Soft sensors

Soft sensors are an important building block for the process industry! With their innovative technology, soft sensors offer an effective solution for optimizing production processes and increasing product quality.

By using soft sensors, you can monitor your systems better, continuously analyze important process parameters and react to deviations at an early stage. This not only leads to greater efficiency and productivity, but also to improved product quality and cost savings.

The possible applications in the process industry are diverse and offer an effective, economical solution for optimizing production processes.

Petra Steinmetz, Lukas Haffner, Softsensoren, CONENGA Group

P. Steinmetz & L. Haffner

We are here for you!

Petra Steinmetz and Lukas Haffner are your experts when it comes to soft sensors!

Would you like to find out more? Should we call you or would you like to receive further information by e-mail?

Let’s talk about it!

WHY soft sensors?

Whether temperature, quantities, masses, pressures, concentrations or other conditions – sensors provide the data required for automation and digitalization. Unfortunately, there is always a lack of basic data for evaluation or regulation. If, for example, physical measurements are not possible or only possible with great difficulty, or if there is no reliable measurement method, we use soft sensors.

What does a soft sensor do and when is it recommended?

Ultimately, a soft sensor provides a reliable online calculation from other, easily available measured values and corresponding software-based simulation of the physical relationship. It is therefore “a piece of know-how captured in software.”

To put it pragmatically: we calculate process values that we cannot measure but need for optimum control from available variables and the physical-process engineering context – online, in real time.
If sufficiently reliable variables are not available or the physical calculation cannot be replicated as a logical algorithm, we also use data-driven AI models as a tool. A soft sensor is therefore an important tool if a physical measurement (valid sensor) is not possible or too expensive. It is also used when existing, less reliable physical measurements need to be checked for plausibility or calibrated.

Examples from our field of application

Soft sensors in combustion technology:

In combustion technology, we use soft sensors, for example, to calculate heat quantities, monitor O2 measuring probes (drift) or detect the firing position in the stove.

For the heat meter, we use a neural network with flue gas-based input data for validation and to bridge short outages.

We also work with neural networks to provide the O₂ value as a central control variable for firing systems; the measurement drift of the physical O₂ sensor is detected and can be corrected to increase efficiency.

For fire position detection, we use either temperature measurements via the course of the grate or the camera image.

Soft sensors in continuous production – example impregnation channel:

For the impregnation of decor papers for coating laminates or wood-based materials, we use soft sensors to ensure the optimum resin ratio and ideal drying. With the resin ratio, for example, we use the weight changes of the storage containers and use these to adjust the channel. This simplifies the manufacturing process and has a positive effect on product quality.

Soft sensor impregnation channel

Here you can find out everything about the typical process of a soft sensor project

Typical project phases in the development of soft sensors

We basically differentiate between the following phases in the design of soft sensors

BEFORE integration (for example in a specific control or visualization) and
AFTER the integration

Requirements analysis / problem definition: Identification of the specific requirements, the “measurement problem” to be solved, the objectives for the soft sensor, including the required accuracy, the possible data sources and the integration requirements.
Preparation, procurement, analysis of data: Acquisition of required / available data from various sources; preparation of data for analysis, including cleansing, transformation and formatting.
Development, selection and training of the model: selection of the appropriate architecture, development of correlations and algorithms, selection of training parameters and training of the model with the available data.
Validation model: Checking the required accuracy / performance of the trained model using test data or test positions in iteration with the previous phase until the desired significance or accuracy is achieved.
Provision, integration, training, commissioning soft sensor: Provision of the trained model, integration into existing systems and data structures or the customer’s operational processes are the basis for commissioning and user training to ensure the dynamically functioning soft sensor.
Monitoring and continuous improvement: The implemented soft sensor is monitored and, if necessary, continuously developed – in step with the further development of the respective system. This includes regular updates and adjustments or retraining of the model and is supplemented by the provision of technical support.