The Design of a Visual Display for the Presentation of Statistical Quality Control Information to Operators on the Plant Floor.
Quality has been defined as "fitness for use" (Juran & Gryna, 1980). To make a product fit for use in a world of complex products, markets, and competition, quality control must focus on preventing defects rather than on responding to them. X-bar and R control charts (also called variable control charts) are tools used in industry to prevent defects. They allow for the control of one variable per piece, per chart and are used when data are continuous (Charbonneau & Webster, 1978). Continuous data include measures of a dimension, a weight, an output, a hardness, and a tensile strength.
A typical X-bar and R control chart is shown in Figure 1; it is a graphical display of a quality characteristic (the outer diameter of a bearing) that has been computed from a sample of three bearings versus the sample number or time. X-bar represents the average outer diameter of the bearings for each sample, and R, the range of the sample, is the difference between the largest and smallest observations in the sample. The X-bar and R control charts contain center lines that represent the average value and average range, respectively, of the quality characteristic when the manufacturing process is in statistical control.
Two other horizontal lines, called the upper control limit (UCL) and the lower control limit (LCL), are shown in each chart. These control limits are determined such that if the process is in control, nearly all of the sample points will fall within them. A point that is outside the control limits is interpreted as a signal or evidence that the process is out of control. This case requires investigation and corrective action to find and eliminate the cause of this behavior (Montgomery, 1991).
Although the term control chart has been universally accepted and used, the chart does not actually control anything. It simply provides a basis for action and is effective only if those responsible for making decisions act on the information that the chart reveals (Charbonneau & Webster, 1978). As decision-making responsibility shifts to those closest to the production of the product, these workers on the plant floor must construct, analyze, and act on the information provided by the control chart. However, a number of companies eager to implement statistical quality control (SQC) on the plant floor have been frustrated by the fact that apparent deficiencies in reading, writing, and arithmetic skills among a significant proportion of their employees inhibit their workers' ability to be trained (Lewis & Kales, 1991; Mast, 1988). Statistical methods must be presented in a simple, straightforward, nonthreatening manner and must be demonstrated to be a practical and helpful tool in the workplace. There is an ur gent need to simplify SQC training and operation by introducing creative methods and involving the illiterate or innumerate operator (Lewis & Kales, 1991).
Graphics and the Plant Floor Operator
Pictures and graphics offer a means to bridge a gap between worker capabilities and the quality control task. There is growing awareness that graphical presentations of data are not only very revealing but also very effective in communicating results. The appropriate data, properly displayed, have a communicative power that can surpass numbers and words. The plotting of points on control charts should be done with explanations and illustrations so that these charts reveal to the operator what is happening within a process (Lewis & Kales, 1991).
The successful design of graphical presentations depends on matching the perceptual and cognitive capabilities of an individual with the specific characteristics of the task domain (Bennett, 1992). Thus all new graphics should be tested for efficacy for different types of information display and for people with different cognitive styles and backgrounds before these graphics are incorporated into decision support systems (Sobol & Klein, 1989). An approach must be developed to present process control information to plant floor operators in a format that they can relate to and use more easily.
The engineering part drawing, or blueprint, constitutes one mode of communication on the plant floor with which operators are familiar. The fundamental purpose of engineering drawings, ranging from design drawings to exploded pictorial drawings, is a common one: They are prepared to convey necessary ideas and facts to others within and beyond the engineering department (Luzadder, 1986). A working drawing tells an operator what needs to be known to make a single part or a complete machine or structure (French, Svensen, Helsel, & Urbanick, 1988). It communicates precisely the shape and size of the part to be manufactured and is the iconic analogy of the actual manufactured part. An engineering part drawing might be used to show the operator the relationship between the control chart and the part that is manufactured. If it were possible to link the control chart to the part drawing, the operator might be better able to visualize the effects of an out-of-control process on the quality of the part. Thus the cont rol chart with the part drawing might help the operator to develop a mental model that leads to better control decisions for a manufacturing process.
