Case story

Predictive Maintenance 4.0: Connecting the Internet of Things with the process control system

Our last case story was about significant energy savings on a recooling system with a total of five pumps. This article describes how digitalization reduces the maintenance costs of much more complex systems. The two most important players in this game: the Internet of Things and the process control system.

Project name
Predictive maintenance and data analysis
Project duration
2 years
Project team members
Collaboration between the IT Service Center and Infraserv Höchst's Data Science & Data Engineering and Refrigeration, Cooling, Water departments in particular
  • Extension of maintenance cycles, reduction of repair costs through damage prevention
  • Machines are relubricated around 10 times a month
  • Gain experience by recognizing hidden patterns through AI
  • Convenient way to analyze very large amounts of data
  • Early detection of damage and therefore predictable and more cost-effective maintenance

Our goal: Condition-based maintenance

Infraserv Höchst's Refrigeration, Cooling, Water department operates and maintains a wide variety of systems at Industriepark Höchst and other sites. What they all have in common is that more than 200 sensors record a wide variety of parameters that must be known to ensure reliable operation. For example, not only the local temperature of individual plant components is monitored. The strength, frequency and direction of vibrations occurring in moving parts also provide important information about the condition of the unit.

For the first time, it was possible to obtain an optimal overall picture by linking this sensor data from the Internet of Things (IoT) with the process control system (PCS): The data sent by the IoT sensors can initially only be used to read factual system statuses and their changes. But not necessarily the cause of their development. It is not possible to see whether a change has occurred as a result of (imminent) damage or deliberate intervention (e.g. a planned increase or decrease in output). This information is documented by the DCS, which is used to control and monitor the systems - e.g. the flow rate of the water, the speed or the power consumption.

Particularly with more complex systems, a large number of parameters come together that ultimately determine the running and vibration behavior of a machine. A compressor running in the partial load range will inevitably exhibit higher vibration values than at the optimum operating point. Increased vibration can therefore indicate a problem (e.g. bearing damage) in one case, while in another it is merely a sign of the current operating mode. In the past, this evaluation and classification of vibration values took up a lot of space, as it required the manual linking of data from different sources or the involvement of several people. This link has already been established in the new app. This increases both efficiency and the experience gained through a more detailed analysis.

Sarah Teizel, Head of Refrigeration Operations at Infraserv Höchst

Previously, shifts in condition caused by human intervention could only be differentiated from wear or damage retrospectively by laboriously comparing data manually. The new system establishes a clear relationship between the data. Various parameters can be specifically selected so that possible causes can be identified using the exclusion method.

One example: Vibrations

Every drive has at least one bearing, and none of them runs without vibration. Depending on the current speed and the condition of the bearing, this results in different vibrations, i.e. very rapid changes in the direction of movement. These are measured either in g, the unit of measurement for acceleration, or in millimeters per second, the speed. The g-values that occur are sometimes surprisingly high - despite the tiny distances traveled, up to 9 g occur here. For comparison: trained astronauts in special suits can withstand 10 g. It is obvious that higher g-values place a strain on the material and vibrations therefore lead to wear in the long term.

There is therefore a frequency spectrum for each bearing at which no material damage can be assumed. The start-up or shut-down of the system is also part of standard operation and is in turn reflected in deviating frequency spectra. However, noticeable deviations from the norm usually indicate damage. The frequency, strength and even the spatial orientation of the vibrations can be used to determine the type of damage.

The new tool: IoT-PLS-Miner

In a new tool, experts Frank Mollard, Benjamin Moll and Patrick Stricker from Infraserv Höchst's Data Science & Data Engineering departments, in collaboration with Dirk Scheid (IT Business Partner) and specialists from Infraserv Höchst's Refrigeration, Cooling, Water department, have combined the IoT data mentioned above with DCS data. This makes it possible to analyze the overall condition of a machine or the condition of individual components with just a few mouse clicks.

This means we have really implemented predictive maintenance consistently. Engineers can use their specialist knowledge to read the condition of a machine from the frequency spectra in conjunction with the DCS data

Frank Mollard, Head of Data Science & Data Engineering at Infraserv

The negative correlation between vibration and speed of the corresponding pump is visible in both the left-hand time series and the right-hand scatter plot. This provides engineers with information about wear-reduced operating modes.

The negative correlation between vibration and speed of the corresponding pump is visible in both the left-hand time series and the right-hand scatter plot. This provides engineers with information about wear-reduced operating modes.

For this purpose, IoT vibration sensors are trained during a learning phase lasting approximately one week. After this, they know the normal state and are able to issue alarms in the event of deviations from the norm. These alarms indicate where a closer look should be taken. In a second step, a more in-depth analysis can be carried out on the screen. This enables the technicians to assess whether there is an immediate need for action at any point or when intervention is unavoidable. Vibrations, frequency spectra and temperatures from the sensors can be combined directly with DCS data such as the speed of the machine, the pressure or the water flow rate in interactive, customizable visualizations.

AI support in finding patterns

The data can not only be examined in the conventional way, but also analyzed using artificial intelligence. This allows patterns to be searched for automatically without the need for explicit instructions.

Unsupervised learning algorithms are used for this. These have the advantage that no prior, possibly resource-intensive, training is required, as the corresponding model is created interactively during the analysis. In this way, a customized analysis can be performed for each combination of data points to identify hidden patterns that would otherwise have remained undetected. Our experience shows that this innovative approach provides unprecedented insights into the condition of machines and systems.

The combination of all this significantly reduces the workload: “We now have around ten maintenance jobs per month on the systems, during which the bearings are lubricated, as well as irregular situations in which imminent damage is averted - in a targeted manner! This extends the service life of the systems and major damage occurs less frequently,” says Mollard.

What does that mean in figures? “In an exemplary damage event, we were able to make significant cost savings: If we hadn't intervened in time, damage of € 12,500 to € 25,000 would certainly have occurred. So we were able to repair the system for around € 3,500.” Mollard estimates that similar cases occur four to six times a year per system. Overall, this extends maintenance cycles and detects impending damage at an early stage. The IoT-PLS-Miner is designed so that it can easily be rolled out to other Infraserv plants.