A computationally efficient method for identification of steady state in time series data from ship monitoring

An increasing number of ships are being equipped with sensors and devices for monitoring of operational behavior, and the amount and access to operational data is gradually increasing. Due to various reasons described in this paper, the operational data may contain erroneous data points that are critical to assess prior to performing data analysis or building mathematical and statistical models. In this paper, a stepwise method for preparation of data for ship operation and performance analysis is presented. The method deals with removing jumps in the time series data, including loss of time synchronization between different measurement subsystems, outlier detection, including repeated samples, dropouts and spikes and data selection and extraction, including stationarity detection. The final result is a data set free from disturbances, distortions and undesired physical effects, that can be used to improve the quality of a ship operation and performance analysis.