Battery Reliability Prediction using Neural Network Methods

The weak link in a standby battery pack emergency system is the battery's characteristic of wearing out, even when not in use. Battery failure can result in more than downtime in a flight situation: it can imperil the mission. The common method to determine battery health was to perform load test, time-consuming physical inspection, and maintenance logs examination. Although reliable, this method often is cumbersome, time consuming, expensive, and risky. The development of a sensor that will calculate in real time the short-term usability (amp-hour rating) as well as the long-term end-of-life reliability of the battery will greatly reduce maintenance costs and can reduce the possibility of a failure in a life-critical situation. The calculation of end-of-life prediction will require advancements in neural network software to linearize a highly non-linear function. Aurora proposes to develop a battery sensor that will calculate reliability in real time. The objective is to apply neural network prediction software in development and demonstrate an operational sensor. Using information about the battery's operational condition the sensor will calculate the battery's state of charge, current amp-hour rating (stationary hangar conditions), project amp-hour rating (under flight conditions), and predict the time of the battery's end of life.