Sensor drift

The MZI interrogation circuit of Fig. 9.13b offers a simple method to eliminate sensor drift due to changes in temperature or other common mode variables. 

After 30 min equilibration at room temperature, the measurement run started. An air flow (room air filtered through active carbon) was conveyed over the sensors at a constant rate (lcm3/s) for 10 s to stabilize the baseline. An automatic syringe then suckled Asiago cheese head-space and conveyed it over the sensor surfaces for 60s. The sensors were exposed again to the reference air flow to eventually recover the baseline. The total cycle time for each measurement was 5 min. No sensor drift was experienced during the measurement period. Each sample was evaluated three times and the average of the results was used for subsequent statistical analysis (principal component analysis (PCA)). 

Sensor drift is often a very difficult problem to solve. Under ideal conditions, a sensor produces an output signal that is related only to the analyte of interest. When presented with a constant reference state, the so-called baseline signal... 

Although the basic principles of type III potentiometric sensors are apphcable for gaseous oxide detection, this should not obscure the fact that these sensors still require further development. This is especially true in view of the kinetics of equilibria and charged species transport across the solid electrolyte/electrode interfaces where auxiliary phases exist. Real life situations have shown that, in practice, gas sensors rarely work under ideal equilibrium conditions. The transient response of a sensor, after a change in the measured gas partial pressure, is in essence a non-equilibrium process at the working electrode. Consequently, although this kind of sensor has been studied for almost 20 years, practical problems still exist and prevent its commercialization. These problems include slow response, lack of sensitivity at low concentrations, and lack of long-term stability. " It has been reported " that the auxiliary phases were the main cause for sensor drift, and that preparation techniques for electrodes with auxiliary phases were very important to sensor performance. ... 

Abnormal III (AIII) The pH value increases slowly and reaches a maximum point, then comes back to 7.0 slowly (region 6 in Figure 7.12), which could be the result of a temporary sensor drift. 

The abnormal condition AI (disturbance) and the abnormal condition AIII (sensor drift) can be recognized clearly (Figures 7.14c, 7.14d). 

Relative Humidity General Eastern 0 to 100% RH -10°C to 100°C gas stream Corrosion resistance sensor drift repeatability, cost... 

In its favour the method would supply a cheap and convenient way of analysing CO2 but there are problems with sensor drift and interference from other gases. 

Abstract. Artificial neural networks (ANN) are useful components in today s data analysis toolbox. They were initially inspired by the brain but are today accepted to be quite different from it. ANN typically lack scalability and mostly rely on supervised learning, both of which are biologically implausible features. Here we describe and evaluate a novel cortex-inspired hybrid algorithm. It is found to perform on par with a Support Vector Machine (SVM) in classification of activation patterns from the rat olfactory bulb. On-line unsupervised learning is shown to provide significant tolerance to sensor drift, an important property of algorithms used to analyze chemo-sensor data. Scalability of the approach is illustrated on the MNIST dataset of handwritten digits. 

The robustness to sensor drift of the method under study was evaluated using a simple synthetic drift model. A gain for each of the 60 sensors was initiated to 1 after which the gain factor was subject for over 100 random-walk steps taken from a Gaussian distribution with = 0.01. In the on-line learning condition while testing drift robustness, the last unsupervised vector quantization step was run continuously. 

In the test of robustness to sensor drift it was shown that when the unsupervised part of the algorithm was allowed to run in on-line training mode drift robustness much superior to SVM and the new algorithm with no on-line learning was demonstrated. This is a promising result, but further characterization of this property is required. Additional evaluation is currently ongoing on a real chemosensor dataset. 

Sensor drift is a first serious impairment of chemical sensors. They alter over time and so have poor repeatability since they produce different responses for the same odour. That is particularly troublesome for electronic noses (Remain et al. 2002). The sensor signals can drift during the learning phase (Holmberg et al. 1997). To try to compensate the sensor drift, three types of solutions were tested for our applications. 

With real-life measurements, it is indeed very difficult to identify a single direction in a multivariate space that is only correlated to sensor drift. So, for each sensor, an individual multiplicative factor was calculated by estimating the drift slope for a standard gas. 

However, several problems still exist. These include sensor drift, which leads to the inability to provide proper calibration. This is of special concern to quality control laboratories and is one of the reasons for the general absence of these instruments in these laboratories [3]. Limitations to the use of the electronic nose include loss of sensitivity in the presence of water vapor and high concentrations of individual components such as alcohol, relatively short life of some sensors, and the inability to obtain quantitative data for aroma differences [72]. Each device also still needs considerable method development, but progress is being made at a rapid rate. Einally, sensor arrays and PR tend to predict the quality of a sample without providing hard data with respect to composition and concentration [74]. 

Despite promising, the reliability of implantable systems is often undermined by factors like biofouling [100, 101] and foreign body response [102] in addition to sensor drifts and lack of temporal resolution [103]. To minimize such problems. 

---