Rocsole’s solutions are based on tomographic technologies, which enable whole volume imaging of a process pipe or tank without using a radioactive source. Tomographic technologies are especially suitable for measuring and controlling multiphase flows. Results in process tomographic imaging are displayed as an image and indices.

The general idea in tomographic measurements is to expose the target of interest to a physical stimulus, e.g. electromagnetic waves, radiation beam, acoustic waves or electrical signals, and measure the response caused by the target. From the response signals it is possible, with the aid of mathematical models, to infer the distribution of different material within the target.

Rocsole's current systems utilize Electrical Tomography (ET).

Bayesian Inversion and Process Tomography

In the majority of real world problems, the interesting quantities cannot be measured directly. Instead, some measurable quantities are usually related to the interesting quantities via mathematical models, and thus, information on the interesting quantities can be obtained. With stable problems, one can perform a more or less straightforward model fitting procedure to gain this information. Technically, this fitting is usually carried out by minimizing the difference between the measurements and the model predictions. With unstable problems, which are also called ill-posed inverse problems, such straightforward model fitting cannot be employed. Such problems can, however, be tackled by using so-called deterministic regularization approaches, or by formulating the problem in the Bayesian statistical framework. The latter approach is feasible also with problems in which the models themselves are only partially known or contain errors.

At Rocsole, we work with tomographic problems, which are one of the largest classes of inverse problems. In particular, we consider a soft-field tomographic modality that probes the unknown object via electric field. In ill-posed problems, small errors in the data or associated models and mappings may cause very large errors in the solution. In deterministic framework, the solution is interpreted as an unknown parameter/vector/function, and we lack the means to statistically assess the reliability of the solutions or any ways to handle uncertainties. Our approach here at Rocsole is to use Bayesian inversion based approach and then we provide point estimates together with uncertainty of posterior. Bayesian inversion allows us to use so called approximation error method to take into account uncertainties and get accurate reconstructions in industrial use.

Typical modeling uncertainties are:

  • Model reduction errors
    Errors due to using a reduced order model, usually for computational efficiency
  • Uncertainties related to the physical models
    Missing boundary and initial data, simplified physical models, uncertainty of geometry
  • Uncertainties related to the measurement system
    Unknown measurement noise variance, covariance structure, electrode contact etc.
  • Prior and other uncertainties
    Generally uncertainties in the prior model (covariance structure and the distribution)
Need more information? Contact Rocsole's sales / technical support »Click here

Multiphase Pipe Flow Visualization with Process Tomography

Multiphase Flow Visualization

Slug Flow and Deposition 3D Imaging with Process Tomography

*If the video is not displayed, visit Product Videos page for a downloadable version


  • Less process chemicals
  • Optimized maintenance actions
  • Better product quality

Also Check

Products System Development Frequently Asked Questions