Controlling Energy Consumption In Production Systems by Utilizing the Digital Twin

Given the environmental impact of industrial energy consumption, there is increasing pressure on manufacturers to decrease the energy consumption in their manufacturing plants. Recent studies have focused on this issue by considering controlling the energy mode of energy intensive resources in manufacturing. Namely, a lower energy consumption can be achieved by putting the machines in idle mode or turning them off when there is few items to process.

Controlling the energy mode of machines can be achieved optimally through threshold-type policies. These policies do not use the actual or predicted arrival times of items to the machine but use only aggregate data. The optimality of threshold-based policies partly depends on having memory-less inter-arrival time distributions. However, this assumption does not hold in all manufacturing settings. Moreover, digital twins create an opportunity for controlling the system with more information. Namely, the digital twin can use the current location of the items in the production system to simulate the future behavior of the system and give predictions for arrival times.

This work considers the problem of an automotive manufacturer that aims to decrease the energy consumption in its paint oven using a digital twin implementation. The paint oven is responsible for a considerable portion of energy consumption in the plant.

The location of the vehicles in the upstream can be used for generating arrival time scenarios and these scenarios can be used for scheduling the oven operations. As the digital twin is not able to predict the arrival times to the oven exactly, we develop a scenario-based mathematical programming approach to solve the problem.

Given the complexity of automotive manufacturing processes, many scenarios are typically needed to reflect the stochasticity in the system. However, solving the problem with many scenarios is not possible with commercial solvers currently. To remedy this, we establish multiple properties for the optimal solution to derive valid inequalities and propose a heuristic algorithm based on the genetic algorithm to solve the problem.

Our numerical results indicate that incorporating the digital twin is very effective in improving the performance of the system by allowing for decreasing the energy consumption by up to 30\%. Additionally, our numerical results indicate that the valid inequalities are very effective for speeding up the process of finding the optimal solution. Our numerical results also show that for large problems, the heuristic method is able to identify close-to-optimal solutions very fast.