本研究主要研究資料探勘運用於橡膠配方預測的可行性,以不同摻合比例及不同原材料的混合,透過混煉、加促及硫化工程,對其成品以分析儀器來測得成品的物理性能、硫化曲線及實體磨耗程度。將這些數據整理,取出較為重要的值,再以資料探勘去預測當配方橡膠比例及橡膠種類改變時,分析其物理性能、硫化曲線及實體磨耗的相關性。以協助工程師在工程進行前可先行預測,減少試作次數與縮短開發時間。 將預測的配方比例,以及膠料種類改變後的配方,進行試作;以同樣的混煉、加促條件,進行成品的製作;同時再以類神經網路進行模擬預測,藉預測模擬值對照驗證配方。實驗結果發現,硫化曲線、DMA差異不大,驗證結果與預測模擬結果非常相近。另外,因橡膠原材料、配合劑及促進劑取得不易,樣品有限,無法試做更多的組合,如改變硫磺含量或促進劑、配合劑含量,只能侷限於橡膠配比與橡膠種類改變。
This study tries to study the feasibility of applying data mining on prediction of PHR (parts per hundreds of rubber) testing. We can obtain the physical properties, vulcanization curve, and aberration after performing the laboratory with different blends, mix of various materials, breakdown and vulcanization process. Consequently, the relationships between the variations of PHR and the experimental data can be analyzed. This can help the engineers to reduce the trial periods. We verify the results of neural networks by performing the trials with corresponding conditions. Experiments show that the vulcanization curve, DMA are very close to the predicted values. Moreover, it is difficult to get the rubber raw materials, rubber ingredient and cure accelerator, therefore, we can only focus on the trials of PHR rather than the different ratio of cure accelerator.