Exploring the impact of big data in economic geology using cloud-based synthetic sensor networks

Abstract:

In a market demanding lower resource prices and increasing efficiencies, resources companies are increasingly looking to the realm of real-time, high-frequency data streams to better measure and manage their minerals processing chain, from pit to plant to port. Sensor streams can include real-time drilling engineering information, data streams from mining trucks, and on-stream sensors operating in the plant feeding back rich chemical information. There are also many opportunities to deploy new sensor streams - unlike environmental monitoring networks, the mine environment is not energy- or bandwidth-limited.

Although the promised efficiency dividends are inviting, the path to achieving these is difficult to see for most companies. As well as knowing where to invest in new sensor technology and how to integrate the new data streams, companies must grapple with risk-laden changes to their established methods of control to achieve maximum gains.

What is required is a sandbox data environment for the development of analysis and control strategies at scale, allowing companies to de-risk proposed changes before actually deploying them to a live mine environment. In this presentation we describe our approach to simulating real-time scaleable data streams in a mine environment. Our sandbox consists of three layers: (a) a ground-truth layer that contains geological models, which can be statistically based on historical operations data, (b) a measurement layer - a network of RESTful synthetic sensor microservices which can simulate measurements of ground-truth properties, and (c) a control layer, which integrates the sensor streams and drives the measurement and optimisation strategies. The control layer could be a new machine learner, or simply a company’s existing data infrastructure. Containerisation allows rapid deployment of large numbers of sensors, as well as service discovery to form a dynamic network of thousands of sensors, at a far lower cost than physically building the network.