Skip to main content

The engine company. DEUTZ®

More efficient drives thanks to artifical intelligence


Oil-smeared workbench? Environment fogged up by exhaust fumes? Not at all! When Volker Smits examines the behavior of an engine, it is usually a clean affair for him. Smits is a software developer and works with mathematical models that he optimizes step-by-step on his computer. For example, such models are used to map emissions of soot or nitrogen oxide in order to later adjust the exhaust gas after-treatment of drives to the respective, increasingly stringent limit values.

Adaptive emission model

In the form of so-called artificial neural networks and evolutionary algorithms, artificial intelligence ensures that these mathematical models match up with reality. The models undergo certain learning processes over and over again and are gradually improved. Smits already programmed such models during his studies before he joined DEUTZ three years ago to apply this method to the calculation of emissions. “There was already an emissions model, but we were not satisfied with the quality,” reports Smits. “That’s why we wanted to improve it.” Therefore, the developers generated a data set that, among other things, takes into account the engine design data and the thermodynamic properties of the fuels – in other words, data that influences the emissions behavior of the engine. In the next step, the developers used the data for an AI algorithm, which constructed a mathematical model of reality from it. Over several hundred runs, the software calculated an authentic image of the emissions output from the corresponding drives.

Real-time post-processing

This is invaluable for exhaust gas after-treatment. Why? If the control unit of a drive is equipped with such an authentic data model, the engine can call up its emission behavior at any time without physical sensors and adjust the post-processing in real time. The fast and high learning ability of AI is a great advantage when creating such data-based models. These models have been refined several times by algorithms, making it possible for them to represent laws that would be too complex for a conventional or physical model. For example, they can virtualize the realistic development of soot, and demonstrate how much stress soot particles place on the installed filters in a certain situation. 

This allows DEUTZ engineers to optimize these filters and ultimately design more efficient drives. “The newly developed model is very powerful and also suitable for demanding emission targets,” says Smits. “The limit values are in compliance, and the engines produce significantly lower emissions thanks to numerous optimizations made by the engineers.” This is also demonstrated by the ongoing digitalization of DEUTZ and the increasing complexity of modern drive systems. “Development in this area is becoming increasingly demanding, and the electronics sector is becoming increasingly important. Of course, this also applies to artificial intelligence in the most diverse areas.”

Preventing errors and problems

The emissions model is one of several AI projects by DEUTZ in which evolutionary algorithms are used. For example, they help identify parameters of purely physical models, for example, for mapping air paths or for simulating catalytic converters in exhaust gas after-treatment. Artificial neural networks, on the other hand, are used in predictive maintenance. “This allows us to predict operational errors based on measurement and production data from the drive,” says Smits. “In this area AI also ensures that drives run more efficiently.”