OverviewQuality control processes are essential for many companies, and the correct detection of errors lead to much lower costs and unplanned work in the future. However, one of the main pains of organizations is the cost and effort to perform quality control manually. For example, in the energy distribution sector companies spend a lot on training technicians in intervention procedures and photographic documentation of energy meters.
ChallengeAll interventions in electricity meters require technicians to document the state of the meter with an image on site. They are stored in a large database which grows by thousands every day. The problem with most of these images is that they are of very poor quality or they even don’t show the necessary information.
OpportunityTo automate the quality control process for energy meters we have developed a solution for classifying images based on neural networks. The main goals were to recognize the presence or absence of an electricity meter and also distinguish between digital and an analog one.
Another identified need was to recognize the numerical record present on the energy meter display. As there is no assured standardization of this type of equipment, the format and technology (digital or analog) of the meters can vary too much depending on each case.