ICCAS2005 June 2-5, KINTEX, Gyeonggi-Do, Korea

1. INTRODUCTION

Every driver on the road is frequently exposed to unhealthy diesel and gasoline exhaust fumes. Especially in population centers with heavy vehicle traffic, high concentrations of these gases penetrate the vehicle interior through the ventilation system.

The air quality sensor(AQS), located near the fresh air inlet, serves to reduce the amount of pollution entering the vehicle cabin through the HVAC(heating, ventilating, and air conditioning) system by sending a signal to close the fresh air inlet door/ventilation flap when the vehicle enters a high pollution area. By performing this function, the air quality sensor provides a key health benefit to drivers and occupants of motor vehicles. The AQS single sensor element was fabricated using normally thick-film technology and consisting of SnO2 and some additives coating on a ceramic substrate. The use of a single sensor element to detect the pollutants of interest enhances the sensor's cost effectiveness. In the presence of reducible gases, the resistance of the sensor element increases, starting from a medium range. When detecting oxidizable gases, the resistance is decreased with respect to the sensor element's resistance range in unpolluted air. Normally AQS sensor was composed of two sensing elements for reducible gas and oxidizable gas separately for effectiveness of different sensing gases.

In our study, we developed one chip sensor module which include above two sensing elements, humidity sensor and bad odor sensor. With this sensor module, PIC microcontroller was designed with back propagation neural network to reduce detecting error when the motor vehicles pass through the dense fog area. And our developed system can intelligently detect the bad odor when the motor vehicles pass through the polluted air zone such as cattle farm.

2. INTELLIGENT AQS SYSTEM 2.1 Intelligent AQS system architecture

Typical AQS system in automobile is an integrated system consisting of a sensor element, signal processing by a microprocessor and an interface. The AQS identifies reducible gases (nitrogen oxides NOx) and oxidizable gases (carbon

monoxide/gasoline/benzene), which occur in road traffic and are harmful to health. Normal sensor element is dual sensor which is mainly composed of SnO2 and some other additives. One sensing layer of the dual sensor is for reducible gases such as NOx and SOx, and the other sensing layer of that is for oxidizable gases such as CO and other hydrocarbon gases. In addition these sensing layers, a bad odor sensing element, a temperature sensing element and a humidity sensor were added to detect the bad odor also and eliminate the misjudgment by humidity effect. And voice alarm function was added to the LCD monitoring that alarms ‘the passage of tunnel’ or ‘the passage of bad odor area such as cattle farm. Figure 1 shoes typical AQS system in a automobile.

Intelligent Air Quality Sensor System with Back Propagation Neural Network in Automobile

Seung-Chul Lee, and Wan-Young Chung * Graduate School of Software, Dongseo University, Busan, Korea

(Tel : +82-51-320-1756; E-mail: no510@hanmail.net) **Division of Internet Engineering, Dongseo University, Busan, Korea

(Tel : +82-51-320-1756; E-mail: wychung@dongseo.ac.kr) Abstract: The Air Quality Sensor(AQS), located near the fresh air inlet, serves to reduce the amount of pollution entering the vehicle cabin through the HVAC(heating, ventilating, and air conditioning) system by sending a signal to close the fresh air inlet door/ventilation flap when the vehicle enters a high pollution area. One chip sensor module which include above two sensing elements, humidity sensor and bad odor sensor was developed for AQS (air quality sensor) in automobile. With this sensor module, PIC microcontroller was designed with back propagation neural network to reduce detecting error when the motor vehicles passthrough the dense fog area. The signal from neural network was modified to control the inlet of automobile and display the result or alarm the situation. One chip microcontroller, Atmega128L (ATmega Ltd., USA) was used. For the control and display. And our developed system can intelligently detect the bad odor when the motor vehicles pass through the polluted air zone such as cattle farm. Keywords: Air quality sensor, automobile sensor, gas sensor, one chip sensor module

Grapic LCD displayer and voice alam Heater Unit

AQS sensor

The Front of Radiator

Frash Mode

Recirculation Mode

Fig. 1 Typical air quality sensor system in automobile

The AQS sensors the unhealthy gases directly, before they enter the cabin. The AQS quickly closes the fresh air inlet door whenever the vehicle enters a cloud of pollution that means the ‘Fresh mode’ changes to ‘Recirculation mode’ in cabin automatically. In the event of potentially harmful gas peaks in the outside air, it is necessary to activate the recirculation function immediately to activate the recirculation function immediately to achieve the best possible air quality in the car’s interior and optimize passenger health and comfort. The AQS triggers the vehicle’s electronic climate control unit to close the fresh air inlet door allows for automatic closure and opening of the fresh air inlet door, thereby more

ICCAS2005 June 2-5, KINTEX, Gyeonggi-Do, Korea

comfortably and reliably reducing the levels of harmful gases in the cabin.

