隨著市場需求轉向少量多樣、生命週期短的趨勢,而產品在加工過程對於品質的要求極為嚴格。由於刀具的磨損會導致工件表面越顯粗糙,而過度磨損的刀具會使刀具超出刀具壽命而使表面粗糙度超出公差界線。為了能得到準確的刀具壽命,許多學者希望利用預測的方式,一方面以監控製程的變異,另一方面則能有效降低其額外成本。而在不同加工環境及設置下,有許多不可控因素是人為無法掌握的。因此,感測技術的發展得以監控製程中所產生的變異,進而分析製程因外在因素所造成的影響,不僅能夠有效降低不良率,更能有效提升預測時的準確度。 本研究將進行預測模型的建構,運用支持向量機快速判別刀具並預測刀具壽命。傳統的預測模型需要大量數據才有辦法進行建模,本研究將投入少筆數建模且達到準確的預測效果。 為了驗證本研究所提出的方法有其準確性及可行性,實驗設置兩組不同加工參數,投入預測系統少量的數據,進行刀具加工筆數的預測。SVM刀具判別模型實驗組結果預測準確率為97.78%及SVM刀具判別模型驗證組結果預測準確率為98.11%,得以證實本研究提出之預測系統的準確性及可行性。
As the market demand shifts to a small number and variety, and the life cycle is short, and the quality requirements of the products in the process are extremely strict. The wear of the tool will cause the workpiece surface to become rougher, and the excessively worn tool will cause the tool to exceed the tool life and to make the surface roughness beyond the tolerance limit. In order to obtain an accurate tool life. Many scholars hope to use predictive methods to monitor the variation of the process on the one hand and to reduce the additional cost on the other hand. In the different processing environments and settings, there are many uncontrollable factors that are beyond human control. Therefore, the development of sensing technology can monitor the variation generated in the process, and then analyze the impact of the process due to external factors, not only can effectively reduce the bad rate, but also can improve the accuracy of the prediction. This study will construct the predictive models. Use support vector machines to quickly identify tools and predict tool life. The traditional predictive models require large amounts of data to modeling, and this study will invest in a small number of models and achieve Accurate prediction results. In order to verify the accuracy and feasibility of the proposed method, two sets of different processing parameters were set up in the experiment, and a small amount of data was put into the prediction system to predict the number of tool processing. The prediction accuracy of the SVM tool discriminant model experimental group is 97.78% and the SVM tool discriminant model verification group results the prediction accuracy rate of 98.11%, which confirms the accuracy and feasibility of the prediction system proposed in this study.