The one speaking with the fan

Data Language and AI – Messengers of Digitalization

Digital transformation has not only been associated with ChatGPT but also with artificial intelligence. The tasks assigned to it encompass both repetitive and highly complex activities. Generative AI can create texts and images from human inputs, descriptive AI can recognize complex patterns and describe current states. AI applications are found in services as well as in production.

Digital transformation also means communication. The means of this include, among other things, the currency of the digital age, data—not only personal data. Every machine, in the case of ventilation technology, every fan, continuously communicates through data generated during operation. Properly utilizing this data means a form of interaction between members of different species – humans and machines – enabling a profound understanding. Because with all data, the machine reveals something about itself and its condition.

The challenge is to hear and understand the language of the fan. Ziehl-Abegg addresses exactly this by using artificial intelligence for evaluation, thus becoming a translator of data-based fan communication.

Traditional Maintenance – Limited Scope for Action

Understanding the fan in its entirety can be particularly important when it comes to potential failures or the need for maintenance activities. The traditional approach waits until components become inoperable. This usually results in unplanned downtime, sometimes at particularly inconvenient times. Follow-up damages to adjacent components are also possible. Interventions at fixed intervals, where downtime is planned, can be helpful, but it can also happen that, for example, fully intact parts are replaced unnecessarily. Additionally, experience or data on operational lifespan do not protect against unplanned and "out-of-sequence" failures of components. Every product is different; different processes do not allow for a 100% accurate assessment of the condition. This means that resources like spare parts must always be available and kept in stock, and service staff must be available at short notice.

It is thus nearly impossible to find the perfect moment for intervention in operation for maintenance purposes. Failures rarely occur suddenly; more often, they give signs beforehand. But if these silent signals cannot be seen and it’s not possible to recognize externally that a component is no longer functioning optimally, actions are based solely on assumptions. If the fan could provide information about itself and its operational state at any time, humans would always be informed. Understanding the data-based language of the fan means knowing exactly how it and its components are doing at every moment. The occurrence of an impending malfunction could be predicted much more accurately.

Embedded Sensors – Gathering Data

Ziehl-Abegg has already taken several pioneering steps in this direction. The challenge initially is to acquire the data. Fans have been equipped with sensors tailored precisely to the requirements of ventilation technology to enable them to "speak." Sensors already present in the fan motor can directly collect vibration and telemetry data on-site; relevant parameters include, for example, vibrations in the axes, rotational speeds, temperatures, and power consumption.

Transmission to the Cloud – Data Historization

The collected data can now be visualized by sending it to the cloud solution ZAbluegalaxy. The fan’s language thus became visible. Measurement data could be monitored, logged, and stored historically. The analyses enabled by this already allowed initial conclusions about the fan’s condition, e.g., warning messages could be issued based on this data.

Embedded AI – Understanding Data in Real Time

The challenge of understanding the language and acting accordingly has thus been addressed. However, several factors make it nearly impossible for human experts to read, evaluate, and interpret data in real time. These include especially the sheer volume of data and the fact that some operational states cannot be deduced from individual data points alone but only from their combinations. Continuous monitoring and analysis of data by humans are not feasible. Even processing a small part of the data would already be outdated. Therefore, real-time data cannot be used meaningfully. In case of a failure, the window for timely action might be missed.

This is where artificial intelligence, in the form of a neural network, comes into play. Neural networks, consisting of interconnected artificial neurons in multiple layers, excel at receiving and analyzing data. Data arrives at the input layer as a vector and passes through one or more hidden layers to the output layer. Each level processes the data. The calculated values are transformed into a human-readable form at the output layer, for example, a percentage indicating the likelihood of a component failure. If a certain probability of damage is detected, a warning or error message is issued. Based on these evaluations, AI can provide recommendations for action if needed. It can also assist in diagnosing problems or failures by identifying possible causes from operational data. AI thus takes on the role of an employee who constantly listens to the fan’s communication, understands, interprets, and suggests actions. Data previously collected can now be processed meaningfully.

The Onboarding of AI – Preparing the Data

But like any employee, AI must also be trained. Ziehl-Abegg has trained it through extensive preparatory work: millions of data points were collected, recorded, and assigned to specific operational states. This required decisions about which data points and features are important for analyzing certain conditions and how they should be weighted. Failures and damages were simulated so that AI learns the fan’s normal behavior and can detect even the smallest deviations immediately, like a tracking dog. Alongside this, vast amounts of data were prepared through feature engineering—standardized and transformed. The data must not only be of high quality but also in a uniform format to enable comparisons. AI is thus given a guide to follow.

