Machine learning for predictive scheduling and resource allocation in large scale manufacturing systems

Abstract

Listen

Translate article

The digitalization processes in manufacturing enterprises and the integration of increasingly smart shop floor devices and software control systems caused an explosion in the data points available in Manufacturing Execution Systems. The degree in which enterprises can capture value from big data processing and extract useful insights represents a differentiating factor in developing controls that optimize production and protect resources. Machine learning and Big Data technologies have gained increased traction being adopted in some critical areas of planning and control. Cloud manufacturing allows using these technologies in real time, lowering the cost of implementing and deployment. In this context, the paper offers a machine learning approach for reality awareness and optimization in cloud.Specifically, the paper focuses on predictive production planning (operation scheduling, resource allocation) and predictive maintenance. The main contribution of this research consists in developing a hybrid control solution that uses Big Data techniques and machine learning algorithms to process in real time information streams in large scale manufacturing systems, focusing on energy consumptions that are aggregated at various layers. The control architecture is distributed at the edge of the shop floor for data collecting and format transformation, and then centralized at the cloud computing platform for data aggregation, machine learning and intelligent decisions. The information is aggregated in logical streams and consolidated based on relevant metadata; a neural network is trained and used to determine possible anomalies or variations relative to the normal patterns of energy consumption at each layer. This novel approach allows for accurate forecasting of energy consumption patterns during production by using Long Short-term Memory neural networks and deep learning in real time to re-assign resources (for batch cost optimization) and detect anomalies (for robustness) based on predicted energy data.