Most products are built to specifications that include some measure of reliability. In the design stage, FEA does a good job of pinpointing the weak spots in parts and assemblies. But the real test comes from field trials when a select set of users get their hands on it.

Users enter data into Weibull Easy through a simple table. F refers to a test that failed and S to one that was suspended or ended without failure. Hitting Calculate or Weibull produces the accompanying plot. |

A plot comes from the input data. The shape factor of 1.3 indicates a nonexponential curve shape. The blue graphed lines (on either side of the red line) are the 5% and 95% confidence levels. The table to the left contains numerical values for the curves. |

The table to the left may be the most useful bit of information for this exercise. The predicted failure rate per year might indicate that a product redesign is in order, or suggest handling failures through a product warranty. |

Even with results from a small sample, Weibull Easy can tell volumes about the product's durability. The software plots a Weibull curve through the data, performs several statistical analyses, and presents a range of statistical indicators.

The software comes with a simple example showing how to input data, so the user manual is almost superfluous. For example, field or lab tests will provide data such as how many hours each prototype or production unit lasted. Even testing a dozen or fewer units, according to developer John Berner, can be quite telling.

A Weibull data distribution is a curve fit of cumulative failures versus hours, cycles, or some other variable that serves as a measure of product or equipment life. Berner says the Weibull acts as a sort of chameleon distribution and takes the shape of the data rather than attempting to force fit the data to a standard distribution, such as exponential, binomial, or normal.Users would input data into a table like this: Unit 1, 140 hr, failure or not, comments, Unit 2, 200 hr, and so on. The unit of measure could be almost any reasonable value such as days, cycles, or trips. Then hit the Weibull button.

The software returns the Weibull capability distribution which indicates the strength of the design in a graph of the percent failed versus hours to failure. This is curious information that gets better with another calculation. The marketing department should have information on the expected use of the product. For example, if broken into quarters or quartiles, the data might say that one-fourth of the users will run the product for 20 hr a year, another quarter for 40 hr/yr, and so on.

Combining this information with the failure rate from tests produces the Combined distributions. A little table on this page tells, for one thing, that about 5% of the devices will fail in the first year, a cumulative 14% will fail by the second year, and so on. Engineering and marketing can then decide whether or not the results are acceptable. If not, then a redesign is in order.

The software has a range of other functions such as different statistical curves for the same data. The shape factors tell a bit more about the data. "For example, a Weibull shape factor of 1.0 represents an exponential distribution," says Berner. "Shape factors correspond to the slope of the regression line on a Weibull plot. If data you've assumed to be distributed exponentially comes out with a Weibull-shape factor of about 3.0, you'd best revisit the assumption. A Weibull shape factor of about 3.25 or above represents an approximately normal distribution."

The software also generates a summary report that includes most of the information in the three accompanying images. Users can download a 30-day free demo from the developer's Web site. The $398 software comes from Applications Research Inc., 4927 St. Croix Ave. N, Golden Valley, MN 55422, (763) 521-0217, www.applicationsresearch.com

*--Paul Dvorak*