Particularly to perform calculations directly on the component itself, it is necessary to break down the data, which involves balancing simplification with the highest possible accuracy. However, this aggregation increases efficiency—too much data and too many features would hinder precise analysis. The trained AI can then be tested with validation data.

Perfect Match – Expert Knowledge and AI

Data and AI must first be prepared for their task. Even after this, AI plays an important role but does not become the sole decision-maker. It translates the fan’s language with excellent analytical capabilities for human experts and provides recommendations. The decision on how to handle a recommendation must, however, be made by human expertise after reviewing the information and considering other factors. Humans must continue to possess technical and contextual knowledge; AI cannot act independently. It is possible to set the fan to shut down immediately in case of an error message— but this feature must be actively requested and is not a standard setting.

Overall, artificial and human intelligence work closely together. The possibility of real-time monitoring and the analysis capabilities of the AI embedded directly at the component create a perfect interplay with humans who possess the expertise, are authorized, and capable of making decisions.

The Revolutionized Maintenance – Action Authority through AI

It becomes evident that humans now have even more scope for action than before. Their collaboration with AI results in a modern maintenance system that can score in multiple ways. Whether a damage occurs predictably or unexpectedly, every damage and potential failure is now announced in advance, providing time for reaction. This shifts human action more toward proactive measures rather than purely preventive or emergency responses. Predictability is key. Unexpected failures also give signs beforehand, and it is no longer necessary to replace intact parts that might last longer than anticipated. The lifespan of a component can be utilized to its maximum, and the ripple effects caused by a faulty part are mitigated.

The announcement generally provides enough lead time to determine a favorable maintenance window. Ongoing processes are optimized, and the reliability of the fan increases. Downtimes are limited to actual maintenance time. Employees no longer need to be on standby, and spare parts do not necessarily need to be kept in stock. Time and costs for maintenance are minimized. Maintenance is no longer a burdensome, surprising disruption but is integrated into the operation.

Besides the advantages of maintenance optimization, other benefits emerge over time. It becomes possible to identify specific problem areas, e.g., in case of recurring similar error messages, or to define the time until the first faults occur, also depending on environmental factors. Operating parameters can be adjusted more easily when needed. Maintenance plans are no longer redundant but can be optimized. Overall, it is possible to learn a new level of operation for the fan, enabling entirely new maintenance strategies.

The Significance of Time and Space – The Embedded System

In addition to the general use of artificial intelligence, the deployment of an embedded system is particularly noteworthy. More precisely, the "embedded system" exists in two senses. The sensors do not need to be purchased separately; they are already integrated into the motor itself—tailored to the requirements of the fan—while external sensors might not capture all relevant data.

AI does not need to be purchased separately either; it is already embedded at the core of the system. Additional external software or infrastructure, such as data analysis platforms, are not necessary.

Connecting the fan to a gateway and transmitting data to the cloud—where data must be actively requested by the user—is possible but not critical for analysis in the case of a single device. The advantages speak for themselves: minimal latency, truly real-time analysis. If visualization or data storage is desired, or if multiple devices are to be networked, the cloud takes on the crucial role. The amount of data transmitted is thus reduced, decreasing network load. This also benefits the data that is transmitted. The cloud can make the AI’s work visible externally, while the AI ensures that only processed, refined, and higher-quality data is sent.

Conclusion

Ziehl-Abegg continues to pursue its path of utilizing existing data that precisely describes the condition of the fan. Artificial intelligence is no longer a nice-to-have but an essential component of the next step. The AI solution has created a tool that will revolutionize maintenance even in ventilation technology. AI translates the data-based language of the fan and becomes "communication intelligence." Its description of operational states allows for predictions of possible failures; descriptive AI becomes predictive.

In the future, it can be installed in every fan as a standard, paving the way for entirely new working methods. It provides the operator with a continuously present service assistant that analyzes optimally prepared data at all times and draws conclusions. Paradoxically, this additional step of assistance leads to greater autonomy through improved action possibilities. Conversely, one step is eliminated: thanks to the embedded solution, data transmission is not necessarily required. The latest digital tool, AI, thus makes constant network connection unnecessary and optimizes the use of the cloud. On one hand, AI stands alone; on the other, it creates an optimal interplay in a hybrid solution with ZAbluegalaxy.