Detecting machine mode can allow smarter process monitoring systems and more accurate fault prediction without external information. A remote machine monitoring system was installed on a cold heading machine in the factory of an automotive fastener manufacturing company. The process monitoring system was non-intrusive and was designed to measure vibration. The end goal of the study was to predict tool wear, but part classification was required first, as the machine produced multiple parts which produced different vibration signals. The collected vibration data was processed using wavelet transform and passed through a convolutional neural network for part classification. This method achieved part classification accuracy as high as 86% when looking at data for a 1-month period. The results show that meaningful classification features are present in the data using the process monitoring system as designed.