Tame the Temp08 Jul - 14 Aug
Predict the temperature control in industrial autoclaves
Predict the temperature control in industrial autoclaves
Newcrest is seeking machine learning models to predict the temperature inside their autoclave. Autoclaves extract gold from rock. The predictive model will enable Newcrest to keep the autoclave better running at peak efficiency over time, allowing Newcrest to reduce costs and emissions.
This challenge invites all teams and individuals with interest in data science, mathematics, analytics, minerals processing, or data in general to participate. Unearthed is trialling a new approach in this challenge to help prepare models to be quickly deployed into production at Newcrest. Details of this new system, as well as contextual information on the challenge and the data, are outlined below.
Can you build an algorithm that predicts the temperature one hour in advance inside an autoclave device?
Newcrest's gold mining site at Lihir uses autoclaves to process the gold ore slurry as part of their overall mineral processing operation. This process is managed by Newcrest's engineers and operators who aim to maximise production output of the plant while ensuring the autoclave temperature remains within a safe range.
Newcrest collects data allowing operators to manage variables such as temperature, water content, oxygen, flow rate, raw ore feed and others. However, factors outside of their control, such as grade and sulphide content can change the operating temperature and have a direct impact on operations.
While Newcrest can control the autoclave operating conditions, they currently do not have an accurate model to predict temperature fluctuations or help inform the operator's decision-making process.
Current Extraction Process
As part of the gold ore extraction process, mined rock is ground and mixed into a slurry. This slurry is fed into large cylindrical heaters, called an autoclave, which heats and continually mixes it.
When the slurry reaches a certain temperature threshold, an exothermic chemical reaction takes place (exothermic reactions release energy) which oxidises waste minerals into compounds which can be removed after the slurry leaves the autoclave. The remaining concentrate is much higher in gold, making it easier to remove gold further along the extraction process.
It is essential to keep the autoclave operating within an optimal temperature range to sustain the oxidising exothermic reaction. If the temperature is too low, the reaction ceases to occur automatically, and the autoclave must be stopped and reheated to a temperature to allow the reaction to restart, which may take several hours. If the temperature gets too high, safety controls may need to be enacted to bring the temperature back down, reducing the amount of slurry that can pass through the autoclave.
The autoclave operators require sufficient time to adjust the controls to keep these factors within safe and optimal operating conditions and to avoid conditions which may lead to needing to shut down the autoclave. A predictive model would allow Newcrest to predict temperature fluctuations and take action to prevent the impact on autoclave operations.
Newcrest is seeking models that can accurately predict autoclave temperature an hour in advance to allow them to reduce the impact of temperature fluctuations and more efficiently operate their autoclave. This model will allow operators to make timely and appropriate decisions in response to changing autoclave conditions and ultimately lead to optimisation of the autoclave process by increasing and stabilising throughput, thereby increasing gold production.
Individuals and companies who participate in this challenge will have the opportunity to the following:
- Try your hand at a new way of working in Data Science - Be a leader in the data science field by building models in an environment that is closer to the production environment.
- Connect with Newcrest leaders – Teams that produce the top models will have the opportunity to virtually meet with a panel of Newcrest leaders for a private discover session. In this forum, Newcrest will seek further detailed background on your model.
- Work with Newcrest to get your model into production – The winning submission may have the opportunity to get their model into the Newcrest production environment
- Win prize money and recognition - Following meetings with Newcrest and successful build of top models within the Newcrest ML pipeline, prize money will be awarded. Top teams will also be profiled across Newcrest operations the wider Unearthed community including other industry representatives.
A New Way of Doing Data Science
During the Tame the Temp challenge you will be the first to use the early access challenge submission platform. A new way to create, submit and track your submission in an Unearthed data science challenge.
Too often data science models are worked on in isolation from the machine learning infrastructure which they are built for. This new way to compete in an Unearthed Challenge is the first step in tackling this industry-wide challenge, of overcoming the data science implementation hurdle.
The Unearthed early access submission platform will bring you closer to the production environment and allow you to score your algorithm directly in the production platform. Your submission will be created inside the container, so the data structure and tooling are all laid out for you.
If you have any questions, bugs, or issues on your journey, please drop a message on the challenge forum, join the slack challenge #tech-support on the unearthed-community group, or contact us directly at [email protected].
Submissions are made through the unearthed-CLI and the submission pipeline. Please head to the "GET STARTED" guide to get set up for initial submission.
The data preparation step in the template is to be a pre-step in the train routine of your files.
- The training data set is from Jan – Nov 2019
The winning model is required to deliver the following:
- All scripts and notebooks in Python only (we are working on introducing additional languages to this process over time).
- The template provides guidance on where to process data, develop models and make submissions - be careful about changing this.
- Clean code, explanation of your code and process in a ReadME document, modularity, and readability are requirements.
Models will be scored on the RMSE accuracy against the validation data set. The validation dataset is more recent than what is provided to competitors and is unseen at the time of competition, to ensure models are well fitted, and competitors are assessed equally.