Multiple binary classifiers to analyse decision of non-compliance: For automated evaluation of piping layout

Abstract

This paper presents an approach to analyse decision from existing framework on automated evaluation of piping layout or design for reason of non-compliance. On top of Histogram of Connectivity and linear Support Vector Machines based approach for prediction if a design is compliant or non-compliant, multiple binary classifiers are trained using linear Support Vector Machines to classify a non-compliant design further according to nature of non-compliance, in space of Histogram of Connectivity. Non-compliant designs in existing dataset of Regulation 12, Annex I, International Convention for the Prevention of Pollution from Ships are further divided into separate categories according to reason of non-compliance. For each sub-category of non-compliance, a binary classifier is trained using linear Support Vector Machines by taking all non-compliant designs belonging to current category as positive and all others as negative class. Existing dataset of 1318 non-compliant designs is divided into seven sub-categories. Developed method has demonstrated encouraging performance on existing dataset of International Convention for the Prevention of Pollution from Ships.