Characterization of the ability of polymeric chemiresistor arrays to quantitate trichloroethylene using partial least squares (PLS): effects of experimental design, humidity, and temperature

Polymeric chemiresistors are a class of chemical sensor that have promise for being practical in situ sensors of volatile organic compounds (VOC) in various environmental monitoring applications. However, these devices may undergo changes in response due to changes in temperature or humidity that must be taken into consideration when developing calibration models. The devices can also display significant hysteresis effects after exposure to VOC vapor. These effects are complicated by the fact that each sensor within an array is coated with a different polymer, each with a different response to temperature, humidity, and VOC exposure. It is shown that partial least squares (PLS) can provide quantitative predictions of trichloroethylene (TCE) using an array of chemiresistors through appropriate experimental design. Effects of humidity and temperature on the response of chemiresistor arrays and predictive ability of PLS are also discussed. It is also shown that to truly assess the quality of a calibration model it must be first tested through prediction of a test set at a time separated from the acquisition of the calibration data. Using only leave-one-out cross-validation results from the calibration can lead to unwarranted confidence in a model that is not stable with respect to changing environmental conditions and device drift.