Fault Detection Method Using Inverse Distance Weight-based Local Outlier Factor

Abstract

In modern complex industrial processes, unexpected shutdown not only shorten the lifespan of the main equipment, but also causes huge maintenance costs. To prevent such a problem, a method for detection equipment failure is required. Therefore, in this paper, we propose a fault detection method using local outlier factor (LOF). Unlike statistical methods such as principal component analysis (PCA) and independent component analysis (ICA), which assume that the data follows a specific distribution (Gaussian, binomial, exponential, etc.), LOF using the density of neighbors does not require distribution assumptions on the data. Thus, it is attracting attention in non-linear system, multimode and non-stationary processes. However, LOF is affected by the distance of neighbors due to characteristic of using density, this paper proposes a method to improve the fault detection performance of an existing LOF in the form of subtracting a weigh proportional to the distance to each neighbor. To verify the performance of the proposed method, it was applied to the Tennessee Eastman process, which is used for the evaluation of fault detect and diagnosis. The experimental results confirmed that the proposed method can properly detect a fault and reduce the occurrence of inappropriate false alarm compared to the conventional PCA and LOF.