Unearthed Cape Town 201606 Feb - 08 Feb
Driving Startup Style Innovation in the Resource Sector
This competition has finished.
Classification of mined material and detection of end of mining based on sound
Problem: Marine diamonds are found in a gravel layer that lies under other sediments and on top of a "footwall" layer that is typically clay. Mining techniques use various forms of suction, such as a suction drill, to remove the sediment and gravel and extract diamonds on board a ship. There is currently no precise way of knowing when the end of the gravel layer has been reached during mining. Manual observations are made of the mined material on board a ship to roughly determine the depth of the gravel, but these observations are subjective and sporadic and thus error-prone. It also takes the material roughly one minute to reach the surface which means that adjustments to mining depth cannot be made in real time. It is likely that significant improvements in mining efficiency can be made if it is known in real-time when the footwall has been reached.
Challenge: Use audio of the mined material traveling through and impacting on a metal pipe to identify characteristics of the mined material in real-time. Ultimately, the goal is to identify the end of the diamondiferous gravel, but there are a number of smaller steps towards this end that would be useful outputs of this challenge. Having this information will allow mining to occur more precisely and avoid situations of under or over mining. The audio is recorded with a hydrophone or accelerometer, attached to the suction pipe, close to the seafloor where the mining occurs. A challenge with this data is that mixing of different geological layers has occurred by the time the material is recorded. So it is generally not possible to detect a discrete footwall event by ear.
Outputs: Useful outputs can be any combination of the following:
1. Measure particle impact frequency (counts/sec) in real-time.
2. Measure particle impact strength in real-time.
3. Determine if distinct classes corresponding to different materials can be identified in the mining audio (e.g. using unsupervised classification). Determine if it is possible for these classes be separated / unmixed from each other and if their relative proportions can be determined over time.
4. Provide a visual representation of the audio (that could combine the above information and or use other techniques) that allows an operator to judge when footwall has been reached.