A novel deep learning approach for event detection in smart manufacturing

In a smart factory, the ability to detect time series events associated with particular conditions of equipment such as peaks, changeovers and failures is an important task that supports process monitoring and drives optimal performance. This task can be formulated as a Machine Learning time-series classification problem (TSC), requiring algorithms that combine good predictive performance and fast training time. In this paper, we propose a novel approach that uses Deep Learning models to predict changeover events from large data streams collected from a metal packaging manufacturing plant. The specific model architecture comprises deep Convolutional Neural Network (CNN) and Bidirectional Long Short-Term Memory (CNN-LSTM) stacked autoencoders. This architecture combines the advantage power of CNNs in automatic feature extraction and the adept sequential learning ability of LSTMs. We empirically evaluate the performance of our proposed model in time-series classification using historical real-world machine speed data. The findings from our experiments demonstrate the applicability of Deep Learning to Smart Manufacturing and support the potential of the proposed approach when compared to state-of-the-art event detection classifiers, especially in the significantly reduced model training time.