Fig 1.
A novel chemo-mechanical actuator.
(A) The structure of leucoemeraldine salt of PANI. (B) A film of polyaniline cellulose acetate (PANI-CA) composite. (C) A PANI-CA film bending in response to gases in the headspace of the conical flask. The solution in the flask has water, acetone, and/or ethanol.
Table 1.
Volatile mixtures tested.
Fig 2.
Image processing and computer vision.
Some steps in the automatic analysis of the video are outlined. (A) RGB images from the video frames are converted to binary black and while images. (B) Binary images are processed to extract a contiguous midline through the filament. This midline path is then used to determine shape features such as angle of turn, curvature, etc.
Fig 3.
Validating automatic filament tracking.
(A) Automatically reconstructed filament shape changes are shown for one trial. The final 25% of the filament length, labeled by the red and black solid circles are used to compute the tip direction (green). (B) Automatic computer-vision-based tracking the filament agrees well with manual tracking of the filament.
Fig 4.
Tracking with deep neural networks.
DNN-based tool DeepLabCut produced tracking well correlated with the manually obtained ground truth. The colored dots are the key point location predicted by the tool. The predicted key points had greater errors in the initial frames of the video compared to the later frames.
Fig 5.
Dimensionality of the bending filament.
(A) We find that one principal component captures about 80% of the variance across a whole thirty-exposure trial. Two principal components capture over 95% and four capture over 99.5%. Curves are shown for specimen 5 (blue) and specimen 6 (red). (B) The shapes of the principal components are shown; linear combinations of the first two PCs produce different orientations and the third and the fourth PCs control bending to the left versus the right.
Fig 6.
The time to maximum deflection changes systematically between exposures to the different acetone and ethanol concentrations but there is a large range in the responses.
Fig 7.
Predictions from filament bending and predictions of filament bending.
(A) A sample series of 30 exposures to varying solution fractions of acetone, ethanol and water are shown. (B) The filament tip angles (relative to some absolute orientation) are shown as a time series. These correspond to the exposures in panel-a. (C) Predicting the acetone fraction in solution using a linear model based on fitting a triple exponential to transients gives about 0.45 R2 value, suggesting some but not perfect predictive ability. (D) Predicting the change in tip angle over a trial using the acetone fraction, ethanol fraction, and the initial angle using a linear model gives about 0.4 R2 value. (E) The initial angle for each trial–which was the angle to which the filament relaxed after the previous trial–changes over time, suggesting hysteresis, or slower time-scale processes over many tens of minutes.