MODELING MOS GAS SENSORS FOR MOBILE ROBOT OLFACTION

Transcription

1 Dep. System Engineering and Automation UNIVERSITY OF MÁLAGA - SPAIN MODELING MOS GAS SENSORS FOR MOBILE ROBOT OLFACTION Javier G. Monroy, Javier González and Jose Luis Blanco

2 INTRODUCTION: e-nose Cyranose 320 SensorFreshQ electronic nose Looking back...first ENose Space Flight launched October 29, 1998! MCE-nose

3 INTRODUCTION: general applications

4 INTRODUCTION: mobile robotic applications Gas mapping Leak Detection Robot Track following

5 GAS SENSORS: What for? What we expect from gas sensors... high sensitivity large dynamic range high selectivity / specificity to a target analyte low cross-sensitivity to interferents perfect reversibility of the physicochemical sensing process short sensor response and recovery time long-term stability "a sensor exhibiting all these properties is a largely unrealizable ideal" Higher-Order Chemical Sensing, A. Hierlemann and R. Gutierrez-Osuna. Chem. Rev. 2008, 108,

6 GAS SENSORS: technologies MOS Electromechanical Infrared Pellistor SAW

7 GAS SENSORS MOS: How they work? Metal Oxide Gas Sensor (MOS) heating element coated with with semiconductor sensing material often tin dioxide sensing material doped with catalytic metal additives e.g. palladium or platinum doping changes operating conditions sensor characteristics Semiconductor Coating (typically SnO 2 ) Heating Element

8 GAS SENSORS MOS: How they work? Semiconductor Coating (typically SnO 2 ) Heating Element

9 GAS SENSORS MOS: Pros & Cons CHEAP HIGH SENSITIVITY Figaro sensors x LACK OF SELECTIVITY x RESPONSE DRIFT(AGE FACTOR) x INFLUENCED BY TEMPERATURE AND HUMIDITY x LONG ACQUISITION CYCLES (SLOW RECOVERY) e2v sensors

10 GAS SENSORS MOS: Selectivity improvement

11 GAS SENSORS - MOS: Other improvements RESPONSE DRIFT(AGE FACTOR) Normalization Signal conditioning INFLUENCED BY TEMPERATURE AND HUMIDITY New materials Humidity and temeprature sensors LONG ACQUISITION CYCLES (SLOW RECOVERY) Hardware MCE-nose Software Modeling

12 GAS SENSORS : MOS long recovery time gas concentration sensor response 1st source sensor response Normalized values time(s)

13 RELATED WORK: exponential model Two phase model Exponential model 100 Smoke source Concentration (%) τ r τ d Ideal Step Concentration Ideal MOS sensor response 25 0 Time (s)

14 PROPOSED MODEL: goal? Time(s) Normalized sensor response Normalized gas distribution Time(s)

15 PROPOSED MODEL Gas (ppm) Resistance (Ohms) Readings (v)

16 PROPOSED MODEL Gas (ppm) TRANSDUCER (ppm V) MOS behaviour (MODEL) Readings (v) Gas distribution estimation (v)

17 PROPOSED MODEL : Experiments (validation) For validation we need. GROUNDTRUTH We need the real gas distribution and concentration!

18 PROPOSED MODEL : Experiments (validation) Chaotic Gas Dispersal - Diffusion -Advective transport -Turbulence [Smyth and Moum, 2001]

19 PROPOSED MODEL : Experiments (validation) KEEP THE GAS LOCALIZED

20 PROPOSED MODEL : Experiments (validation) Odor Airflow Clean Air

21 PROPOSED MODEL : Experiments (validation)

22 PROPOSED MODEL : Compensations 1. Two phase model (rise / decay) 2. Speed compensation (delay) 3. Dynamic time constants Normalized sensor response ? Negative Concentrations Normalized gas distribution Time (s) -0.4

23 PROPOSED MODEL : Compensations 1. Two phase model (rise / decay) 2. Speed compensation (delay) 3. Dynamic time constants

24 PROPOSED MODEL : Experiments 1 1 Normalized sensor response Normalized gas distribution Time(s) 0

25 PROPOSED MODEL : Experiments Robot motion Gas sources Raw readings Robot motion Gas Distribution estimation

26 PROPOSED MODEL : Experiments WHAT HAS BEEN DONE TILL TODAY? Slow down the robot (few cm/s). Pass several times over the same locations but along different directions. Gas source Robot motion

27 PROPOSED MODEL : Experiments 1 Speed = 10cm/s Gas = Acetona One way Modeled Normalized Values Distance (m) Raw Readings

28 PROPOSED MODEL : Experiments 1 Speed = 10cm/s Gas = Acetona Go & return Normalized Values Modeled Distance (m) Raw Readings

29 PROPOSED MODEL : Experiments Speed = 40cm/s Gas = Acetona One way Normalized Values TGS2620 Reading Gas Source Position Modeled Distance (m) Raw Readings

30 PROPOSED MODEL : Experiments 1 Speed = 40cm/s Gas = Acetona Go & return Normalized Values Modeled Distance (m) Raw Readings

31 PROPOSED MODEL : Conclusions Conclusions The proposed model overcomes the long-decay-time problem of MOS sensors. Improves sensing task wit mobile robots (accuracy, robot speed, task time reduction) Future Work Improve the calibration of the model parameters. Exploit the model in mobile robotics olfaction: multiple gas source finding, plume tracking, Compare the results with the MCE-nose

32 THANKS & QUESTIONS THANKS!