Reviewing the EVA Results

Doing the EVA.

When the EVA calculation is done, it is time to review the results.

The purpose of the Goodness of the Fit tests is to investigate how well a given sample of data approximates a given probably distribution. Failing the tests tells us that it is unlikely that the sample data is drawn from a Gumbel type distribution. Many samples will not pass the fit tests. This, however, does not necessarily invalidate the ability to use the EVA results. 

Using the Calculation Table and Charts

The data in the calculation table and charts enables you to assess the goodness of fit (and extrapolation in space).

To review the table and charts:

• Go to EVA Calculation Charts at the bottom of the EVA page.
• For each Strat, select the Strat’s row to show the associated charts.
• Evaluate the p-values (p-KS = Kolmogorov-Smirnov, p-AD = Anderson-Darling goodness of fit test) in the table. Note:  p-AD has more weights for estimating extremes. p-AD > 0.05 is a pass (the higher p-AD the better).
• Select the different tabs in the chart window to review the charts.
• Click on the Full Screen button below the Charts Tabs to view the Charts in full screen.
• If desired, change the EVA Settings (e.g. CI). Note: When a Setting is changed, EVA will be calculated again,
• If the fit is poor, consider defining a new stratification, since a more homogeneous spread will improve the fit. Then Calculate EVA again.
• Remember a poor fit does not necessarily mean that you should not use the EVA results. However, before using the EVA results, you should thoroughly evaluate the results and predicted wall loss, to make sure the corrosion behavior is understood.
• If you decide not to use the EVA results, then you can choose not to select the EVA result in the Calc Summary.

Calculation Charts. 

The Calculation Table

The table shows the following parameters:

The table can be exported with the export button on the top left.

EVA Calculation Charts Table and export button. 

The Calculation Charts

Seven different Charts are available to help access the goodness of fit:

• Variate Plot,
• 2 Probability Plots,
• Quantile Plot,
• Exceedance Probability Plot,
• Return Level Method, and
• Extrapolated distribution method.

These charts are further explained in Assess Goodness-of-Fit and Extrapolation in Space.

Interpreting the Results

Most likely values of the Max Wall Loss

The Variant Plot shows the Most Likely value of Maximum Wall Loss in the whole HX and the different CBs (Confidence Bounds) of the Max Wall Loss estimate.

The Variant Plot shows the Most Likely Max Wall Loss and different CBs of the Max Wall Loss estimate. 

Also, the Extrapolated Distribution Plot indicate the Most Likely value of Maximum Wall Loss, using percentiles of the extrapolated Gumbel distribution. For example, 95 % percentile means that 95% of the estimated Max Wall Losses are expected to be lower that this number. 

The quantile plot shows the Most Likely Max Wall Loss using percentiles of the extrapolated Gumbel distribution.

Minimum Remaining (Wall) Thickness and Corrosion Rate

The results are also expressed in terms of Min RT (Min Remaining (Wall) Thickness) and CR (Corrosion Rate). 

Given the Nom WT (Nominal Wall Thickness) of the HX:

• The Min RT for a given CL (Confidence Level) is calculated as: CL% Min RT = Nom WT – CL% Max Wall Loss. For example: 

Given the T in Service (Total Time In Service), determined from the Start Date of the Equipment:

• CR for given CL is calculated as: CL% CR= CL% Max Wall Loss/T in Service. For example:

For a more in depth explanation of the calculations and charts see Assess Goodness-of-Fit and Extrapolations in Space.

Time Prediction

Similar to extrapolation in space, the EVA methodology can also be used to extrapolate in time. However, this is currently not implemented in IMS. IMS uses the space extrapolation to estimate the Maximum Wall Loss for each Strat. It then uses a simple constant CR (Corrosion Rate) model to predict the Maximum Wall Loss over time (see Minimum Remaining Wall Thickness and corrosion Rate). From the Max Wall Loss over time it can predict a RL (Remnant Life). The MII (Maximum Inspection Interval) and NID (Next Inspection Date) are then calculated. The worst case, i.e., the Driving Strat, is shown in the Calculation Summary and (potentially) takes part in the horserace. (See Remnant Life (RL) and Next Inspection Date (NID) Calculations.)

Note: IMS assumes a long-term CR, and it is possible that short-term CRs may be higher (i.e., CR has not been constant since start date).

Now you just need to implement the EVA results.