Analog pressure gauges are widely used in many industries, and such gauges shall be verified no more than every 6 months according to the JJG52-2013 standard in China. Traditionally, the gauges are verified manually, and this is no easy job due to the number of gauges that need to be checked on regular bases. One of the most important but tedious steps during the verification process is reading the outputs of each gauge accurately when it is pressurized at different levels, and the reading accuracy can be affected due to the fatigue of humans. This paper described a comprehensive machine-learning-based analog gauge reading approach to facilitate the verification process and reduce the workload of humans. A semantic segmentation model was implemented for retrieving the masks of the pointer and the scale area to calculate the angular displacement of the pointer. The numbers and the gauge units were recognized using OCR algorithms. Finally, the actual reading of the pressure gauge can be determined based on the angular displacement of the pointer and its corresponding number and unit. The experimental results showed that the method described in this paper could fulfill the required accuracy of the verification standard.
|