You are on page 1of 2

Progressive Search 1. Families of cause variation, family with the highest variation has the dominant cause.

Within piece magnet manufacturer Piece to piece within the same mfg Time to time day to day Cavity to cavity n/a Machine to machine Reynosa to Springfield Supplier to Supplier JMQ / Jaingman 2. Variation There is no special or common cause spikes in data. The strategy is to look for causes with a large effect. It can be hard to identify a dominant cause if there is an interaction between varying inputs in different families. There is a risk of focusing on fliers/outliers 3. The Shainin Process - It is innately sequential in nature Stage 1 - Diagnostic search is first in the process i. First you should establish a baseline to quantify the problem, taking into account all the variation ii. The baseline is used to verify we have selected the dominant cause Stage 2 - Establish an effective measurement system i. Need to consider not only the output measurement system but the input measurement systems. Stage 3 Progressive Search i. Tools Multi vary studies, stratification, group comparison(chi sq), and scatter correlation plots ii. Setup observational plans early in the stages of the search iii. The dominant cause is eliminated or reduced by changing the product design, the process, or the control plan. Stage 4 & 5 Confirmation of dominant cause i. DOE is used to confirm because of the possibility of confounding data ii. The levels of dominant causes (red x) are set at the extremes, to produce the BOB & WOW parts. Other inputs are held constant during the DOE iii. Full factorial design is recommended so that interactions among the suspects can be identified. iv. With a single suspect, SS recommends a six run experiment, with 3 replicates for each level. v. Full factorial is feasible since we have a short list of suspects vi. Tolerance Parallelogram(linear regression) is used to set process control limits. 4. Shainin Tools Isoplot is used to compare the relative size of the process & measurement system families of variation. i. Make 2 measurements on the same part, 30 parts ii. If the points lie on the 45 degree line then the variation is small

iii. AIAG recommends using ANOVA to evaluate the variation of the 2 components. Multivari i. Using shaft diameter 1. measure 3 places on the part within part variation 2. using 3 parts in a row to evaluate part-to-part variation 3. sampling every hour to capture time-to-time 4. evaluating 3 different machines for machine-to-machine variation Component Search & Variable Search i. Take the BOB & WOW with output values at the two extremes of the baseline distribution. ii. Units are disassembled & reassembled, evaluating assembly process iii. Units are disassembled & reassembled, evaluating components (swap) iv. Variable search is used when 4 or more suspects are identified Group Comparison for destructive testing i. We measure the inputs & find the extremes for each component by either continuous measurement or extreme binary inputs ii. Recommended analysis uses two sample nonparametric testing for each family BOB & WOW

You might also like