For the AQS system test, firstly the gas sensing properties of dual sensor were surveyed. To simplify the humidity effect and reduce some complicated parameters, humidity was defined as two categories, that is, higher than 70% and lower than 70%. Figure 3 shows the gas sensing properties of dual type gas sensor to CO, CH4 and NO at 30 °C in both humidity areas.

3. GAS SENSING PROPERTIES OF THE USED GAS SENSORS

In this AQS system, three commercially available sensor elements were used. A TGS2201 dual sensor (Figaro Gas Sensor Ltd., Japan) which have a reducible gas sensing layer and a oxidizible gas sensing layer on the both sides of same substrate. And a temperature sensor and a humidity sensor were mounted on the sensor module for our AQS system to compensate temperature and humidity effect to SnO2 based dual sensing elements of TGS2201.

4. NEURON NETWORK FOR INTELLIGENT

DECISION 4.1 Error back-propagation

The key features of AQS include dynamic adaptation to various driving environments such as city, rural, traffic jams, tunnels and humid weather. The sensor’s sensitivity should be controlled by software to continuously adapt to ambient pollution levels. The AQS’s software structure allows the microcontroller to use different parameter settings based on the driving environment. During vehicle operation, the AQS adapts to the environment and uses algorithms to provide optimized performance for particular driving situations. An error back propagation neural network was used to reduce the malfunction of the gas sensor by environmental parameters such as humidity, temperature and bad odor.

3.1 Gas sensing effects by temperature and humidity

U4

LM 35DZ

1

2

3+V

GND

VOUT

R53k

U5

808H5V5

1

2

3VCC

GND

VOUT

12V U3

Gas Sensor

1

3

2

4OUT1

OUT2

COM

Heat

R41k

PAD4

OUTPAD

5V

-

+

U7A

LM358

3

21

84

RV1100k

PAD1

OUTPAD

RV1

100k

5V

RV1100k

PAD3

OUTPAD

PAD2

OUTPAD

C3104

7V

5V

Figure 4 shows 3 layers architecture of back propagation neural network. This network was composed of input layer to accept input signal form each sensors, hidden layer to enhance the performance of learning effect and output layer to discriminate the gas kinds. The statistic and the neural network techniques are usually used for gas pattern recognition. Recently back propagation neural network algorithm having nonlinear discriminating ability appeared as a powerful recognizer in the pattern recognition using parallel distribution process.

Fig.2 measuring circuit for gas sensor, temperature sensor and humidity sensor, separately.

Figure 2 shows the measuring circuit for dual type gas sensor,

temperature sensor and humidity sensor, separately. For dual type gas sensor two source DC voltages were used, that is, 12V was for gas sensing elements and 7V was for sensor heater. PV1 and PV2 were the load resistors and the output voltages V1 and V2 could be calculated to the sensor resistances, separately. The signal from a digital temperature chip (LM35D, National Co.) was amplified by 4 times and used digital temperature signal between 0°C and 100°C. A humidity sensor (808H5V5, Sencera Co) was used to evaluate relative humidity between 30% and 80%.

0

1

i

N

0

1

j

L

0

1

k

M

Xp0

Xp1

Xpi

XpN

Input layer Hidden layer Ouput layer

WkjWji

Op0

Op1

Opk

Opm

3.2 Gas sensing parameter extraction from gas sensor, temperature sensor and humidity sensor

Fig. 4 a schematic diagrams of 3 layer error back-propagation Neural network.

2

temperature 30°C±, humidity 70% below, CO

temperature 30°C±, humidity 70% below, CH4

temperature 30°C±, humidity 70% below, NO

temperature 30°C±, humidity 70% above, CO

temperature 30°C±, humidity 70% above, CH4

temperature 30°C±, humidity 70% above, NO

1

2

4

3

5

6

Neuron network with biology remember by experience or knowledge leaning new information with weight value on synapse. In this paper, Error back-propagation algorism is used.. 4.2 Gas sensing effects by temperature and humidity

In this study, 4 sensors were used as a input of the neural network. The nodes in input layer and output layer are already determined by the number of sensors and results to be determined, respectively. Learning rate, which affects learning speed and accuracy, are set properly in a range of 0.1~0.5 in

Fig.3 Responses of gas sensor to CO, CH4 and NO with different humidity level.

