New SPC technique excels for multicavity molds.

A new statistical process control (SPC) method is said to allow operators to

easily identify both "local" and "global" sources of variation in multicavity molds of almost any size and configuration. The technique, known as median/individual (M/I) control charting, was developed specifically to address problems associated with "family" manufacturing processes--i.e., multiple similar or identical-member processes sharing a common source. It's said to have potential in monitoring multicavity injection and blow molding processes. M/I charting reportedly is still new to plastics processing and its use so far has been limited to a handful of injection molders and blow molders.

What's said to be the first commercial SPC software product to incorporate this M/I charting is version 4.2 of "Quality Analyst" for PCs from Northwest Analytical, Inc., Portland, Ore. M/I charting represents a departure from most conventional SPC software packages. Standard control charts are effective in analyzing continuous processes such as extrusion, or those that produce a single part at a time, such as injection molding with a single cavity. In such "single-output" processes, all process factors are "global," explains John van der Veen, president of Integrated Process Management, an SPC/management consulting firm in Nevada City, Calif., who developed M/I charting. "Global" means that the product is acted upon by every variable in the process, and a change in any global factor will result in a corresponding change in the final product.

MULTICAVITY IS DIFFERENT

However, van der Veen adds that multicavity molds, in which each cavity represents a statistically independent "process," are problematic. In addition to global variables, the output of multicavity molds is subject to "local" factors. Each cavity represents a unique, one-part-at-a-time process with its own set of variations based on separate cooling, material temperatures and pressures, gating, and runner flow factors.

Traditional X-bar and R charts, van der Veen says, are insensitive and inefficient in monitoring individual cavities in a multicavity mold. Typically, local variations will affect the X-bar chart only when they are included by chance in the sample. For instance, if five cavities are sampled from a 32-cavity mold, there is an 83.3% probability that a cavity will not be included in the sample. With hourly sampling, production may continue for a full shift or even a day without sampling one of the cavities.

When a faulty cavity is part of a sample that leads to an out-of-control data point, the usual response is to adjust global process factors affecting all cavities, says van der Veen. If the offending cavity is not included in the next sample, the process will appear to have been properly adjusted. In fact, however, a factor may have been changed that was actually in statistical control, causing the process to trend out of control.

One alternative would be an X-bar and R chart using a sample size of 32, or one part from each cavity. This approach would convey information primarily from global effects, so operators would still lack information to easily identify the offending individual cavities.

On the other hand, X-bar and R charts could be kept for each individual cavity. Yet the logistics of collecting even as few as five samples from each cavity make this impractical and inefficient. In addition, charting each cavity does not readily convey information on the global factors affecting the process.

M/I CONTROL CHARTING

Median/Individual charting reportedly solves this dilemma by combining the median and individual control charts. It's based on the assumption that multicavity molds are affected by both global factors (shared or common sources of variation) and by individual cavities' unique local variations.

The median, which equally divides the values of all family members, is used as an estimate of the mean (average) to track the global process. It will detect and show assignable causes that affect all of the cavities simultaneously. Typical events that will cause the median to move out of control include errant cooling-water temperature or regrind percentage.

In M/I charting, individual cavities are also monitored to identify which ones are producing parts with a measured variable near the extremes of the statistically derived distribution. Individual limits are exceeded if there is an assignable local cause unique to any individual cavity. Typical causes are clogged vents, improper gate freeze-off, and short shots.

A benefit claimed for the M/I approach is that analysis is much more efficient because it differentiates between global and local causes of variation. Nonconforming product reportedly can be detected faster and more reliably.

When the median of a sample is in control, it typically indicates that the overall process is in control. An individual member charting out of control suggests a local assignable cause. Conversely, if the median is out of control, the global process is also out of control, and other charted individual limits should be ignored until the global problem is solved. Additional statistical methods can then be used for further analysis and problem-solving.

Identification of individuals is required only when they consistently approach 2.5 sigma (2.5 standard deviations from the mean) of the measurement population. Tracking of these members, known as "outliers," allows an operator to reduce variation without changing the overall process.

The M/I approach works well in multicavity molds, says van der Veen, because for large sample sizes, the median starts to approach the average closely. It's also reportedly easier to calculate the median than the average.

In Northwest Analytical's Quality Analyst, median and individual control limits are calculated automatically for any sample size. Most SPC software programs only have median chart factors for sample sizes up to 10, which limits their use for higher-cavitation molds, according to van der Veen.

TRACKS SENSITIVE FACTORS

M/I charting works by testing attributes in a sample that are most sensitive to change in the process. These factors do not have to be critical to the part. Typical factors in injection molding, for example, include individual part weight or a dimensional characteristic. "We want to identify a characteristic that will be the first thing on the part to change. Then, as you control your process, you will see the variation of that characteristic and correlate it with the dimensional characteristics that you are trying to control," explains van der Veen. In some cases, such as with complex parts, it may be necessary to monitor and plot separately two characteristics that shift independently of each other.