The object of this work was to use human-centered design principles and methodologies to develop a tool for statistical quality control that enhances the quality control performance of operators on the plant floor. It was hypothesized that linking the engineering part drawing to the statistical quality control chart would result in a more useful and usable depiction of the state of the manufacturing process and its relationship to the physical features of the component, and that this depiction would enable the operator to make better quality control decisions in less time.
ITERATIVE DESIGN AND DEVELOPMENT
An iterative design process was used to develop and refine the concept of linking the part drawing to a control chart. Using Gould's design for usability methodology (1988), information was collected from operators on the floor of a manufacturing plant about their work with existing control charts. Furthermore, observations, questionnaires, and interviews with plant personnel yielded a positive response to our proposed pictorial format for representing quality control information as opposed to their conventional control chart format. This supported further development of the pictorial control chart concept.
Four iterations of design and testing were performed, which resulted in the addition or elimination of several features based on recommendations from technical personnel and representative users on the plant floor. These iterations concluded with the identification of three design features of the pictorial control chart for further study (each feature was studied at two different levels): format (single dimension, keyed dimension), plotting grid (absent, present), and position of the data entry table and R chart (above the X-bar chart, below the X-bar chart). These features and their levels are illustrated in Figures 2 and 3. Figure 2 depicts the single-dimension format without a plotting grid and with the data entry table and R chart above the X-bar chart. The single-dimension format presents only one depiction of the part drawing, and this depiction provides only the dimension plotted on the chart.
Figure 3 depicts the keyed dimension format with a plotting grid and with the data entry table and R chart below the X-bar chart. The keyed dimension format includes a second, comprehensive depiction of the part drawing adjacent to the control chart. The dimension plotted on the chart is highlighted using a surrounding box and a larger font size. This dimension is keyed, or linked, to the control chart by a line and arrow.
A preliminary study was conducted to determine whether a particular combination of the aforementioned design features would provide a particularly effective pictorial control chart. Eight chart designs were included in the study. The designs differed only in that each implemented a different one of the eight combinations of levels of the three design features identified for further study. This study, performed in a real-world manufacturing environment, focused on three design details: the single- and keyed-dimension formats, the presence or absence of grid lines for plotting points, and the positioning of the data entry table and R chart relative to the X-bar chart.
The participants consisted of eight operators and quality control inspectors in the press shop of the production facility that produced the part depicted in the control charts. None of the participants had been involved in the prior iterative development of the pictorial chart designs. Each operator measured eight part dimensions and plotted data points on each of the eight pictorial control chart designs. The time taken to complete the quality control task and the accuracy with which the task was performed were measured. Subjective measures of chart effectiveness were also elicited from the operators for each of the eight pictorial control charts. The operators were not given feedback about their performance during or after performance of the tasks.
The results of the preliminary study were used to make the following design decisions.
1. The keyed-dimension format of the pictorial control chart was selected over the single-dimension format because designs incorporating the keyed format resulted in shorter task completion times. The keyed- and single-dimension formats did not appear to produce any differences in error rates.
2. The positioning of the R chart and the data entry table below the X-bar chart was selected over the positioning of the R-chart and data entry table above the X-bar chart because designs combining the former feature with the keyed-dimension format resulted in shorter task completion times. The results for errors did not conclusively support one position of the R chart and table over the other with regard to interactions with the use of a plotting grid.
3. The objective data did not conclusively support either the inclusion or the omission of a plotting grid. Subjective measures elicited from the participants indicated that they perceived the grid lines as enabling more accurate plotting but hindering accurate detection of out-of-control points. On the basis of this subjective data, use of a plotting grid was selected over not including a plotting grid, but it was decided that the grid lines should be de-emphasized through the use of lighter, less saturated line textures.
Thus it was decided that a pictorial control chart with the keyed-dimension format, the R chart and data entry table below the X-bar chart, and a de-emphasized plotting grid would be carried forward for further testing. Grid lines were de-emphasized by using dotted lines instead of solid lines and drawing the dotted lines using a grey-scale texture instead of solid black. The dot-to-dot spacing on the vertical lines was 0.5 mm, whereas spacing on the horizontal lines was 1.5 mm.
An experiment was developed to compare and contrast the quality control performance of operators using a pictorial control chart with the keyed-dimension format, the de-emphasized plotting grid, and the data entry table and R-chart positioned below the X-bar chart (Figure 4) with their performance using a conventional control chart (Figure 5).
