Finding corrosion and fabric defects by close-up visual inspection is time-consuming and inefficient. Moreover, this traditional approach to corrosion inspection cannot track and predict corrosion development over time. In contrast, by leveraging artificial intelligence (AI) algorithms, Inspection² AI software can automatically detect and predict corrosion and coating breakdown.

In this third instalment of the series “AI in Action“, we will explore the detection of corrosion on our customers’ industrial assets.  We will discuss our four proprietary approaches, which can also be applied to many other anomalies:

  • AI Labelling Strategy
  • Tuning of AI model parameters
  • Prioritisation of Results
  • Trending of Results

A corrosion check can take weeks or even years. Using Inspection²’s AI, Computer Vision techniques will enable more efficient monitoring and deliver consistent, faster, and cheaper corrosion detection.

In addition, asset owners will be able to make data-driven decisions about fabric maintenance optimisation, potentially saving them hefty repair costs down the line.

Technique 1 – Data Labelling Strategy

We developed our first Artificial Intelligent models six years ago to detect corrosion in offshore Oil & Gas rigs. These AI models have since been expanded to include onshore assets, lattice towers and other industrial assets.

Instead of applying engineering principles to the design and strategy of labelling corrosion, the most common approach involves labelling large volumes of images. Thus, it has led to the detection of surface rust instead of corrosion and severity classification.

Our approach is different.

  • First, we define the different grades of corrosion. We have devised an engineering classification based on 5 grades of corrosion, where 1 is the lowest and 5 is the worst.
  • We also build AI to detect staining and rust as we are interested in excluding them to create a more accurate AI.
  • We have a set of techniques that include environmental factors in the training and analysis, based on the type of the asset.

While accuracy rates vary depending on the nature of the organic corrosion and the customer’s requirements, we usually achieve between 80-90% accuracy.

Examples of AI in action showing corrosion detection grading based on coloured bounding boxes

Examples of AI in action, showing not only corrosion detection but also grading based on coloured bounding boxes

Technique 2 – AI Model Parameters Tuning

Building and training an AI takes time. Once the AI is accurate enough, it is a simple matter of tuning it to the nature of your assets and objectives. 

When an asset is old and prone to corrosion, such as a North Sea oil rig, detecting every spot of corrosion would yield a large amount of data that was neither useful nor worth analysing.

This is what AI tuning is about, applying AI as a tool to get the results you want, not as a blanket detection mechanism.

In this example, the asset was relatively new and in good condition. Preventive maintenance was the objective here rather than repairing critical anomalies. The AI is finely tuned to detect even the slightest spots of rust on edges or rust patches caused by paint flaking that exposed the base metal.

Detect corrosion on gas tower

Instead of tagging every individual corrosion spot, we tuned the AI to detect only the critical points and highlights them.

Here is an example of adjusting different levels of granularity on the same image to detect only the key corrosion points. Using the same image, we added a lot more sensitivity and granularity to the AI to detect many more points of corrosion and classifications of grading.

Detect corrosion and classify severity by adjusting AI parameters

Technique 3 – Prioritisation Of Results

The next step is to set priorities for the maintenance or corrective actions. The issue is more complex than just addressing the most severe corrosion spots.

To assess any maintenance on industrial assets, the location of corrosion and what component it is on are of paramount importance. The corrosion spotted on pressurised equipment is more concerning than corrosion in non-load bearing elements.

As a result, we also build AI models that identify the critical components of the asset including handrails, nuts and bolts, flanges, valves, etc.

By using AI to detect and grade corrosion, and then to identify what components it is on, we can automate the risk assessment using our Risk Management software.

These outputs and techniques allow you to automatically pinpoint the most critical risks and problems associated with your assets.

AI detection of nuts and bolts on high pressure pipework

Example of AI detection of ‘nuts and bolts’ on high-pressure pipework.

Technique 4 – Trending Of Results

The above section discusses our process to assess the immediate issues and risks that corrosion poses to assets. 

Corrosion is a slow-moving anomaly. Even if it is not a priority to fix, it is beneficial to track and trend corrosion. It is important to consider the rate of corrosion growth in determining whether it should be prioritised. 

If you have the right tool to track corrosion trends, you can perform predictive analytics on your asset. You then have an understanding of the condition of your assets and maintenance spending. 

Using the Inspection² Computer Vision software, you can mark the areas of corrosion that the AI has identified, e.g. with three different colours indicating high, medium, and low corrosion severity.

AI in Action automate Corrosion Detection Classification

We then use reference measurement and pixel count techniques to measure the area of corrosion. Our change detection software will track these tagged corrosion spots over time and calculate the growth rates.

Corrosion trending and change detection

Inspection²’s  AI Computer Vision techniques compresses the time to complete complex corrosion checks from weeks or years to minutes. Enabling ready and rapid access to ongoing monitoring and detection. Empowering asset owners to make data driven decisions to proactively optimise the fabric maintenance. Potential issues can be identified and addressed before they occur, resulting in significant time and cost savings.

Pin It on Pinterest

Share This