Deploying MVA as a Digital Twin to Address Equipment Fouling

Equipment performance is reduced when unwanted material accumulates on the solid surfaces of heat exchangers, furnaces, pipes, reactors, etc.  A fouled heat exchanger typically has a poor heat transfer capability (lower thermal efficiency), which drives the operating cost and energy consumption higher as additional fuel is required for heating.

For context, fouling of heat exchangers in crude refinery preheat trains in the United States is estimated to cost ~1.2B annually.¹

This case study demonstrates how multivariate analysis (MVA) can be used to mitigate equipment fouling  through offline analysis as well as real-time implementation (digital twin deployment).

Project Details

Challenges:

The main challenges with addressing fouling issues are the difficulty to consistently measure fouling and the inability to completely prevent it. The accumulation of material on any surface is a natural phenomenon that is expected to happen over time. Therefore, the primary objective is typically not to entirely eliminate fouling, but rather to decrease the rate at which fouling occurs in order to maintain high throughput, minimize unplanned shutdowns and minimize energy consumption.

The ProSensus Approach:

1- Fouling Quantification & Data Analysis

Fouling is often unmeasured in real time. Rather, operators typically inspect and take note of the state of solid accumulation in the equipment at the end of a production campaign or during other periods of downtime. This lack of quantitative real-time fouling data limits the ability to complete troubleshooting and plant optimization initiatives, but there are still analyses that can be valuable.

ProSensus typically conducts both offline troubleshooting analyses and real-time digital twin deployment as consulting services and utilizes AspenTech’s ProMV suite.

Campaign by Campaign Analysis – real time quantification not available

Campaigns that were operated similarly tend to have a similar fouling outcome. Therefore, the fouling campaigns usually cluster together in the score plot so one region becomes the “fouling region” that should be investigated and avoided. A PLS-DA (discriminant analysis) modeling approach is typically used in this analysis to maximize the fouling classification capability of the historical process campaigns.

Contribution plots are then used to identify the key operating parameters that vary between the fouling campaigns and the stable campaigns. This is a significant outcome since it highlights the combination of operating conditions throughout the process that usually result in higher levels of solid accumulation.

Once the key operating parameters are identified, an optimization can be configured to set each campaign’s optimal operation that minimizes fouling based on its initial conditions. Based on the results of this offline analysis, changes can be made to operation policies to avoid high fouling campaigns.

Real Time Workflow – continuous fouling quantification

Figure from How Braskem Idesa Reduced Reactor Fouling Rate by Using Multivariate Analysis (2021)³

Using the heat exchanger as an example, fouling has a significant impact on the efficacy of the heat transfer system. The material accumulation reduces the cross-sectional area of the tubes and flow channels increasing the resistance of the passing fluid. These side effects lead to an increase in the pressure drop across the heat exchanger which reduces flow rates and can eventually block the exchanger.² Measuring and monitoring these and other related process tags (where available) and their rate of change can provide a continuous indication of the fouling state, which is a significant advantage over waiting for a manual inspection at the end of the campaign.

Once the available measurements that can be used as fouling indicator have been identified, and a meaningful way to combine multiple measurements has been determined, this generated data can be utilized to better understand the correlations between operating conditions and the fouling rate using multivariate PLS models.

This distinction can be made based on the generated fouling rates or indicators, and results should be confirmed by operators and process engineers who are familiar with the process and equipment history.

Figure from How Braskem Idesa Reduced Reactor Fouling Rate by Using Multivariate Analysis (2021)

Coefficient plots of fouling indicators are then used to understand the main fouling predictors. For such complex problems, multivariate models are powerful since the results don’t typically point to a univariate correlation but rather the combination of multiple factors. Furthermore, process experts may have knowledge of particular process conditions that anecdotally accelerate impact fouling, such as running a certain product grade, using a certain additive or operating above a specific temperature. Such insights are valuable to disclose during the data analysis, and may lead to the most useful modeling decisions and results.

2- Real Time Monitoring & Prediction

Once the correlations are understood, multivariate models can be implemented in real time. Current equipment operation is sent to the multivariate model (digital twin) and is compared to ideal, non-fouling operation and can be used for early detection and online monitoring.  ProSensus has successfully reduced fouling rates for multiple clients across different industries, and some clients have reported a 50% reduction in equipment cleaning times as a result of our executed workflow.

Monitoring models are usually built on periods of “good” operation/campaigns, where fouling rates were low and stable. These models represent the ideal operating window that the process should stay within to mitigate fouling. Any deviations from that window trigger an alarm indicating that the fouling rate is starting to increase (or expected to increase shortly) and that operator action is required.

Further, a real-time prediction of fouling can be used to set the optimal schedule for equipment cleaning so that equipment is cleaned on an as-needed basis rather than following a pre-determined cleaning schedule that may be either too frequent or too sparse maintenance.

The state of fouling can be analyzed at the beginning of each campaign to better optimize cleaning schedules particularly for time consuming procedures such as furnace decoking. Implementing this real-time prediction of fouling can maximize time between cleanings and reduce frequent unplanned shutdowns.

Figure from How Braskem Idesa Reduced Reactor Fouling Rate by Using Multivariate Analysis (2021)³