The figure schematically depicts a flow-measurement system based on an acoustic sensor mounted on a pipe, along which flows a single or two-phase fluid. The electrical output of the sensor is processed by an electronic subsystem to estimate various parameters of the flow.
The sensor output is first amplified, then sent through a narrow-band-pass filter with a pass frequency approximately equal to the resonance frequency of the sensor. (A typical acoustic flow sensor has a resonance frequency ≈83 kHz). For the purpose of digital processing to completely characterize the filtered signal, it would ordinarily be necessary to sample and process the signal at a rate of about four times the pass frequency. However, assuming that the sensor resonance is not highly damped, useful data about the acoustic excitation (and thus about the flow that causes the acoustic excitation) can be extracted by measuring the relatively slowly varying amplitude envelope of the filter output.
Accordingly, the output of the filter is demodulated to obtain the amplitude envelope. Because of the relatively slow variation of the amplitude envelope, it is possible to extract sufficient information by digitizing and processing the output of the demodulator at a lower sampling rate (as low as about 33 kHz), which can be implemented much more easily. The digital samples are fed to a computer, which executes special-purpose software to perform a combination of statistical and neural-network processing to obtain flow parameters. The sampling rate is low enough to enable the computer to process the samples in real time.
The statistical flow-indicator quantities calculated by the computer can include the following:
- The average and the standard deviation of the signal amplitude,
- The characteristic autocorrelation time of the variation in the amplitude envelope,
- The average of the absolute value of the difference between successive samples,
- The root mean square of the difference between successive samples,
- The average and the root mean square of the interval between instants when the instantaneous amplitude crosses from below to above the average amplitude, and
- The average and the root mean square of the time between an upward and a downward crossing.
These statistical flow-indicator quantities are fed to a neural network, which calculates the mass flow rate and other flow parameters of interest. A neural network is used for this purpose because of the inherent ability of a neural network to learn nonlinear correlations that involve abrupt transitions, strongly varying behaviors, or complicated interactions among input variables, even in the absence of a mathematical model of the relationships among the input and output variables. As in the cases of other neural networks, this neural network is programmed by training it with numerous sets of examples of inputs (the statistical flow-indicator quantities) and the corresponding outputs (e.g., the mass flow rate in single-phase flow, the presence or absence of a second phase, and separate mass flow rates of liquid and gas in the case of two-phase flow). The analytical methods are described in U.S. patents 5,600,073, 5,717,691, and 5,741,980.
This work was done by Wayne S. Hill and Bruce N. Barck of Foster-Miller, Inc., for Stennis Space Center. For further information, access the Technical Support Package (TSP) free on-line at www.nasatech.com/tsp the Electronics & Computers category.
In accordance with Public Law 96-517, the contractor has elected to retain title to this invention. Inquiries concerning rights for its commercial use should be addressed to:
350 Second Ave.
Waltham, MA 02451-1196
Refer to SSC-00053, volume and number of this NASA Tech Briefs issue, and the page number.