three different concentrations each.
The aim is either to classify the type of gas that is currently seen by the sensor independent of its concentration, or the concentration of one specific gas type.

Attribute Information:

• sensordata.csv
  The data set consists of the measured conductance in nS (nanosiemens) of the sensor. Each row represents one cycle (200-400-200 °C in 40 s) with 4001 data points. 
• targetvectors.csv
  Different target vectors have been prepared manually. All of them include an ‘ignore’ label that indicates cycles where the exact gas concentrations are unknown, e.g. due to a change in concentration which can take several cycles. Cycles during an “init peak” in the beginning to test the setup are labeled as ‘ignore’ as well. These cycles should always be discarded. The objectives of the ten target vectors are as follows:
  1. categorical. A short segment at the beginning is labeled as ‘background’, i.e. no relevant test gas is present. In the following, the presence of a test gas is indicated with its name, i.e. CO, NH3, NO2, and CH4
  2. continuous. A short segment at the beginning is labeled as ‘0’, CO exposures are labeled with their respective concentrations, everything else is ignored.
  3. continuous. A short segment at the beginning is labeled as ‘0’, NH3 exposures are labeled with their respective concentrations, everything else is ignored.
  4. continuous. A short segment at the beginning is labeled as ‘0’, NO2 exposures are labeled with their respective concentrations, everything else is ignored.
  5. continuous. A short segment at the beginning is labeled as ‘0’, CH4 exposures are labeled with their respective concentrations, everything else is ignored.
  6. categorical. Like (1), but the previously ignored background cycles at the beginning and between gas exposures are now labeled as ‘background’.
  7. continuous. Like (2), but the previously ignored background cycles at the beginning and between gas exposures are now labeled as ‘0’.
  8. continuous. Like (3), but the previously ignored background cycles at the beginning and between gas exposures are now labeled as ‘0’.
  9. continuous. Like (4), but the previously ignored background cycles at the beginning and between gas exposures are now labeled as ‘0’.
  10. continuous. Like (5), but the previously ignored background cycles at the beginning and between gas exposures are now labeled as ‘0’.

Target vectors (6)-(10) contain some sensor drift in the background. Additionally, the background class is much larger compared to the gas exposure classes. Target vectors (7)-(10) also label all but one gas as ‘0’, so that a selective quantification of the target gas must be performed.

• gasexposures.png
  A graphical summary of the dataset.
• temperaturecycle.png
  A graphical representation of the temperature cycle used when operating the gas sensor.