Refrigerated Showcase Fault Detection by an Artificial Neural Network using Correntropy with Improved Adaptive Kernel Size Tuning

This paper proposes refrigerated showcase fault detection (FD) by an Artificial Neural Network (ANN) using Maximum Correntropy Criterion (MCC) with an improved adaptive kernel size tuning method. The MCC based ANN can reduce influence from outliers by tuning a kernel size appropriately. However, tuning the kernel size by trial and error is time-consuming. The proposed method utilizes the improved adaptive kernel size tuning method to improve FD accuracy. Effectiveness of the proposed method is verified by comparison with ANN based methods using Least Square Error (LSE) and the MCC with the conventional adaptive kernel size tuning method.