One company that reportedly has used M/I charting to evaluate and improve its process is Afa, an injection molder in Forest City, N.C., that manufactures trigger sprayers. The sprayer body is produced in an eight-cavity mold. When implementing M/I charting, body weight was chosen as a reliable indicator of the process response to material and machine changes. Evaluation using the M/I chart revealed two things: changes in median weights indicated global shifts in the process; and the weight for cavity number six was consistently lower than all of the others.

Evaluation of the global shifts revealed that they were associated with changes in the color of the material. Further examination showed that the most consistent dimensional characteristics could be found with a body weight of 11.3 grams, which became the target weight. In addition, cooling-water temperatures were shown to be lower for cavity number six, thereby restricting its packing. By adjusting its water flow, this cavity subsequently trended near the center of the weights. NWA Quality Analyst costs $695 for a single user and $2495 for a five-user LAN.

MORE SPC PRODUCTS

A number of other new SPC software and hardware products also have become available recently. Here is a brief rundown.

* Quality Monitor from Northwest Analytical is designed as a configurable, off-the-shelf package for PCs that brings SPC charting to the plant floor without any programming. The package consists of a configuration kit that provides the necessary tools to set up data collection, design data input screens, and produce charts. Workstations are customized to their specific tasks and can be adapted to changing requirements. Quality Monitor uses the same charting engine and data structures as NWA Quality Analyst. It can be used with Quality Analyst for further in-depth process analysis and for data transfer to other databases and manufacturing software. Quality Monitor licenses start at $495 per seat.

* Two new PC-based SPC software packages have been introduced by the DataMyte Div. of Allen-Bradley Co., Minnetonka, Minn. Quantum SPC/DC allows IBM-compatible PCs to be used for shop-floor data collection and real-time monitoring. It has Windows capability to display multiple charts and statistics simultaneously. It supports up to 16 serial data inputs, both analog and digital, through interface with DataMyte multiplexers. On-screen prompts and easy-to-use menus guide users through data-collection procedures.

Quantum SPC/QA performs quality analysis on collected SPC data from a client/server network. It employs a reportedly unique "product" concept to keep track of information. Setups can be created to define data-collection preferences and sequences. Individual setups are assigned unique names and are accessible from multiple stations on the network. Prices for either product run $1195-2495, depending on options.

* A new version 2.0 of SQC-pack/Plus SPC software for PCs has been released by PQ Systems, Dayton, Ohio. Among more than 100 user-requested enhancements is SQC Quality Advisor, a sophisticated "Help" program that guides users through common questions to help them get the most out of the software. Another new feature lets users combine several control charts, histograms, and statistics in a single report. Users can now design their own statistical summary reports or use the software's predefined reports. Price is $995.

Also new is SQCimage, a Windows-based package for creating presentation-quality SPC graphics from SQC-pack/Plus.

* A new version of SPC FocusPlus for PCs has been released by Verax Systems, East Rochester, N.Y. New features include Watchdog, which presents an overview of the process and has administrator "alert" icons; warning systems that signal out-of-control status and allow user-definable process adjustment and assignable-cause designation; CAD Drawing, which displays part information; Notepad for text editing and commenting on part numbers; and High Resolution Printer Support. Also new is 10-Way Sorting--by date and time plus eight user-definable entries. Price is $1995. Versions for networked PCs or client/server LANs are available.

* A new software package, SimPliCity Short Run SPC for Windows, has been introduced by Qualitron Systems, Rockford, Ill. It's aimed at companies that run as few as one piece in a lot. In addition to very short runs, it allows the user to monitor different parts numbers as well as different characteristics on the same graph. Electronic (analog or digital) and keyboard entry of data and assignable causes is provided. SPC charts and Cp and Cpk indices are provided in real time, as well as audible and visual alarms. The PC software is multi-user capable and can run on a network. It costs $695.

* The latest version (5.0) of the RQM Real-Time Quality Management system from Automated Technology Associates (ATA), Indianapolis, offers several features for enhanced flexibility. These include the ability to import drawings and illustrations to assist in the production process, an easy-to-use call function to connect outside programs while RQM is on-line, and a data-tagging feature that allows users to attach unlimited additional information such as user ID or comments when entering data. RQM is a fully automated SPC system for monitoring and analyzing factory-floor processes. It alerts operators to problems in real time. RQM runs on workstations and some PCs. Licenses are $1700 per user or $22,000 per server.

* A hand-held SPC calculator, called QC Calc, has been introduced by Calculated Industries, Inc., Yorba Linda, Calif. It's said to provide instant results of process- or machine-capability studies and fast updating of control charts by pressing individual keys marked Cpk, X-Bar, R, sigma, and so forth. QC Calc has built-in table values to reduce look-up time. List price is $79.95.