A Novel Resolver-to-Digital Conversion Method Using Time Delay Neural Networks
Amirhosein Asilian · IEEE Transactions on Instrumentation and Measurement · 2025
This paper presents a novel approach to Resolver-to-Digital Conversion (RDC) utilizing Time Delay Neural Networks (TDNNs) to significantly enhance accuracy in angular position measurements, a critical aspect of precision engineering. the proposed method addresses these challenges by dynamically compensating for these issues, resulting in a more reliable and accurate system. By integrating custom-designed hardware with a TDNN implemented on a high-performance microcontroller, the proposed system not only improves resolution but also simplifies the overall design, reducing costs and increasing efficiency. Experimental evaluations demonstrate a substantial reduction in the average error to 0.0011773, alongside a mean squared error (MSE) of 0.045605 and R² values of 0.99977, 0.99978, and 0.99973 for the training, validation, and test datasets, respectively. These results indicate a significant improvement over conventional methods, highlighting the effectiveness of TDNNs in enhancing precision in RDC systems. The proposed approach holds promise for widespread industrial application, offering a robust solution for high-precision tasks requiring accurate angular readings. This research sets new benchmarks in resolver accuracy, paving the way for advanced control systems in various engineering applications.