Applied AI For Unmanned Oil & Gas Facilities?
In this article, and my next few articles, I will focus on cutting through the hype of AI to explore pragmatic, real-world use cases for AI from an operations perspective as opposed to a data science perspective.
In this article the intent is to explore AI in the context of unmanned aka not-normally manned facilities.
Simply taking an existing facility, bolting on AI and declaring unmanned operations an unmitigated success amid fanfare, smoke and mirrors is the worst possible scenario.
We need to start with the fundamentals. A facility needs to be designed from the ground up, in the case of green-field, or re-engineered appropriately, in the case of brown-field, to support unmanned or not-normally manned operations – period.
What specifically do I mean?
An unmanned facility mandates a highly instrumented solution. The facility has to be designed or re-engineered to be as inherently reliable as possible by design. Single point failures need to be designed-out and redundancy installed for critical functions. The probability of significant trip events needs to be minimised. Utilisation of proven, reliable equipment augmented with contemporary technology.
Assuming we have a fit-for purpose, inherently reliable facility, it doesn’t stop there. We need to properly support the facility throughout its life-cycle. Again, what specifically do I mean?
The facilities need to be supported with highly skilled, multi-disciplined campaign style maintenance crews. Training and retaining highly skilled, multi-disciplined campaign style maintenance crews brings its own challenges.
It is essential to understand and optimise the key limiting factor(s) that ultimately determine/fix the frequency for maintenance and thus impact the planning and scheduling around mobilisation of campaign style maintenance crews.
The facilities and maintenance crews need to be supported with an appropriate repair and spares strategy and focused spare part holding regime commensurate with the unmanned philosophy and campaign style maintenance – this is very different from a manned facility.
An unmanned facility is NOT a low cost option from a Capex perspective.
The two extremes of unmanned facilities that I can think of are a land based remote facility with little or no infrastructure complete with challenging logistics versus a remote offshore facility tied-back to a central processing facility. In an offshore facility the capex overhead of installing additional redundancy for critical functions can be partly offset by not having to engineer a large accommodation module or heavy jacket structure for example. In an unmanned/not normally manned facility significantly fewer people are exposed to risk for significantly reduced periods of time although risks peak during mobilisation and demobilisation activities.
Where does this leave Applied AI? I see Applied AI, Deep Machine Learning and Natural Language Processing being of value in 4 main areas:
- Predictive Analytics
- Decision Support
In my view, the combined technology is about reducing uncertainty in a given process or outcome and correlating, seemingly unconnected, disparate sources of data in a meaningful way to provide useful information as the datum for decision support. Some of this technology and mode of application is available currently and some is future state, the intent is to explore what the future could look like and how Applied AI could help. I will explore specific areas in future articles.
- Very early identification of developing anomalies - both known & unknown.
- Labelling of known anomalies including an estimate of probability/certainty.
- Tagging of unknown anomalies and identification of historical close matches including an estimate of probability/certainty.
- Consolidation of the corroborating data/evidence (condition based, process, machine, unstructured) indicating the number of independent sources of data that corroborate the prediction.
- Estimated warning time until functional failure including an estimate of probability/certainty.
- Recording operational context and anomaly signature.
- Correlation of the timeline & sequence of events with the facility event log.
- Decision support data to support a risk optimised emergency response or required course of action – in the first few Seconds/Minutes/Hours.
- Decision support data to support the safe, continued operation of the plant including safe operation period & estimate of probability/certainty.
- Decision support data to provide a quantified risk of continued operation in terms of safety, environment, image, consequential additional secondary damage/repair costs.
- Decision support data to identify available mitigation options, including: installed spare, cannibalisation, formal deviation, bypass/re-routing/plant re-config, reduced throughput, temporary equipment etc.
- Decision support data to identify the next planned outage date and projected scope.
- Decision support data to identify other opportunistic maintenance that could be pulled forward, planned and scheduled if equipment outage is required, identification of work orders which are currently pending/in backlog.
- Decision support data to identify & support mandatory work process requirements to resolve issue (Formal Risk Assessment / Deviation / MoC / Control / Reportable Incident / Requirement To Inform Authorities / Regulatory Bodies etc).
- Decision support data to identify the impacted Plant/Systems/Packages/Skids/Functional Locations/Equipment/Make/Model/Serial Numbers/Data Sheets/Drawing Refs/Materials/BOMs etc.
- Decision support data to identify spare parts historically consumed for repairs, the availability from stock of the identified spare part and/or estimated lead times.
- Decision support data to identify additional/specialist resources, support, special tooling, chemicals, consumables, deviations/controls required to be put in place.
- Historical data to support the identification of any precedence of a technically similar anomaly occurrence on same/other similar equipment?
- Historical data to support the identification of the risk exposure in terms of safety, environment, image, production, repair, spares, resources etc for any previous technically similar occurrences.
- Historical work order data of technically similar equipment and RCAs to support the path forward?
- Historical data of technically similar anomalies to identify/frame additional resources, support, special tooling, chemicals, consumables, spare parts, deviations/controls that might be required.
- Historical data to quantify which spare parts were previously consumed, their stocks levels and/or estimated lead times?
- The potential for AI to assist Master Data is vast but most data science examples center around the wrong issues or non-existent issues.
- Data is one the biggest issues facing the Oil & Gas Industry going forward and has been getting steadily worse in the last 25 years, not better.
- In order to have a disruptive influence and step change in this area we need to stem the flow of poor quality data in preference to trying to apply corrective measures once the damage is done.
- As an industry we need to fundamentally change how we currently technically specify, collect and use/analyse data.
- For example, one of the biggest challenges in the Oil & Gas industry is spare part data. When a failure occurs and spares are required, more often than not the WMS does not contain any spares/materials information or, commonly, the wrong spares/materials data. The more we know about a tag from source data the better chance we have of identifying the correct spare parts which correlate with the installed base on site.
- The more correlating “meta” or “characteristic” data we have for a tag then the better the chance of accurate identification.
- Taking a very simplistic example, if we have an equipment’s correct tag number and original purchase order information this simple relationship can narrow down the range of possibilities or, in other words, increase the probability that we are correct exponentially. It may be the case that if the tag was supplied under the stated purchase order then only 2 or 3 types of the same device were supplied in the PO therefore the range of options have been narrowed from 1000s to 2 or 3.
- Using AI to develop these intelligent networks and providing the data in an AI friendly way to start with, I am convinced is the way forward.
In an unmanned/not normally manned facility the problems are exacerbated, you have to mobilise to site with a right first time mindset, with the appropriate skilled resources, with the correct spares and a clear, focused scope of work.
AI can be of significant value in this context starting with very early detection of developing anomalies which in turn provides the widest possible planning and scheduling window to give humans the headspace required to make correct, optimal, informed decisions underpinned by meaningful data. Integrating and corroborating disparate historical and predictive data sources in this way helps fully utilise historical data to address the immediate issues but also enhance predictive tools to further improve going forward.
COPYRIGHT DUBAIDUBAIDUBAI.COM 2019