Online Inductance Estimation of PM-Assisted Synchronous Reluctance Motor Using Artificial Neural Network

Karami-Shahnani, Ahamdreza; Dehghan-Niri, Hossein; Nasiri‐Zarandi, Reza; Abbaszadeh, Karim; Toulabi, Mohammad Sedigh

doi:10.1109/pedstc57673.2023.10087136

Cited by 7 publications

(2 citation statements)

References 21 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Compared to [8] which uses Kalman filters with model-based estimation requiring knowledge of Ld and Lq, the proposed method estimates all the parameters without requiring precise knowledge of each other. [14] uses artificial neural networks to estimate the parameters which are trained off-line, requiring additional effort at the training stage and not generalizable to other systems outside of the training dataset. The estimation in [28] uses two-stage RLS by assuming certain parameters that are known from off-line testing, which requires off-line testing and parameter knowledge, which is not the case in the proposed method.…”

Section: Conlusionsmentioning

confidence: 99%

“…The literature is rich with parameter estimation methods for PMSMs, which have a very similar structure to the PMaSynRMs. Among the most popular and still studied parameter estimation methods are model reference adaptive systems [6], Kalman filters [7], [8], ordinary or recursive least square methods (OLS, RLS) [3], [9], [10], [11], [12], Disturbance Observers, and Artificial Neural Networks [13], [14]. Among these methods, the RLS is the most common preferred, as it is easy to implement and robust compared to other methods.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

FFT-Based On-Line Electrical Parameters Estimation for PMaSynRM Considering Flux Harmonics

Tap,

Sarikaya,

Yilmaz

et al. 2024

IEEE Access

View full text Add to dashboard Cite

Permanent magnet assisted synchronous reluctance motors (PMaSynRM) are popular motors in today's industry because their cost, performance, ease of control, field weakening (FW) and robustness are comparable to their competitors. High performance control of these machines requires a priori knowledge of motor parameters and the internal and external effects on the variation of these parameters, such as temperature, permanent magnet (PM) demagnetization, and magnetic saturation. The advantages of on-line identification can be used to reduce the development effort and the use of look-up tables for control, instead of experimental and off-line identification. Data-based identification of dynamic systems is a promising area for electric motor parameter identification. Such methods require a data set containing correctly excited system dynamics -so called persistent excitation condition over the entire identification data set. Although voltage source inverter, magneto-motive force (MMF) and flux harmonics can be used for parameter estimation, this is not sufficient in most cases and usually additional identification signals are injected into the system, causing unwanted disturbances such as torque and speed ripple or acoustic noise. This study proposes a near-zero torque ripple current injection method to provide a persistent excitation for the PMaSynRM to generate a parameter identification data set in real time. The identification data is recorded for two electrical periods and a fast Fourier transform (FFT) based denoising and thresholding is applied to the frequency domain components to retain only the relevant components for identification. Least squaresbased regression is then applied to the identification data on harmonics extended q-axis difference model to obtain the d&q-axes inductances (Ld & Lq) and the stator winding resistance (Rs). The PM flux linkage (λm) is then derived for this operating point using d-axis fundamental voltage equation. The proposed method achieves good estimation accuracy without degrading the speed control with average errors of 3.66%, 4.82%, 1.3% and 3.25% for Ld, Lq, Rs and λm respectively under an equally spaced torque-speed measurement grid. This results in a system capable of estimating all the parameters on-line with very high accuracy compared to the literature.

show abstract

Section: Conlusionsmentioning

confidence: 99%