Breakthrough AI Initiative Drives Minerals Processing Efficiencies
Vibrating machines such as screens and feeders are used at almost every mine site to classify and convey bulk materials that have been extracted. The machines churn through tonnes of material and are integral to the successful operation of mining sites. Therefore, valuable information that drives the efficient operation of these assets is constantly created.
The challenge is that extracting this information involves deciphering cryptic and complex motion raw data, a job that can only be done by the most elite domain and data experts. So, what if this intricate process could be predominantly achieved through artificial intelligent (AI) algorithms?
Schenck Process’ research and development team in Germany, led by Jan Schäfer, is using diverse data mining technologies to solve classification and time series extrapolation problems to create such AI algorithms, signifying a breakthrough in the monitoring of vibrating machines.
Some of the most important information used to sustain these assets fall into two categories, the first being performance parameters that include the total loads, distribution of bulk material and speed. The second, importantly, is labelled condition parameters, which Schäfer emphasises is a ‘cost-saving preventive maintenance strategy’, which for the most part is hidden.
Schäfer insists that the concealment of this information is leading to significant unused cost-saving opportunities that mining companies are missing out on.
“Our customers’ plants handle very high throughputs, within every hour they are producing millions of dollars’ worth of material, so there is potential that major costs could occur for processes that are forced to shut down if a problem with machinery is not detected,” he explains. “Condition monitoring finds the problem before they become failures; therefore preventative maintenance saves significant costs.”
Schäfer explains the importance of condition-based maintenance having the capabilities of taking action on a problem as soon as it’s noticed – the equation is simple. “If you replace the right component at the right time, you can save substantially on maintenance costs,” he says.
It is widely known that most performance and condition parameters correlate with a vibrating machine’s motion pattern. Various industrial sensors that measure the acceleration of the vibrating body are therefore used to capture the motion of a vibrating machine.
While the intricate data being produced by these accelerometers is currently given to a data specialist, Schenck Process has developed AI algorithms to process accelerometer raw data and output specific information. Given the scarcity of these data experts, the algorithms offer a huge opportunity to mining companies, essentially creating the need for humans to decipher this immensely complicated data obsolete.
“The experts who eventually get that data are rarely available, our customers don’t have these people,” Schäfer explains. “These algorithms enable customers to use the same possibilities system as we do, we can make it available to them.”
Schenck Process has found that a combination of using AI learning and human domain expertise allows vibrating machines to achieve an optimum level of reliability and accuracy. To achieve this learning, Schäfer presents the ‘solution approach’ which is based on a technical framework that is coined the ‘object-based training library.’ It allows AI algorithms to constantly learn and develop in order to adjust to a wider range of situations.
“If you want to transfer data to information, you need to develop knowledge – the process to generate knowledge though, is complicated, so we started a library where we archive historic data,” Schäfer says. “Think about all datasets as an experience, which we collect into the database, we label and structure the data.”
Labelling data is crucial for what Schäfer describes as “data mining”, a key process of finding data to then generate knowledge. This is all done by saving and archiving data and then developing apps to explore said data.
“It’s an efficient way to create AI and maintain it so as soon we obtain new data, we are updating the algorithm,” Schäfer says. “Each data that we put in creates a new experience and can consequently optimise your parameters.”
For many, the concept of removing humans completely from the process is frightening. However, this is not what Schäfer is suggesting and instead emphasising that both AI and humans each bring their own benefits.
“Each AI algorithm is able to perform a very specific task, it can do this in an excellent way but it’s very specific,” he explains. “Looking at screen performance or maintenance, it’s a wider range of tasks and things to consider, AI algorithms can deal with specific tasks, but they aren’t able to consider a variety of possibilities to action like humans can. The algorithm will never make the human obsolete, but it certainly will be very helpful.”
Ideally, human experts will still need to interpret the information produced by the AI algorithms, coming to a combined conclusion.
According to Schäfer, the answer to problems associated with these machines is rarely black and white, “in most cases uncertainty remains in the diagnosis and by nature, all predictions carry some grade of uncertainty.”
“Imagine you go to a doctor and they give you the best interpretation possible, sometimes it’s pretty clear and sometimes it’s vague and you have to do further tests,” he explains.
The situation for resolving issues with vibrating machines is similar to diagnosing an illness in humans, however, Schenck Process is closing the gap on the uncertainty. Ultimately, the company’s research offers mining company’s the hidden opportunity to maximise uptime, reduce costs and increase operational efficiency.
While Schäfer admits the AI algorithms won’t be perfect in the first stage of use. Within five years he hypothesises, however, that human interaction will be less required and as algorithms evolve, so will autonomism.
Leave a Reply