Press

Dingo making tracks with new Trakka

Release date: 5/20/2019

Mining Magazine - Driven by artificial intelligence and machine learning, Australian-based Dingo has unveiled its Trakka predictive analytics solution for predicting equipment failures to keep mines’ uptime at a maximum, improve component life and more.

The new solution can help users target issues proactively, meaning corrective maintenance can be performed, allowing for optimised asset life. The company said its algorithms, which work in tandem with a site's existing condition-monitoring data, can result in accuracy exceeding 85%.

"Trakka's machine-learning models are developed by Dingo's team of data scientists using failure data from actual equipment, which is then validated by maintenance experts," Dingo's Kirstin Gaffney told MM.

Dingo has made an anomaly detection model available, as well as one for remaining useful life (RUL). The former has been the culmination of 12 months of development and refinement work with its machine-learning models - work it has performed alongside the Queensland University of Technology.

"[That work has been] to detect anomalies in condition-monitoring data in Dingo's OEM-independent global asset health database," Gaffney noted.

"These models highlight anomalous behaviour in the data and will be available to users of Trakka. As more quality component failure data is added to the data set, the accuracy of the anomaly detection models will improve.

Dingo said that, by detecting anomalies automatically, those who use Trakka will be able to detect developing issues well before traditional engineering limits are reached, find slight changes in data trends which are not discernable to a human analyst, and act faster to correct abnormalities and restore equipment to normal operating condition.

The RUL model, meanwhile, has been designed by the company to incorporate a high level of accuracy to predict an asset's remaining useful life, and Trakka users will have access to this valuable analytical information about both the probability of failure and degradation indexes.

"These models are built by Dingo subject matter experts for common asset-specific failure modes, e.g. final drive gear wear failure," Gaffney said. "They are designed with scalability in mind and can be easily retrained to work with a broad range of asset/failure mode problems experienced by real mining operations, making them highly reusable without further development."

Trakka users with the RUL model will have the benefits of optimising repair costs as components at a site near the end of their lives, having more confidence when planning component replacements, and have the ability to improve related processes such as budgeting and supply-chain logistics and management.

She added that users can realise improved availability and reliability, planning and budget capabilities, as well as a reduction in costs since repairs can be planned. That is in addition, of course, to a reduction in unexpected failures and risk.

Gaffney pointed out that Dingo's solution is bringing an answer to real issues for real clients.

"There are a lot of lofty predictive analytics claims in the market, but most of the companies touting these claims haven't produced any tangible benefits for mining operations.

"There is a big difference between passing data through a generic analytics platform and Dingo's solution. We have captured the problems our customers are actually having and using their data to develop models designed to help them gain better insight into the underlying issues so they can address the root cause."

While some methods and solutions offer ideas, Dingo said Trakka can deliver actionable intelligence - it isn't data science alone that spotlights an issue and leaves its solution open-ended.

"With enough lead time, they can perform corrective maintenance and address the underlying problem, increasing the life of the component and reducing their overall costs," she said. "Dingo already has the work management solution in place, so any resulting actions can flow straight into existing maintenance processes."

That's important because, she noted, sometimes the output needs human expertise; that's why Dingo is available to all of its clients in need of assistance.

"There are a lot of companies who are throwing a lot of data scientists at this problem, but they're missing the mark because they don't understand mining maintenance. The sweet spot is achieved by starting with the end in mind (what problem are you trying to solve, outcome-focused) and then bringing the right blend of data science and maintenance expertise to the table to help miners achieve their desired outcomes," she said.

Dingo is currently working in collaboration with a top 10 miner and has already developed a RUL model for the operator's Caterpillar 789C and D haul truck final drives. The project has focused on discovering gear wear failures for a Central American gold mining operation.

It is also developing a model for another client, a large gold and copper mining operator in North America, to identify con-rod bearing failures in Caterpillar 3516 engines, and it is also developing a model to identify valve failures in Caterpillar 3516 engines for a large gold miner in North America.

Finally, the company is in testing for a model design alongside an Australian client that can predict gearbox and pinion bearing failures at its mill.

For more information on Dingo's industry leading predictive maintenance solutions, email us at info@dingo.com or use our contact form


Donna Schmidt │Mining Magazine │ May 2019