Participants. The eight operators and quality control inspectors who participated in the preliminary study served as participants in this study. Five of them were men and three were women. All had knowledge of control charting with conventional control charts and experience working with part drawings. Only one participant had formal education beyond the 10th grade.
Independent variable. Control chart design was tested at two levels: conventional control chart and pictorial control chart. The conventional control chart was a Shewhart X-bar and R chart (Blank, 1988) using the chart format employed by the company at which the participants worked. The part drawing of the measured component was provided to them using the conventional control chart to ensure that equivalent information was available to the operators in both conditions. The pictorial control chart differed from the conventional control chart in the following respects.
1. The part drawing of the measured part was shown with the control chart on the same sheet of paper. The critical dimension on the part drawing (the dimension to be measured and charted) was enclosed in a box that was keyed to the control chart by a leading arrow.
2. Another view of the part drawing with only the critical dimension noted was depicted vertically just below the X-bar chart.
3. A de-emphasized plotting grid was used instead of a grid with solid black lines.
4. The R chart and the table of observed values were positioned below the X-bar chart and the vertically oriented part drawing.
Dependent variables. The total task completion time and the total number of errors committed were recorded for each treatment condition. Completion time and number of errors for each of four task components were also recorded to better determine the sources of observed effects. Subjective measures were collected at the end of the study.
Apparatus. The charts were presented to operators on 11 X 14 in. (28 X 36 cm) paper. Upper and lower control limits and the center line were calculated and drawn on the charts by the experimenters prior to their presentation to the operators. Operators were also provided with basic four-function and memory calculators to perform the calculations necessary to plot points on the charts and with a pencil to plot the points.
Procedure. The features of the two types of charts were described to the operator before he or she performed the quality control task. One part type (an armature in production at the plant) with two critical dimensions was used for this study. The operator was given a box of 21 parts and a 6-in. (15-cm) digital caliper to measure the dimensions on the parts. The 21 parts were presented as seven subgroups, labeled 1-7, each with a sample size of three. The subgroups were presented to each operator in the same order. An out-of-control condition was defined as a point or points that fell outside the predetermined control limits (Montgomery, 1991).
The task was composed of the following task components:
1. Task Component 1 -- measuring the critical dimension on each of the three parts in a subgroup, writing the observed values on the chart, and calculating and writing the total, average, and range on the chart.
2. Task Component 2 -- plotting the average and range of the subgroup on the chart.
3. Task Component 3 -- detecting whether the points plotted for this subgroup indicated that the process was out of control.
4. Task Component 4 -- determining the total number of points falling outside the control limits on the X-bar and R-charts.
Task Components 1-3 were performed in turn for each of the seven subgroups of parts. After all 21 parts had been measured and the average and range had been plotted for all seven subgroups, Task Component 4 was performed. On one day, the operator inspected one of the two critical dimensions and constructed the control chart using one of the two control chart designs. On the following day, the operator performed the same task using the same box of parts, but inspected the other critical dimension and constructed the control chart using the other control chart design. The two designs were tested on different days with different critical dimensions to minimize any transfer effects. No feedback on task performance was provided to the operators. At the completion of the second session, users were asked to rank the two chart designs on 10 subjective measures, including ease in identifying the critical dimension, user satisfaction, convenience, ease of information transfer to an operator on the following shift, and overall ease in using the chart.
Experimental design. A paired t test was used to determine any significant differences in performance time and the number of errors with the control chart designs. Blocking was used to control operator-to-operator variability, and counterbalancing was used to control variation attributable to the two critical dimensions that were inspected. The order of administration of the part dimension inspected and the control chart design used was completely counterbalanced so that two participants used the pictorial control chart first with the first critical dimension, two used this design first with the second dimension, two used the conventional control chart first with the first critical dimension, and two used this design first with the second dimension. This resulted in a total of four conditions performed by a total of eight participants.
