由於伺服閥控制油壓缸系統為高階非線性,且系統參數常隨使用環境﹑負
荷及干擾的影響而具時變性,欲推導其正確系統數學模式以設計控制器是
相當費時的,由於模糊控制在設計控制器時並不需要系統的數學模式 ,因
此對於具有複雜之非線性數學模式的油壓系統而言,將可節省模式判別所
需的時間。然而一般模糊控制器之控制法則來自專家或操作者的知識與經
驗,所缺乏的就是沒有能力將這些片段知識整合起來,於是本文加入了類神
經網路的學習能力,可對規則不清楚的複雜系統,依其輸入﹑輸出關係進行
適當的規劃,自動形成規則集和相對應隸屬關係;針對伺服閥控制油壓缸位
置而言,吾人以鐘型隸屬函數(Belled-Shape membership function)及
Mamdani 推論法則為基礎,建立類神經模糊控制器網路架構 ,並混合了非
監督及監督學習法(混合式學習法),比起原先只用逆傳遞學習法(
backpropagation),其學習收斂速度快的很多,亦能根據訓練資料的分佈,
自動形成規則集及刪除不必要的規則,進而找出最佳之隸屬函數 ,故節省
了傳統模糊控制器必須花費在推論規則之選取及搭配之時間,並將結果應
用至油壓缸位置控制實驗上,然後給予不同的負荷情況,經由實驗結果顯
示,此類神經模糊控制器具有很好的適應能力。

The electro-hydraulic servocylinder control system is a highly
nonlinear system, which system parameters are time- varying due
to the friction-force,the load , or the unknown disturbances.
It is usually time-consuming to formulate and decide an
accurate system mathematical model for the controller
design. It is unnecessary to derive and bulid up the system
mathematical model for the fuzzy controller design , but the
fuzzy logic rules are obtained from the knowledges and
experience of experts and operators and they lack the
abilities to integrate the segmental knowledges.
Therefore, in this paper we apply the learning ability of
neural network to the fuzzy control system. Such neuro-fuzzy
controller can be constructed from training data to find
the optimal fuzzy control rules and the relative
membership function. In the study , we use the belled
shape membership functions and the type of the Mamdani
inference to construct a neuro-fuzzy controller network
structure , and combine both unsupervised and supervised
learning schemes (hybrid learning algorithm),the learning
speed converges much faster than the original back-
propagation learning algorithm.Through the neuro-fuzzy
controller design process , one can get the fuzzy control
rules , perform the rules elimination and find the optimal
membership functions. So the neuro-fuzzy controller network
structure avoids the rule-matching time of the inference
engine in the traditional fuzzy logic system. Applying the
result and implemented in a microprocessor based controller for
the experiment of the electro-hydraulic servocylinder
position control. The experiment results indicate that the
neuro-fuzzy control technique has good control performance.