• Figures of the UNIMAN and Synthetic Datasets

    The images come from the article submitted to the ISOEN 2011 conference.

    Figure 1: Scoreplot of the Principal Component Analysis (PCA) performed on the first 1500 samples from the experimental dataset of 17 polymeric sensors. Steady-state features were extracted and preprocessed to remove outliers. The covariance structure for classes with the highest concentration is depicted with the confidence ellipses.

    Figure 2: Example of synthetic experiments for gases A and C on different concentration levels and in mixture. The top panel shows the input concentration matrix of three gases injected over time. The bottom panel contains the output signals of 10 sensors over time, including transient and steady-state response.

    Figure 3: A Principal Component Analysis computed on the sensor steady-state readings is shown for: left) No competitive model in mixtures, right) Langmuir isotherm competitive model.
    The synthetic array of 40 sensors was exposed to different mixtures of gases A and B, by varying every component concentration from 0.1 to 1.0 in normalized units, where the binary mixture is coded by a RGB color (the pure A is blue, the pure C is green). On the right panel, one can clearly observe the non-linearity in the response to the mixture.

    Figure 4: Performance of a k-NN classifier (k=3) as a function of the distance in time of the acquisition of the Validation Set from the Training Set. The classification aimed the discrimination of the A, B and C gases on three synthetic datasets with different level of noise in the array of 40 sensors (10%, 100% and 200%, percentual to the default noise parameters in the synthetic sensor array). Three noise sources are included in the model: concentration noise, sensor noise and instrumental drift.

Comments are closed.