 |
 |
| The New Weibull Handbook |
Weibull analysis is the world's most popular method of analyzing and predicting failures and malfunctions of all types. The method identifies the category of failure: infant mortality, random or wear out. Weibull analysis provides the quantitative information needed for making RCM decisions which are often made from a qualitative approach. Weibull failure predictions are accurate even with very small samples of data. Weibull analysis calculates optimal parts replacement intervals for minimizing cost. Weibull analysis is needed for making risk based inspection decisions to take action or defer action on potential failures. The most accurate warranty claims predictions are based on Weibull analysis. Weibayes, invented by Dr. Abernethy and others at Pratt & Whitney Aircraft, provides the most accurate analysis for small samples and Weibayes substantiation tests use the absolute minimum Wherever there are alternative methods like likelihood ratio versus Fisher Matrix vesus Pivotal confidence methods or maximum likelihood versus median rank regression, best practices are indicated. This is a unique feature of The New Weibull Handbook. Fifth edition: Every chapter has been rewritten to present the state of the art technology for life data analysis. Crow-AMSAA (C-A) has been moved to a new chapter because it has become so useful with the new IEC methods. Dr. Abernethy now considers C-A to be as important as Weibull analysis. C-A predicts future failures, analyzes R&D testing, tracks fleets of repairable systems, and provides the most accurate trending of significant events for management. Goodness of fit and confidence bounds are provided. Both Crow-AMSAA and Weibull may predict either failure events or cost analysis. CA tells us if failure rates are increasing, decreasing, or staying constant with time. CA is the most robust of all our methods. It handles mixtures of failure modes, missing data, batch problems and still produces accurate results. Three methods for improving the accuracy of interval and inspection data are presented and for the first time accuracy comparisons are available. Which method is best; the inspection option, Probit or interval MLE? Best practices are indicated. Bob Rock of PACCAR has updated the logic diagram. It provides guidance step-by-step guidance to the best practice as function of the type of data and the objective of the analysis. It simplifies finding the best method for your analysis. A version in Spanish is available. The concept of "power" of a goodness of fit is presented in a new appendix on the subject. Power is the probability that the parent distribution will be correctly chosen in a distribution analysis. The four methods of goodness of fit can be compared. Warranty analysis has become more and more important. Joe Wauben of Honda authored the new section on warranties. Forecasting by the age of the system and also by calendar month is included. There are new case studies and one of particular interest was contributed by Randy Pirtle of Honeywell on their development of a Weibull library using the SuperSMITH automatic library input. MLE-RBA, reduced bias adjustments for maximum likelihood estimates, for the Weibull, normal and log normal distributions have largely replaced MLE as they provide significant improvements in accuracy. Maximum likelihood estimates can now be made for small samples as accurately as median rank regression. Similarly, the reduced bias adjustment for the likelihood ratio test and the likelihood contour comparisons of life data sets have become the standard for comparing data sets. Is the new design significantly better than the old design based on testing four units of each design? The accelerated test methodology has been extended to produce SN curves, (stress-cycles). Data may be entered at several stress levels and analyzed as a group to produce SN plots. Is that first point or the last point really part of the data set or is it a wild point, an outlier? Is there statistical evidence that a data point is suspect? The analysis should not proceed until the data sets are tested for suspect outliers. For substantiation & reliability testing which design is best, zero failures, zero-one failures or Sudden Death? Which design is best for your situation to minimize test time and cost? Our associate, Paul Barringer, has provided a new section on Life Cycle Costs explaining the concepts like net present value, discount rates and interest rates in simple terms. Extreme value statistics have new applications in forecasting boiler tube and pipeline erosion, metallurgy and wind gust loads using the Gumbel distribution. You can download the Cover, Preface, and Table of Contents from The New Weibull Handbook as a 12 page PDF file--the file size is 183KB. Also, you can download Chapter 1 from The New Weibull Handbook is available as a 11 page PDF file-?the file size is 550KB. This book is used for Weibull Analysis Seminars conducted by Dr. Abernethy and Wes Fulton. The subtitle for The New Weibull Handbook 5th edition is: Reliability & Statistical Analysis for Predicting Life, Safety, Survivability, Risk, Cost and Warranty Claims. The 5th edition is identified by: ISBN 0-9653062-3-2 for the 350 pages of fresh information. The book weighs 1.7 lbs = 0.8 kg and has clear plastic flexible covers. If you own earlier copies of The New Weibull Handbook should you upgrade to the 5th edition?