Predictive Maintenance for Heat Exchangers

How to find the right balance to maximize the overall cost efficiency?

In the pulp and paper industry energy costs are a significant part of the overall operational
costs. Studies show that in a typical paper mill there are dozens, or hundreds of Gigawatt hours of energy lost yearly. This means Millions of Euros in lost profits per plant.

The first and second laws of thermodynamics states that heat energy can’t be destroyed. In pulp & paper processes this means, that more efficient the secondary heat circulation is, the less primary energy is needed. Heat exchangers play a crucial role in this.

Heat transfer in heat exchangers typically consists of convection in both fluids and
conduction through the heat exchanger plate. These phenomena are directly affected by fouling. The cleaner the heat exchanger is, the more efficient it is.

Studies show that heat exchangers operate typically with a 40 – 70 % efficiency. This means that there is a lot to be improved by better maintenance of heat exchangers.

This article investigates different maintenance approaches to keep the heat exchangers working efficiently. We will study the fundamental challenge in maintenance planning. Too much maintenance is expensive. Too little maintenance increases energy costs.

Preventive vs. Predictive Maintenance

There are three different basic approaches to maintenance: reactive maintenance, preventative maintenance, and predictive maintenance. The fundamental difference is what triggers a maintenance task and how you make maintenance plans.

In reactive maintenance or ‘corrective maintenance’ you go and fix the piece of equipment when it is broken. You do not try to eliminate the failure. In the application of heat exchangers this means, that you clean a heat exchanger when it already creates problems.

The other two approaches, predictive and preventive maintenance, are often referred to as ‘scheduled maintenance’. Both aim to eliminate equipment failures by scheduling maintenance tasks before the failure. In the application of heat exchangers this means, that you clean the heat exchanger before it starts to create problems.

Even though having the same goal, the two approaches sound similar but are very different by nature.

Preventive Maintenance

Preventive maintenance aims to prevent future failures. In preventative maintenance you have a pre-defined schedule that determines work orders and maintenance tasks. In the application of heat exchangers, this means that you would have an annual clean-up schedule for each heat exchanger. For example, a process critical heat exchanger might be cleaned in the beginning of each month and a small non-critical heat exchanger every September.

Preventive maintenance has major challenges, however. It works well in a stable and predictable processes. But especially in pulp & paper industry the fouling is very unstable and unpredictable by calendar time.

With preventive maintenance in pulp & paper processes you are likely to do one of following. Either you do expensive over-maintenance, since every clean-up has a cost. Or you do under-maintenance and let the heat exchangers to foul. This increases energy cost. In the worst case you do a combination of the two: you clean them often but at a wrong time.

Predictive Maintenance

Predictive maintenance or ‘condition-based maintenance’ aims also to prevent future failures. But instead of calendar time, in predictive data-based maintenance you determine the need of maintenance based on the actual measured condition of the device.

With heat exchangers, this means that you measure the fouling-% of each heat exchanger. You also know the cost of each maintenance as well as your energy cost and demand. Based on this information you determine the most cost-effective time to clean each heat exchanger.

This can also be called as ‘prescriptive maintenance’. In literature, the term ‘predictive maintenance’ is often used when sensors are used to collect data to determine and predict the condition of a device. The term ‘prescriptive maintenance’ goes one step further by also recommending correct actions to take.

Economic Impact of Different Maintenance Approaches

Whichever term you want to use, predictive or prescriptive maintenance, the benefit compared to reactive or preventive maintenance come in the form of lowered maintenance cost and improved process efficiency. Or is it that way always?

Sure, it would be nice to implement a modern digitalized predictive maintenance program, prevent fouling and maximize cost-efficiency. But since it is based on actual heat exchanger condition, you must collect the data. For this you need sensors. You also must collect all other data, such as energy prices, clean-up costs and uphold a clean-up task list. For this you need special software.

Implementing all of these come with a cost. Does the investment and work time ever pay itself back?

