Multisensory data fusion-based deep learning approach for fault diagnosis of an industrial autonomous transfer vehicle

The integration of Industry 4.0 concepts into today's manufacturing settings has introduced new technology tools that have already started providing companies an increased level of efficiency in certain operations. Autonomous Transfer Vehicles (ATV) are one of these new tools that are popular in today's manufacturing settings. As these tools become an integral part of the manufacturing ecosystem, accurate diagnosis of ATV faults and anomalies will also be crucial in manufacturing settings. Similar to any other intelligent detection of machinery faults, analyzing and utilizing signals measured from attached ATV sensors may reveal any uncovered operational faults or critical operational/safety concerns. In this context, this paper focuses on an intelligent fault detection of an ATV tool utilizing signals measured from multiple attached sensors. A novel Convolutional Neural Network -based data fusion approach, utilizing short time Fourier Transform, is proposed for the detection and identifi-cation of operational faults occurring in an ATV. The approach is tested on an experimental dataset, consisting of two motors' sound and vibration signals, collected as an ATV operates for a specific task under three different conditions. The diagnosis results indicate that the proposed deep learning-based multisensory fault diagnosis approach is able to diagnose operational conditions with significantly high accuracy compared to single or dual sensor approaches.

Automated summary

The integration of Industry 4.0 concepts has brought new technology tools, such as Autonomous Transfer Vehicles (ATV), that are increasingly being used in manufacturing settings. Accurate diagnosis of ATV faults and anomalies is crucial, and this paper proposes a deep learning-based multisensory fault diagnosis approach using signals from multiple attached sensors. The approach shows significantly higher accuracy compared to single or dual sensor approaches in diagnosing operational conditions.