Case Study: Image Analytics Proof of Concept Results in External Visual Inspection Planning and Execution Efficiencies

In the traditional way of conducting an inspection, an API 570 inspector would use a work package to inspect a particular pipe segment. Still, as they go out to inspect, they may see multiple issues in the surrounding area. Because of the systems in place, that inspector most likely will not document the problems they see and will only write a report for the assigned piping segment. Typically, additional reporting will only happen if something significant is observed, such as a leak.

With the solution described above, the cameras and computers capture everything in the frame, not just the pipe being inspected. An inspector can capture multiple assets simultaneously and perform a basic screening in one click. From there, an SME can evaluate it to pinpoint precisely what assets need additional attention.  Other important aspects to highlight are:

• Screening can be conducted on an entire plant in just a matter of days, depending on plant size.
• Personnel can capture the images in just a fraction of the time it would take to do planned inspections.
• This methodology can be used to screen hard to reach areas with ease, such as pipe racks and high- temperature assets that would normally require shutdown or scaffolding to be built.

This POC created a heat map to identify areas of interest. It pulls up the associated image along with your GPS coordinates and gives a graph percentage of what kind of damage it sees. Computer imaging software and image analytics help eliminate human subjectivity. For example, an inspector may go into the field and make a recommendation to code align to a damage mechanism. Still, even when missing paint and light surface oxidation is observed during an SME review, the SME’s recommendation may be not to do anything.

Throughout the POC, SMEs validated the results, and in many cases, they agreed with the corrosion prediction from the computer vision models. Over 75% of the time, SMEs said that the results from ML algorithms met their expectations for external corrosion detection. Almost 50% of the time, the SMEs didn’t agree with each other, such as an inspector recommending coating replacement for low instances of corrosion and coating breakdown, but another SME disagrees with that recommendation. This leads us to believe that such a solution can help us standardize corrosion detection and drive the right actions to optimize cost and resources. In this case, accuracy is defined as what a human would identify as corrosion, and the computer was able to identify it as well. At the end of these POC projects, each customer was provided with the following deliverables:

• Corrosion Segmentation Model- outlines the detection of corrosion and rust stains on images of the facilities taken by a panoramic camera.
• Texture Classification ML Model- outlines the detection of paint defects, rust staining, light, and severe corrosion on close-up images of the stationary equipment.
• User interface (UI)- browser application hosted on the secured Virtual Machine (VM), allowing users to tune processing parameters, run ML models in real-time on the selected images, and see the big picture on the map. The UI enables the user to process images one by one with a preview or batch processing of the entire directory passed as a parameter.
• Technical Report- the report outlines the chosen approaches and techniques for building the pipeline and incorporates the results of the main experiments with the conclusions provided and potential improvements for future development.