A new real-time statistical analysis system for injection molding reportedly provides a new level of fault detection and process quality. The system analyzes and displays process data in a way that can help in automatic part-quality monitoring and faster correction when the process drifts. It can decrease time to troubleshoot processing problems, provides automatic rejection of defective parts, measures process consistency, and can be used for preventive maintenance scheduling.
The reason for a new approach is that current statistical process control (SPC) methodologies have significant limitations as a quality assurance approach, says Daniel Hazen, product manager for advanced industrial applications at MKS Instruments, a supplier of data-acquisition and analysis tools. This is because the interactions of the many variables in the injection molding process are not captured and evaluated simultaneously by conventional SPC, said Hazen at the Molding 2009 conference, held in January in New Orleans and sponsored by Executive Conference Management, Plymouth, Mich.
“Conventional SPC looks at one variable at a time, which does not provide enough information to accurately show a reject part,” says Hazen. Evaluating single process variables one at a time, called univariate analysis, does not capture all of the variables and interactions influencing part quality.
Combining multiple single-variable SPC charts into a single chart has been tried, but that makes it difficult to determine when the process moved out of control or to effectively pinpoint the causes. Hazen says, “SPC may show the occurrence of a problem, but it has trouble showing all of the variables responsible.”
Hazen says injection molders require a system that analyzes multiple process variables simultaneously—called multivariate analysis (MVA). MVA systems have been used in the chemical, pharmaceutical, and semiconductor industries. “MVA not only looks at multiple process variables simultaneously but it can show a relationship between variables. It goes beyond SPC by determining a correlation between variables,” Hazen said. For example, melt viscosity is a complex function of the material type, melt pressure, ram velocity, and melt temperature.
An MVA system takes process signals, such as melt pressure, screw position, barrel temperature, melt temperature, mold closing, filling, pack/hold, and cooling. The variables are compared in real time to a historical model that represents ideal operating conditions. When the various conditions are all concurrently within their respective ranges, then the molder is relatively assured of making good parts. If a process is out of statistical control, MVA can show exactly which variables are out of tolerance and ranks the variables by order of significance to the problem.
MVA can reject parts based on the value of a variable, or on a change in the correlation between a few or multiple variables, says Hazen. “The detection of change in the relationship between variables provides a tremendous improvement over typical univariate SPC methods.”
In one example, a medical device was being produced in a 32-cavity mold that experienced short shots in one cavity. MVA showed that the defects were caused by a decrease in mold temperature, which increased the melt viscosity. MVA found the cause even though no mold sensors were used.
A study conducted at the University of Massachusetts—Lowell showed that an MVA approach detected twice as many real process changes vs. false positives as conventional SPC, and it missed fewer changes. MVA also missed fewer defects.
INJECTION MOLDING MVA
MKS Instruments offers its SenseLink QM data-acquisition system with an MVA engine developed for injection molding. SenseLink collects the data and builds a model around an acceptable process window established from a design of experiments (DOE). The most important features, such as mass and dimensions of the part, or short shots and flash, are defined and analyzed to create the best MVA model of the process.
New production data are compared in real time to the alarm limits developed by the model. The MVA system shows which variable or interaction of variables was responsible for an alarm and ranks the variables in order of importance.
The SenseLink QM system has all necessary data-acquisition, MVA processing, and control functions in a self-contained, compact unit that mounts on the injection machine and connects to existing machine and mold sensors. Users view results through SenseLink’s web browser interface.
MVA is performed on all molding variables in real time on each molding cycle. Process variables contributing to poor quality can be identified, so the system can predict flash, short shots, voids, burning, contamination, bubbles, and surface defects. SenseLink provides real-time process control with filling/packing switchover capabilities. It also provides automatic reject containment and production reporting.
Process models can adapt to acceptable process changes resulting from changes in material viscosity or the plant environment. SenseLink can show a slowly shifting or drifting process, using adaptive techniques that allow the model to adjust to acceptable process change while minimizing alarms. Users can reset the model baseline.
The system can detect process drift toward an out-of-tolerance condition. SenseLink QM displays the MVA results along with a ranking of the variables that are main contributors to the alarm condition. The visual display assists rapid troubleshooting since the variable at the top of the list is the most likely culprit.