AI is no longer an emerging technology.
It is an essential asset.
Exploration & production
Renewable power & fuels
We focus on the energy industry in areas of energy production & storage, energy transport & logistics, and industrial processing & manufacturing.
Networks & smart grids
Mining & metallurgy
AREAS OF EXPERTISE
We focus on the most important issues across many sectors.
Industrial analytics supported by machine learning models are key to migrating from time-based to condition-based maintenance (CBM). Rotating equipment vendors use sensors to create proprietary databases and platforms to increase equipment reliability and to predict when maintenance is required. Retrofitting sensors to “dumb” equipment to enable CBM is equipment agnostic and enables a transition to CBM for asset owners using legacy equipment.
Oil rigs typically have 20,000+ data tags for various purposes which are often inadequately utilised. A feasible challenge for machine learning is to create CBM models from these existing sensors to reduce OPEX and to reduce the need for expensive proprietary vendor systems. Selecting “just enough” data from existing sensors and then investing in proper data preparation is vital, as incomplete and error-prone time series data can result in inaccurate and unreliable models.
Corrosion detection using computer vision, essentially identifying the number of pixels containing anomalies in an image, was a natural progression from using human judgement. More advanced deep learning techniques provide much faster results with better accuracy - an example where incremental advancement in AI technology from other industries can help to solve asset integrity challenges.
We can manage all the steps in developing deep learning models using image data. Determining what is rust, what is paint and what is dirt is easy for the human eye, but even a modern data model requires thousands of correctly tagged images to obtain the same level of accuracy. Tagging these images has to be supervised by an expert, whether the challenge is rust on structures, weld defects, gouge positions or radiographic images. Once this laborious data preparation stage has been completed, the development of AI algorithms becomes a straightforward process.
The use of physics based models to optimise the many different stages of the drilling process can be supplemented with machine learning models such as Artificial Neural Networks and Support Vector Machines. The increasing availability and ease of sharing of structured (e.g., WITSML) and unstructured data allows machine learning models to be improved, resulting in increased drilling efficiency and more accurate well placement.
Future collaborative data sharing and model development at a basin level is increasingly key to gaining sustainable improvement. The use of Natural Language Processing for digitising historic well and drilling data is an area where machine learning increases the volume of decision support data available to well planning teams.
The highest standards in industrial safety and environmental protection are paramount for responsible asset owners. HSE data is often narrative, with routine inspection reports, near misses and incident reports written post event to drive future improvements. Natural language processing techniques allow deep analysis of large volumes of report data to find trends, risk factors and cause/effect relationships which are often hidden from human interpretation.
The energy industry has seen the need for increased visibility and sharing of information as well as providing remote access to critical equipment. This has led to the connection of traditionally isolated facilities to corporate networks and the internet. The risks from cyber-attacks can include plant downtime, equipment damage, HSE incidents and reputational damage.
A growing concern is that deep learning neural networks are sometimes vulnerable to ‘adversarial attack’. This means that carefully chosen inputs can cause AI to provide a wrong result whilst the input data appears to be perfectly normal to manual examination. The concept of adversarial attacks needs to be considered for projects where this kind of misclassification, leading to wrong results, could generate adverse outcomes to health, safety or the environment.
Members of our team have strong industrial cyber security expertise, and proper risk management will always be an integral part of our projects.
Commercialised machine learning applications to improve well productivity have been successfully deployed in US shale assets, where homogeneous geological conditions, simplicity, volume and single entity data ownership have allowed large and high-quality data sets to be developed. Elsewhere the challenges of assembling suitable data sets from multiple operators with different production monitoring systems in highly sophisticated off-shore environments have been restricting the development of successful machine learning models.
The challenge for operators and service providers alike is in data understanding and preparation – working collaboratively to pool data, invest in data cleansing and normalisation will be key to efficient AI roll-out. Endila is well equipped to create and manage multi-client projects.