Variable selection method for fault isolation using least absolute shrinkage and selection operator (LASSO)

Abstract

In statistical fault isolation, contribution plots are most commonly used tools, whose results are often influenced by the smearing effect. To solve this problem, reconstruction-based methods were developed. Unfortunately, the conventional reconstruction-based methods rely on the knowledge of fault directions or abundant historical fault data which are seldom available in industrial applications. The branch and bound (B&B) algorithm can be adopted to relieve such limitation. However, the computational burden of B&B is usually heavy, especially for a large number of variables. In this paper, a fault isolation method based on variable selection is proposed to overcome these shortcomings of the existing methods. The fault isolation problem is transformed into a quadratic programming problem with a sparsity constraint, which has a unified form for different monitoring statistics and can be solved efficiently using the least absolute shrinkage and selection operator (LASSO). The effectiveness of this method is illustrated by case studies.