Natural gas is a very flexible source of energy that can help us bridge the gap between our current high-carbon economy and our zero-carbon future.
World’s largest producer of Liquefied Natural Gas (LNG)
Maximize production at lower cost to keep profit margins in an ever increasingly volatile market
Few environments are mission critical at the level of gas liquefaction, taking place in plants of high technological complexity where everything must work as intended all the time without fail, as a glitch can quickly turn into an accident with catastrophic consequences. Moreover, to keep profit margins in an ever increasingly volatile market, with prices suffering from periods of oversupply, the world’s largest producer of Liquefied Natural Gas (LNG) needed to maximize production at lower cost.
The critical success factor became Predictive Maintenance of LNG Plant Assets though digital innovation, which allows evaluating the condition of LNG operational assets and predicting the related Maintenance requirements and actions, in order to achieve optimum performance and prevent malfunctions.
Agile Lab produced a comprehensive feasibility study of the solution, based on the new Digital Twin concepts currently rising in Oil and Gas operations. This extended well beyond the boundaries of “traditional” maintenance technologies, thanks to a solution centered on the adoption of the witboost platform as a technological stack composed by innovative and robust open source modules within a micro-service architecture.
The feasibility study provided by Agile Lab enabled the Client to define a detailed Plan for its “Predictive Maintenance Digital Revamping” initiative and to successfully follow-up this path.
Our Client is a major global energy operator in the development, production, and sales of hydrocarbons. It is the largest producer of Liquefied Natural Gas (LNG) in the world, with a total annual capacity of 77 million tons. It is considered a LNG industry benchmark in terms of safety and reliability of energy supply to its worldwide Clients. To maintain its upstream assets in several non-associated gas fields – including the world’s largest – it has 13 LNG train plants for gas liquefaction and purification.
In order to be transported from one country to another in a commercially viable manner, the natural gas has to be condensed in liquid form, thus reducing its volume by about 600 times. This allows the storage and transport (by sea vessels over long distances) of great quantities of energy at competitive costs. To achieve liquefaction, the gas must be cooled town to – 161 °C, and this process requires strict safety precautions during all stages due to its flammability. Moreover, the gas must be purified before it can be liquefied, as impurities, common in raw gas, freeze at different temperatures and they would block the cryogenic section of the plant.
Each LNG plant consists of one or more lines (trains) to convert the gas into liquid, and a typical train comprises several dedicated areas:
- treatment (dehydration and removal of carbon dioxide, hydrogen sulphide, mercury etc.),
- extraction and fractionation of heavy components,
- compression and storage,
- dock for loading vessels,
- auxiliary areas.
The LNG Asset Maintenance Operations are therefore very complex and intensive, with related high operating costs.
Moreover – although a LNG process plant is designed to operate optimally at a design production level – any variations in feedstock and gas composition, changes in ambient conditions, and degradation of operating assets can impact throughput and optimal processing conditions, affecting operating profit. Stricter regulations for environmental protection, gas quality and safety have also led to an increase in operating costs, adding pressure to maintain profit margins.
In the last few years, worldwide government policies encouraging a transition to clean energy have generated a rise in demand for natural gas. In particular, natural gas plays an ever growing role in replacing more polluting fuel, such as coal, for electricity generation.
At first this created a favourable situation for the established producers, with profitability guaranteed by demand-driven prices and long-term purchase agreements. More recently, however, new fracking technologies to extract shale gas have resulted in an unprecedented boom in gas production, dramatically altering the market scenario and liquefied natural gas (LNG) prices have fallen due to oversupply. To survive and thrive in this new economic environment, maximizing LNG production at lower cost is imperative and, as a result, Predictive Maintenance of LNG Plant Assets though digital innovation has become a critical success factor. Economic slowdowns in various areas of the world are also increasing hydrocarbons’ price volatility, giving an additional impulse to the adoption of Predictive Maintenance technologies. These help cut back on operational expenditure, by optimizing maintenance scheduling, and drive productivity, preventing unplanned breakdowns and guaranteeing the maximum flexibility of LNG trains, also preventing bottlenecks where gas needs to be liquefied for overseas transportation.
In this context, our Client’s priority was to quickly define the technical and project feasibility for a “Predictive Maintenance Digital Revamping” solution, together with the specific underlying key components. In conjunction with condition monitoring by sensors, new Predictive Maintenance tools should be able to exploit value from historical data taken from machinery and equipment, spotting signs of deterioration as well as estimating asset components longevity. All this with the overall goal of reducing costs and time spent on maintenance, scheduling corrective maintenance and preventing failures, breakages and consequent production bottlenecks, preparing in advance by sourcing parts, equipment and labor for scheduled maintenance.
Agile Lab produced a comprehensive feasibility study of the solution. This is based on the new Digital Twin concepts currently rising in Oil and Gas operations and extended well beyond the boundaries of “traditional” maintenance technologies, in order to provide a reliable tool for plant asset integrity monitoring. The solution is focused on Predictive Maintenance specific Models and tools, aimed at evaluating the condition of LNG operational assets and predicting the related Maintenance requirements and actions, in order to achieve optimum performance and prevent malfunctions. This is to be achieved through the adoption of automated condition monitoring and advanced data analytics, to gather vital equipment statistics – such as vibration, temperature, sound, electric current, and so on – and to compare them with historical records of analogous equipment for the detection of signs of deterioration.
