The dynamic nature of manufacturing production environments, along with numerous machines, their unique activity states, and mutual interactions render challenges to energy monitoring at a machine level. To this aim, a machine learning framework is presented, to predict the machine-specific load profiles via energy disaggregation, and these machine-specific load profiles are in turn used to predict the machineâ€™s activity state as well as their respective production capacities. Various supervised machine learning algorithms such as GBDT, XGBoost, LightGBM, LSTM and BLSTM were evaluated on their capacities to predict load profiles and production capacities of four machines investigated in this study. LightGBM and EnBLSTM were identified as the respective best performing algorithms with an average MAE and RMSE of 0.035 and 0.105 for disaggregation studies and 1.64 and 11.41 for production capacity estimation. Four unsupervised machine learning algorithms, namely K-means, minibatch K-means, HMM and GMM were evaluated to cluster the machines activity states from their disaggregated load, where the GMM algorithm had a superior performance with the V score and Fowlkes-Mallows index of 0.85 and 0.98, respectively. The framework and methodology developed in this study are purely data-driven, cross-deployable and serve as promising catalyst to foster smart energy management practices and sustainable productions in the manufacturing industry.
Keywords smart energy management, energy disaggregation, smart manufacturing, machine learning, big data, data-driven analytics