Pushing Distributed Vibration Analysis to the Edge with a Low-Resolution Companding Autoencoder: Industrial IoT for PHM

Abstract

The Industrial Internet-of-Things (IIoT) has disrupted theway of collecting physical data for predictive maintenancepurposes. At present, networks of intelligent wireless sensorsare pervasive, finding success in many environments andindustries, including the railways. However, when it comesto data-intensive applications like vibration monitoring thatrequire the delivery of large amounts of records, the limitationsof these devices arise. The shortfalls are mainly drivenby the low-bandwidth transmission capacity of their radio interfaces,and the low-power features of their battery-operated(and/or energy-harvested) electronics. In sight of these limitedresources, this article explores a vibration data compressionstrategy for diagnosis purposes. To maximise the amountof transferred information with the least amount of bytes thismethod works in three stages: first, it extracts the most usefulfeatures for vibration-based analytics. Then, it compressesthe raw signal waveform using an Autoencoder neural networkwith an undercomplete representation, assessing its optimumregularisation approach: the denoising, sparse, andcontractive configurations. Finally, it reduces the resolutionof the compressed data by quantising all the resulting real valuesinto single-byte unsigned integers. The proposed strategyis evaluated with a dataset of railway axle bearings with differentlevels of degradation. The results of the analysis provethat with compression rates up to 10 the vibration signals arepractically unaffected by this procedure, allowing for manydiagnosis goals like anomaly detection, fault location, andseverity appraisal. This approach yields a wide range of businessopportunities for on-board predictive maintenance withIIoT technology.