Bryce Bjornvick (Midstream Process Solutions
- MPS) has developed an expert-trained AI/ML solution to recover key
metadata from scanned pipeline documentation. Operators of the 3
million mile US natural gas pipeline network need to constantly access
and review documentation covering construction, operating pressure,
maintenance, and demonstrate compliance and realize financial value.
However, documentation is frequently of poor quality or incomplete.
Accessing these large document sets requires expert knowledge and is a
‘tedious and repetitive’ process.
MPS is developing AI/ML-based recognition technology to extract data
and context from documents such as ‘as-built’ welding records,
radiographic inspections and material test reports. The models were
trained with input from engineers. The system provides a measure of
data completeness and outputs the results to a map. Even with very poor
quality scans or handwritten documents, the system can capture around
half of the information required, speeding subsequent human auditing.
Overall, the system has reduced material test report processing by 85%.
MPS is now developing an application for data gathering during
construction and maintenance.
Yingchao Zhang demonstrated Singularity Systems ‘SingùAI’ platform that sits between data in human-readable formats (Word, HTML, PDF...) and applications such as SAP, Appian, Pega, UIPath and others. SingùAI combines computer vision, optical character recognition, machine learning and natural language processing into a data pipeline. Models are trained through a proprietary human/AI interface, driven by domain specialists with ‘no coding and no data science knowledge required’. Zhang concluded with some remarks on process automation. Few commercial business process solutions allow complex use cases to be fully automated. A key metric here is ‘straight through processing’, the percentage success rate of end-to-end data capture from document to source. 100% STP is almost impossible to achieve. The question then arises, what level STP can be considered success and how much effort and expense is required to achieve it. Zhang believes, ‘the single biggest factor is whether you are using a world class solution, a legacy tool, or people dressed as robots!’
Watch the presentations on Youtube and visit the conference home page and Petroleum Trade Network.
© Oil IT Journal - all rights reserved.