Thu, 10 December, 2020
The IEEE Conference paper of NSIRC and University of Sheffield PhD student Hesham Yusuf, has been published in the 2020 IEEE International Conference on Fuzzy Systems (FUZZ-IEEE).
Multi-criteria decision making using Fuzzy Logic and ATOVIC with application to manufacturing
Read the conference paper
The paper investigates interpretable machine learning (ML) for an advanced manufacturing case study featuring ultrasonic pipe inspection.
Interpretable machine learning is a type of modelling which has the potential for being “explainable”. Explainable modelling has more potential for providing linguistic explanation to the user.
The importance of Explainable-ML lies in applications where risk to assets and/or human life are high, such as: medical, aviation or nuclear fields.
Explanation becomes vital in these applications because it is challenging to achieve justification without it, moreover without justification there can be no decision. Hence, in the past, non-interpretable ML models in critical applications were used merely as a guide in the sense that the expert would have to come up with their own decision supported by their explanation/justification.
In advanced manufacturing, NDT experts are tasked with inspecting products to ensure they are non-defective. Manual NDT has proven to be a tedious task especially taking into consideration the large number of images that have to be viewed to test each manufactured part. Hence, this is where Automatic NDT comes in to play, a method of NDT that relies on models to find those defects. The paper investigates an interpretable machine learning framework for classifying ultrasonic images of plastic pipes.
The paper demonstrates an extended version of Amended fused TOPSIS-VIKOR for Classification (ATOVIC), Fuzzy-ATOVIC, to assess its performance and potential for linguistic explanation.
Hesham's PhD research is sponsored by The Lloyd's Register Foundation and focuses on autonomous defect classification for ultrasonic non-destructive testing in advanced manufacturing processes.
To contact Hesham about his research, email email@example.com.