Home Health Case Studies in Maintenance and Reliability: A Wealth of Best Practices
Fire Water Pumps
If a fire pump fails to start, it can have serious consequences. Laws or insurance regulations will often prescribe a test frequency. In many cases, these tests are done fortnightly, so each pump is started at least 26 times a year. Failure to start at first attempt is what matters. There will be plenty of data points, so computing failure to start MTBFs for these items should be fairly easy. However, this data for computing MTBFs is not usually available in the CMMS, and has to be obtained from the operating log.
Evident Failures—Weibull Parameters
So far, we have considered hidden failures. The bulk of the recorded data relates to evident failures, such as those of bearings, seals, and couplings. In our case, there were about 900,000 records to examine. If these had been on paper, the stack would have been nearly 300 feet tall! It would have taken a person 4 to 5 years to analyze such a large volume of data. There would be many errors, especially if we put in many analysts to speed up the work. The analysis would in any case be out of date by the time it was complete. We had to look for a more efficient solution. Since the data was available electronically, a software-based search appeared viable.
In order to understand the method used in searching for the applicable work orders, some of the vagaries involved in data entry have to be understood. Typically, we may find one or more of the following difficulties in searching in the history.
We found data mining software that had a built-in lexicon and used context sensitive searches. It could identify word order, such as oil seal vs. seal oil or words which were spaced differently but meant the same thing such as 'bearing seized' and 'bearing examined and found seized'. It could recognize synonyms and homonyms. It had error-forgiving rules built in, so that it recognized P 1234, P-1234 and P:1234 as the same item. With all this sophistication, we could locate the work orders pertaining to a single item or of a group of items—such as pumps—rapidly and sort them out by failure modes. All this could be done in days rather than years, and repeat searches produced fairly consistent results. About 35,000 failure data points were identified, but less than 20% were useable. The rest were about failures that were trivial or otherwise uninteresting.
Reliability engineers have found that many failures are distributed according to a distribution called Weibull, named after a Swedish mathematician who introduced it. The distribution itself allows us to approximate many other common distributions by changing the values of some of the parameters in the Weibull equation. We found good Weibull software fairly easily. Once the failure data was entered, we could get nearly 1000 sets of Weibull parameters.
|< Prev||CONTENTS||Next >|