Applications of neural networks for transducer calibration and signal processing of transducer data containing periodic interference

Abstract

Two practical industrial neural network applications are presented. A method of calibrating quartz transducers is described in which a neural network is used to accomplish a dynamic calibration of a transducer pair for a range of temperatures and pressures. In order to compensate for the undesirable transient thermal effects on the pressure transducer performance, tapped delay lines containing previous and subsequent temperature and pressure sensor outputs are used as inputs to the neural network. This method allows the time-dependent qualities of the quartz-oscillator transducer to be included in the calibration scheme. A second application in which real-time adaptive neural network filtering is used to enhance the detection of subsurface explosives detonation in an oilwell is also presented. In the oilfield, accelerometers are used to monitor the surface structure of an oilwell to detect the detonation of subterranean explosives which are used to perforate well casings and formations so oil production from the perforated zones may commence. The method presented shows how neural network filtering may be used to dramatically improve the detection of weak subsurface signals in the presence of heavy noise contamination in the monitored signal.