This blog post is a continuation of the ‘ Research at the KAIST University ‘ post where we talked about our arrival to Korea and the work plan for the project. This entry focuses on the last days at KAIST, the work achieved in the Biorobotics Laboratory and the outcomes of this research. The goals of the project have been updated and now the parameter testing will only include the dimensional effects. This comprises both width and length of the sensors.
The fabrication process of the samples used in the testing involves different steps. To obtain a broad and comparative dataset, a total of twelve rectangular shape sensors were fabricated with lengths of 30 mm, 60 mm and 90 mm and widths of 5 mm, 10 mm, 15 mm and 20 mm. The thickness of the samples was maintained constant at around 1 mm. First, the nylon skin adhesive elastic fabric was prepared and cut in the corresponding shape. Then, the fabric was placed and aligned on the dispenser table-bed, ready for the deposition of the composite material. For the fabrication, a custom pneumatic dispenser mounted to a 3-axis table was used. The dispenser consist in a syringe, a specialized pneumatic plunger and a pneumatic pump. The composite material is made out of multiwalled carbon nanotubes (MWCNT) in a polymeric matrix of Ecoflex 00-30 (Smooth On, Inc) with a weight percentage of 3.5%wt. The premixed material was introduced in the syringe and then this was mounted to the table and connected to the pneumatic line. The nozzle size, motion speed, air pressure and height of the syringe were controlled to achieve optimum quality of the samples. Once the material was extruded, the samples were introduced in the oven at 70 °C for curing. Finally, the samples were tested using a multimeter to ensure that the specimens were within range.
Figure 1. a. Fabricated sensor samples using the composite and stretchable fabric. b. Custom automatic dispenser.
To examine the effects of the dimensional changes in the properties of the carbon nanotube-composite sensor, a parameter testing was performed on different size specimens. The test was performed using the custom extensometer, measuring strain, stress and the electrical resistance of the specimens. The sensor signal was obtained by attaching a 5 mm wide strip of copper tape to both ends of the sensor covering all the width of the composite material. An optimal contact point between the copper tape and the sensor was achieved using a two piece screwed grip. The resistance of the sensors was measured using a constant voltage Wheatstone bridge calibrated with a ratio of 1:4. The positive voltage divider consisted of two resistance R3 and R4 with an impedance of 25kΩ and 100kΩ. The negative voltage divider was composed by the composite strain sensor and a variable resistor (type) that was used to generate balanced Wheatstone bridge. A voltage source of 12 V was used to excite the circuit. To calibrate the Wheatstone bridge the variable resistor was adjusted for each of the samples to maintain the equivalence. The test chosen to study the characteristics of the sensor was a saw tooth dynamic cycle test. This test allows to comprehensively study four characteristics: hysteresis, non-linearity, sensitivity and range and dynamic and oscillatory response. To obtain a broad range of results for the application the test was divided into two parts. A frequency range test was performed at multiple frequencies (0.25 Hz, 0.5 Hz, 1 Hz and 2 Hz) at a fixed strain of 40% of the length of the sample. The second test consisted in a constant frequency test at 0.25 Hz with different elongation percentages (10%, 20%, 30% and 40%). The data was obtained using a cork board (Quanser Q8-USB, Quanser) and the results were processed and analyzed using a MatLab program.
Figure 2. Extensometer used for testing and its different components.
Results and Outcomes
For the testing of each sensor, 3 key performance indicators (KPIs) were analyzed: sensitivity, linearity and hysteresis. These indicators were chosen as together; they allow for the sensor’s overall performance to be measured. A brief explanation of the method used to calculate each of these indicators is below:
- Sensitivity: A linear regression line was plotted against the loading and unloading voltage paths (on the graph). From this, the gradient of the linear approximation was calculated and taken as the sensitivity (ds/dV) of the sensor.
- Linearity: Also from the linear regression line, an R2 value was calculated. This represents an abstract measure of the fit of the regression line to the data and hence a measure of linearity.
- Hysteresis: The difference in the measured voltage, over the sensor between the loading and unloading path. This represents the dependence of the sensor current value of previous values.
Whilst each of these values doesn’t offer any particular insight into the sensor performance, these measures combined allow for a better picture to be developed on how varying of dimensions of the sensor affect the sensor’s ability to produce accurate measurements when subjected to a range of frequencies and strains.
The results of the testing were very clear; in general, the longer and the thinner the sensor, the better the sensor performance in all of the KPIs and over the testing ranges.
Figure 3. Graph of the sensitivity against sensor dimension for the different frequencies.
Figure 4. Graph of the sensitivity against sensor dimension for the different elongation percentages.
Unfortunately, in this blog post, only sensitivity will be considered as the other KPIs are outside of the scope of this blog post. That being said, the sensitivity results effectively convey the overall trend of the data: with increasing length and decreasing width, the performance of the sensors improved.
There are of course limitation of this analysis; the first of which being that not a wide enough range of the test parameters nor sensor dimensions were tested. As a result, it’s difficult to know how far into the extremities of dimensions this trend holds. Furthermore, in relation to the analysis methods; the use of linear regression line in not necessarily the most precise approximation method due to the non-linearity of some of the results. Because of this, error is introduced into the results due to the bad fit of the line to the actual data and it becomes harder to accurately characterize the sensor.
Effects like these are discussed further in the official report on the research with the blog post mostly summarizing and providing some insight into the testing that was carried out whilst Daniel and I were at KAIST.
The next steps for the project are to finalize the results of the testing and write up a detailed discussion of the results such that it can be included as a section within a journal paper. Until that point, the focus will be on analyzing the quality of the data and highlighting any shortcomings in the testing and analysis methods that offer explanation to any outlets found within the data.
The opportunity to travel to South Korea and undertake research at such a prestigious university with such passionate and inspiring researchers has been incredible. We are both incredibly grateful to our research supervisors: Dr. Ali Alazmani and Dr. Peter Culmer, Professor Jung Kim at KAIST and the University of Leeds for putting in so much work to supporting us before, during and after our time at KAIST. As well as this, a big thank you to those researchers within the Biorobotics who made us feel at home and always challenged us to go further in our research.