--YES because of the new research findings in most chapters of the book. At this web site, you can review WinSMITH Weibull software for performing the Weibull analysis and WinSMITH Visual software for performing the Crow-AMSAA reliability growth plots which are described in The New Weibull Handbook. Dr. Robert B. Abernethy is the author and publisher of The New Weibull Handbook and welcomes E-mail questions about the material in the Handbook. Dr. Abernethy also offers EagleEYE(SM) service to quickly review your data, using concepts detailed in the handbook, and provide comments on what he observes in your data sets. A book review from the Royal Statistical Society is shown below for the earlier second edition, and a copy of the review can be downloaded as a PDF file. The New Weibull Handbook, is published and distributed by the author for US$98 plus free shipping and handling for domestic addresses, outside the USA the shipping charge is US$20). For pricing and shipping details along with bundled packages of books and software click here. Order The New Weibull Handbook by PayPal using American Express, Visa, or MasterCard. Download the PDF file for the TABLE OF CONTENTS About Dr. Bob Abernethy Dr. Abernethy is THE leading world authority on the principles and application of Weibull analysis for solving failure problems in a number of different fields. Dr. Abernethy was a practicing Mechanical Engineer prior to earning his Ph.D. in Statistics as a Fulbright Scholar to England which eventually involved him in Weibull statistics as the tool of choice for reliability engineers.. Dr. Abernethy (affectionately know around the world as Dr. Bob) is the retired Chief Statistician for Pratt & Whitney Aircraft. At Pratt & Whitney Aircraft, he gained practical use of Weibull techniques for solving reliability problems. He is given credit for inventing Weibayes, failure forecasting, zero-failure substantiation testing, the MLE Reduced Bias Adjustment, the Likelihood Contour Test and many studies to determine best practices. Dr. Bob has over a half century of experience with Weibull analysis. He was the manager of reliability, safety, maintainability and statistical analysis of Pratt & Whitney's military engine division. He holds the patent on the J58 engine cycle used in the SR-71 "Blackbird" supersonic spy plane that still holds all seventeen world speed records. He also has the patent on the afterburner control system used in the F15 and F16 aircraft. He was the principal author and project manager for the US Air Force Weibull Analysis Handbook, the original edition of The New Weibull Handbook. Since his early retirement from Pratt & Whitney in 1987, he has conducted extensive research to develop new and improved Weibull methods as in the Handbook. Dr. Bob's experiences as a consultant include medical devices, automotive, electrical/electronic equipment, and nuclear reactor components as well as aerospace. His client list includes NASA and many of the major companies around the world. He is the foremost lecturer in Weibull analysis worldwide. He is a fellow of The Royal Statistical Society, SAE, ASQ, ASME and Associate Fellow of the American Institute of Aeronautics and Astronautics. For advanced technical questions about Dr. Bob's book or queries about his consulting or training schedules you can reach him by e-mail at Dr. Bob. Dr. Abernethy also offers EagleEYE(SM) service to quickly review your data, using concepts detailed in the handbook, and provide comments on what he observes in your data sets. The New Weibull Handbook is the reference text for theory and methodology for the SuperSMITH software written by Fulton Findings for solving reliability on a personal computer. The SuperSMITH software packages include WinSMITH Weibull and WinSMITH Visual as Windows programs. More details on software prices and features are available from the SuperSMITH web page. A tutorial book call PlayTIME with SuperSMITH is available. The tutorial has step-by-step instructions for solving over thirty problems illustrating all the features and options in the software. This tutorial with The New Weibull Handbook, the New Methods CD, and the SuperSMITH software from Fulton Findings provides a complete home study package for engineers and scientist. The tutorial includes a diskette with the PlayTIME data-sets to avoid the drudgery of hand input except for a few examples to illustrate input methods. The New Weibull Handbook undergoes continuous improvement. Each printing is substantially improved based on the latest findings of Dr. Bob's research. The original USAF Weibull Analysis Handbook published in 1983 set the standard for the analysis of reliability data. The New Weibull Handbook undergoes continuous improvement. Each printing is improved based on the latest findings of Dr. Bob's research. Where several methods are available, best practice is indicated. The first edition of The New Weibull Handbook was published in 1993. It had two printings and represented a substantial upgrade from the original Weibull Analysis Handbook. The second edition of The New Weibull Handbook was published in 1996. It had four printings in 1996, four printings in 1997, and two printings in 1998. A review of this edition follows below. The following book review was described in the Journal of the Royal Statistical Society/ A (1997) 160, Part 2. pp 367-380. Permission was given on October 1, 2001 via email to Dr. Robert B. Abernethy to publish this review on the Internet by Martin Owen (m.owen@rss.org.uk), Executive Editor of the Royal Statistical Society, 12 Errol Street, London, EC1Y 8LX, UK http://www.rss.org.uk and by Malcom A.J. Allison (Malcolm.Allison@blacksci.co.uk), Permission Assistant, Blackwell Publishing, Osney Mead, Oxford OX2 0EL, UK http://www.blackwellpublishing.com. The original copyright was 1997 by the Royal Statistical Society. [Hyperlinks have been added to the review for use as Internet pages at http://www.barringer1.com] The New Weibull Handbook, 2nd edn. By R. B. Abernethy. ISBN 0 9653062 