Segmentation of shot peening patterns using k-means clustering

Abstract

Shot peen forming is an industrial process for shaping thin and large metal plates. The process consists in treating a plate with a stream of rigid shot that are projected through a moving nozzle according to a peening pattern. The existing methods for the peening pattern computation either prescribe a continuously varying peening intensity, which is not reproducible in practice, or require having pre-defined intensities. Here, we present a segmentation strategy that enforces practical constraints to any peening pattern by dividing it into uniformly treated zones. In addition, the strategy automatically identifies the best treatment parameters for each segment. The strategy consists of the clustering algorithm, which splits the pattern into segments, and of the noise filtering algorithm, which removes too small segments from the pattern. The clustering algorithm is inspired by the $\mathtt{k}$-means method, and the filtering algorithm is based on cellular automata. Both algorithms were tested numerically using 200 randomly generated test cases.