'CQI' software goes beyond SPC.

If your molding process isn't getting better all the time, continuous quality improvement (CQI) might help.

Supported by specialized software and data-acquisition hardware, CQI takes a molding process through evolutionary steps aimed at creating and maintaining a truly optimal processing region. Compared with traditional SPC, which lets a process hover within broader specification or control limits, CQI takes a far less tolerant view of process variation - sometimes to the point of trying to eliminate it completely. "You need specs, but you still have to understand how to arrive at the best process settings. 'Specsmanship' is not quality improvement," says Stith Bennett, a CQI advocate and president of SynchroStat Systems Corp. in Boulder, Colo. To shepherd molders through CQI, SynchroStat recently rolled out a new release of its Streamer1 software.

HOMING IN ON ZERO DEFECTS 

When applying Streamer1, users first embark on a "process characterization," building an empirical model that accounts for relationships between process parameters. For this task, which usually takes place during mold start-up, Streamer1 offers a suite of tools, including regression analysis and formal experimental design methods. According to Bennett, the process characterization is what establishes target values for all the key process and product variables - melt pressure, cavity pressure, and melt flow, to name just three. Bennett refers collectively to the target values as "a robust operating region."

Next comes the optimization process, which is where Streamer1 starts to deviate from traditional SPC approaches. "SPC won't optimize," Bennett claims. "It won't steer the process to those target values." Depending on user needs, Streamer1 optimizes for product variables, process variables, or even more abstract concepts such as loss - that is, the financial loss associated with any deviation from absolute quality (also known as Taguchi loss). Often, but not always, it optimizes for variation reduction.

The CQI methodology culminates in the ability to predict process trends. "Once you've run large variations out of the process and you're in that robust region where losses are low, you can get into predicting when the process will leave the best processing region," says Bennett. This capability rests on time-series predictions rooted in an exponentially weighted moving average of process parameters, yielding an early indication of a drifting process.

MULTIVARIATE ANALYSIS IS KEY

CQI's cornerstone - finding the best process settings - is easier said than done mostly because of injection molding's multivariate nature. Bennett has seen many cases where staying in the "robust region" requires users to adjust three or more variables in different directions. "Much of the failure of quality improvement so far has been the lack of ability to deal with molding's complex multivariate nature." Bennett contends that the best shop-floor personnel traditionally have shown an intuitive understanding of these multivariate relationships. Streamer1 tries to puts that knowledge into a formal problem-solving tool.

Streamer1's decision-support engine consists of many different statistical tests. "Streamer performs up to 300 tests on every variable for every parameter of every shot and provides the results in real time," says Bennett. Behind the scenes, it determines "P-values" for each statistical result. This on-going assessment of statistical significance indicates whether monitored variation is random or warrants consideration of corrective action. Here's where the system's continuous-improvement aspect enters the picture: Streamer1 users choose whatever probability levels they deem appropriate for meeting their quality goals. "It's like a high jump, where you set the bar at a certain level," explains Bennett. Actually, it's more like a limbo contest: With each passing shot, the process homes in on its target values as users "lower the bar" on significant variation by considering increasingly minor variations to be significant.

SPC'S FLAWS

On its most controversial level, the Streamer1 package is based on the argument that common SPC analysis - the kind that revolves around x-bar and R-charts - doesn't work for injection molding. Why not? For one thing, Bennett argues, molding data frequently violates the normal-distribution assumptions of ordinary statistics. Many molders, however, assume a normal distribution exists without checking to see if the data fits. "Injection molding data does not always fall into a normal distribution. If it doesn't, calculations made from Gaussian [normal-distribution] statistics will be in error. People end up calculating Cpk from wrong values," Bennett says.

As the process improves, Bennett continues, traditional SPC becomes even more ill suited. "Eventually none of your data will fall into a normal distribution as you center around the target. Changing the shape of the distribution to a higher peak is what you want to do," he says.

False SPC assumptions manifest themselves in an over-reliance on x-bar and R-charts. "People think they've done quality if they do control charts," says Bennett. "These methods are good for certain problems but are a poor choice as a single tool. Charts are only visual representations of statistical algorithms. The underlying stuff, in terms of the math, is what's really going on."

Streamer1's underlying math incorporates data-driven statistics that make no distributional assumptions. Thus, the software's statistical measures include not only traditional Gaussian Cpk, but several non-Gaussian Cpk counterparts. "We compare the results from these parallel tests," Bennett says. "It gives you more than one kind of result on which to base your decision." For example, Streamer1 shows control limits in a "hybrid chart" that combines standard statistical charts with "box plots" whose control limits do not rely on distributional assumptions.

Fortunately, Streamer1 does not force users into any arcane statistical debates. Its graphical user interface can dispense advice as simple textual messages and indicate changes with color-coded icons. For example, if injection pressure goes out of whack, Streamer1 can give a plain-English notice of the condition. It can also recommend a solution or even trigger the adjustment automatically.

INSTALLING THE SOFTWARE

Though SynchroStat typically supplies turnkey quality workstations as an integrated hardware/software package, its Streamer1 software can work with a variety of data-acquisition hardware platforms. "Name a protocol, and we support it in and out." Bennett says, citing RS 232, RS 485, direct digital, DDE and OLE (two Microsoft Windows standards) as examples. Bennett cautions that some data-acquisition systems will not fully support Streamer1's real-time capabilities without modification. Barring hardware limitations, Streamer1 operates in an asynchronous mode. This way, processing events can trigger responses when they occur rather than when the central computer polls distributed microprocessors.

A stand-alone copy of Streamer1 costs $7395, a price that includes 20 hours of telephone support. Super1 supervisory software for multi-cell applications costs $4995. Adding in hardware, training, consulting, instrumentation, and installation costs, Bennett says a typical Streamer1 CQI system would cost about $830,000.

RELATED ARTICLE: Reaping CQI's Benefits

Technology Products Inc., a medical custom molder in Longmont, Colo., has used Streamer1 to make some quality improvements in molding a two-piece handle for a laparoscopic device. Though TPI had been molding it successfully for two years, the part had been plagued by short shots and warpage.

According to process technician Dave Saenz, Streamer1 helped TPI get a grasp on the multivariate nature of these problems. The non-intuitive solution involved changing packing pressure, packing time, mold temperature, and screw speed simultaneously. Soon after putting Streamer1 to work creating multivariate control charts, TPI was able to bring the part down to zero defects by eliminating short shots and reducing warpage to within half if its allowable specification limit. Cycle time, meanwhile, improved by 22%.

TPI has started applying Streamer1 to many new molds "to narrow down the variation right away," Saenz says.