Comparative Analysis of Supervised, Unsupervised and Hybrid Models for Time Series Segmentation in Sustainable Manufacturing

Machine learning techniques can help in meaningful segmenting of data acquired from manufacturing processes and thereby aid in achieving better process knowledge and control towards sustainability. In this work, supervised, unsupervised, and hybrid machine learning models are developed for segmentation of time series cutting force data in sustainable manufacturing and a comparative analysis of the models are performed. Drilling operations are performed to capture cutting force data at different conditions. Supervised learning model based a Random Forest classifier gave high overall accuracy of 89% but underperformed for the retraction class which has lesser data points. Unsupervised learning model based on K-Means clustering gave an improved overall efficiency of 91% but gave inaccurate identifications of the tool-entry class. Hybrid model adopted a novel approach combining unsupervised and supervised models by incorporating the cluster prediction of unsupervised model as a feature of supervised learning model. The hybrid model gave a high overall accuracy of 89% while ensuring good class wise accuracy for each class.