The Recognition Study On The Combustion States Of Oxyacetylene Flame Based On Hidden Markov Model

The combustion performance is an important factor related to production efficiency, safety and environment protection in the production process. The combustion performance indicators can be measured using special equipment, but with the increasing of the types of measuring equipments, the cost increased, and it is not easy to realize automation. For the oxyacetylene flame is widely used in the industrial production, the automatic monitoring of its combustion states is played more and more attention. If the combustion states of the oxyacetylene can be identified by computer, it would make the auto-control of the combustion process better and the productivity improved. It is needed to realize the classification and identification of the combustion process based on the machine in order to realize auto-monitoring of combustion process. Flame images are the most direct reflection of combustion. First pretreatment is done on flame images, and parameters that represent combustion states are picked up, and then auto-identification is realized, and better effects can be gotten. The combustion process of the oxyacetylene flame is a random one, it obtains good result using the Hidden Markov Model(HMM) which has good imitation ability on the random process. This study is done in this paper.In this study, CCD camera is used as optical sensor, and the oxyacetylene flame images and image sequences gotten from CCD are used as monitoring values. In the pretreatment process of the images, the advantages and disadvantages of traditional methods of filter and segment are analyzed, and AVMF and a new way which segment images in the RGB color channels are proposed, then characteristic parameters that can reflect the characteristics of the oxyacetylene flame images most are picked up.During the study, Artificial Neural Network (ANN),1-d HMMith single characteristic vector and 2-d HMMre used to train the model of the combustion process of oxyacetylene flame, and the identification accuracy and computation speed are analyzed. Based on the conclusion above, a new technique called 1-d and multi-step HMM was proposed to build HMM of the combustion process of oxyacetylene flame, and the characteristic sequences of the images are used as the training data instead of the images. For the image sequences to be identified, their combustion state sequences can be obtained using viterbi algorithm after picking up the characteristic parameter vectors, so the combustion process of oxyacetylene can be classified and identified. The experiment under MATLAB6.5 proves that, according to the characteristics of the oxyacetylene flame image, adopting 1-d and multi-step classification method can cut down time, improve the real-time and the identifying accuracy.