Back to : msc04.pdf | Home

HAVING CONFIDENCE IN SPECIFICATIONS

David Deaver

Fluke Corporation

PO Box 9090

Everett, WA 98206

425-446-6434

[email protected]

Abstract: This paper presents some of the unique specification would have to be assigned

challenges in creating instrument to each instrument, taking into account its actual

specifications from a manufacturer's performance and the operating environment. In

perspective and some of the difficulties practice, this is done by many laboratories when

using specifications as an estimate of characterizing a standard but is seldom done by

uncertainty from a user's perspective. the manufacturer.

Despite the difficulties, specifications are the

most common means of determining the Manufacturing Yield Targets

Type B uncertainty for the standard when

making a GUM (Guide to Uncertainty of Though specifications may be stated with a

Measurement) compliant uncertainty confidence, typically between 95% and 99.7%,

analysis. An explanation is presented for the all the points tested on the production line must

practice of providing 95% and 99% be measured within the specification limits

confidence specifications, warranting only before the product can be shipped. For a

the 99% specifications, and verifying the 99% standard resistor, whose value is specified at a

specifications to 95% confidence. 95% confidence level to be within limits centered

about the nominal value for the resistor, we

THE BASIS OF SPECIFICATIONS could expect to have to re-trim 5% of the

standards before they could be shipped.

Simply stated, specifications are an implied Specifying a complex instrument at the true 95%

contract between a buyer and seller of a piece of confidence level for each point would be a

equipment. The seller expects to be paid if the manufacturing disaster. For example, each

product performs within its specifications and the Fluke Model 5520A Multiproduct Calibrator is

buyer expects to receive an instrument which tested at 552 points on the production line prior

lives up to the promises made in the data sheet. to shipment. If each of the points has a 95%

Specifications, however, may or may not be a probability of being found in tolerance, there

552

very good representation of the product's actual would only be a 0.95 = 0.00000000051%

performance. The manufacturer considers a chance of finding all the points within the

number of factors when establishing the specification limits if the points are independent!

specifications: Even if we estimate 100 independent points

(about 2 per range for each function), we would

100

Actual Instrument Performance still have only a 0.95 = 0.6% chance of being

able to ship the product. Figure 1 shows the

One would hope there is some correlation probability of an instrument passing all points

between the actual performance of an tested if the points are independent and if all

instrument and its specifications. Ideally, the points have normal probability density

specification would correspond exactly to the distributions with the same confidence interval,

uncertainty. To approach the ideal, however, a 95%, 99%, or 99.7%.

Competitive and Market Needs

1E+00

99.7%

1E-01

If there is not a compelling need to make

1E-02

specifications as tight as absolutely possible,

1E-03

99% there is considerable benefit to the manufacturer

1E-04

to keep them loose in order to enjoy high yields

1E-05

95%

and low warranty costs. Similarly, for the user, it

1E-06 is nice to have higher confidence the unit is

1E-07 performing within the specification limits and

1E-08 0 200 400 600 800 1000

enjoy long calibration intervals. Often, however,

1E-09 manufacturers have specsmanship battles with

1E-10

Num ber of Independent Points Tested (N)

their competitors, each wanting to have the

better specifications. And users ask for tighter

Fig. 1 Probability of all N Points In-Tolerance specs on test equipment to maintain low

uncertainties and high TURs (Test Uncertainty

For our example of 100 independent points, on

Ratios) for their calibrations. As we have seen

average, we would have to design the individual

however, along with tighter specifications comes

points to a 99.95% confidence level to have 95%

the adverse impact on the yields, warranty

confidence that the instrument would pass the

costs, and percent-in-tolerance numbers.

complete verification test. Table 1 lists the

confidence levels for individual points (P)

required for an overall instrument yield (Y) of Environmental Conditions

95%, 99% and 99.7%.

Specifications for an instrument must take into

N Y=95% Y=99% Y=99.7% account its operating environment. Though most

instruments are specified for a broad

1 0.95 0.99 0.997 temperature range such as 0-50