—We present a data augmentation technique for generating location variant audio samples using ray-traced audio in virtual recreations of the real world. Hardware Audio-Based Location-Aware Systems are capable of locating audio sources in relation to mobile devices. This is a relevant technique in the context of location-based and person tracking in ubiquitous environments. However, this solution is limited in collecting vast data to train the machine learning model reliably. To overcome this problem, we constructed a virtual environment using the audio ray-tracing solution, NVidia VRWorks Audio in Unreal Engine 4, to simulate a real-world setting. The environmental sounds in the real-world scenario were imported into the virtual environment. This strategy could augment data for training Hardware Audio-Based Location-Aware Systems machine learning models with the necessary calibration of the unreal and real data sets. Our results show the audio ray-tracing framework could simulate real-world sound in the virtual environment to a certain extent.
Recent progress in Sensorics and Internet of Things (IoT) enables real-time data analytics based on data from multiple sensors covering the target industrial production system and its manufacturing processes. Diagnostics and prognosis can be implemented using the neural network approach on top of vibration and other sensed data. Neural network methods lead to high accuracy in fault detection and fault evolution. Nevertheless, transferring a neural network model to edge devices leads to performance issues and platform limitations. In this paper, we discuss the edge computing opportunities for diagnostics of industrial rotary machinery using well-known neural network methods. Keywords–Fault diagnosis; convolutional neural network; edge computing; vibration diagnostics.
