Sensitivity Study on Quantitative Methods by Analysis of Pressure Fluctuations in a Fluidized Bed

Abstract

The sensitivity and reliability of 23 data processing methods are studied by analyzing pressure fluctuation time series in a fluidized bed. First, the concept of minimal stationary segment length is proposed to study signal stationarity. Second, three indexes estimating the sensitivity of the methods are proposed and their algorithms are designed. The sensitivity problems of some methods are studied. The results show that the root causes of the SKEW (skewness) method sensitive to segment length include outliers and a small value as the denominator, and the STD (standard deviation) method is sensitive to data normalization because it depends on the range of time series. Finally, a common framework is proposed based on the three sensitivity indexes and harnessed to evaluate the 23 methods. The two most reliable methods are picked and proved appropriate for identifying the flow regimes. Additionally, the result also indicates that the time-domain methods have better reliability than frequency-domain and state-space methods.