Big Data and Maritime Predictive Maintenance

 

 

Introduction

From 2000 onwards, an unprecedented data explosion followed, where it is still ongoing. With the advent of the fourth industrial revolution, the computational capacity to collect data has now increased signifcantly. The sources of these data streams come from all sectors: automotive, medical, IT, etc. The volume of data generated on a daily basis is increasing at a geometric rate and thus the term Big Data was created. Big Data (BD) refers to any form of data generated on a continuous basis from various sources (e.g. sensors).

Big Data mainly refers to large sets of complex data, structured or unstructured, on which traditional processing techniques and corresponding algorithms cannot work. Therefore, new procedures and methods are required to process them in order to reveal hidden patterns, thus leading to the evolution of scientifc analysis and modelling from the traditional model-driven scientifc model to a data-based scientifc model.

On the other hand, new technologies emerging from the fourth industrial revolution integrate people, machines and products, allowing for faster and more targeted information exchange.

The majority of data collected from industrial systems contains information about the processes, events and alarms that take place on an industrial production line. Also, when processed and analyzed, this data can reveal valuable information and insights from the production process and overall system dynamics. Thus, by using statistical-mathematical approaches based on the data, it is possible to fnd explanatory results for strategic decision making, providing advantages such as reducing maintenance costs, reducing machine failure, reducing spare parts storage, increasing production, improving operator safety, repair verifcation, overall proft, etc. (Peres et al., 2018). These advantages are related to industrial equipment maintenance processes. In industries, equipment maintenance is an important key that affects the equipment uptime (life of the machinery) and its efciency.

Therefore, equipment failures must be identifed and resolved, avoiding the shutdown of production processes, with negative consequences on the economic operation of the industrial plant (Wan etal., 2017).

Basic Defnitions

They also have a wide range of applications, from industry, medicine, transport to military technology (Austin and Kusumoto, 2016). However, although Big Data has been around for some time, it still remains one of the biggest challenges, with a focus on data collection and preparing it for analysis. Different systems store data in different formats, even within the same company or organization. Gathering, standardizing and cleaning irregularity data remains a signifcant challenge. For example, the digital data generated is partly the result of the use of Internet-connected devices. Thus, smart phones (smartphones), tablets (tablets) and computers transmit data about their users. Connected smart systems transmit information about the consumer’s use of everyday objects. All these sources are the reservoir of big data generation and create the challenges for the future, such as how to store, pre-process, analyze in real time, etc. (Riahi and Riahi, 2018).

Big Data has characteristics that can bedescribed by seven (7) Vs: Volume, Variety, Velocity, Value, Visualization, Variability and Veracity. This huge volume of unprocessed Big Data data has limited value without undergoing proper processing and analysis. In classical data analysis it is necessary to utilize appropriate statistical methods to maximize
the value of the information. In addition to classical analysis, different methods of analysis may be followed depending on the data. In particular, Big Data analysis can be performed either in real time (on line/real time analysis) or in non-real time (ofine analysis). The analysis of MS can be classifed according to the level of memory, business activity and also on a mass level. In addition, the analysis of MDs can be done according to the complexity of the algorithms used (Emani et al., 2015).

Overall the processes of big data analysis (Riahi ans Riahi, 2018):
- Predictive type or model (Categorization-Classifcation, Regression or Interpolation-Regression), Time series analysis-Time series analysis, Prediction-Prediction, forecasting.
- Descriptive type or model (Clustering-Clustering, Summarization-Summarization or Generalization-Generalization, Finding Association Rules-Association Rules, Discovery of
Associations in Sequences-Pattern Discovery in Sequences).

In addition, in the processing of big data, data pre-processing is required. And the main tasks performed in data pre-processing are as follows (Yaqoob et al., 2016): In addition, in the processing of big data, data pre-processing is required. And the main tasks performed in data pre-processing are as follows (Yaqoob et al., 2016):

- Data cleaning (flling in missing values, smoothing data that have noise, identifying or removing outliers, resolving inconsistencies).
- Data Consolidation (consolidating multiple databases, data cubes, fles, etc.).
- Data transformation (normalization by scaling to a specifc range, aggregation).
- Data reduction (obtaining reduced representations in volume but producing the same or similar analytical results, data discretization and data aggregation, with dimensionality reduction, data compression and generalization).

On the other hand, industrial maintenance is divided into (Susto et al., 2015; Thyago et al., 2019):

- Corrective Breakdown Maintenance (R2F). It refers to maintenance after a machine failure and requires high repair costs. It occurs only when an equipment (e.g., machine) stops working. It is the simplest maintenance strategy since it is necessary to both stop production and repair the parts that need to be replaced, adding direct costs to the process.

Preventive / Scheduled Maintenance (PVM)

It refers to the periodic maintenance of machines based on manufacturer’s instructions, the acquired experience of maintenance personnel and recommendations of relevant international organisations. In particular, it is a maintenance technique that is performed periodically on a planned schedule to anticipate failures. It is generally an effective approach to avoid failures and major breakdowns on the production line. However, unnecessary corrective actions are taken, leading to an increase in operational costs.

Conditional / Predictive Maintenance

This is the maintenance procedure that is carried out only when a machine needs repair. It is the most economical maintenance method compared to the previous categories. Artifcial intelligence is used to detect wear and tear and faults. From the measurements (data) with proper analysis, classifcation, and prediction, various equipment failures that will occur during future use can be diagnosed.

It is based on the continuous monitoring of a machine or production line, allowing maintenance to be carried out only when necessary.

Big Data and Predictive Maintenance

The philosophy of Predictive Maintenance is to create transparency in the state of the machine and the use of available information to make maintenance decisions. In the next Figure, the framework of a big data platform for predictive maintenance is designed to more closely integrate data acquisition and the maintenance decision support system (MDSS), which highlights the data fow process in modeling diagnostic and predictive analysis for predictive maintenance (PdM).

The benefts of leveraging big data in predictive maintenance include: maximizing equipment uptime and uptime, delaying/reducing activity In general, predictive maintenance strategy deals with faults or failures before they occur. in industrial equipment, and in this, big data and its analysis techniques play an important role (Lee et al., 2017).

Maritime Applications

The use of big data in predictive maintenance is also useful in a maritime environment.

The reasons are their utilization in HUMS (Health and Usage Monitoring System) systems for

- Optimization of machines
- Minimizing costs
- Improving operational readiness.

For example, HUMS on warships, records the condition of their critical systems and components, so that early detection of progressive defects or indications of such defects is possible and therefore, correction can be achieved before they have a direct impact on operational safety (Shen et al., 2003).

Therefore, the utilization of big data in preventive maintenance is a panacea for optimal performance of HUMS systems.

 

By Dr. Nikitas Nikitakos,
Professor, Dept. of Shipping Trade and Transport, University of the Aegean | Head of Maritime Logistics & SC Management, Sharjah
Maritime Academy