Table 1 presents the results of the paired t tests for the overall task completion and component times. A significant difference in the overall task completion times was observed for the two formats tested, t(7) = 3.45, p = .011. Mean task completion time was 11.73 min with the pictorial control chart, which was 1.50 min less than the 13.23 mm with the conventional control chart. The effect of chart format on the first task component (taking measurements, writing the observed values, and calculating and writing the total, average, and range on the chart) was also significant, t(7) = 3.54, p = .009.
An average of 9.28 min was required to complete this task component with the pictorial chart, which was 1.12 min less than the 10.40 min required with the conventional chart. No significant effects were observed for the time to plot points on the charts or the time to detect out-of-control points.
The effect of chart format on the time to determine the total number of points that were out of control approached statistical significance, t(7) = 2.20, p = .064. An average of 0.20 min was required to complete this task component with the pictorial chart, which was 0.19 min less than the 0.39 min required with the conventional chart.
No significant differences in the number of errors committed using the two charts were observed for the overall task or for any of its components. The effect of chart format on the total number of errors committed did, however, approach significance, t(7) = 2.11, p = .073. An average of 8.38 errors were committed in completing the overall task with the pictorial control chart - 3.12 fewer errors on average than the 11.50 errors committed in completing the overall task with the conventional control chart.
Subjective data were analyzed using the sign test, a special case of the binomial test (Daniel, 1978), using an alpha level of .05. All the subjective measures demonstrated a statistically significant preference for the pictorial control chart relative to the conventional control chart. Specifically, users ranked the pictorial control chart above the conventional control chart in terms of (a) ease of finding the critical dimension (p [less than] .005), (b) satisfaction with the amount of process control information presented (p [less than] .005), (c) convenience of use at the workplace (p [less than] .005), (d) likelihood of measuring the correct dimension (p [less than] .005), (e) ease of information transfer to the next operator (p [less than] .005), (f) ease of learning (p [less than] .05), (g) ease of detecting out of control points (p [less than] .005), (h) usefulness on the plant floor (p [less than] .005), (i) ease of plotting points (p [less than] .005), and (j) overall ease of use (p [less than] .00 5).
The results favored the pictorial control chart over the conventional control chart for use in a quality control task. The use of a human-centered, iterative design process to develop the pictorial control chart resulted in a design that achieved better operator performance than did the conventional control chart.
The results provide the basis to conclude that integrating the part drawing with the control chart favorably influenced operators' performance of the task component involving measuring parts, writing the observed values, and calculating and writing the total, average, and range on the chart. This task component accounted for 70%-80% of the overall task completion time.
The strong preference for the pictorial control chart along all of the subjective measures suggests greater acceptance of this chart on the plant floor. The de-emphasized grid and the open space between the X-bar and R chart on the pictorial chart were perceived by operators as less distracting and confusing than the solid grid lines and the positioning of the R chart just below the X-bar chart on the conventional control chart. The pictorial control chart addresses the need for effective information transfer on the shop floor. Operators perceived the pictorial control chart as effectively presenting quality control information to those knowledgeable about the product and process and to less knowledgeable operators.
The pictorial control chart eliminates the need to use an additional blueprint drawing during performance of quality control tasks. The inclusion of the part drawing on the chart helps to ensure that the operator is aware of how the process is affecting a critical dimension on the part. This information also becomes easier to convey to the operator on the next shift. In addition, on receipt of a batch of parts and the associated pictorial control charts, operators at the next workstation are able to determine more quickly which operations have been performed and which dimensions were monitored during the operations.
The development of the pictorial control chart through this research and the achievement of more effective quality control performance with the chart compared with the conventional control chart demonstrate the successful application of human-centered design methodologies to statistical quality control and to the manufacturing plant floor. More generally, this work serves as a case study of the application of human-centered design methodologies in the real-world context where the tasks of interest are carried out. As industry worldwide moves from a system of mass production to one of lean production (Womack, Jones, & Roos, 1990), the development of effective human-system interfaces in production systems will become increasingly critical to industry competitiveness. Our work suggests that although quality improvement tools are useful, workers on the plant floor might not find them particularly easy to use. The contextual approach to the design of the pictorial control chart demonstrates the usefulness and imp ortance of developing tools that are relevant to the domain, users, and task performed. In this instance, features such as the keyed part drawing were developed to link new, unfamiliar processes and technologies to the domain with which users were familiar. This is especially significant as new tools are rapidly introduced into the work environment to support advances in technology.
