Data-driven and safety-aware holistic production planning

Equipment degradation can lead to reduced mechanical integrity and a lower level of process safety. It is desired to improve system safety and increase productivity through decision planning. However due to system complexity and failure non-linearity, quantifying how different process decisions and equipment conditions affect system safety and performance is challenging which renders production and maintenance planning highly non-trivial. This research provides a Safety-Aware Sustainable Maintenance and Process Optimization (SASUMAPRO) paradigm to help improve system performance. It demonstrates the incorporation of an artificial neural network failure prediction model into a mixed-integer nonlinear programming production and maintenance planning model. It simultaneously maximizes the expected productivity of the plans and minimizes their power consumption while satisfying safety constraints. The methodology is illustrated with a biodiesel production process and results show that the neural network was able to predict equipment mechanical failure with an accuracy of 82.7%. Furthermore, it was seen that the planning model that incorporated the equipment failure model was able to recommend a sequence of decisions that increased expected overall productivity by 31.7% relative to a commonly adopted industrial approach. SASUMAPRO is process-agnostic and can be used to obtain multiple solutions to help improve system performance.