CNN-Based Automatic Tablet Classification Using a Vibration-Controlled Bowl Feeder with Spiral Torque Optimization.

Area of Science:

• Pharmaceutical Technology
• Artificial Intelligence in Medicine
• Robotics and Automation

Background:

• Accurate drug identification is critical for patient safety and effective treatment.
• Existing drug classification methods may lack efficiency or precision.
• Automated systems are needed to handle the increasing variety and volume of pharmaceutical products.

Purpose of the Study:

• To develop and evaluate an automated drug classification system.
• To integrate convolutional neural network (CNN) technology with rotational pill dropping.
• To optimize the performance of a bowl feeder for stable pill handling and classification.

Main Methods:

• Captured images of 4080 pills across 102 drug types.
• Trained a convolutional neural network (CNN) for image-based classification.
• Utilized a bowl feeder with optimized parameters (voltage, torque, PWM, tilt angle, vibration amplitude and frequency).
• Conducted performance tests at specific operating conditions (5 V, 20 rpm, 20% PWM, 1.5 mm vibration amplitude).

Main Results:

• Achieved an 88.8% classification accuracy using the CNN model.
• Demonstrated stable, sequential pill movement without loss or clumping with optimized feeder parameters.
• The bowl feeder structure successfully tolerated oblique angles up to 75° for precise alignment.

Conclusions:

• The proposed system effectively classifies drugs using CNN and rotational pill dropping.
• Optimized bowl feeder parameters are crucial for reliable pill handling and system performance.
• This automated approach offers a promising solution for accurate and efficient drug classification.