Attitude UAV Stability Control Using Linear Quadratic Regulator-Neural Network (LQR-NN)

Abstract

The stability of an Unmanned Aerial Vehicle (UAV) attitude is crucial in aviation to mitigate the risk of accidents and ensure mission success. This study aims to optimize and adaptively control the flight attitude stability of a flying wing-type UAV amidst environmental variations. This is achieved through the utilization of Linear Quadratic Regulator-Neural Network (LQR-NN) control, wherein the Neural Network predicts the optimal K gain value by fine-tuning Q and R parameters to minimize system errors. An online learning neural network adjusts the K value based on real-time error feedback, enhancing system performance. Experimental results demonstrate improved stability metrics: for roll angle stability, a rise time of 0.4682 seconds, settling time of 1.3819 seconds, overshoot of 0.298%, and Steady State Error (SSE) of 0.133 degrees; for pitch angle stability, a rise time of 0.2309 seconds, settling time of 0.7091 seconds, overshoot of 0.1224%, and Steady State Error (SSE) of 0.0239 degrees. The LQR-NN approach effectively reduces overshoot compared to traditional Linear Quadratic Regulator (LQR) control, thereby minimizing oscillations. Furthermore, LQR-NN can minimize the Steady State Error (SSE) to 0.074 degrees for roll rotation motion and 0.035 degrees for pitch rotation motion. ABSTRAK: Kestabilan perubahan Pesawat Tanpa Pemandu (UAV) adalah penting dalam penerbangan bagi mengurangkan risiko kemalangan dan memastikan kejayaan misi. Kajian ini bertujuan mengoptimum dan menstabilkan perubahan kawalan adaptif penerbangan UAV jenis sayap terbang di tengah-tengah variasi persekitaran. Ini dicapai melalui penggunaan kawalan Rangkaian Linear Kuadratik Pengatur-Neural (LQR-NN), di mana Rangkaian Neural meramal nilai perolehan K optimum dengan meneliti parameter Q dan R bagi mengurangkan ralat sistem. Rangkaian neural pembelajaran dalam talian melaraskan nilai K berdasarkan maklum balas ralat masa nyata, ini meningkatkan prestasi sistem. Dapatan kajian eksperimen menunjukkan metrik kestabilan lebih baik: bagi kestabilan sudut gulungan, masa kenaikan sebanyak 0.4682 saat, masa kestabilan 1.3819 saat, lajakan 0.298% dan Ralat Keadaan Mantap (SSE) 0.133 darjah; bagi kestabilan sudut pic, masa kenaikan 0.2309 saat, masa penetapan 0.7091 saat, lajakan 0.1224%, dan Ralat Keadaan Mantap (SSE) 0.0239 darjah. Pendekatan LQR-NN berkesan mengurangkan lajakan berbanding kawalan tradisi Pengatur Kuadratik Linear (LQR), dengan itu mengurangkan ayunan. Tambahan, LQR-NN dapat mengurangkan Ralat Keadaan Mantap (SSE), sebanyak 0.074 darjah bagi gerakan putaran guling dan 0.035 darjah bagi gerakan putaran anggul.

Keywords

• UAV
• Flying Wing
• Optimal Control
• Adaptive