A coevolutionary optimization approach with deep sparse autoencoder for the extraction of equipment degradation indicators

We present a coevolutionary optimization approach for the automatic and unsupervised extraction of industrial component degradation indicators from a set of signals collected during operation. It embeds a deep sparse autoencoder (SAE) for the extraction of the degradation indicators, into a multi-objective coevolutionary optimization algorithm, which maximizes the SAE's performance by optimizing its architecture and hyperparameters. The effectiveness of the proposed approach is shown by its application to a synthetic dataset, which mimics the operation of a degrading component in an environment affected by seasonal changes.