An Unsupervised Reconstruction-Based Fault Detection Algorithm for Maritime Components

Abstract

In recent years, the reliability and safety requirements of ship systems have increased drastically. This has prompted a paradigm shift toward the development of prognostics and health management (PHM) approaches for these systems' critical maritime components. In light of harsh environmental conditions with varying operational loads, and a lack of fault labels in the maritime industry generally, any PHM solution for maritime components should include independent and intelligent fault detection algorithms that can report faults automatically. In this paper, we propose an unsupervised reconstruction-based fault detection algorithm for maritime components. The advantages of the proposed algorithm are verified on five different data sets of real operational run-to-failure data provided by a highly regarded industrial company. Each data set is subject to a fault at an unknown time step. In addition, different magnitudes of random white Gaussian noise are applied to each data set in order to create several real-life situations. The results suggest that the algorithm is highly suitable to be included as part of a pure data-driven diagnostics approach in future end-to-end PHM system solutions. INDEX TERMS Automatic fault detection, deep learning, maritime industry, prognostics and health management, unsupervised learning.

A Step-wise Feature Selection Scheme for a Prognostics and Health Management System in Autonomous Ferry Crossing Operation

“…The selected features will be applied to the PHM systems for autonomous ferries, such as the detection and classification of a fault, the performance evaluation and the estimation of remaining useful life [5]. In this paper, the selected features will be applied to the purpose of both fault detection and fault classification.…”

“…The first one is the ferry crossing during normal operation, while the second one is the ferry crossing when the fault is introduced. Then, the already developed unsupervised fault detection algorithm [5] is trained on the normal operation data and employed on the faulty degradation data to detect the fault time step automatically. In other words, the algorithm estimates the time step which separates normal operation data from faulty degradation data.…”

“…As described above, the number of all input features is 47, the number of features selected by HDK is 22, and the number of features selected by JMI, RF, and SA is 12. The algorithm, which is proposed in our previous paper [5], is used for unsupervised fault detection. Then, a one-layer LSTM with 100 hidden nodes and a softmax classifier is trained on the constructed training targets and used for fault classification, namely, classifying normal operation conditions and fault conditions in the second data set, as described in Section III-A.…”

“…Recently, more and more attentions have been paid to deep learning (DL)-based PHM systems [5], [6]. The core idea of DL-based PHM system is to develop a specific DL algorithm to detect unforeseen degradation trends in the sensor data.…”

Automatic Fault Detection for Marine Diesel Engine Degradation in Autonomous Ferry Crossing Operation

“…However, due to a general lack of fault labels for critical components in the maritime industry [9], an appropriate fault detection algorithm should not depend on a supervised classifier. An alternative approach is the utilization of unsupervised reconstruction-based fault detection algorithms [10], [11]. Usually, these algorithms train a Variational Autoencoder (VAE), in an unsupervised practice, to reconstruct normal operation data.…”