When assessing the total cost of maintenance, we need to include prevention cost and failure cost. In pulp & paper industry heat exchangers this means the cost to perform clean-ups and maintenance tasks as well as the cost of lost energy. In the literature, balancing between these two costs is referred to as ‘traditional vs. modern view of total cost of maintenance’.

In the traditional view, the discussion is about finding the optimal prevention level. It is assumed that the ‘prevention cost’ of preventing everything has an extremely high cost. On the other hand, not preventing anything with reactive maintenance is also considered to have extremely high cost. So, the task was to find a balance somewhere in the middle, as seen in FIGURE 1 A.

In the modern view, the assumptions are different. The game changer comes from technological development, such as IoT-sensors, mobile data networks and cloud computing. Today you do not need to perform measurements, do constant manual inspections, or perform human-based analysis to detect fouling heat exchangers and prevent failures. Technology does it for you. Thus, the modern view in the literature states that lowest total cost of maintenance is achieved with a 100 % prevention level, as seen in FIGURE 1 B.

In the complex pulp & paper mills transitioning to modern predictive maintenance has reduced total costs by Millions of Euros. This means Million of Euros in increased profitability per mill.

Thus, selecting the right maintenance approach shouldn’t be only a technical question. It should be a fundamental business question.

Challenges in Predictive Maintenance for Heat Exchangers

Predictive maintenance will benefit you in many ways. First, you will save energy costs. You will also prevent failures in heat exchangers and other process equipment. This reduces downtime and lowers your maintenance costs. You might see improvements in your process quality since the process is running with designed parameters. You will also benefit from issues related to environmental responsibility and regulations.

However, predictive maintenance does come with a bunch of challenges. It may not be as straight forward to implement effectively. Predictive maintenance consists of three elements: measurements, analysis, and prediction. Prescriptive maintenance includes the fourth element: recommended action. All these elements need to be tightly interconnected to make predictive maintenance work effectively.


Measurements in heat exchangers come from process automation and by individual sensors. The data input for calculating heat exchanger efficiency is simple. You need four temperature and two flow measurements. Rest is mathematics.

The first challenge is to ensure that your measurement values are correct. Since expensive calibration and validation is usually done only for the most important measurements, you will have inaccuracies in your data. The question is how much? And is it too much?

The second challenge in most of the pulp & paper mills is the lack of measurements. Each heat exchanger very rarely has all the needed measurements. For example, you might have only one or two temperatures. Or your flow measurement is before several valves that split the flow to multiple heat exchangers.

You can overcome the lack of measurement data in two ways. First, you can use ‘soft sensors’. This means that you calculate the missing measurement from other existing data. As an example, flow measurements can be re-constructed based on valve position data. Or temperatures can be re-constructed from energy balance.

The other way is to install wireless IoT-sensors to positions where crucial measurements are missing.

A Elsys ELT-2 LoRaWAN sensor, which you can use for temperature and flow measurements.
A Elsys ELT-2 LoRaWAN sensor, which you can use for temperature and flow measurements.


After you have all the data, you need to turn that into information. Showing the temperature and flow data does not take you anywhere. First, you need to calculate the power of the heat exchanger, which is straight forward.

After that you need to determine the heat transfer coefficient as well as the production volume dependent corrections to it.

Now you are ready to calculate the efficiency ratio of your heat exchanger. From this you can determine the actual power loss. The condition-corrected power loss is your first truly usable information for determining whether a clean-up is necessary or not. FIGURE X shows basic mathematics needed for your analysis.

In all these steps you have needed to clean-up your data and remove false measurements. Also, a centralized data pool is highly recommended. This will make it easier when you calculate and analyze the correlation between different data sets.


If you get this far, you know the current condition of your heat exchanger. This is good for ‘reactive maintenance’ and might help you to catch problems quickly.

However, the predicting the future is where the magic happens. For this, you need a predictive fouling model. In addition to the earlier mentioned calculations, you need your production forecast and the information on how the production volume affects the fouling.

