Much has been said as to the “key importance” of data from past data collections and how much this data should be kept for future reference. Is this really the case? What, in fact, is of primordial importance, and what might simply be done away with, given a little reflection on the matter?
Most vibration data becomes quickly irrelevant, UNLESS it contains pertinent traces or indications of a developing problem. A ceaseless repetition of waveforms and spectra showing minute and harmless change simply clutters a database, but most PdM software fails to adopt a design enabling quick and accurate management of such issues.
We want baselines. We want them to be accurate not only in terms of resolution, but also in terms of machine operation (load, temp, product, etc.) documentation. Multiple baselines helps us considerably, inasmuch as we may then assign different operating-conditions or perhaps post-maintenance (repair, replacement, etc.) “slots” to these reference measurements. We probably want these to be accessible in multiple parameters. Baselines are, so to speak, a “keeper” in the context of this article.
Regular, run-of-the-mill, “no defect found” data should be considered as having a limited shelf life. Identical spectra within a waterfall (or cascade) take up space for no good reason. They do, however, provide continuance of a pattern, which could easily be managed as a “replicator” function once time proves that no review is forthcoming (reviewing or self-auditing of data should occur when a failure occurs in spite of no defect having been detected by the predictive measurements, their programming and/or the analyst’s assessment of such data). Any changes to the measurement set-up should be documented as well, particularly if a change arises from the conclusions brought about by an investigation in an undetected developing failure or “crash”.
A predictive maintenance program accumulates a knowledge-base that seemingly, and quite sadly, evaporates over time. The principal cause for the volatility of this knowledge rests in the inability to “grab” a data set from a specfic date (or successive dates when trends are relevant) and form a diagnostic subset. In fact, it has long been my recommendation to our clients to keep a book of problems detected (and a hardcopy of the pertinent data) on monitored assets, simply because it becomes a critical reference for the future and a BRIDGE to pass the program over to somebody else if and when the analyst is ever called away from this condition-monitoring posting.
The answer to any question implicating old data relevance becomes obvious. Measurement set-ups, baselines, measurement reviews, problem identification documentation form the basis for a reference book that will over the years become a fabulous treasury, a fountain of wisdom for both the technicians and analysts involved in PdM work and management. Except for the previous and recent, still believed to be time-relevant to potential review, data, all the other accumulating data creates unnecessary and unwanted clutter.
© 2007 by François Gagnon