Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Transfer learning of state-based potential games for process optimization in decentralized manufacturing systems

Steve Yuwono, Dorothea Schwung, Andreas Schwung

TL;DR

The paper tackles the challenge of slow learning in distributed, self-optimizing manufacturing systems by introducing TL-SbPGs, an online transfer-learning framework for state-based potential games. It proposes two predefined-similarity approaches (SW and MOM) and a similarity-inference approach using RBF networks, plus an adaptive weighting scheme guided by exploration and state-visitation similarity ($JSD$). Theoretical results establish SbPG convergence under transfer, and extensive experiments on a lab bulk good system (BGS) and a larger LS-BGS demonstrate faster convergence and meaningful gains in power efficiency and throughput compared with vanilla SbPGs. The work advances practical, scalable multi-agent coordination in decentralized manufacturing by enabling knowledge reuse across similar modules with minimal centralized coordination.

Abstract

This paper presents a novel online transfer learning approach in state-based potential games (TL-SbPGs) for distributed self-optimization in manufacturing systems. The approach targets practical industrial scenarios where knowledge sharing among similar players enhances learning in large-scale and decentralized environments. TL-SbPGs enable players to reuse learned policies from others, which improves learning outcomes and accelerates convergence. To accomplish this goal, we develop transfer learning concepts and similarity criteria for players, which offer two distinct settings: (a) predefined similarities between players and (b) dynamically inferred similarities between players during training. The applicability of the SbPG framework to transfer learning is formally established. Furthermore, we present a method to optimize the timing and weighting of knowledge transfer. Experimental results from a laboratory-scale testbed show that TL-SbPGs improve production efficiency and reduce power consumption compared to vanilla SbPGs.

Transfer learning of state-based potential games for process optimization in decentralized manufacturing systems

TL;DR

The paper tackles the challenge of slow learning in distributed, self-optimizing manufacturing systems by introducing TL-SbPGs, an online transfer-learning framework for state-based potential games. It proposes two predefined-similarity approaches (SW and MOM) and a similarity-inference approach using RBF networks, plus an adaptive weighting scheme guided by exploration and state-visitation similarity (). Theoretical results establish SbPG convergence under transfer, and extensive experiments on a lab bulk good system (BGS) and a larger LS-BGS demonstrate faster convergence and meaningful gains in power efficiency and throughput compared with vanilla SbPGs. The work advances practical, scalable multi-agent coordination in decentralized manufacturing by enabling knowledge reuse across similar modules with minimal centralized coordination.

Abstract

This paper presents a novel online transfer learning approach in state-based potential games (TL-SbPGs) for distributed self-optimization in manufacturing systems. The approach targets practical industrial scenarios where knowledge sharing among similar players enhances learning in large-scale and decentralized environments. TL-SbPGs enable players to reuse learned policies from others, which improves learning outcomes and accelerates convergence. To accomplish this goal, we develop transfer learning concepts and similarity criteria for players, which offer two distinct settings: (a) predefined similarities between players and (b) dynamically inferred similarities between players during training. The applicability of the SbPG framework to transfer learning is formally established. Furthermore, we present a method to optimize the timing and weighting of knowledge transfer. Experimental results from a laboratory-scale testbed show that TL-SbPGs improve production efficiency and reduce power consumption compared to vanilla SbPGs.
Paper Structure (22 sections, 4 theorems, 28 equations, 13 figures, 6 tables, 1 algorithm)

This paper contains 22 sections, 4 theorems, 28 equations, 13 figures, 6 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related work
  3. Self-learning in production systems
  4. Transfer learning for automation in manufacturing
  5. Potential games
  6. Problem statement
  7. Basics of state-based potential games
  8. Transfer learning of state-based potential games
  9. TL-SbPGs with pre-defined similarities
  10. Sliding window approach
  11. Momentum-based approach
  12. Adaptive weight parameter
  13. TL-SbPGs without pre-defined similarities
  14. RBF networks
  15. Similarity measures implementations
  16. ...and 7 more sections

Key Result

Lemma 1

Zazo2016 A game $\Gamma(\mathcal{N}, \mathcal{A}, \{U_i\}, {S}, P, \phi)$ is a DPG if the players' utilities satisfy the following conditions: $\forall i,j \in \mathcal{N}$, $\forall m \in S^{A_i}$ and $\forall n \in S^{A_j}$.

Figures (13)

  • Figure 1: Curriculum for transfer learning in GT-based learning.
  • Figure 2: Considered a distributed system structure including several subsystems with their own control and communication systems.
  • Figure 3: Production chain schematic with transfer learning networks on serial and parallel connected sub-systems.
  • Figure 4: An overview of TL-SbPGs for distributed optimizations.
  • Figure 5: An example of probability distributions of $S^i_{m}$ and $S^j_{m}$, where number of samples = 10,000 and m denotes the number of discrete states. The x-axis denotes the indices of discrete states, while the y-axis indicates the probability of each state being visited.
  • ...and 8 more figures

Theorems & Definitions (9)

  • Definition 1
  • Definition 2
  • Definition 3
  • Lemma 1
  • Lemma 2
  • Lemma 3
  • Proof 1
  • Lemma 4
  • Proof 2