{"title":"Analysis of Variational Autoencoders for Imputing Missing Values from Sensor Data of Marine Systems","authors":"C. Velasco-Gallego, I. Lazakis","doi":"10.5957/josr.09210032","DOIUrl":null,"url":null,"abstract":"Of all the causes of accidents to ships, 14% pertains to damage due to ship equipment. Accordingly, the maritime industry is currently considering state-of-the-art maintenance and inspection processes, an example of which is condition-based maintenance (CBM). This is a strategy that hinges on the condition monitoring (CM) of assets. CM has proven to increase efficiency, reliability, profitability, and performance of vessel. To enable this maintenance strategy, sensors need to be installed along the most critical ship components and around the environment where these assets are operating through the application of Internet of Ships (IoS). IoS has demonstrated to be effective for collecting data in real time as well as performing diagnosis and prognosis to assess the current and future health of machinery to assist instant decision-making. The employment of IoS presents several challenges, an example of which is the imputation of missing values. Data imputation is a compelling preprocessing step, the aim of this is to estimate identified missing values to avoid underutilization of data. This data preparation step has gained popularity over the last few years due to its importance when dealing with Industrial Internet of Things (IIoT) sensor data. Although some articles presented new methodologies to impute missing values from sensor data of marine machinery based on machine learning methodologies, deep learning models have not yet been considered. For this reason, variational autoencoders for imputing missing values from sensor data of marine systems are analyzed in this article. To assess the performance of variational autoencoders as imputation methods, a comparative study is performed with widely implemented imputation techniques. Mean imputation, Forward Fill and Backward Fill, and k-Nearest Neighbors are considered. To that end, a case study on marine machinery system parameters obtained from sensors installed on a diesel generator of a tanker ship is performed. Results demonstrate the applicability of variational autoencoders when dealing with missing values of marine machinery systems sensor data, achieving a coefficient of determination of 0.99 when imputing missing values of the diesel generator power parameter.","PeriodicalId":50052,"journal":{"name":"Journal of Ship Research","volume":" ","pages":""},"PeriodicalIF":1.3000,"publicationDate":"2022-09-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"2","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Journal of Ship Research","FirstCategoryId":"5","ListUrlMain":"https://doi.org/10.5957/josr.09210032","RegionNum":4,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q3","JCRName":"ENGINEERING, CIVIL","Score":null,"Total":0}
引用次数: 2
Abstract
Of all the causes of accidents to ships, 14% pertains to damage due to ship equipment. Accordingly, the maritime industry is currently considering state-of-the-art maintenance and inspection processes, an example of which is condition-based maintenance (CBM). This is a strategy that hinges on the condition monitoring (CM) of assets. CM has proven to increase efficiency, reliability, profitability, and performance of vessel. To enable this maintenance strategy, sensors need to be installed along the most critical ship components and around the environment where these assets are operating through the application of Internet of Ships (IoS). IoS has demonstrated to be effective for collecting data in real time as well as performing diagnosis and prognosis to assess the current and future health of machinery to assist instant decision-making. The employment of IoS presents several challenges, an example of which is the imputation of missing values. Data imputation is a compelling preprocessing step, the aim of this is to estimate identified missing values to avoid underutilization of data. This data preparation step has gained popularity over the last few years due to its importance when dealing with Industrial Internet of Things (IIoT) sensor data. Although some articles presented new methodologies to impute missing values from sensor data of marine machinery based on machine learning methodologies, deep learning models have not yet been considered. For this reason, variational autoencoders for imputing missing values from sensor data of marine systems are analyzed in this article. To assess the performance of variational autoencoders as imputation methods, a comparative study is performed with widely implemented imputation techniques. Mean imputation, Forward Fill and Backward Fill, and k-Nearest Neighbors are considered. To that end, a case study on marine machinery system parameters obtained from sensors installed on a diesel generator of a tanker ship is performed. Results demonstrate the applicability of variational autoencoders when dealing with missing values of marine machinery systems sensor data, achieving a coefficient of determination of 0.99 when imputing missing values of the diesel generator power parameter.
期刊介绍:
Original and Timely technical papers addressing problems of shipyard techniques and production of merchant and naval ships appear in this quarterly publication. Since its inception, the Journal of Ship Production and Design (formerly the Journal of Ship Production) has been a forum for peer-reviewed, professionally edited papers from academic and industry sources. As such, it has influenced the worldwide development of ship production engineering as a fully qualified professional discipline. The expanded scope seeks papers in additional areas, specifically ship design, including design for production, plus other marine technology topics, such as ship operations, shipping economic, and safety. Each issue contains a well-rounded selection of technical papers relevant to marine professionals.