Abstract:
A remote air pollution monitoring system is developed in this project. The system consists of 3 main parts; measuring unit, data transfer unit and data processing unit. The system can work in 2 modes. One is for a not too long distance which connect the measuring unit the data processing unit with an electrical wire. And the other connetcs the measuring system and the data processing via modem which can be worked as far as telephone line can be reached. Up to 5 gas sensors could be installed in the gas measuring unit to collect data for both qualitative and quantitative analysis of the samples. The gas sensors are made from semiconductor material. The system is a flow system which could be measured samples in both gas and liquid phase. 2 types of data transfer unit were developed. The first one is for the case which is applied for not too long distance and the modem was not used. The latter is for the long distance measurement which modem is used to transfer data at a speed at a speed of 9600 bps. The unit could be sampled as fast as 52 microsec per 1 data. However in this project, the sampling speed of 0.6/data is sufficient for data collection. The data processing unit will control the data transfering in the system and also analysis the samples. There are 3 types of analytical methods which could be applied; a analysis using back-propagation neural networks, a analysis using radial basis neural networks and chemometrics. In the analysis procedure, the effective using of response characteristics of sensors (rise-time, peak and fall-time of the response curve) is proposed. The system was tested in 2 modes; qualitative and quantitative analysis of the samples. 5 types of samples, namely de-ionized water, alcohol, amonia, acetone and acetadehyde, with the concentration in the range of 0.01%-0.1% by volume, are used. The analysis results shows that the using of radial basis neural networks provided 100% accuracy in quantitative analysis. The qualitative analysis with accuracy upto 100% could be obtained from the back-propagation neural networks.