Mahalanobis classification system (MCS) integrated with binary particle swarm optimization for robust quality classification of complex metallic turbine blades

On the Influence of Reference Mahalanobis Distance Space for Quality Classification of Complex Metal Parts Using Vibrations

“…The current authors have recently proposed the IMCS algorithm [17], which is an extension of the earlier proposed two-stage MCS approach [16] for improving classification performance. The IMCS approach incorporates the decision-making process directly into the feature selection process, thus avoiding the need for extra user-dependent inputs and enhancing classification performance.…”

“…In order to avoid overfitting for sparse training datasets, the cross-validation of k-folders is considered. The downside of this fully integrated MCS approach is the higher computational effort needed compared to two-stage MCS [16].…”

“…The blades have been labeled as good/bad using alternative inspection procedures (e.g., X-ray Computed Tomography, ultrasonic, penetrant testing, microscopic sectioning, operational performance, and operator experience). For more details on the turbine blades, readers are referred to [16]. In the following context, MD normal and MD abnormal stand for the Mahalanobis distance based on normal and abnormal observations, respectively.…”

“…The feature selection process, however, is seen as not optimal by several scholars [8][9][10]. Therefore, some scholars have carried out in-depth and extensive research and have achieved promising results, making important contributions to the development of MTS [11][12][13][14][15][16][17]. The majority of those methodologies are based on metaheuristic algorithms such as the genetic algorithm (GA), binary particle swarm optimization (BPSO), binary ant colony optimization (BACO), and the binary gravitational search algorithm (BGSA), effectively improving the feature selection strategy.…”

“…A novel two-stage Mahalanobis classification system (MCS), as well as its integrated version (IMCS), has been recently proposed by the current authors [16,17]. IMCS integrates the decision boundary searching process, which is based on the particle swarm optimizer algorithm, directly into the feature selection phase (based on BPSO or BGSA) for constructing the MD space.…”