Continuous modelling of multidimensional data streams

Real-world data monitored using only the traditional, oversimplified measurement types, a few temperature and pressure sensors, a few accelerometers, light measurements or cameras with only a few wavelengths of light is often not adequate, since it may produce unrealistic cognitive strain on the observer, and may well be overwhelming.

Expanding to whole arrays of temperature, pressure, mechanical or electromagnetic vibration spectra sensors allows scientists to gain deeper insight, and in turn operators to better control processes. Adequate data modelling, along with high-dimensional measurements can give a valuable overview and add sensitive alarms.

However, without such tools, the volume of output data from the system becomes overwhelming.  For each new sensor channel added when monitoring a process, it has been tradition to assign a new graphical display. For instance in industrial process monitoring, ship control rooms and in medical intensive care units.  This gives cognitive overload for the operators, and the many false alarms bothers the operators, who may lose respect for alarms in general.  Moreover, the size of raw- data files to be transmitted, stored and later interpreted become staggering:


OnTheFly model-based data processing

Idletechs develops software for handling of large amounts of data to aid customers in understanding and utilizing data flows coming from modern sensors like hyperspectral video cameras. From raw streams of data, systematic patterns and relationships are automatically discovered and modelled. The data can be stored in a highly compressed format.

Using our software, we can reduce the cognitive load of operators, give fewer but more sensitive alarms and, reduce file sizes. This gives the users better overview of the process being monitored. The OTFP learning model automatically detects any new co-variation patterns arising in the data stream and will remove much of the noise in the data.

Together, the compressed state variables and residuals statistics from the OTFP form the basis for few compact, reliable alarms. Reacting to subtle trends over time, to too high or too low levels of known phenomena and to abnormal, unexpected process upheavals or sensor failures.

When associated with external information about the inputs to the process, the graphical representation becomes even more understandable. In process control, the compact state variables generated from the process may be used as inputs for improved prediction and control. In multichannel instrumentation, such as hyperspectral cameras, the external information may be used by the OTFP system to give selectivity enhancement and thus more meaningful measurements:

The OTFP software version 1.0, is available in 2016 Q3.