以多層感知類神經網路輔助複合式節流器設計

節流器為液靜壓軸承的重要關鍵模組，複合式節流器是由自補償式節流器與不同深度之溝槽式節流器搭配而成，本文藉由探討複合式節流器工作原理，建立其物理模型，並推導出相關統御方程式，接著進行數值模擬來探討複合式節流器在液靜壓軸承系統中，不同節流比時的剛性表現，同時利用單向墊液靜壓實驗平台進行理論及模擬驗證。然而，複合式節流器之數學模型為非線性，在建立其物理模型時又做了許多假設條件，再加上實驗過程中，來自於環境干擾的誤差因素，造成理論模擬無法與實驗結果非常吻合，使工程師未來在快速選擇出適合的結構參數以達軸承最佳剛性時會面臨相當大的難題。本論文使用的多層感知器(MLP)模型為人工智慧的一個分支，具有可以接受多輸入、多輸出參數、並能處理非線性問題等特性。本文以實驗所取得之數據做為訓練資料，建立一MLP網路模型，該模型會比理論模擬更接近實際狀況，在預測液靜壓軸承系統搭配不同節流器時，隨著供油壓力、負載、節流器結構參數等數值變化時其他相對應參數可能的改變，並依此結果輔助節流器設計時理論模擬的不足，達到最佳化設計的目標。

關鍵字

液靜壓軸承；複合式節流器；人工智慧；多層感知器

並列摘要

Hydrostatic bearings are widely used in precision machine tools because of their superior features of high stiffness, high load-carrying capacity and long life. To obtain the high stiffness characteristic a hybrid-type flow restrictor, which is composed of a groove restrictor in series with a self-compensation one, is usually employed in hydrostatic bearings. However, engineers always face challenges in quickly obtaining a proper design of the hybrid-type restrictor because of too many design parameters involved. This paper mainly employs multilayer perceptron (MLP) to improve the stiffness of the hydrostatic bearing to a level of near infinite by optimizing the design parameters of the hybrid-type restrictor, such as the groove depth, preload and stiffness of springs. The equations governing the relationship among bearing stiffness, preload of hybrid-type restrictor, and various key parameters are first derived. The experimental rig is constructed comprising a hydrostatic slide with a single pad as well as an opposed pad together with several hybrid-type flow restrictors. Pressure, flow rate, load, and oil-film thickness are simultaneously measured. The numerical simulation of the hybrid-type flow restrictor based on the derived equations is performed and compared with experimental results. The MLP model, a branch of artificial intelligence (AI), consists of at least three layers, an input, an output, and a hidden layer. According to the collected data, the MLP model is trained effectively. By comparing the simulation with the MLP model predicted results, the MLP model can optimize the design parameters of flow restrictors to improve the stiffness of hydrostatic bearings.

並列關鍵字

hydrostatic bearings ； hybrid flow restrictor ； AI ； multilayer ； perceptron

參考文獻

[1] Girard, L. D., 1862, Application des Surfaces Glissantes, Bachelier, Paris.

Google Scholar

[2] Hirn, G. A., 1854, "Sur les principaux phenomenes que present les frottements mediats et sur les diverses manieres de determiner la valeur mecanique des matieres employees au graissage des machines," Bull. Soc. ind. de Mulhouse, 1954, Vol. XXVI, pp. 188-277.