Aize aim to cut asset surface inspection cost by 50% using computer vision AI solution

Aize aim to cut asset surface inspection cost by 50% using computer vision AI solution

Anoop Sharma
Anoop Sharma,

Rust detection in the heavy-asset energy industry is a cumbersome and costly process. Aize’s new asset management SaaS solution, powered by machine learning and computer vision, speeds up the process by auto-detection of rust anomalies.

Inspecting assets is an expensive, laborious process that requires flying an engineer, typically on a helicopter, to an oil rig or other heavy asset along, with a drone pilot. Some of the issues that energy operators face include how to – in a timely and efficient manner – identify anomalies such as surface coating breaking down, paint peeling off, corrosion and cracks, mechanical damage or underwater flora growing onto an asset.

Partnering with Cognizant for new solution

Performing rust detection without the benefit of any technical support can take many months. This means that getting ahead of maintenance issues, before something breaks down, can potentially save people’s lives and prevent disasters, such as oil spills and explosions.

To optimize this time- and cost-intensive manual asset monitoring, Aize, a Norway- and UK-based company specialized in project execution and operation in heavy-asset industries, partnered with Cognizant. The aim was to create a solution for automatic detection of rust anomalies for sub-surface and topside assets, with the help of a software platform using advanced machine learning. 

Machine learning in action

Cognizant helped Aize develop a novel machine learning solution that combines computer vision techniques with a methodology that combines a rules-based approach with fuzzy labeling known as 'weak-supervision.’ This process automates the rust detection process without requiring labor-intensive labeling (tagging).

Key capabilities of the new asset management platform include:

  • Feature engineering: This capability uses classical image processing to standardize images, and create basic and complex pixel-level attributes that also take into account neighboring pixels.
  • Weak supervision/fuzzy labeling (Rust Yes/No): This capability employs methodologies proposed in academic papers and Cognizant-developed hypotheses to determine if a particular pixel is part of a rust patch, e.g. is the set of pixel-level labels rust, not rust or unknown. Some of the labeling functions are straightforward, using common image processing and some are sophisticated multistep machine learning sub-processes. Some of the fuzzy labels focus on what is not rust to clear misclassification of run-offs.
  • Consolidation of fuzzy labels: This capability takes into account the interrelationships between the weak labels, combining them into a weakly labeled training set that is almost as good as those labeled manually by an expert.
  • Classification (Rust Yes/No): This capability uses fuzzy labels to train a machine learning classification, filling in for the gaps in the weak labeling through pattern recognition. 

 

The platform holds a digital twin, or digital representation, of an oil rig, vessel, gas container or any other heavy asset typical of those in use by Aize clients. At the heart of the solution, is the ability to auto-annotate suspected rust areas using still images extracted from video streams collected during drone flights over land-based storage tanks. This gives the possibility for engineers to work with the digital representation of the asset to plan inspections in detail, in advance of going to a location.

Towards autonomous inspections 

Aize sees its automated detection solution as the first step towards autonomous unmanned inspections using sophisticated sensors backed by weak supervision and powered by machine learning. 

All in all, the solution aims to help businesses save and achieve the following benefits:

  • Cuts down the effort required for integrity inspection by more than 50%, which translates to potential 50% cost savings for this operation
  • Automates inspections of assets for surface corrosion to prevent future issues
  • Lowers the number of resources needed for inspections 
  • Dramatically reduces cycle time and cost of rust detection, with improvement in the quality and integrity of operational work processes
  • Automatically suggests to engineers where to focus their attention by cutting out irrelevant scenes in the video stream


Aize plans to use the new solution to provide highly sophisticated capabilities that give clients a full picture of their heavy-asset operation at all times. With better intelligent data management and analytics, Aize endeavors to help customers keep energy assets running smoothly, and faster, with smarter decision-making.

To learn more, please visit Cognizant's Oil and gas section of the web.
 

Related Publication

Cloud eBook: Potential, pitfalls and best practices in the Nordics

Whether it’s migrating the entire core infrastructure to one or more cloud service providers, aka hyperscalers, or just simplifying a handful of processes using Software as a Service (SaaS) tools, the cloud represents the digital future. The pandemic has accelerated the ongoing cloud adoption and during 2021, 51% of Nordic organizations are predicted to reach IDC’s highest level of cloud maturity.