The offshore oil & gas industry is exposed to extreme levels of risk. Companies and equipment are subject to volatile price fluctuations, extreme weather conditions and operational hazards like explosions, spills and workplace injuries. This is why the importance of predictive maintenance of the drilling equipment is much more intensified and puts additional stress on offshore maintenance providers as minor errors in predictions may lead to massive scale consequences.
Julian Zec, the Former Chief Engineer & Manager CM, CBM, Predictive Maintenance & Reliability Engineering at National Oilwell Varco (now Senior Program Manager at Maersk Drilling), shared lessons learned and challenges they’ve encountered with providing condition-based maintenance in the offshore oil & gas industry and digitalisation of maintenance at the Maintenance Analytics Summit 2019.
Offshore maintenance analytics
When we talk about analytics, Julian states that it is always a simplification of reality. Analytics is not almighty, over natural force – failures shall still happen and are part of the daily operations. We hope to improve our situation. He advises that validating and measuring the effectiveness of analytics should be towards real observations, and not idealized or isolated sets of procedures.
“What we don’t want to have is an illusion of control,” highlights Julian. Predictive analytics that is deployed should be taken as a learning point about how to improve certain functionality, not treat factors such as bad weather, that predictive analytics cannot cover, states Julian.
However, when it comes to measuring performance and maintenance, there is a strong relationship between failing and process improvement. Failures of the offshore equipment are very relevant KPIs for condition-based monitoring.
Through their CBM programme, they collects all kind of failure data that are caused by:
Through their CBM programme, they collects all kind of failure data that are caused by:
- Design and operation – that can be predicted, and
- Natural phenomena, which can’t be predicted.
The insight into the overall failure statistics provides them with opportunities in predictive maintenance.
Maintenance analytics should be towards real observations, and not idealised or isolated set of procedures.
The traditional maintenance model in the offshore industry that they had worked with previously relied on the calendar-based or scheduled maintenance model. It involved adhering to OEM manuals and disassembling the rigs and machines every five years. However, they experienced a steep increase in spending after the first five years, which could amount to 100 million dollars, explains Julian.
This was a significant concern for the offshore industry, but no concrete actions were taken until the oil crisis in 2014. In 2016 the price of oil plummeted to 20$ and it was the wake-up call and an occasion for National Oilwell Varco to start really implementing other maintenance models.
The new business model they introduced was condition-based monitoring and they identified three use cases where they can create value:
- Ensuring operational uptime of equipment
- Optimisation of maintenance
- Removal of 5-years scheduled inspections
The lessons learnt with the new CBM model
1. It’s not only about failure detection – Only detecting a failure doesn’t provide any value; it only includes cost if something is not done with it, states Julian.
They had to gain control of all three pillars of predictive maintenance:
- Getting the data and analytics right
- Ability to conclude and recommend
- Ability to execute.
The execution part is more complicated for the offshore industry because the mobility of the assets – drillships are sailing all over the globe, and manufacturing facilities can’t follow.
One crucial thing they realised when they started with the digitalisation of their maintenance was that people had misconceptions about how simple the data analytics process is.
It was believed it was as simple as getting the data, doing analytics, presenting results via data visualisation, pushing a button and the job got done. Even this takes days to go from the first to the last step.
But in reality, maintenance analytics in the offshore industry with ships all over the world is a bit more complicated than that. First of all, there are a lot of data sources, checklists, connected and not connected instruments. The data coming from them is fed into many different systems provided by different vendors and presented on various portals. And because there are many machines and equipment involved, more people are needed to analyse the data who discuss and make decisions. Then they prepare a purchase or work order and documentation and mobilise operators to send them to the location where intervention is needed. So considering all these, going through all these stages from detection to intervention takes not days, but weeks. Improvement has to be implemented on all levels.
So the primary learning was that they need a system that would detect events weeks before to create value.
2. Software development – To work with the data, they needed software. So believing they can do it alone, they started creating software applications – which was a big challenge and a lesson learned, reveals Julian.
Like in all software development, Julian’s team started with requirements and user stories that they need to build the software. But soon things got out of control and different teams pulled in different directions, and no one made any real value producing progress.
Eventually, there was a software application, but analytics was in another application, and Julians’ reliability team was still using spreadsheets.
The first mistake they made was they didn’t start with the business goal, but instead, they focused on technology.
3. Collaboration is the key – National Oilwell Varco had issues with communication at the beginnings because they had many actors involved in the process:
- Design engineers
- Aftermarket expertise & process
- Infrastructure & data science
- Customer goals
- Customer experience
- Regulation bodies
They had to make sure all these actors work hand in hand on the same goal, and for that, a use case had to be created that benefits everyone. The solution was for everybody to start sharing the data they had. Time-series data, in combination with customers’ maintenance records, enabled them to produce algorithms to detect failures and do failure analysis. They could pinpoint what exactly was going on with a machine and prescribe actions.
4. Integration and scalability – However, one successful use case that worked for one ship had to work for all 73 ships they had. They had to scale the process. The different kinds of knowledge originating from humans, data and documents, and different kinds of information shared in a different manner. Additionally, they had been planning their own, 3rd party and customer systems interconnected in their data architecture across which they had to ensure cybersecurity, data quality, system responsibility and service level agreement to get quality results on the receiving end.
The future of data-driven analytics and predictive maintenance
Going back to everything they went through in establishing their maintenance analytics, we need to state that there are significant challenges in leading and implementing it. Many companies dove into analytics but didn’t see any practical results, especially from machine learning. But probably it is the way of progress.
The pilot may also blind organisations, while producing some local results , but ending nowhere, state Julian. That’s because pilot projects are not a digital strategy. Oil & gas industry has not improved itself because they basked in a long, fruitful period of good oil prices. Julian states that even today in the data-driven era, companies are exploiting low-hanging fruits. And only after these simple optimisation methods are done, the industry can move towards the domain of artificial intelligence, emphasises Julian.
What should we do to move towards AI?
The first step that companies can take to start fully utilising AI into their operations is to look beyond equipment failures, and instead focus on digitalisation and building comprehensive AI-driven asset management strategy, advises Julian. That entails connecting the condition with the global spares, manufacturing, supply chain management, logistics, and creating interconnected micro-incentives and operational windows.
Another thing organisations should look into is AI-driven project management. If a smart strategy does not regulate smart predictive systems, they are creating recommendations which are difficult to accept as they go against our human gut feeling. Oil & gas companies are operating defensively by putting out fires; they’re identifying problematic areas and acting based on intermediate danger. But intelligent systems are prescribing strategic actions that may not have an apparent intermediate benefit. So gaining trust in the AI systems should be the next frontier for companies, Julian states. They should gradually adopt these “non-logical recommendations” which go against the human gut feeling.
Julian sees many opportunities for analytics in maintenance, such as information management, reliability management and asset management. And he hopes that there will continue movement towards building smart functionalities in the industry.