Ensemble-Based Model-Agnostic Meta-Learning with Operational Grouping for Intelligent Sensory Systems.

Area of Science:

• Robotics and Industrial Automation
• Machine Learning and Artificial Intelligence

Background:

• Predictive maintenance (PdM) is crucial for robotic arms in assembly lines, requiring accurate and rapid fault classification.
• Model-agnostic meta-learning (MAML) shows promise for PdM but suffers from parameter hypersensitivity and limited generalization.
• Existing meta-learning frameworks struggle with few-shot learning and cross-domain generalization in complex industrial settings.

Purpose of the Study:

• To develop an improved meta-learning framework for enhanced fault classification and generalization in robotic arm PdM.
• To address the hypersensitivity and limited generalization challenges of traditional MAML in PdM applications.
• To enhance few-shot learning capabilities for robotic arm fault detection using digital twins.

Main Methods:

• An ensemble-based meta-learning approach integrating majority voting with Model-Agnostic Meta-Learning (MAML).
• Operational grouping implemented via Latin Hypercube Sampling (LHS) to improve few-shot learning and generalization.
• Validation using synthetic vibration signal datasets of robotic arm faults generated via a digital twin.

Main Results:

• The proposed ensemble MAML approach demonstrated superior accuracy in classifying a larger number of defective mechanical classes.
• Significant improvements were observed in cross-domain few-shot (CDFS) learning scenarios.
• The methodology maintained stable output while enhancing few-shot learning ability and generalization.

Conclusions:

• The ensemble-based meta-learning approach effectively overcomes limitations of standard MAML for robotic arm PdM.
• The integration of LHS and majority voting enhances robustness and generalization in few-shot fault classification.
• This framework offers a more accurate and reliable solution for predictive maintenance in industrial robotic systems.