Strategies for synchronizing chocolate conching batch process data using dynamic time warping.

Area of Science:

• Chemical Engineering
• Process Control
• Data Analysis

Background:

• Batch processing control often assumes synchronized, equal-duration batches, which is unrealistic.
• Dynamic time warping (DTW) offers a solution for synchronizing and aligning variable batch durations.
• This study evaluates DTW methods for process control in a real-world application.

Purpose of the Study:

• To compare three dynamic time warping (DTW) methods for batch process synchronization.
• To assess the effectiveness of DTW in classifying conforming and non-conforming batches using k-nearest neighbor.
• To identify key process variables for improved batch alignment and synchronization.

Main Methods:

• Applied three dynamic time warping (DTW) algorithms to a milk chocolate conching dataset.
• Utilized k-nearest neighbor (KNN) supervised classification for batch categorization.
• Analyzed four process variables from 62 batches with varying durations (495-1170 min).

Main Results:

• All DTW methods successfully synchronized and aligned the batch data.
• The KMT method demonstrated superior performance with 93.7% accuracy, 97.2% sensitivity, and 90.3% specificity.
• The main motor's drive current was identified as the most consistent and important variable for alignment.

Conclusions:

• DTW methods, particularly KMT, are effective for synchronizing non-uniform batch process data.
• Accurate batch classification is achievable by aligning process trajectories.
• Process variable consistency, like motor drive current, is crucial for robust batch control.