An Optimization-Based Sample Selection Method Considering Sample Redundancy and Usefulness

Data pre-processing is a crucial step for data management, communication, and modeling in Prognostics and Health Management (PHM) framework. However, most existing studies have primarily focused on feature selection while disregarding the significance of sample selection. In practice, effective sample selection not only decreases data redundancy but also enhances computational efficiency and model performance. This research introduces a new offline sample selection method that employs an optimization algorithm based on simultaneous sparse recovery. We formulate the sample selection problem as a linear programming model that is solvable using most standard convex solvers. The solution of the model not only identifies the important samples but also establishes the mapping relationships between each sample and the important ones. The method can identify a small subset of samples that are crucial by considering their usefulness and freshness. We demonstrate the effectiveness of our proposed method in a case study that employs the dataset from the 2016 PHM data challenge.