Blogs

Abstract
Corrosion under insulation (CUI) is a common and safety-critical type of corrosion in aged chemical plants. Currently, CUI inspections are mainly carried out by setting up scaffolding and removing the thermal insulation because there is no efficient and cost-effective nondestructive inspection method for CUI. The cost is enormous, and in many cases, corrosion may not be present even after the insulation is removed for inspection. Therefore, it is important to properly estimate the areas to be inspected. Several CUI estimation methods have been proposed, but the accuracy is not sufficiently high. CUI inspections have been conducted for many years, on the other hand, by removing heat insulation. With the support of several Japanese chemical companies, we collected and analyzed the results of CUI inspections by peeling off heat insulators
and information on the specific conditions of the equipment.

Developing the Algorithm to Determine the Possibility of CUI
Approximately 12,000 inspection data of piping were collected from chemical companies in Japan, in a format according to collection rules for inspection results of CUI depth by removing insulation and related equipment information (thickness of the part of inspection, period of use, part, temperature, etc.). Based on this information, the probability of CUI damage occurrence was calculated using machine learning, and an algorithm was developed to determine the likelihood of CUI occurrence into four ranks, as shown in Table 1.

From this result, it is evident that the majority of data points are classified as rank A, and only limited points classified as rank C. This CUI model clearly identifies the necessary inspection points and leads to a reduction in the inspection area.

Similar evaluations were conducted with some supporting companies, and it was verified that the developed CUI prediction model was more accurate than the conventional CUI prediction methods used by each company.

Furthermore, artificial data under various conditions were generated to examine how the ranks of CUI changes with operating conditions; some of the results are introduced below.

Figure 2 illustrates the change in CUI ranks by specific piping components and parts (assuming the insulation jacketing plate is in good condition with an initial allowable thickness of 3 mm). In common with all piping components and parts, the result shows that the longer the operation period and the lower the service temperature, the higher the rank of CUI. Additionally, the ranks of CUI occurrence around the support components, elbows and T-components are higher than those of straight piping components.

Figure 3 displays the CUI occurrence ranks of each piping component under the damaged condition of the jacketing plate. From this figure, it is clear that the possibility of CUI rank in each component has changed to the severe side compared to the case with the good jacketing plate condition shown in Figure 2. In addition, the corrosion tendency around the support and the elbow and T-component is greater than that in the straight pipe component, which is the same tendency as Figure 2.

Figure 4 presents the results from evaluating the effect of initial allowable thickness on the CUI rank using the straight pipe component (good jacketing plate) as an example. The thicker the initial allowable thickness, the lower the rank of CUI.

As demonstrated in these examples, the developed model can quantitatively evaluate the possibility of CUI occurrence for various piping conditions. Furthermore, since the present model can evaluate the probability of CUI occurrence based on visual observation results of the jacketing plate, it can be used for inspections limited to certain piping positions, as schematically illustrated in Figure 5.

Developing Software from the CUI Model
The model is currently being converted into software and distributed to several supporting companies to verify the cost reduction effect of CUI inspections and the possibility of improving inspection efficiency when this model is applied. We have also started distributing the model for a fee to chemical companies that wish to use it.

In the present state, the quantity of data is not sufficient, and so the estimation of the possibility of CUI occurrence may not be appropriate under certain conditions. In addition, the CUI data was collected at some plants located on the coast of Japan, and there are still questions as to whether this CUI model can be applied under different climatic conditions. In the future, we hope to collaborate with supporting companies to increase the number and variety of data and improve the accuracy of the CUI model. Moreover, while the CUI prediction model was introduced this time, it would be possible to develop a similar prediction model for material damage phenomena like corrosion at pipe supports that are highly common among chemical companies, if similar accurate inspection data for the damage from actual equipment could be collected.

Acknowledgements
Part of this study was supported by the Japanese Ministry of Economy, Trade and Industry and NEDO, Japanese government-run fund. In addition, several Japanese chemical companies cooperated in providing data and verifying the model. We express our deep gratitude to these organizations and companies.

By: MASAO NAKAHARA, DESIGNATED PROFESSIONAL OF MTI

This article was originally published in MTI CONNECT 2024, Issue 1.