The chemical industry has been undergoing structural change for years. In 2025, the annual average capacity utilization rate for plants in Germany was only around 72.5% – a historic low [1]. High energy and raw material costs, volatile supply chains, and international competition are putting additional pressure on competitiveness. Under these circumstances, improving the efficiency of plants and processes is essential to ensuring the long-term viability of many sites.
The situation is exacerbated by the shortage of skilled workers. While economic conditions have led to a decline in demand for personnel in most professional fields in the chemical and pharmaceutical industries, it continues to rise in engineering and maintenance, ironically. Recent calculations indicated a shortage of approximately 46,500 skilled workers in this sector [2]. So anyone who wants to keep their plants running must make do with fewer staff – and plan resource deployment right down to the very last detail.
One way to address this tension between cost pressures and labor shortages is to consistently use plant data, and this can be done flexibly and seamlessly with Beckhoff’s PC-based control system. Many field devices used in process analysis technology already provide significantly more information than is actually used in practice. In existing plants in particular, status data that allows conclusions to be drawn about wear or remaining service life is not being read out – either because the infrastructure, some of which is decades old, does not allow for it, or because this specific aspect of automation has simply not been implemented. Even where this data is collected, the automated analysis needed to draw the right conclusions and make maintenance decisions proactively rather than reactively is often lacking. Performing plant maintenance based on actual need rather than on a routine schedule, for example, would not only save on labor hours but also cut the cost of raw materials and maintenance.
Which data creates genuine added value?
For predictive maintenance, the most relevant data is the data that provides insights into the wear and tear and life cycle of a field device – a sensor’s internal temperature or measured impedances, for example. Data from previous years provides additional context for correctly interpreting deviations from the norm.
A typical example from the field of process analysis technology is the pH sensor. Its measuring electrode is subject to continuous wear in media and must be maintained weekly or monthly, depending on the application. The actual effort required is minimal, but without reliable status data, the only options are usually maintenance at fixed intervals or on-site inspections – with the result that electrodes are replaced too early or personnel are sent out unnecessarily to perform inspections. This could be avoided by continuously monitoring the sensor’s status data: the “slope” – the rate at which the measured voltage increases in relation to a change in the pH level – as well as the electrode’s impedance change measurably as wear progresses. Ideally, thresholds are defined so that a maintenance alert is triggered automatically if the values fall below or exceed these thresholds.
However, the relevant parameters cannot always be determined so clearly for every device in the field. Depending on their complexity, modern devices provide a wide range of data, so the real challenge lies in filtering out the values that are actually meaningful. Relevant norms and standards provide guidance: IEC 61987 defines a manufacturer-spanning glossary for process automation device parameters and assigns them unique, standardized designations [3]. However, the PA-DIM information model provides the key component, drawing on a set of core parameters that have been predefined by users in the process industry and making them available in a standardized format that is independent of communication protocols and manufacturers [3].
End-to-end greenfield data collection
Once the relevant data has been defined, the question remains: How can it be collected in practice? When designing new plants, this can be implemented consistently from the ground up: Ethernet-APL provides a communication standard that implements end-to-end communication in process technology plants on an Ethernet basis – from the field level to higher-level control systems. Power and data are transmitted via the same cable, allowing status data to be retrieved directly from the field device. The transmission rate of 10 Mbit/s over cable lengths of up to 1,000 m ensures seamless communication of all relevant values. Beckhoff uses the EL6233 and ELX6233 EtherCAT Terminals to integrate Ethernet-APL into the I/O system on a modular basis, facilitating combination with other digital and analog signal types in the same I/O station.
Whichever physical integration method is selected, many modern field devices already provide significantly more information than was available in the past. Smart sensor technology with advanced diagnostics – sometimes including a graphical user interface for direct configuration and evaluation – is now available for many device classes, independent of communication protocol.
Harnessing the benefits of status data for existing systems
In existing systems, the situation is different. However, the majority of field devices already installed have far greater diagnostic capabilities than have been utilized to date. Many existing sensors are already “smart” and provide additional status parameters that can be retrieved and visualized using manufacturer-specific tools or FDT/DTM (field device tool/device type manager). HART commands can also be used for cyclic reading of values. However, the processing logic required for this is often simply not yet implemented in existing systems.
