A good sample should include both good and bad actors. For example, pumps that fail frequently as well as those that do not fail or fail infrequently should be included. This ensures an unbiased sample. Normally the analyst takes two points in calendar time a few to several years apart. Then within that time window he selects all the equipment units that are similar for the purposes of reliability analysis.
The analysis will attempt to discover correlation between failure and candidate condition indicators. These might include vibration amplitudes in various frequency bands, start-stop frequency, oil analysis results, operational factors, seasonal factors, or whatever else could be relevant.
A sample reflects the failure behavior of a physical asset in a sufficiently wide calendar period, usually several years. Applying such a time window of interest and other filtering criteria, the reliability engineer generates a sample in the form of an “Events table”. The sample generation process is illustrated in the following slide presentation.
In performing reliability analysis, the analyst attempts to “fit” the data in the Events table and the historical CBM data to a Proportional Hazard Model (PHM). A PHM is an equation that relates failure probability to working age and to CBM variables likely to be useful. The result of this model fitting exercise is a list of CBM variables relevant to a failure mode and their precise relationship to failure probability.
The PHM is then combined with business data to build a predictive optimal decision model. This model’s purpose is to support maintenance decisions by assessing the latest CBM data as it arrives in day-to-day operation. The model predicts when the asset will enter a potential failure state.
Software tracks actual work order observations in order to assess predictive performance. This is necessary for continuous improvement. Model performance indicators will include the standard deviation (i.e. predictive confidence) as well as changes in cost, availability, and profitability associated with the occurrance of each modeled failure mode.
© 2011 – 2016, Murray Wiseman. All rights reserved.
[…] η are the shape and scale parameters. To use a reliability model we require a “good” sample. A sample is a set records of instances of failure modes, their ages at the time of occurrence, and […]
[…] Extract a sample of age data covering a period of 3 or 4 or more years. The format of the data should be as […]