3-D FeatScape: Feature Landscape for Fuel Cell Diagnosis

With the advancement of automotive fuel cell (FC), there is a growing demand for longer service life and higher dynamic performance. However, in various complex transient decay conditions, it is prone to issues such as insufficient gas supply and fluctuating water content, which can lead to failures. Therefore, a rapid diagnosis method is necessary. Current research has made certain achievements in the accuracy and computational time of fault diagnosis using electrochemical impedance spectroscopy, but it still retains a significant amount of redundant information, preventing further reduction in computation time. Additionally, prevailing method is limited to the reading of one-dimensional (1-D) signal or the recognition of 2-D images, lacking exploration of high-dimensional feature spaces. From the traditional perspective, the amount of information contained in the features is quite limited. Hence, this paper proposes the 3-D feature landscape (featscape) to enrich the information quantity in FC fault diagnosis. By employing a novel ultra-fast impedance measurement method with emphasizes-response subdues-disturbing (ERSD), the frequency spectrum is obtained. The 2-D scale-invariant features are then reconstructed into a 3-D featscape, followed by featscape projection to create a unique high-information-density state barcode. Then a barcode reader that integrates virtual photoelectric recognition and machine learning algorithm is designed to achieve rapid and accurate status tracking for flooding, drying, hydrogen starvation, and oxygen starvation. The total time from front-end data acquisition to the final state identification is 1.9 seconds, with an identification accuracy of 99.6%. This demonstrates the efficiency and reliability of the method in fault diagnosis for automotive FC.