Molding Process Monitoring Using Nearest-Neighbor Quality Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing molding process monitoring systems face challenges in predicting the quality of molded parts due to the complexity and non-linearity of injection molding processes, which are exacerbated by changes in external conditions and machine settings, often requiring large datasets that are not readily available, especially at the start of production, leading to unreliable predictions and inefficient monitoring.
Innovation Solution
The method determines the nearest neighbors in past cycles with similar data to the current cycle, checks a predictability criterion for quality variation, and if not met, issues a notification and performs an anomaly check to gather quality data or actuate a reject gate, ensuring reliable prediction by requesting or providing quality data when necessary.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If statistical models are used for quality prediction, then prediction capability is improved, but large amounts of data are required which are not available at the start of production
Solution Approach 1:
The system performs preliminary actions by collecting and storing cycle data and quality data in a data collection during past cycles before actual production starts. This preliminary data accumulation enables the model to make reliable predictions even when data is scarce during early production phases, resolving the contradiction between needing large datasets and starting production with limited data.
Solution Approach 2:
The system uses itself to improve predictions by continuously accumulating its own operational data and quality data from past cycles. This self-service approach allows the model to progressively enhance its prediction capability without requiring external large datasets, enabling reliable predictions even with limited initial data availability.
2Reliability
If complex inspection systems are used, then monitoring effectiveness is improved, but device complexity and effort increase significantly
Solution Approach 1:
The system uses feedback from past cycle data and quality data to continuously improve prediction accuracy. By analyzing historical data patterns and feeding this information back into the prediction model, the system achieves effective monitoring without requiring complex inspection hardware, resolving the contradiction between monitoring effectiveness and system complexity.
Solution Approach 2:
Instead of using complex physical inspection systems, the system creates a virtual copy or model of the quality prediction process using statistical models and historical data. This digital twin approach enables effective monitoring through data analysis rather than complex physical inspection equipment, reducing device complexity while maintaining monitoring effectiveness.
3Measurement precision
If models are retrained continuously to adapt to process changes, then prediction accuracy is improved, but productivity and time are reduced due to retraining requirements
Solution Approach 1:
The system performs preliminary data collection and model training during past cycles before production changes occur. By preparing the model in advance with historical data, the system can adapt to process changes without requiring continuous retraining during production, thus maintaining both prediction accuracy and productivity.
Solution Approach 2:
The system maintains continuous operation by using the trained model for predictions without interrupting production for retraining. The model continuously processes new cycle data and quality data in the background, enabling uninterrupted production while maintaining prediction accuracy through continuous data accumulation rather than continuous retraining.
Data Source
AI summary
A method for monitoring a molding process carried out in cycles includes determining at least two nearest neighbors in the form of cycle data from at least two past cycles, such that the cycle data of the at least two nearest neighbors lie nearer to the current cycle data than the cycle data which do not pertain to the at least two nearest neighbors. Only those past cycles for which quality data are contained in the data collection are used for the determination of the at least two nearest neighbors. A predictability criterion is checked to determine whether a quality variation of the quality data of the cycles of the at least two nearest neighbors is smaller than a maximum variation and/or larger than a minimum variation. If the predictability criterion is not met, a first notification that a quality and/or a quality datum of the molded part is not reliably predictable is issued.