Toward Explainable AutoEncoder-Based Diagnosis of Dynamical Systems

Autoencoders have been used widely for diagnosing devices, example, faults in rotating machinery. However, autoencoder-based approaches lack explainability their results and can be hard to tune. In this article, we propose an explainable method applying autoencoders diagnosis, where use a metric that maximizes the diagnostics accuracy. Since autoencoder projects input into reduced subspace (the code), define theoretically well-understood approach, principal angle, over possible fault labels. We show how approach both single-device (e.g., machinery) complex (multi-device) dynamical systems. empirically validate theoretical claims using multiple architectures.