Probabilistic Condition Monitoring of Azimuth Thrusters Based on Acceleration Measurements

Abstract

Drill ships and offshore rigs use azimuth thrusters for propulsion, maneuvering and steering, attitude control and dynamic positioning activities. The versatile operating modes and the challenging marine environment create demand for flexible and practical condition monitoring solutions onboard. This study introduces a condition monitoring algorithm using acceleration and shaft speed data to detect anomalies that give information on the defects in the driveline components of the thrusters. Statistical features of vibration are predicted with linear regression models and the residuals are then monitored relative to multivariate normal distributions. The method includes an automated shaft speed selection approach that identifies the normal distributed operational areas from the training data based on the residuals. During monitoring, the squared Mahalanobis distance to the identified distributions is calculated in the defined shaft speed ranges, providing information on the thruster condition. The performance of the method was validated based on data from two operating thrusters and compared with reference classifiers. The results suggest that the method could detect changes in the condition of the thrusters during online monitoring. Moreover, it had high accuracy in the bearing condition related binary classification tests. In conclusion, the algorithm has practical properties that exhibit suitability for online application.

On Training Data Selection in Condition Monitoring Applications—Case Azimuth Thrusters

“…The algorithm for identifying CM models was originally presented in [6] and a flowchart of its stages is shown in Figure 1. The training subset selection process is highlighted there.…”

“…In industrial maintenance and condition monitoring (CM), machine learning (ML) methods have become an intriguing option for data analysis because they could recognize the health states of machines automatically [3]. However, the research on this field is widely realized with simulation data and data from precisely controlled laboratory tests, which are commonly free from the characteristics of industrial data, such as noise, inconsistency, outliers, incomplete records, irrelevant samples, unfavorable and varying operating areas and the lack of labeled samples [4][5][6]. In addition, the data for model training are commonly selected manually and the introduction of their characteristics may not be explicit [7].…”

“…On the other hand, all the data from such a distribution are not equally useful in model training [17] and random selection is then impractical. In industrial applications, training data are typically hand-picked based on the status codes given by the monitoring system [4,12] and based on expert knowledge [6]. The fault data of rotating machines, collected into data repositories, typically have drawbacks such as the narrow coverage of the complete operating area or poor quality.…”

“…Moreover, each unique fault may show new, previously unseen symptoms in indirect measurements, such as acceleration signals. Therefore, the data of fault-free operating conditions solely are commonly applied for model training [4][5][6]12,18]. The one-class classifier is then used for anomaly detection.…”

“…The study was made by applying the probabilistic condition monitoring algorithm introduced in [6] on the datasets collected from a single azimuth thruster. While the previous work [6] was focused on describing the algorithm for anomaly detection, this work clarifies the effect of training data on the system identification procedure of the algorithm. The performance is studied based on relatively large datasets which were not available in the previous work.…”