ICCAS2005 June 2-5, KINTEX, Gyeonggi-Do, Korea

this study. Figure 5 shows the neural network architecture of this study. The gas sensor signal was converted to digital signal through A/D converter, and the digital signal introduced in input layer. Using the synaptic weights learned by the back-propagation neural network and input signals, the objective output node in output layer is selected and the determination is displayed on the LCD or a monitor.

One chip microcontroller, Atmega128L (ATmega Ltd., USA) was used for the control and display.

In our study, automatic and manual buttons was designed for the driver can choose the manual opening the air inlet or automatically opening by the signal from AQS. Figure 8 shows the designed intelligent AQS microcontroller PCB layout and the photograph of fabricated intelligent AQS main board. The microcontroller alarm using recorded human voice and displayed the opening or closing status of the air inlets. The voice recordable chip (ISD1420, Voice chip Co.) was used for this system.

Firstly, synaptic weight was investigated by the computer simulation using the extracted parameters of Fig. 3. The synaptic weights are listed following the error rate in shown in Fig. 6. Figure 9 shows test apparatus for AQS controller and heat

unit.

Interface

Interface

NO xxx ppm

NO2 xxx ppm

CO xxx ppmH2 xxx ppm

HC xxx ppm

S1

S2

S4

Gas1

Gas2

Gas3CO2 xxx ppm

Input layer

Hidden layer

Output layer

Neuron network

Gas Input Output display

Weight update

Error back-propagation

A/D

Input patten

A/D

Integer change

(a)

Fig 5. Neural network architecture for intelligent AQS system

Fig 6. Synaptic weight value for minimum Error range (b) Fig. 8 Intelligent AQS microcontroller board.

5. INTELLIGENT AQS CONTROLLER BOARD (a) Intelligent AQS main board layout (b) Photograph of fabricated microcontroller board.

Air Pollution Sensing Sensor

Neuron NetworkAlgorithm

A/D Converter

Input port

Passive Button

AQSOPEN/CLOSE

Automatic Button

Real Time Clock

12V Patteri7 and 5V DC voltage conver circuit

Voice1

Voice2

Voice3

AQS Control

AQS system is action Automatic

AQS system is inflow to passive the out side air

AQS system is intercept to passive the side air

Inflow Hole Open/Close

12V Power Input

Display panel

RS-232 serial Communication

micro processor ATmega 128

Voice4the out side air pollution is the middle danger

Fig. 7 Block diagram of AQS Controller board.

The signal from neural network was modified to control the inlet of automobile and display the result or alarm the situation.

Fig.9 Test apparatus for AQS controller and heat unit.

ICCAS2005 June 2-5, KINTEX, Gyeonggi-Do, Korea

6. CONCLUSIONS

One chip sensor module which include above two sensing elements, humidity sensor and bad odor sensor was developed for AQS (air quality sensor) in automobile. With this sensor module, PIC microcontroller was designed with back propagation neural network to reduce detecting error when the motor vehicles pass through the dense fog area. Using the sensitivity signals from a sensor array as multi-dimensional input patterns, a gas pattern recognizer using a multi-layer neural network with an error-back-propagation learning algorithm was then implemented. And our developed system can intelligently detect the bad odor when the motor vehicles pass through the polluted air zone such as cattle farm.

REFERENCES [1] Dae-Sik Lee, “SnO2-based sensor arrays for Monitoring

Combustible Gases and Volatile Oranic Compounds,” Ph, D, thesis, Kyungpook national university, pp. 114-127, 2000.

[2] Chang-Hyun Shim, “Fabrication of Thin Film Gas Sensor by Oxidizing Sn/Pt Double Layer and Its Application for Gas Classification,” Ph, D, thesis, Kyungpook national university, pp.32-33, 2003.

[3] Figaro engineering Inc., gas sensor TGS2201. [4] Hung-gee Hong, “A portable Electronic Nose System

Using Gas Sensor Array and Aritificial Neural Network,” Ministry of science and Technology, pp.41-43, 2001.

[5] J. H. lee and S. j. Park, “SnO2 powder preparation from hydroxide and oxalate and its characterization”, j. Kor. Ceram. Soc. J. Kor. Ceram. Soc.,pp274, 1990

[6] W. P. Carey et al., “Selection of adsorbates for chemical sensor arrays by pattern recognition”, Anal. Chem., Vol. 58, pp149-153, 1986.

ACKNOWLEDGEMENT “This work was supported by Korea Research Foundation Grant (KRF-2004-042-D00129)”