- 學位論文
探討砂輪黏屑現象與其監控技術之研究
Study of Grinding Wheel Loading Phenomenon and its Monitoring Technology
摘要
磨削屬於超精密加工中重要的加工方式，在於追求優異的表面品質。實際加工中，影響磨削件表面粗糙度的因素除了加工參數之外，砂輪的表面狀態尤其重要。隨著砂輪的使用，黏屑與自銳相互發生，造就砂輪表面形貌的變化，黏屑程度反應磨粒的切削能力，進而影響工件表面粗糙度。當嚴重黏屑造成堵塞時，易伴隨著工件燒傷的發生，且黏屑程度可視為砂輪修整時機的判斷，故黏屑比的監控在磨削加工中有其必要性。本研究探討參數、訊號與加工結果三者之間的關係，建立砂輪黏屑比監控與工件表面粗糙度控制。 實驗結果顯示，對工件表面粗糙度與砂輪黏屑比而言，皆與工件線速度、磨削深度呈一正相關；與砂輪線速度呈一負相關。並將砂輪線速度與工件線速度取比值，顯示在本研究加工參數組合下，兩加工結果與線速度比值呈一負相關。 本研究嘗試透過振動與聲射兩種訊號來監控黏屑比。實驗結果顯示振動訊號經由移動平均法；聲射訊號經由正規化，而後皆取其方均根值，兩種訊號與黏屑比普遍有0.75以上的相關係數，正相關性質高。而因訊號產生的機制不同，適用的加工條件則有差異。實驗結果統整出，振動訊號適合用於加工參數組合下磨削力較大的條件，越能以振動強度反應黏屑比的變化趨勢；聲射訊號則適用在磨削深度淺的加工條件，表面晶格、差排些微的變化越能由聲射的應力波表現。 並透過機器學習建立類神經網路模型，以加工訊號為輸入，預測砂輪黏屑比；再以黏屑比為輸入，預測工件表面粗糙度。實驗結果顯示模型之平均絕對百分比誤差落在4.49~8.91%，並以此應用於加工結果的預測。當訊號預測到砂輪嚴重黏屑而危害到工件表面品質時，本研究使用降低砂輪轉速的手法改善，實驗結果顯示降低砂輪轉速至原轉速度的90~70%，能有效降低磨削比而使黏屑比大幅下降，恢復砂輪切削能力與工件表面粗糙度。使本研究達到探討參數對加工結果、黏屑比監控、表面粗糙度控制與延長砂輪使用時間等目的。
並列摘要
Wheel loading and self-sharpening occur interactively during the grinding process. Loading will affect the grinding ability of the wheel, and influence the surface quality of the workpiece, but also can be regarded as the judgment time of dressing. Hence, the monitoring of the wheel loading situation is important and has its necessity. In this study, grinding parameters, processing signals and grinding results will be discussed, to establish the monitoring of wheel loading and workpiece surface roughness control. The experimental result shows that table speed and grinding depth are proportional to the wheel loading and surface roughness; wheel speed is inversely proportional to them. Then evaluate the ratio of wheel speed to table speed, shows that both processing results are inversely proportional to this speed ratio. Acceleration and AE signals are applied in this study. The experimental result shows that signal processing by moving average in the vibration signal; normalization in AE signal, and then takes the RMS value. Both generally have a correlation coefficient greater than 0.75 with the wheel loading. Because of the different mechanisms of signal, vibration is suitable for monitoring in the conditions of higher grinding force situations, the trend of loading can be response effective by the vibration strength. AE is suitable for shallow grinding depth and lower grinding force situation, the slight changes in the surface lattice and dislocation can reflect strongly by the AE intensity. Then establish a neural network by machine learning to predict the wheel loading by the signals, and predict the surface roughness by the wheel loading. The experimental result shows that the MAPE value generally in the range of 4.49~8.91%, showing that the neural network has a highly accurate prediction ability. When the loading is too high to endanger the surface quality of the workpieces, in this study adopt the method of reducing the wheel speed. The experimental result shows that by adjusting the wheel speed to 90~70% of the original rotation speed can effectively reduce the grinding ratio, to make the wheel tend to shed. The loading area and glazing grains will detach the wheel surface, and then the wheel can restore its grinding ability, to optimize and control the surface roughness of the workpieces.