CNC Machining Quality Prediction Using Variational Autoencoder: A Novel Industrial 2 TB Dataset
Abstract
The purpose of this paper is to present and to describe a novel dataset acquired entirely in an industrial environment during multiple regular scheduled production runs. In the monitoring, prognostic and fault detection literature, researchers are often faced with work based on the same popular datasets; for instance, the milling dataset, the Pronostia Bearing Dataset, the IMS Bearing Dataset or the Turbofan Engine Degradation Simulation Dataset. On the one hand, these datasets are the results of either simulations or acquired in a laboratory under controlled environment. On the other hand, a real industrial context might not be adequately represented within these datasets due to less controlled parameters or increased complexity. Consequently, it becomes critical to have access to a way to test and validate research work on both experimental and industrial data. In that mindset, to accelerate the technological transfer to the industry and to ensure that it can quickly profit from the benefits that the monitoring, diagnostic and prognostic research area can provide them, a new dataset acquired at an industrial partner: a machining company located in Quebec City (Qc, Canada) is presented.