Fields of Research and Development
HawkSpex® Embedded integrates developed soft sensor solution in industrial equipment and systems – scaled to the hardware available. Apart from simply implementing the necessary functions, the validity of the system and the data processed are monitored, e.g. when process conditions change.
High-performance hardware and software that can rapidly process large quantities of training data, test different types of models, and optimize hyperparameters are employed to create a soft sensor model. Once a valid soft sensor model has been created, e.g. with our HawkSpex® Flow, it must be run reliably on our client’s system.
We designed HawkSpex® Embedded to run models. It is a lean, stable, exhaustively tested and long-lived software.
Transfers between systems are done in a file that stores the soft sensor model’s functions in an abstract mathematical form. Should the model have to be altered or modified for new conditions, no new program has to be installed and nothing has to be updated or recoded. Only the model file has to be replaced. Model files can also be easily versioned and added to the documentation.
With HawkSpex® Embedded, the new model can also initially run concurrently in test operation and then be smoothly transferred to regular operation.
HawkSpex® Embedded is available for different target systems and configurations. The container solution can be used in the cloud too. The special thing is that a model created with HawkSpex® Flow is always identical and guaranteed to produce identical results on any system.
Does the soft sensor model still function? Does it still deliver valid results?
Many users ask themselves these questions. Determining if all the sensors are still working, the operating point is drifting or the model is computing data for which it was never trained is not easy, especially when the soft sensor model obtains its results from many inputs.
We provide extensive monitoring functions to answer such questions during operation. Static and dynamic parameters and parameter maps for every single sensor, for sensor groups and even for all sensors can be monitored together. The necessary parameters can be automatically extracted from the model’s training data or ascertained from experiential knowledge or data sheets. The model outputs and the model’s sensitivity can be monitored as well.
What happens if a test is not passed? Our clients can individually decide in advance whether a warning signal suffices, the system must be stopped or there is a second, more robust (fallback) model that can be switched over to. The integrated monitoring system ensures that our clients are always on the safe side.
Fraunhofer Institute for Factory Operation