Artificial Intelligence Boosts the Reliability of Your Measurement Data

Machine learning supports accuracy and exposes potential discrepancies in test results at an early stage.

The growing share of renewable energy sources leads to an increasing decentralization of the electrical energy production. Therefore, more and more regulated loads appear in the power grid and with them the risk of dangerous effects such as short circuits, harmonics, and transients. The changing operating conditions for the assets make it advisable to carefully check for possible changes of measurement data.

Rapidly changing load conditions in the power grid also affectthe expensive power transformers. Appropriate maintenance measures are needed now to ensure installed power transformers' long and healthy life cycles. The SFRA method is one commonly used offline procedure proven to be the most sensitive and non-invasive method for detecting mechanical and electrical faults. It is one important procedure for condition-based transformer assessments.

What does SFRA stand for?

SFRA means Sweep Frequency Response Analysis. and is based on a comparison of actual and reference measurements (fingerprints).

What can SFRA do?

The analysis can perform measurements on power transformers in a frequency domain.

What are the advantages of the SFRA principle?

It is robust against broadband and narrowband noise and thus, it can achieve a high signal-to-noise ratio.

Why is SFRA such a popular measuring method?

Compared to other methods SFRA is a non-invasive measurement method and can be performed fast.

Why are SFRA results comparably reliable?

The measurements are based on a comparison of actual and reference measurements (fingerprints). Therefore, deviations due to measurement errors can be detected quite easily.

 

Machine learning with the SFRA method

The SFRA method provides results about the mechanical integrity of the core, windings, clamping structure, and electrical integrity, such as shorted windings and turns.

How does the SFRA measurement work in detail?

A low-voltage signal with variable frequency is injected into one power transformer terminal and measured at the other terminal. Comparing the output and input signals gives a frequency response that can be compared with previous measurements such as the reference measurement.

In which way machine learning can give additional value here?

With the help of machine learning / artificial intelligence it is possible to compare test results with the help of algorithms based on expert knowledge and a large amount of data. Those models help detect device- and application-related systematic measurement errors and quality issues very early.

Which external factors can influence the measuring results?

Changes in the measurement setup, transformer state, transformer configuration, and external factors can affect the measuring results and thus lead to a misdiagnosis. Therefore, checking the validity and quality of the measurement results before evaluation is important.

What are the evaluation references for the SFRA measurements?

Measured SFRA curves are evaluated with reference data by using one of the two standard algorithms, the NCEPRI algorithm (NCEPRI, North China Electric Power Research Institute) or the DLT algorithm (DLT, the Electric Power Industry Standard of the Peoples Republic of China (DL/T 911 – 2004)).

How can machine learning / artificial intelligence be connected to SFRA measurements?

For the machine learning process 19.787 SFRA curves representing over 2000 power transformers were available to train the algorithms. The results show that artificial intelligence-based algorithms are suitable for automatic qualification and validation of SFRA measurements. Currently, the algorithms are very well suited for auxiliary purposes. Users can pre-validate their own SFRA curves based on three different classes. The measurement results qualify as OK, Investigate, and Error. This possibility enables an enormous increase in the quality of your data and brings your asset into sharper focus.

 

Learn more about OMICRON’s SFRA test system FRANEO 800 and it’s profound knowledge about SFRA-tests.

Read more

Listen to our podcast

We have produced a podcast episode in which you can hear how important the validation of measurement data is for more reliable assessments.

In this episode, OMICRON data transformation experts David Gopp and Lukas Klingenschmid talk about the validation of measurement data and the role it plays in achieving a successful digital transformation in the power industry. Listen now:

Energy Talks features various episodes related to power system testing. You can find all episodes on our podcast landing page.

Discover our Coverstories

Listen to our podcasts

You are using an outdated browser version.
Please upgrade your browser or use another browser to view this page correctly.
×