AI Predictor for Equipment Icing
Predicting and responding to freezing and icing accidents of main equipment in harsh external environments through real-time analysis.
The Problem of Vaporizer Icing Management Is Not Simply Whether "Ice Is Forming"
In actual fields, it's impossible to predict when and how quickly icing will progress, and the manager must watch that uncertain time all day long.
This case story shows how EVA is transforming field icing monitoring methods into a data-centered predictive management system.
Background: Icing Was Visible, but its Progress Was Unpredictable
A facility site storing oxygen and nitrogen. For the engineers here, 'icing' was not an unfamiliar sight. It was common for frost and ice to form on the surface of vaporizers. The problem was that it was difficult to predict when and to what extent icing would progress.
Even if the vaporizer freezes, it doesn't stop immediately. However, when icing covers a certain area (about 50% or more), heat exchange efficiency drops sharply, and the safety risk due to pressure imbalance increases. This is 'the point where the golden time for response has already passed.'
The problem is that no one knows when that moment will come. Icing progresses irregularly depending on time, temperature, and humidity, and although it can be seen with eyes, it was virtually impossible to keep watching those changes in real-time.
Existing monitoring systems stayed at simple video recording functions, and anomalies could only be recognized after icing had built up to a dangerous level. So, managers had to repeatedly watch dozens of screens every day and judge "should we respond at this level?" by intuition.
Ultimately, although 'visual information' about icing was sufficient, there was no signal for 'when to respond' — that was the biggest frustration in the field.
The Clue to the Solution: Judging Visible Icing with Data Now (Why EVA?)
The customer summarized the essence of this problem like this: “Icing is visible, but we cannot keep watching its ‘progress’.” So a new approach was needed. Instead of a person monitoring the screen, what if AI could notify the area and progress speed of icing?
The solution highlighted then was EVA. EVA utilizes Few-shot learning-based object recognition AI technology to learn icing on the vaporizer as an object and track the area in real-time.
In other words, it goes beyond recognizing whether icing 'is/is not' present, and can capture the moment when icing expands to about 50% or more by analyzing the area change rate of ice. Through this function, EVA was able to transmit a signal in real-time that "the time to respond has come."
The Process of Problem Solving: EVA Observes the Speed of Icing with Data
In the PoC stage, EVA was directly linked to the field vaporizer CCTV video. Without separate sensors or additional equipment, real-time analysis started with only existing camera video.
EVA learned the collected icing image data in a Few-shot manner. This was a process to stably recognize the icing area regardless of vaporizer surface patterns, light reflection, or temperature changes.
As a result, every time icing progresses, EVA automatically tracks it by displaying the detection area in a specific area on the screen and provides an immediate warning notification when the icing area reaches about 50% of the total.
This function completely changed the approach to the field. Previously, workers had to check the camera periodically, but now EVA accumulatively observes the progress speed of icing and accurately informs the threshold point. Field engineers didn't need to keep monitoring "when ice crosses the limit line," and were able to take proactive de-icing measures according to EVA's notification.
The Possibility Proved by EVA: Now Icing Is Data, Not a Concern
PoC clearly showed EVA's effectiveness. Although icing was visible to the naked eye, it was difficult to judge when the actual danger stage was, but EVA visualized this with specific figures and area analysis.
During actual testing, EVA captured patterns where the icing area increased sharply and provided quick notifications at the point of reaching the 50% threshold. Because of this, the manager was able to respond immediately before the icing spread stage, and as a result, was able to minimize efficiency degradation and unnecessary inspection.
Field engineers evaluated it like this: “Thanks to EVA, we can now manage icing 'with data, not intuition'.”
As a result of PoC, it was also confirmed that the icing monitoring process was simplified and the response time in the field was shortened. Above all, EVA was recognized as an 'AI assistant that accurately informs when the vaporizer becomes dangerous' by field managers.
“You don't need to keep watching for icing. Now EVA watches instead.”
Icing is visible to eyes. However, 'when to respond' is not visible to eyes. To eliminate that uncertainty, EVA automatically measures the progress speed and area change of icing and notifies you right before crossing the danger threshold.
Now you don't need to keep watching the vaporizer. EVA sees, judges, and notifies instead.
EVA – The AI solution that protects the golden time of industrial equipment. Check the change your site needs now.
