以機器學習輔助應用於液靜壓雙向墊系統之自補償式節流器設計

Design a Self-compensating Restrictor for Hydrostatic Opposed-Pad System by Machine Learning

鄭兆鈞(Chao-Chun Cheng)

指導教授 ： 宋震國

國立清華大學/工學院/動力機械工程學系/碩士(2021年)

摘要

節流器為液靜壓軸承系統中重要的關鍵零組件，其構型會影響整個液靜壓系統的剛性及承載能力。本研究著重於自補償式節流器應用於雙向墊液靜壓系統之情況，於理論方面，本文藉由自補償式節流器的運作原理與雙向墊液靜壓軸承的基本理論，推導出相關的物理理論模型，並進一步利用雙向墊液靜壓實驗平台，量測不同負載下其壓力、流量及油膜厚度之變化，進行理論與模擬之驗證。然而，雙向墊系統的理論模型為非線性，在進行理論模擬時需做許多假設條件，再加上實驗過程中誤差的影響，使得模擬結果與實際情況有所偏差。
為解決此一問題，本研究結合物理模型與機器學習中的多層感知器模型，建立一混合模型(Hybrid model)，並以實驗數據訓練該模型，透過深度學習讓該模型學習物理理論模型未包含的訊息以修正物理模型之誤差，讓預測結果更貼近實際狀況；此外由於結合物理模型之緣故，此一模型相較於單獨使用MLP模型，具有較佳的外推能力，故在探討規格品外之設計參數(如彈簧剛性或溝槽流阻)時，能夠提高不同設計參數下預測液靜壓軸承剛性表現之準確性，最終達到輔助最佳化節流器設計之目標。

關鍵字

液靜壓軸承 ； 自補償式節流器 ； 機器學習 ； 多層感知器

並列摘要

The flow restrictor is a key component in the hydrostatic bearing, which configuration will affect the stiffness and load-carrying capacity of the hydrostatic bearing. Thus, the goal of this paper is to improve the stiffness of the hydrostatic opposed-pad system to a level of near infinite by optimizing the design parameters of the self-compensating restrictor. However, the design of a self-compensating restrictor is always a challenging subject because the theoretical model has been greatly simplified and the equations are coupled and non-linear.
This thesis proposes a hybrid model, which combines the theoretical model and multi-layer perceptron (MLP) model, to predict the stiffness of the hydrostatic opposed-pad system. Since the hybrid model contains the MLP model and theoretical model, it not only can deal with non-linear problems but also has good extrapolation ability. The training data are collected from a hydrostatic opposed-pad system experiment, which measures pressure, load, flow rate, and oil-film thickness. Based on training data, the hybrid model predicted results are closer to the actual situation.
By comparing the MLP model with the hybrid model predicted results, the hybrid model has better accuracy of the prediction results and extrapolation ability, so it can optimize the design parameters of self-compensating restrictor to improve the stiffness of hydrostatic bearings.

並列關鍵字

hydrostatic bearings ； self-compensating restrictor ； machine learning ； multilayer perceptron

參考文獻

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

[2] Slocum, A. H., 1992, Precision Machine Design, NJ, Englewood Cliffs:Prentice Hall

[3] 黃華志，2009， “液靜壓軸承技術清大演講”，工研院機械所.

[4] Telingater, V. S., 1972, “Hydrostatic Slideways Using Standard Bearings ”, Machines and Tooling, Vol. 43, Iss. 2, pp.15-20

[5] Raimondi, A. A., Boyd, J.,1957, “An Analysis of Orifice and Capillary-compensated Hydrostatic Journal Bearing ” , Lubr. Eng, Vol. 13, Iss. 1, pp.28-37