[ Log In | Sign Up | Lost password? ]

Sensor fusion & soft-sensors

Although in many application redundant and correlated sensor types exist, only few of them tend to complete leverage the possibilities of fusion techniques. My research interest focuses the autonomy of sensorial inputs, the prediction of sensor failures and according isolation or mitigation strategies. From my view, it is possible to hierachically classify following techniques: data fusion, sensor fusion or information fusion in general.

Data fusion

Data is hereby combined from homogenous or distributed sources. From the statistics of our data we know important indicators like e.g. the covariance and using techniques like covariance intersection we can easily fuse data from different inputs. Covariance intersection is a very simple and more ore less introductory methods, but it illustrated the key tasks of algorithms for data fusion, which is to make sure only healthy information content is regarded in the data evaluation.

I apply several data fusion principles in order to analyze the data quality of distributed sensors continuously.

Sensor fusion

While data fusion restricts to fusion techniques that use the statistical nature of the data itself to fuse it, sensor fusion extends this view to models of the process. This means, that the dependencies between different sensors are known in advance and a mathematical relationship between the sensors can be used to improve the data quality and, more important, to measure physical quantities that are principally not measured by just using one of both sensors.


A soft-sensor is a virtual new sensor, built by fusion multiple input sources. These inputs could be two sensors or, which can be encountered quite frequently, a sensor and some database with prior knowledge values. Nevertheless, the soft-sensor takes the data of its inputs and delivers a new measurement value.

Examples of soft-sensors

The cognitive filter and the strontium-ID module are in fact soft-sensors. They are used in a wider scope for improving the function of modern radiation equipment.

Isolation strategies for sensorial faults

Typically soft-sensors give a very important inputs for other processes such as control. Soft-sensors can also help to identify sensorial malfunctions. Some methods to do this are e.g. inspired by the human immune system. The defect sensor can then be isolated and all subsystems depending on the sensor can be informed about a problem.

Mitigation strategies

In the best case, the defect of a sensor can be removed by smartly re-interpreting its results. This is called mitigation.