Traditional feature extraction and selection is a labor-intensive process requiring expert knowledge of the relevant features pertinent to the system. This knowledge is sometimes a luxury and could introduce added uncertainty and bias to the results. To address this problem a deep learning enabled featureless methodology is proposed to automatically learn the features of the data. Time-frequency representations of the raw data are used to generate image representations of the raw signal, which are then fed into a deep convolutional neural network (CNN) architecture for classification and fault diagnosis. This methodology was applied to two public data sets of rolling element bearing vibration signals. Three time-frequency analysis methods (short-time Fourier transform, wavelet transform, and Hilbert-Huang transform) were explored for their representation effectiveness. The proposed CNN architecture achieves better results with less learnable parameters than similar architectures used for fault detection, including cases with experimental noise.

Multi-sensor systems are proliferating in the asset management industry. Industry 4.0, combined with the Internet of Things (IoT), has ushered in the requirements of prognostics and health management systems to predict the system’s reliability and assess maintenance decisions. State of the art systems now generate big machinery data and require multi-sensor fusion for integrated remaining useful life prognostic capabilities. When dealing with these data sets, traditional prediction methods are not equipped to handle the multiple sensor signals in unison. To address this challenge, this paper proposes a new, deep, adversarial approach to any remaining useful life prediction in which a novel, non-Markovian, variational, inference-based model, incorporating an adversarial methodology, is derived. To evaluate the proposed approach, two public multi-sensor data sets are used for the remaining useful life prediction applications: (1) CMAPSS turbofan engine dataset, and (2) FEMTO Pronostia rolling element bearing data set. The proposed approach obtains favorable results when against similar deep learning models.

With the availability of cheaper multisensor suites, one has access to massive and multidimensional datasets that can and should be used for fault diagnosis. However, from a time, resource, engineering, and computational perspective, it is often cost prohibitive to label all the data streaming into a database in the context of big machinery data, that is, massive multidimensional data. Therefore, this article proposes both a fully unsupervised and a semi-supervised deep learning enabled generative adversarial network-based methodology for fault diagnostics. Two public datasets of vibration data from rolling element bearings are used to evaluate the performance of the proposed methodology for fault diagnostics. The results indicate that the proposed methodology is a promising approach for both unsupervised and semi-supervised fault diagnostics.

Multi-sensor systems are proliferating the asset management industry and by proxy, the structural health management community. Asset managers are beginning to require a prognostics and health management system to predict and assess maintenance decisions. These systems handle big machinery data and multi-sensor fusion and integrate remaining useful life prognostic capabilities. We introduce a deep adversarial learning approach to damage prognostics. A non-Markovian variational inference-based model incorporating an adversarial training algorithm framework was developed. The proposed framework was applied to a public multi-sensor data set of turbofan engines to demonstrate its ability to predict remaining useful life. We find that using the deep adversarial based approach results in higher performing remaining useful life predictions.