SPC data is usually used to react to problems that have already happened. How the same data can reveal trends that predict quality problems, from tool wear to material changes, before defects reach customers.
A moulder had used SPC charts on a critical dimension for years. They worked as designed: when a point went out of control, the operator stopped and investigated. Then an engineer looked at the data differently, plotting months of measurements against shot count for each tool. A clear, slow trend emerged: the dimension drifted predictably as a core insert wore. By the time the chart signalled, the tool was near the end of its window. With the trend known, insert replacement could be scheduled before the dimension drifted far, as described in SPC data predicts tool wear.
SPC data holds far more information than most plants use.
From reactive to predictive
Traditional SPC asks: is the process stable right now? Predictive use of the same data asks: where is the process heading, and when will it become a problem?
Patterns that predict problems
- Trends with shot count or time: tool wear, gate erosion, screw and barrel wear.
- Shifts associated with material lots: changes in melt flow or moisture.
- Seasonal or daily patterns: ambient temperature and humidity affecting process or measurement.
- Increasing variation: the spread widening before the average moves, often indicating developing machine or tool problems.
- Correlations between process parameters and quality: injection pressure, cushion or fill time changes that precede dimensional changes.
Making it practical
- Store SPC data with context: tool, cavity, machine, material lot, shift and shot count.
- Combine quality data with process data from machines.
- Review trends regularly, not only when points go out of control.
- Use simple analysis first: run charts by shot count, comparisons by lot and correlation plots.
- Set early warning limits for trends, inside the control limits.
Where AI and analytics help
With enough data, machine learning models can identify complex relationships between process parameters and quality, predicting defects from process signals in real time. These approaches work best when built on clean, well-structured data and sound process understanding, starting from the fundamentals described in SPC without the jargon.
Turning prediction into action
- Schedule tool maintenance or insert replacement based on predicted drift.
- Adjust process settings proactively for known material lot differences.
- Trigger extra inspection when patterns indicate rising risk.
- Feed findings back into tool design, material specification and process windows.
Measuring success
Track defects reaching customers, scrap rates, unplanned tool interventions and the number of problems detected before they produced bad parts.
Start with what you have
Most plants already collect SPC data. Organise it with context, look for trends and patterns, and start acting on them. Predictive quality does not have to begin with advanced technology; it begins with looking at existing data in a new way.
