Real-Time Monitoring of Cables Based on Network Interface Controllers for Predictive Maintenance

Conventional cable diagnostic methods require an interruption of the communication link and are therefore unsuitable for real-time quality monitoring and predictive maintenance. In this article, we propose a novel diagnosis method, which can be implemented on regular low-cost network interface controllers by just implementing dedicated signal acquisition and processing. The novel method passively monitors the state transitions between voltage levels, during the running communication, by undersampling and sorting the collected data into a histogram. By fitting the collected histogram to a sum of three Gaussian distribution functions, robust features can be extracted, having more independence on noise and the transmitted signals. Based on the extracted features, an early fault detection can be realized using a linear support vector classifier (SVC). The novel method has been validated for fast Ethernet cables using STM32-based microcontroller measurements. Open circuit and short circuit faults could be accurately identified. Classification accuracy of 100% could be realized even by
$k$
-fold cross-validation and testing. Furthermore, the proposed method can easily be applied to other cable types after adaptation to the corresponding communication signals.