0 8. Abernethy, North Palm Beach, 1996. $88.00. [Please note, The New Weibull Handbook is now at the 5th edition. The price is US$98 + shipping and handling] The handbook is used as a manual for training by the author who took early retirement in 1987 to undertake full-time lecturing in Weibull analysis. The print is in a large fount and the equations are in even larger fount so that those of us whose sight is poor will have no excuse for not absorbing the message. There is no doubting the author?s enthusiasm for the Weibull distribution and indeed, as I read on, I found myself influenced increasingly by the author?s persuasiveness. Some of the claims made for its predictive ability for small samples may seem extravagant but, in this respect, I think that we must bow to the wealth of experience gained by Dr. Abernethy over many years. There is a bewildering variety of software from Fulton Findings to accompany the handbook, namely WeibullSMITH™, VisualSMITH™, BiWeibullSMITH™ and MonteCarloSMITH™, none of which were presented with the manual for review. True, other life distributions are mentioned such as the log-normal but, in the context of the handbook, they are subservient to the Weibull. Probability plots are strongly emphasized and sufficient introductory theory is relegated to appendixes. The plotting position recommended, particularly for skew distributions, is the median plotting position. Not having pursued the merits of one plotting position versus another I fail to understand the logic given for this but the author does quote a remark made by Wayne Nelson that such arguments are just as fruitless as arguing religions since they all point to heaven. Both nonparametric and parametric methods are explained and well illustrated. These include the Kaplan-Meier, maximum likelihood and likelihood ration approaches. Confidence intervals for parameters and for quantiles such as the B10 life are calculated but the author is not in favour of their use in presentations to management because they are so little understood (in this respect he quotes a 'Forthsooth' from RSS News in 1992). I agree with his opinion on this but I take strong exception to his answer to one of the question from readers where he states that even most statisticians do no understand the confidence concept. However, it may be true that many who profess to be statisticians do not understand the concept. Particularly in the field of reliability there is much confusion and lack of understanding of classical statistics as enshrined in military standards. Engineers without previous exposure to statistics more readily take to the Bayesian approach. To a certain extent this is catered for by WeiBayes in the handbook, but it is not a full Bayesian approach since a single value is established for the shape parameter on the basis of past experience. There is no natural conjugacy for the Weibull distribution but there is gamma conjugacy for a scale parameter conditional on discrete prior values for the shape parameter (Martz and Waller, 1982). This should not be at all difficult for engineers to assimilate, especially when displays of priors and posteriors are easy to produce. The difficulties in estimation for the Weibull distribution with a threshold parameter t0 are not sufficiently emphasized. The technique of subtracting values from failure times and suspension times until the Weibull plot 'straightens' is easier said than done since substantially different values for t0 can make very little difference to a straightened plot. Another manifestation of this difficulty is that the likelihood function is quite flat over wide ranges of t0. There is a section on reliability growth and the Duane model written by D. P. Weber, a reliability consultant. My only criticism of this section is in his use of the term instantaneous failure rate. There is thus no distinction between its use here and its previous use for non-repairable items. Many researchers now use the rate of occurrence of failures for the repairable case. I detected few errors in the handbook. On p.2-10 the author states that as time progresses the reliability of components surviving infant mortality increases. To be precise reliability should be replaced by conditional reliability. On p. 2-22 the Weibull closure, or self-locking property, is the reciprocal of the correct answer and in Appendix G-23 the second moment about the origin is incorrect and is also incorrectly signified by µ2 rather than by µ'2. There are a few inconsistencies such as the use of both s2 and σ2 to denote a theoretical variance and the author talks about inverse ranks when he means reverse ranks. In summary the handbook, together with the previously mentioned software (or other), will be an invaluable aid to reliability engineers. Statisticians would probably prefer a more concise theoretical treatment with more sophisticated applications such as Weibull regression models. As a statistician who is interested in the application of statistical methods in industry I found the many diverse applications and case-studies extremely interesting. Reference Alan Winterbottom The 5th edition of The New Weibull Handbook is now available with the latest findings from the Weibull community of experts around the world—shipments begin on November 1, 2006. You can download the cover, preface, and table of contents as a PDF file, and Chapter 1 is also available as a PDF file. Order The New Weibull Handbook by PayPal with a credit card or by sending a check to the address shown on the home page of this web site. All prices are subject to change without notice. Domestic shipping & handling is free if you mention this website. Overseas shipping is half price, $20.00 instead of $40.00. |
 |