Use this URL to cite or link to this record in EThOS:
Title: Soft-computing and human-centric approaches for modelling complex manufacturing systems
Author: Baraka, Ali
ISNI:       0000 0004 6058 6699
Awarding Body: University of Sheffield
Current Institution: University of Sheffield
Date of Award: 2017
Availability of Full Text:
Access from EThOS:
Access from Institution:
In systems engineering and especially in manufacturing systems, first-principle models have been widely used for systems modelling. However, advanced manufacturing systems are often complex and information intensive rendering conventional modelling approaches via first-principle models inconvenient for use due to their high computation cost and on some instances limited accuracy. The main objective of this thesis is to develop parsimonious transparent, interpretable and computationally efficient soft-computing techniques and human-centric systems to address challenges associated with modelling complex manufacturing systems such as high-nonlinearity, measurement imprecision and sparsity as well as low process repeatability. A new data-driven modelling framework based on granular computing (GrC), radial basis function neural fuzzy (RBF-NF) system and conflict measure is proposed in order to allow for the quantification of the uncertainty present during the initial structure identification of a RBF neural fuzzy system. Such framework can be easily translated into human language via simple linguistic rules in order to describe the underlying dynamics behaviour of complex industrial processes with good generalisation capability, tolerance to input imprecision and low computational cost. A new perpetual learning approach for neural-fuzzy systems is proposed. The proposed perpetual learning framework combines more advanced system's features such as the ability to continuously learn from batch data and periodically update its structure to accommodate new data/information without significantly disturbing the previously gained knowledge and, therefore, allowing for the ability to have an open structure while taking into consideration the trade-off between interpretability and accuracy. To confirm the effectiveness of each of proposed frameworks in this thesis, a rigorous set of simulation results is presented on well-known benchmark functions as well as real industrial case studies. The perpetual learning concept has great potential for successful implementation in systems where lifelong learning is required.
Supervisor: Panoutsos, G. ; Mahfouf, M. Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available