There’s no room for subjective opinions when the reliability of your FRP industrial assets is on the line.
UTComp’s UltraAnalytix® non-destructive evaluation (NDE) system provides objective, data-driven Fitness For Service assessments of FRP pipes, tanks and other equipment used to store, process and transport hazardous substances.
The system is powered by a patented AI algorithm that draws on millions of inspection data points to process and analyze raw ultrasonic data to identify changes in polymer condition.
By tracking these changes over time, inspectors and asset owners are able to make informed decisions about equipment maintenance and repairs while avoiding unplanned shutdowns and unnecessary replacement costs.
But what do we mean when something is “powered” by AI?
AI doom or boom?
Depending on who you ask, artificial intelligence (AI) will either destroy or save humanity.
The truth lies somewhere in between.
AI is already having a profound impact in a wide range of industries, from agriculture to finance, health care, and manufacturing. In the world of FRP, AI and machine learning have enabled major advances in material design and manufacturing of advanced FRP composites.
But there are plenty of examples of the hype racing ahead of reality.
A recent Apple research paper found that even the most advanced Large Reasoning Models (LRMs) that power AI agents like Chat GPT-4 and Deep Seek have serious limitations and can suffer from “complete accuracy collapse” when performing tasks that are too complex.
In the FRP integrity management business, any “accuracy collapse” could have terrible consequences.
So what does the future hold for AI in FRP inspection and Fitness For Service Assessment?
AI is a powerful pattern recognition tool
Fundamentally, AI is a broad and very useful pattern recognition tool.
That’s all – AI is not magic, or even “intelligent” in the way that humans are.
It’s a computer program or algorithm that has been trained on a dataset to recognize or identify patterns, trends and probabilities and use that information to make predictions or decisions about new sets of data. For example, the UltraAnalytix algorithm takes raw ultrasonic data, as well as simple structured data from external visual inspections, and transforms it into actionable insights about the condition of FRP equipment.
AI tools enhance FRP inspection and assessment in a number of ways, including:
- Automated analysis of large quantities of data: AI can analyze vast amounts of data (images, sensor readings, text, spreadsheets and more) obtained from various NDT methods to efficiently detect and characterize flaws in FRP equipment.
- Superior accuracy and sensitivity: AI can identify micro-cracks, delaminations, voids and other defects that manual inspection and analysis might otherwise miss.
- Reliable results: AI systems, while not perfect, reduce the potential for human errors in visual inspection and data interpretation.
- Predictive analytics: AI can analyse historical and real-time data to provide insights into FRP condition and performance, predict future failures, and optimize equipment maintenance and replacement schedules.
AI is only as good as the dataset you feed into it
There are lots of conversations these days about the potential benefits of AI in the mechanical integrity field.
But progress has been slowed by the lack of published or publicly available datasets that could be used to train AI inspection and assessment tools for FRP. There is data out there, but most of it is not well-assembled or is proprietary information held by materials manufacturers and other private interests.
If you don’t have a good dataset to start with, it’s going to be really difficult to use AI for anything.
This lack of publicly available data is both a source of frustration and an opportunity for our industry.
However, from the beginning, UTComp has approached FRP inspection and assessment with the intention of automating the process as much as possible.
As a result, over the past couple of decades, we’ve assembled a relatively large dataset that contains observations and calculations from thousands of inspections describing FRP behavior, mechanical properties and changes in those properties over time.
This growing dataset now contains approximately 2 million individual data points, and many of those are the result of hundreds of calculations needed to attach a numerical value to a particular flaw or condition.
This provides the mathematical foundation needed to convert an ultrasonic reading into a Polymer Damage Status (PDS) value, which can then be used to calculate Remaining Strength Factor (RSF) of the FRP material being inspected.
AI also improves how we serve our clients in other ways. For instance, it dramatically speeds up Fitness For Service reporting: typically, in the process of producing one inspection report, we do several million calculations. In fact, even getting the data from the field into a file that we can analyze takes several hundred-thousand calculations.
All of this would be impossible without AI. Well, not impossible, but it would take a very, very long time.
Consensus needed to speed AI adoption for FRP
When it comes to building and deploying AI, more high-quality data is always better than less.
For example, AI tools have become indispensable in the field of medical imaging and diagnosis. Microsoft announced recently that its medical AI system successfully diagnosed 85% of cases in the New England Journal of Medicine. The AI was four times better at diagnosing complex cases than human doctors.
How is that possible? AI is incredibly useful in medical diagnostic imaging because doctors around the world use the same measurements that can be combined into a common database.
Large, high-quality datasets have been widely available for years to train the algorithms and there are agreed-upon standards, methods and tools for doctors to gather and interpret diagnostic information.
When it comes to FRP Fitness For Service evaluation, we’re not there yet.
The FRP industry would benefit from an agreed-upon platform and methodology to identify and characterize FRP damage in numerical terms.
This would enable access to the high-quality data needed to develop reliable, trustworthy AI models, encourage adoption of the technology and drive the industry forward.
At UTComp, we’ve been able to create quantitative inspection data – including visual observations – so that we can apply these pattern recognition and AI techniques to provide quantitative reliability assessments.
The next step to making this available to the industry is to create the first coherent, complete and consensus in-service inspection standard for FRP equipment to dovetail with requirements of the new part being developed for the API 579-1 / ASME-FFS-1 Fitness For Service assessment standard.