True Predictive Maintenance Thanks to Innovative Technology

Chen Linchevski: Precognize makes it possible for us to comprehensively analyze the large amounts of data that are generated in an industrial plant. We are now able to detect and remedy faults and anomalies that are deeply rooted in the plant structure.

Dr. Andreas Widl: In addition, the technology allows us to compile the enormous know-how that the plant owners and operators possess on the topology and processes in their plants and reproduce it in a clearly structured digital model.

Dr. Thomas Steckenreiter: We offer the first solution that brings together this complex topology of different field units with the alarms and error messages documented in their data.

Linchevski: Our solution offers true predictive maintenance. We analyze the data, detect deviations and are capable of relating them to certain field units.

Linchevski: There is no way of reliably predicting malfunctions or failures of a valve, for example. What we can see though is that there are changes in the measured values and data recorded by the sensors of this valve. The deviations – we call them anomalies – accumulate.

Widl: Yes, that's true. But the problem with these tools is that they show the anomalies to the plant operators without any filtering. Plant operators realize that something isn't working the way it should do but they still cannot say what really causes the problems. With our solution, we can pinpoint the exact cause of the anomalies – right down to sensor level.

Linchevski: With SAM GUARD, a software tool based on Precognize, we can cover the topology of a single machine, a plant or an entire works site. We scan the plant for about two weeks and learn how the plant operator runs which processes. After that, we have an exact idea of which valve and which sensor is involved in which process.

Linchevski: I would rather call it a 'digital map'. We relate this digital topology to our algorithm, which permanently analyzes the constant stream of data. In reality, this enables us to single out anomalies and, based on the digital model, we can indicate whether they point to a specific problem.

Linchevski: Exactly. The core strength of our technology is to predict which of the sometimes thousands of field units in a plant will have to be inspected, replaced or repaired in the future.

Linchevski: The technology is based on an algorithm, which assigns the anomalies and relates or links them to certain field units. To give the algorithm the capabilities it needs, we need to feed it with historical data. Based on these data, the algorithm generates normality clusters. This means it collects more and more details to detect the limits of normal operation, which allows for a quick and reliable identification of any new malfunctions or failures that occur. In a second step, we can assign the anomalies based on the digital model of the plant and visualize them using a human-machine interface (HMI).

Steckenreiter: You could say that. By solving these problems, our technology can profoundly optimize plant operation: the topology linking reveals previously unknown dependencies between field units and machinery to the plant operators, who can prevent malfunctions in plant operation in the future.

Linchevski: In addition, Precognize learns with every new anomaly and demands operator feedback to develop further. The system learns something new from every failure, regardless of the type of field unit that is being used.

Steckenreiter: No, the system can be used regardless of the device types to be monitored: You can use it to monitor sensors, valves, pumps or entire plants with several thousand field units of different quality levels. The employed methods are always the same.

Widl: Another great advantage is that different user groups can work with different areas of the tool. For example, it is possible that the sensor service team only has access to the sensor monitor in the HMI while the valve team only sees the valve monitor.

Linchevski: Absolutely. We need to speak the language of our customers. Otherwise, our tool just wouldn't be accepted in the market. We are not limited to certain semantics, neither when creating the digital model nor for the analyses. Our software supports all common machine languages.

Linchevski: Yes, this is definitely an option. It is really not a question of technology but rather of the business model. Precognize lets us interrelate so many data and creates so many opportunities that it will take us another few years to fully exploit them. We are just at the beginning.

Linchevski: Our hope for the future is that, in addition to predictive maintenance, we will also be capable of providing prescriptive maintenance. This would be the next step in our evolution. The tool would not only detect an anomaly and assign it to a specific field unit, it would already start the necessary processes as well./p>