MANUFACTURING PROCESS

Given that the primary response variable is the back-coverage measurement (Figures 5.8b) and the staff seemed to be using what looked like a very basic measurement system, it was decided to test the precision of the measurement system of back-coverage measurement of the product (Thongs), using a Gauge repeatability and reproducibility (R&R) study.

Verifying the Precision of the Measurement System via a Gauge R&R 5.4.1.

Study

Robust design approaches can only be used to minimise the variation in a stable system where no variation exist due to special causes. However, the total variation inherent in any system is a combination of the process (or product) variation and the measurement system variation, as shown in Figure 5.11. Understanding the measurement system variation as a percentage of total variation is important before checking the process stability and process capability because these results do get obscured by measurement system variation (Montgomery, 2009; Ryan, 2011). Therefore a Gauge R&R test was used to assess the process variation caused due to the measurement system. Repeatability is the variation that results in repeated measurements of the same measurement taken by the same operator using the same measurement gauge while reproducibility is the variation caused due to different operators taking the same measurement using the same gauge (Montgomery, 2009).

The existing method used to measure the back-coverage measurement was taking two points at a perpendicular distance (judged without using any device such as a set square) of 1 ¼'' from the top of the waist at the back side of the garment and then measuring the distance between the two points by placing a measuring tape across the two marked points (parallel to the waist).

The study variation from total Gauge R&R study for the existing measurement system was 56.92% which cannot be accepted as a satisfactory level of precision of the measurement system based on the values prescribed by the American automotive industry action group (Wheeler, 2009). Also the contribution from Part-to-Part was 67.61%. Hence a third (32.39%) of variation was due to the variation of the measurement system. For these reasons it was decided to improve measurement system.

In observing how the staff measure back coverage, the researcher observed that there was not only some guess work involved (in taking two points perpendicular to the waste, which was not a straight line), but there was disturbance to the fabric (stretching), which varied from operator to operator. The researcher therefore instructed the staff to use a template (cut from plastic, having a width of 1 ¼'') (Figure 5.12) to locate the two points of measurement and use a measuring device that does not disturb

(stretch) the fabric. The operators were asked to use a Vernier calliper to measure the distance between the two points, simply because it did not disturb (stretch) the fabric. The new measurement gauge that was developed to measure the back-coverage measurement was called “Gamage’s Improved Gauge”, for charting results. Consistency and accuracy of the improved measurement system is discussed in section 7.2.1.

153 F igure 5.11 : Cate g o risat

ion of total var

iation in a s y stem (sour ce: H are , 20 12)

Figure 5.12: Placing the template to get the 1 1/4'' of displacement from the waist

Figure 5.13: Coding the garments

Generally Gauge R&R studies are conducted with two or more operators (otherwise reproducibility cannot be assessed), with a number of samples plus a measurement gauge; the operators, the measurement gauge and the method of measurement is collectively called the measurement system (Montgomery, 2009). The study with the improved measurement gauge study was conducted using two quality controllers using15 pieces of garments (Thongs); the Vernier calliper used for taking measurements was calibrated in inches. The 15 pieces of garments of same waist size (medium) were randomly selected from the finished good area; the finished goods are subjected to all

potential sources of process variability (machine, time, shift, job rotation). The 15 pieces of garments were coded as shown in Figure 5.13 in order to identify the measurements taken on each piece of garment.

Verifying Process Stability Using Control Charts 5.4.2.

Control charts are used to visually document the process variability over time to understand whether or not the process is stable (under statistical control). A stable system shows a random variability of its output (or performance) in the control chart. If the process is under statistical control, the observations will lie within predictable limits (control limits) nearly all the time—99.74% of the time within three standard deviations from the process mean under the assumption of normality. In addition, a stable system is not expected to show any non-random patterns such trends, oscillations and long runs (Hoyer & Ellis, 1996; Montgomery, 2009; Nelson, 1984; Rao, Carr, Dambolena, Kopp, Martin, Rafii, & Schlesinger, 1996; Swift, 1995). The most well-known control charts are the control charts advanced by Walter Shewhart, commonly known as Shewhart control charts.

A number of Shewhart control charts are available for monitoring attribute data (p chart and np chart for defectives; c chart and u charts for defects) and variable data (X and R

chart, Xbar and S chart, and X and MR chart). Montgomery (2009, p. 246), citing the work of Schilling and Nelson (1976) states that “sample sizes of 4 or 5 are sufficient to ensure reasonable robustness to the normality assumption”. Since the main noise factor of this study was voted as the handling technique (Figure 5.10), the frequency of change of this main noise factor (labour) was taken into consideration in deciding the sampling interval. Based on the input received from work-study officers, it was decided that five samples drawing on an hourly basis31 was optimal.

The X and R chart was the most appropriate control chart type to test the stability of

the production process (for the quality characteristic back-coverage), given the subgroup size (see Swift, 1995, p. 208 for a flow chart on control chart selection). Nelson (1984) prescribes 8 possible tests for detecting special cause variation fromX

31

The sampling frequency can also be determined based on the average run length (Karlsson & Åhlström). For details see Wadsworth, Stephens, and Godfrey (2002).

and R chart although the first four (according to Nelson) tests are the most important. These four tests for special cause variation (Table 5.7) were used in this study to test for presence or absence of special cause variation.

Table 5.7: The Control Chart Tests Used in the Study (adopted from Nelson, 1984)

Test # Indication of a Potential Special Cause Variation

1 One point more than 3 standard deviations from centre line 2 Nine points in a row on same side of centre line

3 Six points in a row, all increasing or all decreasing 4 Fourteen points in a row, alternating up and down

DESIGNING AND CONDUCTING THE OPTIMISATION EXPERIMENT