An MCDM framework for the classification of features from ultrasonic images of plastic pipe welds
H Yusuf1, 2, G Panoutsos2 and C Nageswaran1
1TWI Ltd, UK
Email: hesham.yusuf@affiliate.twi.co.uk / channa.nageswaran@twi.co.uk
2The University of Sheffield, UK
Email: hahyusuf1@sheffield.ac.uk / g.panoutsos@sheffield.ac.uk
Multi-criteria decision-making (MCDM) is a set of computational and mathematical techniques for assessing a set of choices, based on often several conflicting criteria such as various costs and benefits. MCDM can be used to aid the decision-making process and attain a better understanding of the problem. MCDM as a discipline covers a wide range of topics, including mathematics, psychology and economics. The literature shows that a great deal was done in the effort of improving MCDM for various everyday decision-making applications, such as selecting the best supplier or deciding which software package to install. However, not a great deal of literature demonstrates MCDM being used for classification applications such as: defect recognition or medical diagnosis.
One of the applications for defect recognition is plastic pipe welds. In the industry, BF welds are inspected using phased array ultrasonic testing (PAUT) where a non-destructive testing (NDT) operator will collect the data and an NDT analyst would check the data for signs of defects. The process is time-consuming and can be fairly subjective. To reduce the time required for analysing the data, a model will be developed that will be able to classify the features as either defective or healthy. The model will use an MCDM technique that is data-driven and based on expert knowledge. The classification model will use extracted features from ultrasonic images of BF welds. After training the model using a training dataset, a testing dataset will be used to assess its performance.
One of the applications for defect recognition is plastic pipe welds. In the industry, BF welds are inspected using phased array ultrasonic testing (PAUT) where a non-destructive testing (NDT) operator will collect the data and an NDT analyst would check the data for signs of defects. The process is time-consuming and can be fairly subjective. To reduce the time required for analysing the data, a model will be developed that will be able to classify the features as either defective or healthy. The model will use an MCDM technique that is data-driven and based on expert knowledge. The classification model will use extracted features from ultrasonic images of BF welds. After training the model using a training dataset, a testing dataset will be used to assess its performance.
