PTQ Q4 2024 Issue

-

Go

1W 1M 2M 6M 1Y 2Y 5Y 10/04/2021

30/06/2023

0.9130

0.8874

0.8618

Measurement time

Minimum Thickness

TC Thickness Temperature mm in All Thickness mm in

2020

2021

2022

2023

2024

-

Go

1W 1M 2M 6M 1Y 2Y 5Y 07/27/2021

03/02/2023

0.9168

0.9067

0.8966

Sep ‘2021

Nov ‘2021

Jan ‘2022

Mar ‘2022

May ‘2022

Jul ‘2022

Sep ‘2022

Nov ‘2022

Jan ‘2023

Mar ‘2023

Measurement time

Minimum Thickness

TC Thickness Temperature mm in All Thickness mm in

2020

2021

2022

2023

2024

Figure 5 Comparison of thickness trends: Static gates vs dynamic gates

High-stress periods should be avoided when optimising the temperature coefficient, as they can skew the results (see Figure 4 ). Figure 4 displays a time-series of temperature and thick - ness over approximately six months. While temperature variations generally correlate with changes in thickness, stress (evident in the highlighted section) can overshadow this temperature dependence, resulting in an anti-correla - tion between thickness and temperature. These periods of anti-correlation should be excluded from the optimisation of the temperature coefficient.³ Dynamically adjust for corrosion effects As previously discussed, corrosion can create an irregu - lar and rough surface, leading to reduced amplitude and broader ultrasonic echoes. This deterioration can affect the determination of ToF reference points using static gates, potentially resulting in a step-function-type thickness trend (Figure 2) or, in more severe cases, an increasing thickness trendline (Figure 3), which may be incorrectly interpreted as material deposits. To overcome this issue with permanent sensors, machine learning algorithms can adjust the gates in real-time to match changes in the ultrasonic waveform. Figure 5 shows how dynamic gates can avoid misleading thickness trends. The top section of Figure 5 illustrates how static gate algorithms produce unusable step-function-type thick - ness trends, while the bottom section demonstrates how

machine learning algorithms provide accurate and usable thickness trends. User-friendly circuit-level dashboards Machine learning enhances the accuracy of thickness measurements, but artificial intelligence (AI) is key to turn - ing vast amounts of data into actionable insights. This ena - bles effective management of corrosion monitoring across many ultrasonic sensors. Traditional dashboards provide detailed views of indi - vidual CMLs, focusing on corrosion, material properties, or sensor health. However, they often lack a broader view of the entire circuit or unit. To address this, a new type of dashboard is needed for a clear and comprehensive over- view of asset status. Figure 6 illustrates the new dashboard. AI processes data from trend lines and A-Scans to create a visual sum- mary at the circuit level. Coloured markers indicate each sensor’s location and health. If a sensor’s measurements exceed a user-defined threshold for thickness, tempera - ture, corrosion rate, or remaining life, its colour changes from blue to red, providing an easy-to-understand over - view of the entire circuit. Additional details about CMLs, such as corrosion rates and remaining life, are listed on the right side of the dashboard. The figure includes a graphical map of the sensors (blue dots) and a list of CMLs ranked by highest long-term corrosion rates and shortest remaining lifespan.

103

PTQ Q4 2024

www.digitalrefining.com

Powered by