Exploring disassembly challenges: jamming, compliance effects, and reinforcement learning in complex peg-holes processes

Goli, Farzaneh ORCID: 0009-0001-6304-3054 (2025). Exploring disassembly challenges: jamming, compliance effects, and reinforcement learning in complex peg-holes processes. University of Birmingham. Ph.D.

Preview

Goli2025PhD.pdf
Text - Accepted Version
Available under License All rights reserved.
Download (9MB) | Preview

Abstract

Disassembly is a prime activity in the remanufacturing chain, promoting recoveries for reuse, hence contributing to sustainable manufacturing. However, being largely a manual and labour-intensive operation, disassembly has poor economic viability. In this thesis, we have focused on investigating and developing robotic disassembly techniques, with special consideration of issues related to dual peg-hole and rectangular peg-hole structures. By incorporating mechanics-based analyses, compliance strategies, and artificial intelligence, this work offers possible solutions for increasing the efficiency, adaptability, and reliability of automated disassembly systems.
Chapter two first explores dual peg-hole disassembly, a typical yet sophisticated operation in remanufacturing. During the extraction process of the peg, the identification of contact states and their geometric conditions are systematically identified to explore the jamming phenomenon, which often disturbs this process. The boundary conditions of jamming are derived based on geometrical and quasistatic analyses, and active compliance strategies are analysed as a solution. Simulation results are presented to investigate the effects of some critical variables such as the amount of compliance, the location of the centre of compliance, and initial position errors. Experimental investigations confirm these results and illustrate how appropriate compliance configurations can reduce the effects of jamming and optimise disassembly.
In the third chapter, rectangular peg-hole structures in three dimensions are examined. The present study finds 26 possible contact states, of which 16 conditions are related to jamming conditions. States are classified according to surface-surface contact interactions with comprehensive interpretations of jamming mechanics. Results indicate that locating the compliance centre at the end of the peg is particularly effective in minimising the risk of jamming and produces a 77.1% reduction in maximum extraction force compared to the less optimised configurations. This therefore highlights the role of compliance strategies in enabling strong and efficient robotic disassembly.
Chapter four will discuss how reinforcement learning was integrated with compliance strategies to overcome the challenges of disassembling complex peg-hole structures. Reinforcement learning enables the robot to adaptively learn disassembly motions through recurrent training and feedback from environmental sensing, such as force and torque measurements. The proposed Reinforcement learning (RL)--based compliance approach reduces extraction forces by 29.6% compared to manual methods, showcasing the benefits of combining machine learning with mechanical insights. Experimental results confirm that the system dynamically optimises disassembly motions with considerations of material compliance and geometric constraints, reaching a new level of efficiency and adaptability.
In general, this thesis will contribute to the development of intelligent robotic systems for disassembly by incorporating compliance strategies with Artificial intelligence (AI). The research study, through theoretical analysis, computational simulations, and experimental validation, demonstrates how compliance strategy can be employed with AI-driven methods to increase the adaptability and efficiency of robotic disassembly processes, with particular emphasis on reinforcement learning. The compliance strategy used in an innovative way further optimises force control and enhances the system's capability to handle complicated disassembly tasks. These findings provide valuable insights into the potential for automation in remanufacturing and recycling and contribute to more sustainable and economically viable production systems.

Type of Work:

Thesis (Doctorates > Ph.D.)

Award Type:

Doctorates > Ph.D.

Supervisor(s):