Fault Diagnosis of Complex System via Quantification of Complexity With Lossless Compression

Data-driven fault diagnosis methods for complex systems, based on machine learning and deep learning, have shown competitive performance. However, these parameterized approaches face two main challenges: 1) the need for large amounts of labeled data, and 2) sensitivity to changes in data distribution. To address both issues simultaneously, a novel data-driven fault diagnosis method is proposed for the first time, which is unconstrained by the change of data distribution and the amount of labeled data. Specifically, the health state of complex systems is evaluated by complexity, which is quantified by information entropy. Subsequently, the Kolmogorov complexity is estimated via a lossless compression approach to approximate the lower bound of information entropy in complex systems. Fault diagnosis is then performed by employing normalized information distance together with the k-nearest neighbor algorithm. The experimental results demonstrate that the proposed method outperforms traditional data-driven fault diagnosis methods in fault diagnosis of complex systems. Additionally, it achieves excellent diagnostic performance in severe few-shot and variable working condition scenarios. This paper opens up a novel methodology for complex system fault diagnosis, holding significant implications for the practical application of data-driven fault diagnosis methods.