Research On AOD Furnace Splash Forecasting And Measurement Analyzer Based On Multi-sensor Fusion

In the process of iron and steel melting,the splashing occurrence is the disease in iron melting industry. Especially it will endanger human’s life and property safety, also damage environment hardly, and cause serious economic loss. As the pillar industry of our nation, it faces the serious actual problem that huge production cost, excessive capacity, highly pollution. Therefore the study of splash forecast analysis technology is bound to relieving the above-mentioned problems.In this paper we focus on research on AOD furnace splash forecasting analyzer based on multi-sensor fusion method. This paper analysis the characteristic of time-frequency about the sound, vibration, flame image signal which was acquired when the splashing occurrence. On the end the paper put forward a method which is based on character layer multi-sensor fusion and used the BP neural network algorithm. And through large of exercise we correct the data used by forecasting, we prove this method is useful to complete the forecasting task accurately. Finally I design a splashing forecast platform based on Lab VIEW software, provides a platform for analysis and research.First of all, I confirm the fusion structure of character fusion layer. Then analyse the method about drawing the right splashing forecast information. For the one-dimensional signal like sound and vibration, I divided it by the normal and splashing signal. And analyzed the FFT spectrum on them firstly, to find the difference of frequency on their whole signal between them. Then analyzed the time domain wavelet packet frequency spectrum,to find their time-frequency character; Secondly, according the analysis above selected the frequency range which have the obvious difference between splashing and normal signal. And used the db10 wavelet packet to decompose the two signal on three and four layers. Analyzed each packet energy percentage and FFT analysis further. According all the analysis results confirmed a set of vector which could fully distinguish the normal and splashing signal. For the three different signals I select different eigenvalues through large amount of experiment. Eventually using similarity method got three thresholds to divide out the splashing signal.For the flame image signal, Using the image recognition technology, similar to one dimensional signal analysis method. Eventually the eigenvalues I had extracted can complete the splashing forecast task accurately.Then according the extraction of splashing eigenvalue complete the fusion algorithm research. Finally determine to use the fusion algorithm based on BP neural network algorithm. The forecast level can be divided into five. The four splash forecast values of characteristic signal make up of the input elements, and according to actual splashing seriousness, the output elements is divided into five kinds. Eventually according to a kind of BP network to train the network, and test it, the result shows that the network I constructed can complete the task of splash forecast.At the end of this paper I design a splashing forecast software interface designed by the LabVIEW software, and introduce the main functions of the prediction platform, including data acquisition, storage, waveform playback and real-time forecast, and so on. Eventually this article achieve the purpose of the thesis research, provides a software platform for splash forecast and analysis.