|
|
| Re: How MDBF is Figured (was:Worst NYC subway car MDBF rating ) (432628) | |||
 |
|||
| Home > SubChat | |||
|
[ Read Responses | Post a New Response | Return to the Index ] |
|
||
|
Re: How MDBF is Figured (was:Worst NYC subway car MDBF rating ) |
|
|
Posted by aem7ac on Sat May 19 01:41:43 2007, in response to Re: How MDBF is Figured (was:Worst NYC subway car MDBF rating ), posted by Jeff H. on Sat May 19 01:03:54 2007. The mean distance between failure is NOT calculated by keeping a distance between failure counter on each individual car, and then averaging the readings of those counters. Not to mention that you can't actually feasibly do that. Because the cars are clocking up so few failures, it's almost impossible to get a MMBF figure for one individual car. If the average car fails about twice a year, then monthly MMBF figures would look like nonsense, because some of the cars would not have had a failure that month and you will have divide by zero problems. It is, as TD described, the total car mileage for a fleet divided by the total number of chargeable failures. Statistics theory says it is OK to do this, but I think only if the PDF is gaussian, I don't remember, where's Baumann when you need him? Baumann would either confirm or deny. My sense is that statistics theory doesn't say anything about this because it's three different measures: 1. Averaging all the actual mileages between failure events would give you a number that is the MMBF since the when the mileage counts began, i.e. the expected miles run until failure event for the whole fleet, historically speaking. But this is kind of a useless statistics in terms of indicating how well a shop has done in a given period. 2. Averaging all the miles-ran-since-last-failure would give you a number that is less than the MMBF, and not the expected value of mileages run until failure event. If the mileage-to-failure distribution is random, then averaging the miles-ran-since-last-failure would give you half the value of the MMBF. 3. Taking the mileage ran in a given period and dividing it by failure events in a given period would give you the expected number of mileage-to-failure during that time. That is different from (1) because it only applies to the given period. This is the most useful statistics because it tells you how productive the shop forces are in terms of how many failures are being 'allowed' to happen normalized over the usage of the fleet. But the measure itself has very little to do with the actual mileage-to-failure for each individual car because (a) it only takes into account of failure events and mileages in the specified time period and (b) just because the expected mileage-to-failure is say 200,000 doesn't mean one particular car isn't going to sit down after 40,000 and the next guy isn't going to run for 800,000 miles. MMBF is often driven by a number of bad actors in the fleet. The maintenance manager's job is to identify those bad actors and either fix them or park them. I once read some statistic that found 20% of a certain freight car fleet caused 80% of the train reliability problems. Some freight cars got sold off real fast at a discount! I have some references on this. If anyone is interested email me and I can send you a PDF. AEM7AC |
|
|
|