TESPAR HELPS YOU SEE IN THE DARK

It could be `the key to commercial DSP’. It is almost 40 years since the first seeds were sewn which would later bear fruit with the development of TESPAR – Time Encoded Signal Processing And Recognition. Now the `new’ digital language for describing waveforms has a multiplicity of applications ranging from condition monitoring to security. […]

It could be `the key to commercial DSP’.

It is almost 40 years since the first seeds were sewn which would later bear fruit with the development of TESPAR – Time Encoded Signal Processing And Recognition. Now the `new’ digital language for describing waveforms has a multiplicity of applications ranging from condition monitoring to security.

The simple but elegant algorithm developed by a military scientist is capable of a level of signal analysis integrity formerly only dreamed of.

To process control, it already offers significant improvements in the analysis of data captured from condition monitoring regimes.

TESPAR is cheap and easy to put onto silicon, or to incorporate in microchip programming.

In sharp contrast with the more conventional approach of linear transformations based on `amplitude’ sampling at regular intervals, TESPAR is a time domain process.

It was sight of a 1948 technical paper by Licklider and Pollack that led to Professor Reg King, then head of the MCRU at Shrivenham, and now chairman of Domain Dynamics, to develop TESPAR.

It showed that by amplifying a radio signal to cause signal clipping, the speech intelligibility of the signal improved significantly.

Now the technology offers startling performance and commercial advantages to a whole range of condition monitoring and diagnostic applications previously considered uneconomic or infeasible, according to Martin George, marketing manager of Domain, the company which now holds patents in the field.

So what does it mean for condition monitoring? Domain says its key benefits are the ability to pinpoint specific component failures while requiring only a small amount of data processing.

`For parameters such as temperature and pressure, such interpretation can be quite straightforward. For parameters such as vibration and noise from rotating and reciprocating machinery, the problem is more difficult.

Large scale plant and process monitoring systems have the computing capability for the complex analysis and interpretation algorithms needed to extract meaningful and reliable data from noisy signals, but portable instruments are somewhat less capable in this respect and can suffer short battery life.

What is more, signals from sensors monitoring plant such as gearboxes, bearings, engines and compressors consist of many superimposed elements that originate from varying rotational speed, gear teeth meshing, or valve reciprocation. The result is a highly complex signal, usually accompanied by wideband noise from mechanical impacts and surface slip.

Typically, Fourier – frequency domain – analysis of these signals has been used to extract sets of spectral features or signatures associated with a particular condition, and this has been the case for some 50 years.

The process’s limitations, include a rather lengthy, tedious procedure for extracting reference signatures, high computational complexity, and a vulnerability to the effects of noise in the signals.

Domain’s patented approach to event classification involves taking a `snapshot’ of a signal emission, and combines novel TESPAR waveform coding procedures and Fast Artificial Neural Networks (FANN), developed initially at the University of Bath and latterly at the RMCS Shrivenham campus of Cranfield University.

Trials confirm a number of performance advantages over frequency domain based methods:

* reduced number of false alarms

* suitability of low-cost piezo-type sensors

* typically two orders of magnitude less computer power required

* suitable for use with portable instruments, smart sensors and parallel architectures.