FastText-Based Framework for Real-Time Fault Log Classification in Industrial IoT: Model, Engineering, and Validation

Abstract

The volume and variety of failure logs produced by automated production systems have increased dramatically in recent years due to the Industrial Internet of Things' (IIoT) rapid development. In order to maintain regular operation and quickly address maintenance issues, the following real-time classification of these logs is required. In this paper, a stable framework based on FastText—which is comparatively fast and accurate—is introduced for the automatic classification of industrial defect records. For the framework test, a complete log dataset of 1.2 million entries from over 2,000 industrial equipment and 15 different types of operating problems was obtained. The dataset has been annotated by subject experts and preprocessed in a reasonably methodical manner. According to the experiment results above, the FastText-based classifier has a median inference latency of 3.1 ms per sample and a high accuracy of 91.8%, making it appropriate for high-speed deployment. The system has been found to be both resource-efficient and general-purpose based on comparisons with widely used statistical and neural models. Deployment in the industrial process has decreased the amount of unscheduled production downtime, decreased the frequency of false alarms, and cut the reaction time for maintenance. For large-scale industrial applications that require quick integration and real-time operation, lightweight neural networks are therefore very useful. The experiment mentioned above demonstrates how the two types of IIoT production environments will become more dependable and efficient.