Reading the Waves

Contents Preview

When the first operational data historians (i.e. process historians) emerged nearly four decades ago to begin replacing pen and paper chart recorders, they focused on relatively slow-changing process variable data from batch and continuous processes. Data from such processes did not typically need to be characterised in the historian with resolution better than every few minutes or seconds. Indeed, while the underlying control loops may have needed to function at sub-second speeds, summary data at 5 or 10 sec. intervals was often considered more than adequate and the concept of breaking the ‘one second data’ barrier became the process industry’s equivalent of the 4 min. mile. Given the requirement for tens, or even hundreds, of thousands of process points at collection speeds measured in seconds or minutes, it is not surprising that process historians focused and evolved along a path of handling a large number of points at intervals measured in seconds rather than milliseconds, sampling speeds for individual points.

At the same time, online software for historising and analysing machinery vibration was evolving along a parallel, but separate, path. Unlike process historian software, its focus was on a much smaller number of total measurement points (or ‘tags’ in process historian parlance) — often fewer than 500 — but at much faster sampling rates. To put this in context, adequately capturing a waveform from a typical vibration sensor requires sampling rates of 40 kHz, or, as Nyquist dictates, 20 000 times faster than 1 sec. of process data without aliasing. An uncompressed, high-fidelity audio signal requires approximately the same sample rates as a typical vibration signal and is, therefore, a useful proxy when thinking of vibration data requirements. A conventional compact disc (600 MB) can hold approximately 60 min. of two-channel (stereo) audio data. Likewise, it can hold approximately 60 min. of uncompressed vibration data from two sensors.