Font Size: a A A

Information Processing Techniques In Artificial Olfactory Systems

Posted on:2001-08-24Degree:DoctorType:Dissertation
Country:ChinaCandidate:J L QuFull Text:PDF
GTID:1118360002451603Subject:Precision instruments and machinery
Abstract/Summary:PDF Full Text Request
Cross sensitivity in gas sensors has led to great interest in developing the artificial olfactory system in recent years. Selectivity can, in principle, be improved by the use of an array of gas sensors with an appropriate pattern recognition technique. First, a complete review of the development and the existing problems of the artificial olfactory system is given in this thesis. Then, more efforts have been put on the new information processing techniques based on neural network, fuzzy logic and genetic algorithm, which present new ideas and new methods for the designing of the artificial olfactory systems. The main efforts and achievements in this study are as follows: 1. An artificial olfactory system, which is composed of a metal oxide semiconductor sensor array with a feed forward neural network, is designed to identify three kinds of gases(CO, H2 and CH4). After the neural network is trained with the back-propagation algorithm, it can classify the unseen patterns. 2. The influences of different signal pro-processing algorithms on the above gas identification system are compared. The array normalization algorithm can reduce or eliminate the effect of the gas concentration on the system and improve its performance. 3. Another artificial olfactory system, which combines a metal oxide semiconductor sensor array with a self-organizing feature map network, is designed for gas identification. When the network is trained with a competitive algorithm, it can cluster the samples of three kinds of gases respectively and recognize the unseen ones. 4. A method for gas qualitative identification based on fuzzy C-means clustering algorithm is proposed. For the pattern unknown samples, given the total number of their classes, c, and the fuzzy weighting exponent, m, the set point of every class and the fuzzy membership of every sample are calculated by optimizing the objective function, J(U, V). Therefore, the class of every sample can be determined by means of using the maximum membership rule. Experiments show that the fuzzy C-means clustering algorithm can distinguish gas samples of different classes, and can be used in real time detection of gas. 5. The genetic algorithm based on back-propagation neural network is III ABSTRACT employed to select gas sensors in distinguishing gas experiment. By evaluating the fitness of sensor arrays, a new method of choosing gas sensors is built. 6. The influence of the environmental temperature and humidity on the gas sensor is analyzed, and two knowledge-based methods for temperature and humidity compensation are proposed. One applies fuzzy logic and another artificial neural network. These two methods can use the existing experience and knowledge and experiment shows that they both work well. 7. A gas mixture analysis system based on artificial neural network is designed, and the primary factors of low precision for gas analysis are investigated. They are the sample quality and the local minimum of network training. To solve these problems, a sample pre-treatment system and improved network training algorithms are proposed. The experiment results show that the precision of the gas analysis have been improved greatly. 8. The partially connected network is used in gas mixture analysis system. It...
Keywords/Search Tags:cross sensitivity, artificial olfactory system, gas sensor array, signal pre-processing, pattern recognition, feed forward neural network, BP algorithm, self- organizing feature map network, fuzzy C-means clustering algorithm, genetic algorithm
PDF Full Text Request
Related items