How Newtrax can help your journey with AI in Mining
Michel Dubois, Vice President AI for Newtrax has been making waves in Artificial Intelligence for the mining industry. He will be speaking in Quebec City about real-time data validation through artificial intelligence in underground mines.
When: Wednedsay, November 21, 2018
Where: Centre des Congrès room 306-B
Time: 2:50 PM
The Quebec Mines and Energy conference is an exciting platform for industry professionals to come and learn about AI in mining. It’s an emerging topic, getting a lot of attention as the market is adopting technologies to improve safety and processes for the future of the industry.
The following is an excerpt from Mining Magazine called “Live to Learn” by Carly Leonida, Mining Magazine Editor on why machine learning and predictive analytics are so important to the mining industry. Carly explores algorithms, quality data, where to start and lessons to learn.
“Another company making waves in the machine-learning space is Canada-based Newtrax Technologies. The firm is developing an expertise in data quality for machine learning (specifically for underground mining) to improve predictive maintenance and shift optimisation with techniques for detecting and measuring problems with unstable input variables like sensor failures, data integrity, conflicting data, biased data, sparsity, business conformity, outliers, high cardinality, or out-of-order and out-of-date data.
Michel Dubois, vice president of QA at the company, and his team believe that one of the main lessons that miners could learn from other industries is about the value of collecting quality data for training machine-learning algorithms.
“We have seen many other industries, like the internet companies for example, where data collection started much earlier, build enormous value out of the information collected through time. By starting data collection now, miners are building up value for tomorrow,” he says.
Newtrax’s solutions collect data across customers’ underground operations using an IoT-enabled sensor network to create what it calls “an underground mining nervous system“. This includes measuring KPIs in real time for drills, trucks, bolters and LHDs with coverage all the way to the face and without the need for operator input. This passive method of data collection enables the identification of productivity bottlenecks and early warnings for safety, health and environmental hazards. The power of machine learning lies in connecting real-time data from vehicles, personnel and the underground environment. Customers host their data in a Newtrax server on their premises before passing it to Newtrax for processing.
Dubois, who is currently studying data quality for machine learning as part of his PhD in engineering, explains: “Scalability and precision [in machine learning] depend mostly on computing power and quality of data since a large number of algorithms are shared publicly. Also, it is common to hear in the machine-learning industry that 75% of the effort involved in a project is spent on data gathering, cleaning and transformation (the remainder effort is spent on algorithms and infrastructure). That is why Newtrax has decided to specialise in data quality for machine learning.”
Since March 2017, Newtrax has been working with the Institute for Data Valorisation (IVADO), Canada’s largest researcher consortium in data science, artificial intelligence and operations research. Their combined focus is on machine-learning pilot projects for underground mines using open-source algorithms.
Louis-Pierre Campeau, research engineer at Newtrax, explains: “Newtrax has constant interactions with IVADO. For example, one project that we are currently working on with Professor Samuel Bassetto is about the visualisation of complex data to help understand trends and logics in apparently random data.
“An example of an application of this is in predictive maintenance where, given the input of a hundred sensors, we are trying to find out signs of defects in different mechanical parts. This kind of project helps us figure out which sensor combinations are more indicative of the different possible failures, so that machine-learning algorithms can predict more types of vehicle failures.”
Another key area of research for Newtrax lies in harnessing the benefits of machine learning for short- to long-term planning at underground operations.
Campeau is leading the project. “The mathematical model for underground mining that I have developed during my PhD studies provides the tools for optimisation of mine planning given fixed inputs,” he explains. “The combination of machine-learning techniques with such a model will allow for a greater precision and reliability in these fixed assumptions.”
“For example, say that a round has to be drilled in the next shift in a given ramp. A fixed assumption could be to plan for the drilling activity to last the average drilling time of 3.5 hours. Machine-learning techniques on the other hand could use information like the level of experience of the miner drilling the round or the rock conditions in the area to predict that the drilling will actually last closer to five hours. In turn, this prediction could be used to optimise the whole planning accordingly.”