A new framework of sensor selection for developing a fault detection system based on data-envelopment analysis

Amir Eshaghi Chaleshtori,Abdollah Aghaie

2024-03-29

Abstract:Several methods have been proposed to identify which sensor sets are optimal for finding and localizing faults under different conditions for mechanical equipment. In order to preserve acceptable performance while minimizing costs, it is crucial to identify the most effective set of sensors available. Nevertheless, some sensor sets can behave differently in fault detection because of uncertainty in the measurement data contaminated by noise interference. This paper develops new sensor selection models using Data Envelopment Analysis (DEA), which has proven helpful as a management approach for determining an optimal number of sensors, associated costs, and sensor performance in the fault diagnosis. We propose four linear optimization models for sensor selection to design the fault detection system. The validity of the presented models is demonstrated using a vibration dataset collected from a gearbox. Based on the case study results, the proposed methods are effectively superior to the other comparison sensor selection methods in fault detection accuracy.

Signal Processing

What problem does this paper attempt to address?

This paper aims to address the problem of selecting sensors to build a fault detection system based on Data Envelopment Analysis (DEA). In the fault recognition of mechanical equipment, selecting the most effective combination of sensors is crucial for reducing costs while maintaining acceptable performance. However, due to the uncertainty and noise interference in measurement data, different sensor combinations may perform differently in fault detection. The paper presents four linear optimization models for sensor selection to design a fault detection system and validates the effectiveness of these models through a vibration data set of a gear box, demonstrating that the proposed method outperforms other compared sensor selection methods in fault detection accuracy.

The main contributions of the paper include:
1. Proposing a data-driven approach considering sensor data attributes and related costs.
2. Developing sensor selection models using DEA models to consider system states and sensor requirements.
3. Experimental results demonstrate the high efficiency of the proposed sensor selection method in fault diagnosis.

The paper first reviews the literature on sensor selection methods and then introduces the theoretical basis and implementation steps of the DEA-based sensor selection method. The method consists of two stages: the DEA model is used to select important sensors, and then the selected sensors' data are used to train machine learning algorithms for fault detection. The paper also defines several performance parameters, such as monotonicity, robustness, trend, detectability, and variance, to evaluate the performance of sensors.

In the experimental section, a data set generated by a gear box fault diagnosis simulator is used to compare the proposed DEA model with other traditional sensor selection methods such as correlation analysis, mRMR algorithm, and Extra Trees classifier, demonstrating the superiority of the proposed method in fault classification. The experimental results show that the best classification performance can be achieved by using the minimum number of sensors, such as combining the CCR model with the KNN algorithm. Finally, the paper discusses future research directions, including incorporating dynamic factors into modeling, estimation, and sensor selection.