More specifically, the results suggest that use of the pictorial control chart in statistical quality control would reduce the frequency with which the wrong dimension is measured on a part, help operators understand how a process is affecting the physical or geometric features of a part, and lead to more effective implementation of statistical quality control. Thus use of the pictorial control chart should result in enhanced quality control performance by operators on the plant floor.
We thank Carl Klebe for providing us with the opportunity to conduct this research at Ryobi Motors in Pickens, South Carolina. We appreciate the cooperation and valuable suggestions of the operators at Ryobi Motors who participated in this research. We also thank Andy Lee, Danny Fahey, and Junius Smith for their support during this research.
Melroy E. D'Souza is a senior human factors engineer in the Gateway Business Products Division of Gateway, Inc. He received his Ph.D. in industrial engineering from Clemson University in Clemson, South Carolina, in 1996.
Joel S. Greenstein is an associate professor of industrial engineering at Clemson University He received his Ph.D. in mechanical engineering from the University of Illinois at Urbana-Champaign in 1979.
Bennett, K. B. (1992). Representation aiding: Complementary decision support for a complex, dynamic control task. IEEE Control Systems, 12(4), 19-24.
Blank, R. E. (1988). Multivariable X-bar and R charting techniques, In ASQC Quality Congress Transactions (pp. 488-491). Dallas, TX: American Society for Quality.
Charbonneau, H. C., & Webster, G. L. (1978). Industrial quality control. Englewood Cliffs, NJ: Prentice-Hall.
Daniel, W W. (1978). Applied nonparametric statistics. Boston: Houghton Mifflin.
French, T. E., Svensen, C. L., Helsel, J. D., & Urbanick, B. (1988). Mechanical drawing. New York: McGraw-Hill.
Gould, J. D. (1988). How to design usable systems. In M. Helander (Ed.), Handbook of human-computer interaction (pp. 757-789). Amsterdam: Elsevier.
Juran, J. M., & Gryna, F. M., Jr. (1980). Quality planning and analysis. New York: McGraw-Hill.
Lewis, D. A., & Kales, P. A. (1991). A need to adapt SPC training for innumerate operators. In ASQC Quality Congress Transacdons (pp. 103-107). Milwaukee, WI: American Society for Quality.
Luzadder, W. J. (1986). Fundamentals of engineering drawing. Englewood Cliffs, NJ: Prentice-Hall.
Mast, G. W (1988). Effective statistical training in the work place. In ASQC Quality Congress Transactions (pp. 680-683). Milwaukee, WI: American Society for Quality.
Montgomery, D. C. (1991). Introduction to statistical quality control. New York: Wiley.
Sobol, M. G., & Klein. G. (1989). New graphics as computerized displays for human information processing. IEEE Transactions on Systems, Man and Cybernetics, 19, 893-897.
Womack, J.P., Jones, D. T., & Roos, D. (1990). The machine that changed the world: The story of lean production. New York: Harper Collins.
Overall Completion Time and Component Times (Means) Conventional Pictorial Task Chart Chart Difference Component (min) (min) (min) Overall task 13.23 11.73 1 50 [*] Measure [a] 10.40 9.28 1.12 [**] Plot [b] 2.07 1.95 0.12 Detect [c] 0.39 0.31 0.08 Count [d] 0.39 0.20 0.19
(a.)Measure parts, write the observed values, and calculate and write the total, average, and range on the chart. (b.)plot points on the X-bar and R charts. (c.)Detect out-of-control points. (d.)Count the total number of points that are out of control.
(*.)Significant at an alpha level of .05.
(**.)Significant at an alpha level of .01.
|Printer friendly Cite/link Email Feedback|
|Author:||D'Souza, Melroy E.; Greenstein, Joel S.|
|Date:||Dec 1, 1999|
|Previous Article:||A New Class of Auditory Warning Signals for Complex Systems: Auditory Icons.|
|Next Article:||Young Novice Driver Subtypes: Relationship to High-Risk Behavior, Traffic Accident Record, and Simulator Driving Performance.|