The measurement of water-levels is critical within hydropower production and with already existing camera surveillance in abundance for the purpose of manual supervision. The allure of automatic visual reading to replace the need for manual oversight is significant in the pursuit of fully data driven solutions within hydropower systems. Could images of water level scales along with machine learning functionality produce a reliable and feasible solution? There are many aspects of visually reading any water-level in practice, such as lighting conditions, environmental interference. Great water level fluctuation needs to be overcome by providing an expansive and diverse dataset based on high resolution image capture. The provided solution is based on machine learning algorithms such as two- dimensional convolution, computationally performed and trained by a high power desktop computer. This algorithm is deployed in the field on a low power System-on-a-Chip (SoC) computer with dedicated in system high resolution camera. Basic image manipulation is performed in algorithm to eliminate image noise and to focus on level scale region of interest. The provided solution overcomes the issues at hand and results in a tested proof of concept system capable of ±5mm level reading accuracy with reliability of up to ≥ 99%, within a predefined data range. The results prove that the solution is feasible and a system implementing it or a derivative solution is practically implementable for real life use cases at edge locations.
| Identifer | oai:union.ndltd.org:UPSALLA1/oai:DiVA.org:ltu-93741 |
| Date | January 2022 |
| Creators | Moregård, Jakob |
| Publisher | Luleå tekniska universitet, Institutionen för system- och rymdteknik |
| Source Sets | DiVA Archive at Upsalla University |
| Language | English |
| Detected Language | English |
| Type | Student thesis, info:eu-repo/semantics/bachelorThesis, text |
| Format | application/pdf |
| Rights | info:eu-repo/semantics/openAccess |
Page generated in 0.0095 seconds