壓力閥供壓之閉式液靜壓軸承類神經網路膜厚控制

Neural Network Controlling Supply Pressure by Pressure Control Valves for a Close-type Hydrostatic Bearing

修宇志(Yu-chi Shiou)

指導教授 ： 康淵

中原大學/工學院/機械工程研究所/碩士(2019年)

https://doi.org/10.6840/cycu201900767

摘要

與其他類型軸承之工作原理不相同，以靜壓軸承壓力補償調節油腔壓力來支撐著負載變化或不斷移動工作平台，在等流量補償之油腔壓力之液靜壓軸承中，有承載、剛度皆大於等節流之優點。本文涉及對液靜壓軸承在暫態維持等油膜厚度的類神經網路壓力控制，並能以感測器數值與類神經網路來對油膜厚度縮減在負荷變化時傳統控制的暫態收斂過程與振盪幅度，令工作台可在隨時保持穩定且不會發生過大的位移變化。
本文目標為閉式液靜壓軸承工作台之壓力閥閥控節流器之供壓穩態時保持等膜厚以及油膜厚度在負載變化之暫態抑制膜厚的振盪，穩態膜厚精度由節流器與軸承之流量連續方程式控制，暫態響應由類神經網路學習架構建立油膜動態模型之PID控制，及減少膜厚隨負荷改變到油膜平穩的暫態時間。本文以閉式液靜壓軸承工作台上的供油壓力、工作台壓力以及位移量訊號回饋給類神經網路，運用工作壓差與膜厚差穩態公式得供壓變化作為補償目標值，並以比例壓力閥補償液靜壓工作台來建立控制迴路，以達到等膜厚之目標值。
參考理論的數學模型，修改並運用在等膜厚壓力補償系統中，再用類神經網路訓練輸入輸出之關係，修正PID控制器系統參數使其隨網路輸出和位移差變動，令暫態快速收斂、穩態保持精準且穩定的補償系統。透過實驗暫態變化找出最佳的類神經網路學習速率與控制器參數修正關係以。本研究的等膜厚類神經能保持穩態次微米精度並能在動態改變補償參數縮減暫態不穩定之膜厚變化，且通過數值模擬和實驗測量驗證此演算法。

關鍵字

液靜壓軸承 ； 類神經網路 ； 膜厚控制 ； 等膜厚

並列摘要

As studies show, restriction or constant flow compensation have better load capacity, higher stiffness than constant restriction in hydrostatic bearings. This study involes Neural Network of pressure control to maintain constant oil film thickness in temporary state, and the ablility using sensors’ data and Neural Network to reduce temporal-state convergence time and the vibration amplitude, to make worktable has the ability to sustain stable without severe displacement.
This study’s goal is to suppress the oscillation of a close-type hydrostatic bearing worktable’s oil film in temporal-state, and discuss the effect of the Neural Network’s learning structure to pressure valve’s supply pressure control, and the reduced time from the variates of load to oil film stabilized. In this paper, in order to reach constant film thickness. Supply pressure, worktable pressure and displacement signals will feedback to Neural Network through sensors, using steady-state function with pressure and displacement differences to obtain compensation target, worktable is compensated by proportion pressure valves to build up a control loop.
Referencing theoretical mathematical model, modified and applied in the constant film pressure compensation system then through Neural Network training the connection between input and output, adjust PID’s control system parameter along with displacement deviation so that compensation system can quickly converged in temporal-state and precise in steady-state. The fitting learning rate of Neural Network and parameter adjustment relatives. Both the steady-state accuracy in few sub-micrometers and temporal-state reduction can be achieved by Neueal Network dynamically adjust PID controller’s parameters using the algorithm of this study, which have been verified by numerical simulation and measurement in experiments.

並列關鍵字

Hydrostatic bearing ； Neural network ； film control ； constant thickness

參考文獻

參考文獻

[1] Rippel H C 1964 Cast Bronze Hydrostatic Bearing Design Manual (Cleveland Ohio: Cast Bronze Bearing Institute).

[2] Bassani R and Piccigallo B 1992 Hydrostatic Lubrication (Amsterdam: Elsevier Science).

[3] Bassani R 2001 Hydrostatic systems supplied through flow dividers, Tribol. Int. 34(1) 25-38.

[4] Rowe W B 2012 Hydrostatic, Aerostatic and Hybrid Bearing Design (London: Butterworth-Heinema類神經網路) pp 83-113.