Abstract: In view of the nonstationary feature of the pressure difference fluctuation signal in the gasliquid twophase flow, a flow pattern identification method was proposed based on the characteristic vector from the singular value of the matrix formed by the multi-dimensional continuous wavelet transform values of the fluctuation signal. The continuous wavelet transform was applied to the pressure difference fluctuation signal in the gasliquid two-phase flow to form the initial characteristic vector matrix, from which the singular value of the matrix could be obtained through the singular value decomposition. The decomposed singular value may serve as the flow pattern characteristic vector and the input to a radial basis function neural network(RBFNN) to realize an intelligent identification of the flow pattern. The proposed identification method can precisely identify the four flow patterns of the air-water twophase flow in a horizontal pipe, providing an effective new method for the flow pattern identification.

Key words: fluid-dynamics, gas-liquid two-phase flow, flow regimes recognition, continuous wavelet transform, singular value decomposition, radial basis function neural network