After you have the data available, you can create the mathematical model using statistical approach or machine learning. In statistical models the different data are assumed to follow statistical distributions. The first question is to determine, whether you need a linear or non-linear model. After this Excel has all the tools you need.

The machine learning algorithms learn from past data and try to fit best working mathematical models to fit the incoming data. You can use Matlab, Azure, AWS or even machine learning plug-ins for Excel. Whatever is your preferred tool.

Keep in mind that the practical usability of your model is more important than the accuracy. As the saying goes: all models are wrong; some are just better than others.

Recommended action

If you get this far, you should be able to determine the condition of your heat exchanger months in the future. This provides you with the needed data to plan correct clean-up dates long in the future. The only challenge left at this point is to determine which kind of clean-up and what is the economic impact of this clean-up. For this you need skilled people. 

If you also take this last step, you will have the recommended actions calculated ready for you. This eliminates the constant need of humans analyzing the correct actions.

For this you need to spice up your predictive model a notch. In addition to the earlier steps, you need to have information of your energy cost, the effectiveness and cost of different clean-up types as well as the process downtime costs during the clean-up. You can collect this information from past experiences. For example, you can compare data after a CIP and mechanical clean-up. You can also collect the data after using acid or alkali detergents. 

When you add these to your predictive model, you will have a model that could look like the examples in FIGURE X and FIGURE Y. You see the clean-ups planned for one year ahead, you see the payback time of each clean-up, and you will see the reclaimed energy profits in Euros. Presenting the performance in Euros will ensure that your maintenance program is economically profitable for the mill. This also motivates the management to support you in further developing your maintenance program.

If you get this far, you just let your predictive maintenance program to collect the lost millions of Euros from your process.

Predictive Maintenance Sounds Great – But Where to Start?

Building up a full-scale predictive maintenance program for your entire pulp & paper process might sound a bit complicated. And it is exactly that. That’s why it is important to follow a couple of important guidelines.

First rule is to start simple and start small. Select one heat exchanger you can start to learn with. Don’t try to build the entire thing in one step. Rather, start building your house from the ground floor. Make sure your six measurements are correct. This is the base for everything. 

Second rule is to do your calculations in steps. And double-check your data after each step. First calculate the power, then the heat transfer coefficients. Finally calculate the heat exchanger efficiency-% and power loss. Now, perform a clean-up and see if your data shows the improvement. If not, find which measurement data is bad. And correct that. 

Third rule is to accept imperfection. You don’t need a perfect predictive model to improve your energy efficiency. Learning to build a working predictive model can require several rounds of learning, but each round will make it better. 

Fourth rule is to learn to scale. After you are successful with your first heat exchanger, you can start scaling up by adding heat exchangers one by one. Each new heat exchanger increases the amount of data to manage. Building a good foundation for good data management is key for scaling up.

So, learning in small steps is the magic word. If you want to build it up at once, you need to combine and analyze data from hundreds or thousands of different sensors. Not to mention the selection of correct mathematical models for every heat exchanger. This requires highly skilled engineers.

Availability of skilled experts is one of the greatest challenges associated with implementing predictive maintenance programs. It is estimated that there are currently less than 18,000 qualified experts in this area worldwide. If you add knowledge of pulp & paper processes, you are looking for a needle in a haystack.

But there is good news as well. First, there are increasing amount of education programs in the field of data sciences. Participating in some on-line courses and practicing the use of the data tools will soon give you the needed skills to build a predictive maintenance program. In the end, predicting heat exchanger efficiency and energy cost savings is not rocket science. If you follow the guidelines above.

The second good news is that you can outsource this task to the HeatHamster team. HeatHamster provides energy profits as a service. It is the world’s only turn-key service for predictive clean-up planning.

Which ever way you want to go, don’t hesitate to contact us. We are happy assist you in the first steps.

Curious about Predictive Maintenance for Heat Exchangers? We can help!

HeatHamster is a IndMeas Oy Ab product.

IndMeas Oy Ab, Vaisalantie 6, FI-02130 Espoo, Finland
© 2021 IndMeas Oy Ab. All rights reserved.