Diagnosing Extrusion Process Based on Displacement Signal and Simple Decision Tree Classifier.

Area of Science:

• Manufacturing Engineering
• Materials Science
• Signal Processing

Background:

• Traditional extrusion process diagnosis relies on complex sensors like strain gauges or accelerometers.
• Existing methods often lack simplicity and real-time decision-making capabilities.
• Literature review reveals limited exploration of displacement-based signals for extrusion diagnostics.

Purpose of the Study:

• To develop a novel, simple, and computationally inexpensive method for diagnosing extrusion processes.
• To utilize punch displacement signals for real-time anomaly detection.
• To validate the proposed method's effectiveness and efficiency.

Main Methods:

• Analysis of literature on extrusion process diagnosis.
• Development of a new method based on punch displacement signal observation and polynomial approximation.
• Utilizing polynomial coefficients and Sum of Squared Errors (SSE) as input features for a classifier.
• Employing a decision tree algorithm for anomaly detection.

Main Results:

• The proposed method successfully diagnoses extrusion processes using punch displacement signals.
• Anomaly detection achieved a high accuracy of 98.36% based solely on SSE values.
• Decisions were made rapidly, within 0.44 seconds, averaging 26.7% of the total extrusion duration.

Conclusions:

• The novel method offers an effective and simple approach to extrusion process diagnosis.
• Real-time decision-making is achievable with minimal computational resources.
• Punch displacement signal analysis presents a viable alternative to traditional sensing methods.