The solution is centered on the adoption of the witboost platform as a technological stack, composed by innovative and robust open source modules within a micro-service architecture. This guarantees the needed flexibility to be able to apply various and complex Predictive Maintenance Modeling techniques on the wide domain of data produced daily in the Client’s upstream LNG operational processes.
The solution focus is on extracting the maximum possible value from the Plant Operations generated big data and fully exploiting it to obtain predictive operational insights and real-time monitoring capabilities. Smart analytical models and AI Machine Learning are to be applied to LNG Predictive Maintenance gathering the Industrial Internet of Things (IIoT) data – describing the condition of assets and equipments – in real time, through sensors and sensor network technologies.
Traditional IT systems were not designed to manage and analyze the volume of big data produced by the Industrial IoT sensors. The powerful analytical features provided by witboost Data Engineering Boost modules, on the other hand, offer a flexible and open solution for processing, analysis and control of these kind of large data sets. witboost is a feature-rich, highly scalable framework, ideal for a IoT / IIoT context: it performs batch processing with ease, but its forte is real time streaming data analysis at scale. witboost Data Streams (formerly Wasp) is the core module of witboost for these kind of processes and it provides Streaming Analytics and Machine Learning at scale. It’s a pre-integrated framework based upon Kafka and Spark Streaming (or Flink), fully compliant with all most used Big Data technologies.
The feasibility study includes all the relevant Project dimensions, starting with a preliminary Exploratory Data Analysis and Data Quality Assessment related to data from sensors environment and current systems. This is followed by Requirement gathering, design of Data Architecture and Data integration interfaces with the existing systems (like, for instance, Osisoft PI and SAP), ML Modeling techniques and related data ingestion requirements, design of required Dashboards, and so on.
It defines the scope and the detailed size of both the Client’s and Agile Lab’s resources to be involved in each agile project activity (e.g. IT Architect Data Engineers, Data Scientists, Machinery Control Subject Matter Experts, Scrum Master, and so on). Agile Scrum is to be adopted as project methodology, as it is widely used on similar projects in complexity and scope.
Because of the central importance of Machine Learning Modeling for the specific LNG context – in particular for the Predictive Maintenance of LNG Turbine-Compressors components – the study dedicates a particular focus to identifying and deepening the proper range of applicable statistical and/or data-driven Modeling Techniques.
With this aim, Agile Lab involved Subject Matter Expert academic resources from its university network in the study.
Amongst the techniques that underwent a deep study were:
- Data pre-processing methods (e.g. missing value analysis, missing data handling, data noise reduction/removal)
- Sensor Fault detection based on Improved-Detection, Classification and Integrated Diagnostic of Gas Turbines Sensors Methodology (I-DCIDS). This is because sensors installed on gas turbines are fundamental in monitoring the turbine’s current health state, but the sensor itself can potentially be error prone, thus leading to unreliable measurements and consequently to erroneous diagnostic interpretations
- Multiple techniques dedicated to Predictive Maintenance for Failure Prediction:
- Data-driven methods. Data-driven anomaly detection, for detecting anomaly based on historical data of gas turbines (“health” data and fault / degradation data) without requiring accurate physical mathematical models of the systems.
- Model-based / physics-based methods. These methods require domain expertise to build accurate mathematical models of gas turbines for fault diagnosis, for example linear gas path analysis (GPA), non-linear GPA, Kalman filters, and so on. They attempts to estimate the evolution of gas turbine degradation by means of simulation models based on physics and thermodynamic principles
- Data-driven methods and Asset prior Knowledge Fusion, taking into account the unique characteristics of specific assets and systematically combining the data-driven and model-based methodologies
- Specialized methods for Predictive Maintenance (e.g. Bayesian Hierarchical Models for Gas Turbine Health State Prognostics, Anomaly Detection with Normal pattern extraction, Early Notification, Early Notification as Remaining Useful Life Estimation, Slow-down anticipation & Transition Monitoring, Trip Detection through Gas Turbine Transients Clustering).
Achieved results & benefits
The feasibility study provided by Agile Lab enabled the Client to define a detailed consequent Plan for its “Predictive Maintenance Digital Revamping” initiative and to follow-up this path, taking in account the key underlying elements:
- Open Solution Architecture with related scalability features
- Data Architecture and Integration Architecture
- Industrial IoT Big Data Ingestion / Integration requirements
- Preliminary solution requirements related to multiple stakeholders: Operations, Business Users, IT & Data Engineering, Data Scientists
- Knowledge of appropriate Predictive Maintenance Modeling Techniques to be applied – in particular ML Techniques for LNG Turbine-Compressors components
- Agile Project blueprint and roadmap
- Profile of skilled resources and specialized competences needed (very valuable but at same time very scarce within the plants)..