Identification of critical inspection samples among railroad wheels by similarity-based agglomerative clustering

Abstract

This paper proposes an unsupervised analysis methodfor identifying critical samples in large populations. The objective is to identify data features which help to pinpoint the critical samples that require the most inspection resources, namely time and money. Typically the data available for deriving the optimized inspection schedules in industry include both numeric and nominal features, and most clustering and classification algorithms are tailored for either numeric or nominal data. For this work, we adopt the Similarity-Based Agglomerative Clustering (SBAC) algorithm that has beenshown to be effective in clustering data with mixed numeric and nominal features. We present the effectiveness of this approach by applying it to an important problem in the railroadindustry, i.e., the inspection of railroad wheels.