PV-Arc Detection - Circuit supervision using DC-Arc Detection in PV systems on dwellings.
Starting point / motivation
Photovoltaic systems installed on buildings nowadays are designed to maintain high levels of DC-voltage. In addition to that the necessary inverter is usually installed closely to the „point of common coupling" (PCC). As a result the DC wirings in urban PV systems are getting more and more complex.
One of the drawbacks of DC wiring is that electrical arcing can occur at connectors and other kinds of contact wiring. Caused by the high energy maintained by the PV generator severe fire hazards can result from such arcs.
Based on this, development requirements have been established by regulation authorities in the USA. They require a mean to detect and interrupt occurring arcs defined under very narrow conditions of ignition.
The detection of more realistic arcs is a task that is far more challenging to achieve since the variety of conditions of ignition is huge. Therefore conventional threshold detection algorithms are insufficient for a reliable detection.
Contents and goals
One Goal of this project is to research fundamentals of electrical arcing with PV sources in
order to find classification methods including low energy arcs and unclear conditions of
ignition in a serial and parallel arcing situation. From our point of view the increased fire
safety as an outcome of this project is one of the key requirements necessary for the
sustainable and save implementation of photovoltaic in urban communities.
The approach to reach our goal is to gain the knowledge to find the key relevant factors for detection of realistic arcs by establishing test setups and parameters. Finally the main goal is to find pattern recognition methods based on laboratory- and field-evaluated data.
Our experience from the USA shows that the main challenge is to detect a big variety of possibly occurring arcs and keeping the number of false positives as low as possible at the same time. The basic research we are conducting enables us to avoid false positives and increase fire safety in PV systems in Europe, regardless normative requirements which may arise.
In addition to literature search on arc ignition, a knowledge transfer from Fronius branch Perfect Welding was performed. Several environmental stress tests were conducted. The validated high resolution data material is key to verification of arc models, simulation, field data test infrastructure and pattern recognition methods.
Construction and implementation of field data test infrastructure employed available design and production processes, i.e., circuit design and simulation, PCB layout, electronic production and testing as well as EMC and environmental tests.
A software toolchain was implemented for research behaviour of different pattern recognition algorithms on the detection of arcs. The resulting toolset included data management, central machine learning engine (consisting of pre-processing, feature extraction and classification training and testing), analysis and visualization of results.
In order to extract meaningful features an analysis of arc signals was performed. A Matlab/Simulink model of a PV system and inverter was implemented. The Ayrton arc model was incorporated into the simulation.
Machine learning based classifiers allowed to increase the detection hit rate on drawn arcs with weak energy from 16% to 85%. At the same time false positive rate with module electronic could be reduced from 54% to 2.5%. The resulting classifiers have been included in field data logging modules and field tests have been started.
Prospects / Suggestions for future research
Within this project the foundations for classification of arcs at weak energies or resulting from unclear ignition events (e.g. contact glowing, power fluctuations or re-ignition) have been investigated. From our perspective the resulting safety increase of PV plants is key to a sustainable usage within urban areas and should be pushed further.
Fronius International GmbH