Condition-based maintenance program on lithium-ion batteries using artificial intelligence for aeronautical operations management

On 2013, all Boeing 787 were grounded due to events of deflagration in lithium-batteries installed in these aircraft, it generated subsequently changes in the flight itinerary, dissatisfaction in customers and expenses in maintenance costs in many companies around the world, losing about $22,000 per hour. For this reason, condition-based maintenance program was performed using State of Health and Remaining Useful Life indicator. A new technique Machine Learning was used for solves regression problems in non-parametric data, called Gaussian Processes, this emerging algorithm of Artificial Intelligence generates predictive models based on previous knowledge, giving a probability distribution that follows the current state, allowing interpret the reliability of the component in different cycles of useful life. The paper used the dataset from the NASA repository, due to it has the same internal composition and is tested run to failure. Kernel mixed Matern1.5 + Matern2.5 got good results versus other mixtures during the different test, mapping the real behavior of the battery. The health status diagnostic was quantitatively evaluated and it got results of 98.34% and 1.13% in R2 and in RMSE respectively, likewise the model served to forecast the remaining useful life of the battery, predicting 64 cycles with a minimum error of 1.53% in reference to the real data. Finally, it helped development a condition-based predictive maintenance program that generated a return on investment (ROI) of 173% and a profit of $331,360 during the first year.