Data Analytics and Visualization Application for Asset Health Monitoring

Abstract

Much of the research on predictive maintenance has focused on statistical and machine learning techniques, while there has been significantly less focus on the human computer interaction or visualization aspects of PHM. Human computer interaction and visualization techniques can quickly help identify interesting data sub-domains from assets, time periods, and sensors provided the data can be queried, retrieved, and displayed in a timely manner. Augmenting visualization and interaction with a visual, aggregative fleet-based query system adds a further dimension, highlighting the ability of the fleet to carry out its mission. Visualizing data from an asset with a multitude of sensors in a way that fosters human understanding and decision making is challenging from the standpoint of dimensionality. That difficulty is significantly compounded as overwhelming numbers of assets of varying type are added to comprise a hyperdimensional dataset. In this paper, we propose a scalable framework that is capable of visualizing past, current, and prognosticated health from the individual sensor up to the fleet or group level. In addition to viewing near real time sensor data, maintenance logs, fault information, and data aggregations will be merged with the sensor data to make the analysis and visualizations more valuable. This framework is scalable regarding how much data can be collected, stored, and processed, and the different organizational levels within a fleet of assets. The framework is built as a web-application primarily using the following visualizations: a collapsible tree structure for asset information; 2D charts for temporal sensor data, fault data, and maintenance data; and 3D digital twins of critical components. These components combine to optimize human-computer interaction for decision making across several phases of operations and support for Army ground vehicles. The dataset used to build and demonstrate the capability of the web-application contains sensor readings from over 3000 vehicles and comprises approximately 9TB of data. Vehicle information such as model, make, sub-component, and fleet organization are presented in a configurable, collapsible tree structure. This allows the user to visualize the fleet and to select the asset and sensor combinations needed to display temporal sensor data to answer a nearly infinite number of questions using individual or combined statistics. Information regarding each vehicle’s health status is displayed then aggregated and displayed for each of the higher tree nodes. A 3D digital twin also highlights sensor locations and current health status of assets and components. These component models can be viewed and manipulated with or without a virtual reality headset to provide diagnostic and repair support. As health status monitoring for asset subcomponents are developed, they can be added to the system, allowing for complete health status reporting.