Monitoring Mixing Processes Using Ultrasonic Sensors and Machine Learning

Area of Science:

• Process Engineering
• Materials Science
• Data Science

Background:

• Mixing is a critical unit operation in food, chemical, and pharmaceutical industries.
• Real-time, in-line monitoring is essential for optimizing mixing processes.
• Ultrasonic sensors offer a low-cost, non-invasive method for characterizing opaque systems.

Purpose of the Study:

• To investigate the use of ultrasonic sensors for real-time monitoring of mixing processes.
• To develop and compare machine learning models for predicting mixing status and completion time.
• To evaluate the effectiveness of single-sensor versus multi-sensor data fusion approaches.

Main Methods:

• A non-invasive, reflection-mode ultrasonic measurement technique was employed.
• Two model systems, honey-water blending and flour-water batter mixing, were studied.
• Various machine learning algorithms (ANNs, SVMs, LSTMs, CNNs) were tested with time- and frequency-domain features.
• Comparisons were made between single-sensor and two-sensor data fusion.

Main Results:

• Classification models achieved up to 96.3% accuracy for honey-water and 92.5% for flour-water batter.
• Regression models yielded R² values up to 0.977 for honey-water and 0.968 for flour-water batter.
• Optimal performance varied across algorithms and feature engineering methods for different tasks.

Conclusions:

• Ultrasonic sensing combined with machine learning provides an effective solution for real-time mixing monitoring.
• The study demonstrates the potential for high accuracy in classifying mixture states and predicting process completion.
• The choice of algorithm and feature engineering is crucial for maximizing performance in specific mixing applications.