1. Introduction
Networks of chemiresistive gas sensors can be used to continuously monitor the different gases in areas of interest with a considerably high spatial density [
1]. One major drawback of using this technology can be its stability over long time scales. As different faults can occur over time, the measurement accuracy can degrade consecutively. Such defects can be caused by different processes on the sensor, which can affect the signal output in different ways. Examples would be a loss of sensitivity, sensor stopping, signal jumps, and battery loss, as is shown in
Figure 1.
In order to repair or replace individual faulty sensors inside the network, it is necessary to have screening algorithms which continuously evaluate the current state of the sensor network in terms of possible defects. Hence, for general wireless sensor networks, such algorithms have already been investigated [
2]. In order to generate and assess such algorithmic approaches for chemiresistive gas sensors, however, simulation data exploring different sensor network scenarios are necessary. For the specific case of sensor drift, such frameworks have already been investigated to evaluate calibration algorithms, e.g., in [
3]. Therefore, in order to study other fault types, we want to present a framework based upon a stochastic sensor simulator [
4], which can provide sensor network simulations specifically feasible for a number of different sensor defects.
2. Methods
Our sensor network simulation framework consists of three different parts, which are depicted in
Figure 2a. In the Concentration Model, an array of different gas sources is simulated in order to calculate the spread of the emitted gases across the simulation area and thus simulate their local distributions. The output of this model is the time-dependent concentration series which would be measured at each of the sensor locations. These concentration time series are then the input of the sensor response model. An illustration of the concentration distribution calculated by the concentration model is shown in
Figure 2b.
The sensor response model then translates the input concentrations at each sensor location to the expected sensor signal measurements. This is done by using a stochastic sensor model which has been developed in previous research [
4].
In particular, the sensor response model simulates the processes which are directly occurring on the sensor surface by modeling its adsorption and desorption processes on a microscopic level. Finally, the sensor fault model generates faults in the synthetic signals. Due to the flexibility of the sensor model, two different approaches can be followed here, as shown in
Figure 3.
On the one hand, the faults can be generated after the signal simulation by manipulating the output signal (post-simulation faults). On the other hand, the faults can be generated intrinsically in the sensor simulation as well (intrinsic faults). Here, sensitivity loss can be modeled by switching off sensor sites in the sensor surface simulation, whereas battery loss can be generated by changing the heating properties in the model parameters. Both these methods have advantages and disadvantages. Post-simulation faults tend to be rather efficient, since they can be generated spontaneously from a normal signal with respect to the needed sensor fault. However, since a formula is applied here after the simulation, these defects seem to be less accurate. This is the main advantage of the intrinsic approach, since the defect is simulated on the sensor surface, leading to a higher accuracy. This approach is less efficient, however, since every case has to be simulated beforehand.
3. Results and Discussion
The different defect types have been implemented intrinsically in the stochastic sensor model by adjusting the heating parameters for battery loss fault simulation and changing the amount of responsive binding sites on the simulation grid for sensitivity loss simulation. Signal jumps and sensor stopping have been implemented as post-simulation faults. In order to test the model, a set of four concentration pulses followed by clean air have been simulated to show the impact of the different faults on the signal. These can be seen in
Figure 4.
It can be seen that the sensitivity loss leads to a lower response to the concentration pulses, which is in-line with the physical expectations. Additionally, for the battery loss, which should lead to lower heating capabilities, a signal change can be seen. Here, the recovery process for the sensor signal is not as efficient as for the original signal, which leads to an additive drift caused by slow recovery. Additionally, the processing defects such as sensor stopping and signal jumps are represented well in the simulation experiment.
4. Conclusions and Outlook
It can be noticed that our simulation framework is suited for simulating various defect effects which can be used for algorithm development for fault detection. There are different design choices to be made which are influenced by computational efficiency and fault accuracy. Therefore, additional research has to be done in this area as well.
In future research, other faults might also be considered for simulation. For example, interfering gases might be an effect which should be studied in more detail for this kind of sensor.
Author Contributions
Conceptualization, S.A.S.; methodology, S.A.S.; software, S.A.S.; validation, S.A.S. and C.C.; formal analysis, S.A.S.; investigation, S.A.S.; resources, S.A.S.; data curation, S.A.S.; writing—original draft preparation, S.A.S. and C.C.; writing—review and editing, S.A.S. and R.W.; visualization, S.A.S.; supervision, C.C. and R.W.; project administration, C.C.; funding acquisition, C.C. All authors have read and agreed to the published version of the manuscript.
Funding
This research received no external funding.
Acknowledgments
This work has been partially supported by the LIT Secure and Correct Systems Lab funded by the State of Upper Austria as well as by the BMK, BMDW, and the State of Upper Austria in the frame of the COMET program (managed by the FFG).
Conflicts of Interest
The authors declare no conflict of interest.
References
- Somov, A.; Baranov, A.; Savkin, A.; Spirjakin, D.; Spirjakin, A.; Passerone, R. Development of wireless sensor network for combustible gas monitoring. Sens. Actuators A Phys. 2011, 171, 398–405. [Google Scholar] [CrossRef]
- Mahapatro, A.; Khilar, P.M. Fault Diagnosis in Wireless Sensor Networks: A Survey. IEEE Commun. Surv. Tutor. 2013, 15, 2000–2026. [Google Scholar] [CrossRef]
- Delaine, F.; Lebental, B.; Rivano, H. Framework for the Simulation of Sensor Networks Aimed at Evaluating In Situ Calibration Algorithms. Sensors 2020, 20, 4577. [Google Scholar] [CrossRef] [PubMed]
- Schober, S.A.; Carbonelli, C.; Roth, A.; Zoepfl, A.; Wille, R. Towards Drift Modeling of Graphene-Based Gas Sensors Using Stochastic Simulation Techniques. In Proceedings of the 2020 IEEE SENSORS, Rotterdam, The Netherlands, 25–28 October 2020; pp. 1–4. [Google Scholar] [CrossRef]
| Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations. |
© 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).