SRC Forum - Message Replies
Forum: Reliability & Maintainability Questions and Answers
Topic: Reliability & Maintainability Questions and Answers
Topic Posted by: Reliability & Maintainability Forum
Organization: System Reliability Center
Date Posted: Mon Aug 31 12:47:36 US/Eastern 1998
Posted by: John
Date posted: Thu May 31 23:40:56 US/Eastern 2001
Subject: Field Data
When analyzing field data for specific part failures (like an electric motor)in a large population - Do you use time to first failure for unique units and then do Weibull plots? Do you ignore any follow up failures of this part in these units? The sample size would be several hundred units.
I would like to use this information in determining base failure rates for future reliability predictions.
Subject: Field Data Analysis
Reply Posted by: Patrick Hetherington
Date Posted: Fri Jun 1 7:15:38 US/Eastern 2001
It really depends on the purpose of the field data analysis. Using Weibull (or other distribution analysis such as normal, lognormal, exponential, etc) the items are being treated as a non-repairable item. Looking at the times to first failure is the only applicable method of Weibull analysis for your parts (i.e. electric motors). This will tell the failure pattern and thus failure distribution of the part as a whole (manufacturing defects, random, wear out). Remember to treat non-failed units as suspensions. This may not be enough information to determine how to improve the reliability of the part, but will provide information to the contribution or degradation to overall system reliability based the part. Separating the time to first failure by failure mode and doing a Weibull analysis on each data set will provide further insight to the greatest areas for improvement. If you were looking for trends in a repairable system, applying the Non-homogenous Poisson Process would be applicable. If you would like us to look at the data and make recommendations, please feel free to e-mail me directly.
Subject: Estimate renewal process
Reply Posted by: Larry George
Organization: Problem Solving Tools
Date Posted: Tue Jun 5 13:44:11 US/Eastern 2001
Consider modeling a sequence of repairs as a renewal process, modified renewal process, or a sequence of independent non-identically distributed lives. These models are more general than a Poisson process, even nonstationary Poisson process. The more general the model, the more likely it might be right. On the other hand, the more general the model, the less precise the estimates will be.
http://web.utk.edu/~asaqp/newsletters/1299newsletter.pdf page 13 shows the age-specific return rate estimates for first warranty return of 1988 Ford V-8 460 drivetrains. It was modeled as a modified renewal process. Subsequent times between warranty returns had similar but not the same distributions indicating problems weren't fixed.
http://members.home.net/pstlarry/tired/frame.htm shows Firestone tire reliabilities. Some of the complaints data didn't specify whether tires were original or replacement. Assuming a renewal model led to estimates that showed Firestone tire reliability didn't differ depending on whether tire was OEM or replacement. That presentation briefly describes the max likelihood estimator with unknown renewal counts.
Generally accepted accounting principles require ships and returns data, which are sufficient even to estimate different age-specific field reliabilities for new or repairs; e.g. http://web.utk.edu/newsletter/1299newsletter.pdf. The input data was monthly production and warranty returns per 1000. Send data for free samples to email@example.com or enter it into http://members.home.net/pstlarry/Table.htm.