衝床的工作原理是利用模具對材料施加壓力，讓材料依據模具的形狀產生型變，達到加工製造的目的，然而隨著使用次數的增加，模具會逐漸損耗造成加工品的良率降低，為了避免模具的損耗影響到加工品良率，工廠會針對衝壓出來的加工品做隨機抽樣檢查，藉由量測加工品的尺寸外觀是否在標準差之內來判斷模具的損耗情況，由於人工抽樣費時費人力，且會因為量測人員不同而有誤差，因此本論文提出利用以下方法來改善這點。 當衝床在運作時，機台本身會產生大量的振動訊號，振動訊號會根據機台的狀況、生產情形、生產環境等因素而有所不同，本論文嘗試利用多顆加速度感測器來收集這些訊號，並以Google 的開源API–Teachable Machine來訓練所收集到的資料，學習的結果與工廠實際量測到的數值做迴歸分析比較，比較結果顯示，振動訊號會隨著模具使用次數的增加而產生變化，並且此變化是具有趨示性的，且該趨勢性與工廠量測結果相符合。 ;The working principle of the punching machine is to use the mold to apply pres-sure to the material, so that the material can be deformed according to the shape of the mold, and the purpose of processing and manufacturing is achieved. However, as the number of uses increases, the mold will gradually wear out and the yield of the pro-cessed product will decrease. In order to avoid the loss of the mold and affect the yield of the processed product, the factory will perform random sampling inspection on the stamped processed product by measuring processing. Whether the size of the product is within the standard deviation to judge the loss of the mold. When the punching machine is in operation, the machine itself will generate a large number of vibration signals, and the vibration signal will vary according to the condition of the machine, the production situation, the production environment and oth-er factors. This paper attempts to collect these signals using multiple accelerometers and train the data with Google′s open source API, the Teachable Machine. The results of the study are compared with the actual measured values of the fac-tory for regression analysis. The comparison results show that the vibration signal will change as the number of times the mold is used increases, and this change is directional, and this trend is consistent with factory measurements.