1Proposal: Odor Tracking with anElectronic NoseCPSC 483 Spring 2003R. Gutierrez-OsunaNinh DangJason HamorGreg AllbeeSimon Saugier2Table of ContentsI - Introduction --------------------------------------------------- 3A. Problem background --------------------------------------------------- 3B. Problem statement --------------------------------------------------- 4C. Design objective --------------------------------------------------- 5II - Method of Solution --------------------------------------------------- 8A. Survey of literature, market, and existing solutions ------------------------ 8B. Originality and creativity ------------------------12C. Design constraints and feasibility study ------------------------14D. Consideration of alternative solutions ------------------------16E. Design validation ------------------------18III - Other Consideration ---------------------------------------------------20A. Economic analysis ---------------------------------------------------20B. Societal, safety, and environmental analysis ------------------------20C. Management, scheduling and team work ------------------------21IV - Bibliography ---------------------------------------------------253I. IntroductionA. Problem BackgroundOdor tracking by electronic means has been a field of much research in recentyears. Many advances have been made by trying to mimic the olfactory tracking abilitiesof animals such as dogs or moths. These behaviors have been reproduced in robots byusing electronic noses, or “e-noses”. An e-nose is a collection of sensors connected toelectronics used to detect changes in odor concentrations in an area. The e-nose takes insamples of the environment air, where the sensors translate the odor concentration intoelectrical signals that are then processed to identify the odor.The idea of applying the sense of smell to robotics came about due to the need foran objective, quantifiable method to locate and identify odors in potentially hazardousareas. Since humans and animals have varying and subjective senses of smell that canfatigue over time, theoretically a robot could be used in situations where odors need to belocated without such organic time constraints. The hardware for locating odors alreadyexists and is manufactured by a number of companies. However, one problem withcurrent technology is the response time of odor sensors. The metal-oxide sensors in thedetectors take time to heat up and provide a useful signal to process odor concentrationwith. This slower response time by a robotic agent however, is still preferable to usinghumans or animals to locate potentially dangerous odor sources such as natural gas.Current research has already produced effective, accurate e-noses. Such devices are usedas simple stationary sensors that monitor air quality in order to preemptively detectunwanted gases. Handheld devices have also been developed that are used extensively inagriculture and food production. Quality assurance personnel use these to monitor foodquality and to ensure that food is still perishable by humans or livestock. E-nose devicesare also found on the space shuttle mission to monitor air quality. Research on astronauts4discovered that the change in blood pressure in space affects the astronauts’ sense ofsmell, and they are not able to detect odors that they would be able to on earth. Thenewest area of odor localization and identification is applying such technology to mobilerobotic agents.B. Problem StatementThe goal of this project is to develop signal-processing procedures to reconstructthe changes in concentration of odor mixtures through the response of a gas sensor array.The main problem in this is how to instruct the robot how to react to the sensor responseprovided to it by the odor dilution system. These includes the development of analgorithm for odor location and tracking, and also signal processing algorithms forrecovering the gas concentration profiles once the source is determined.5C. Design ObjectivesFig. 1The objectives for this project are to simulate the dispersion of an odor source inan area, and to simulate a robot’s behavior in tracking the odor source. This model willbe composed of three interrelated modules, as shown in Figure 1. The modules are therobot simulator, the dispersion model and the e-nose model. The robot simulator is givena map of the geometry of the room, so that it knows how to navigate around walls and6other obstacles. The robot passes its’ current coordinates to the dispersion model. Thismodule receives a sensory response from the e-nose model and uses this to determine itsnext course of action. The dispersion model is given an arbitrary map of a room, one ormore odor sources, and one or more airflow sources. This model simulates the dispersionof the odor sources based on the room geometry and the airflow currents. The dispersionmodel produces an odor concentration at the robot’s current location, and passes thisvalue to the e-nose model. The e-nose model takes the current odor concentration fromthe dispersion model and produces a sensory response, which the robot uses to determinewhere to move. This e-nose model will first be simulated in software as a simpleequation, in order to easily test the other two modules interactions. After the dispersionmodel and robot simulation are confirmed as working properly, the e-nose model will bereplaced by an actual dilution system. A team from a previous class has already designedand provided a dilution system, which we will use in our project. This dilution system iscomposed of three diluters, a sensor array, a pump and some circuitry to control themixing levels. The system takes specified concentration levels and mixes the designatedodors with air to produce the correct concentration. This concentration is passed throughthe sensor array, which produces the response for the robot simulator.Our team will begin with simplified system inputs and gradually increase systemcomplexity as time permits. We will start with simple room geometry, a single odorsource and an airflow source. As time permits, we will consider the task of adding morecomplex room geometries, with multiple odor and airflow sources. We may also add newsensor heating profiles to the LabVIEW TM interface. These profiles control the heat ofthe sensors, and provide for greater accuracy. By allowing more profiles, we will furtherincrease the accuracy. Currently, only a sine wave