Dynamic Production System Identification for Smart Manufacturing Systems.

Area of Science:

• Manufacturing Systems Engineering
• Artificial Intelligence
• Operations Research

Background:

• Traditional process mining methods often overlook infrequent exceptional events crucial for understanding system reliability.
• Existing models lack robust validation mechanisms and dynamic updating capabilities for real-world production environments.

Purpose of the Study:

• To present a new methodology, production system identification, for creating manufacturing system models from operational logs.
• To address limitations in process mining, specifically regarding exceptional events, causal validation, and model adaptability.
• To improve production scheduling decisions through more comprehensive and dynamic system modeling.

Main Methods:

• Utilized genetic programming (GP) combined with Petri nets and probabilistic neural nets (PNNs).
• Developed a colored Petri net formalism for log interpretation and identification of exceptional system states.
• Employed a novel formulation of PNNs for learning state relations and a generalized stochastic Petri net for validation.

Main Results:

• Successfully generated a production system model from operational logs, incorporating exceptional events.
• Demonstrated the ability to validate the model using causal understanding and PNNs.
• Showcased the methodology's effectiveness with an automotive assembly system example.

Conclusions:

• Production system identification offers a more robust approach than standard process mining for manufacturing system modeling.
• The integrated methodology effectively handles exceptional events and provides a validated, adaptable model for production scheduling.
• This approach enhances insights into system capabilities and reliability for informed decision-making.