Beckhoff’s HART-capable EtherCAT Terminals can be easily integrated into these existing systems as a retrofit solution. If the cabling and control logic are to remain completely untouched, the EL6184 is available. It reads HART signals in parallel with the existing wiring without affecting the existing 4–20 mA loop – making it ideal for retrofitting HART integration. For example, the NAMUR Open Architecture can be implemented by combining an embedded PC as an edge device with a suitable TwinCAT software project that provides field device data in a structured format via OPC UA in accordance with PA-DIM.
Data preparation and visualization
Simply collecting status data does not create added value – what matters is what happens with it afterwards. In most systems today, measured values are primarily visualized as graphs, time sequences, or in relation to defined thresholds. On this basis, trends like a gradual decline in the slope of a pH sensor can be identified, along with anomalies that deviate from expected values. If values exceed or fall below predefined thresholds, an alarm can be triggered to notify maintenance personnel. The long-term availability of data is equally important. Wear patterns can only be tracked and conclusions drawn about future behavior if values are stored and remain accessible over an extended period of time.
For the visualization and analysis of signal curves, Beckhoff offers TwinCAT Scope, a software oscilloscope that displays data streams graphically – from short time windows in the microsecond range to long-term recordings spanning several days. In addition, TwinCAT Analytics facilitates an end-to-end workflow from data acquisition and historization through to analysis in web-based dashboards, including preconfigured algorithms for limit value monitoring, statistical metrics, and correlation analyses – to name a few examples. Database connections are also available for the structured long-term data storage, ensuring that historical data remains accessible for future analysis even beyond the original data collection period.
The processing of status data provides an important foundation for predictive maintenance, but it does not replace the actual decision-making process. For the most part, it is still a person who interprets trends, evaluates alerts, and determines the appropriate maintenance action based on this information. Although the data is available in a structured format, the actual analysis and the derivation of actions remain manual processes.
How can data analysis be automated?
The next step is to automate this manual evaluation, at least in part. In condition-based maintenance, the focus is often on a threshold or trigger value. If this threshold or value is exceeded or not reached, action is required (a manual inspection or the replacement of a component, for example). Today, this value is usually determined based on expert knowledge, empirical data, or manufacturer specifications from the data sheet. In a traditional implementation, it is entered into the system manually. Status data is continuously recorded, and an alarm is triggered when the threshold value is reached. However, the underlying cause and the resulting maintenance action still need to be determined.
AI opens up additional possibilities here. Threshold values can be determined based on existing operating data, for example. In machine learning, representative inventory data is used to train a model to identify typical patterns and anomalies. Instead of setting a fixed threshold for every situation, this approach also allows the normal operating behavior of a device to be modeled. TwinCAT 3 Machine Learning offers functions for analyzing signals and time series; with the Machine Learning Creator, users can create AI models without needing specialized data science knowledge.
One example is the flame ionization detector (FID), which is used for the continuous measurement of hydrocarbons. A key status parameter here is the detector temperature, which should be above 150°C during ignition to prevent condensation forming [4]. If the temperature is continuously monitored, trends can be specifically identified and analyzed. If, for example, the temperature drops continuously over several days toward the critical range, a proactive alarm can be triggered before the temperature falls below the threshold and an error is reported. An AI-powered analysis can also help classify the alarm and draw on additional information – such as operating data, documentation, or stored expert knowledge – to assist further diagnostics.
This is exactly where the AI assistant TwinCAT 3 CoAgent for Operations comes in. It supports service personnel and maintenance engineers during ongoing operations by accessing configured operational data and analyzing process values, log files, and key performance indicators. On this basis, it can support a structured diagnostic process, test hypotheses based on the available data, and provide specific recommendations for action along with troubleshooting instructions. The employees in charge retain control over the results and their implementation. In this way, an initially isolated status report is transformed into a context-based recommendation for action, and the analysis of the data evolves from a simple alarm to a supported decision-making process.
In the long term, the combination of continuous condition monitoring and the use of AI could even lead to the automated generation of maintenance orders that specify which component on which piece of plant equipment should be inspected or replaced by a certain date, for example. This reduces the workload associated with routine inspection and analysis tasks and allows existing maintenance personnel to be deployed in a more targeted manner. Especially in view of the increasing shortage of skilled workers, the combination of condition monitoring, automated data analysis, and AI can play a key role in ensuring that plants operate efficiently and reliably, even with limited personnel resources.
[2] Fachkräftecheck Chemie 2025: Branche vor Herausforderungen - KOFA
[3] https://www.profibus.com/technologies/profinet/information-models/pa-dim