A hybrid CNN-LSTM model for joint optimization of production and imperfect predictive maintenance planning

• Developing a dynamic model that integrates production and PdM planning. • Implementing a RHP approach to address the dynamic nature of production systems. • Introducing constraints to the model that account for imperfect maintenance actions. • Adopting a novel hybrid DL method that combines CNN and LSTM networks. • Validating the performance of the proposed model based on the NASA data set. This paper deals with the problem of dynamically integrating tactical production planning and predictive maintenance in the context of a rolling horizon approach. At the production level, a set of items need to be produced in lots over a finite planning horizon. It is assumed that the system is in as-good-as-new condition at the beginning, and then it is degraded over time because of operating. The system operating state is predicted by a data-driven predictive maintenance approach. The system can be maintained at the beginning of each period. We introduce a novel hybrid deep learning method based on a combination of a convolutional neural network (CNN) and long short-term memory (LSTM) to improve the prediction accuracy of the remaining useful life. The CNN-LSTM method is used to determine the optimal maintenance action based on the data collected by sensors. A maintenance action is assumed to be perfect or imperfect. Imperfect maintenance places the manufacturing system in an operating state that lies between 'as-bad-as-old' and 'as-good-as-new'. A benchmarking dataset is used to validate the proposed integrated production and predictive maintenance planning approach. Comparison results highlight the advantages of the proposed framework in reducing the total production and maintenance cost.