Artificial Intelligence (AI) is making waves by disrupting almost every aspect of our lives, business and industry, and mining is no exception.
Miners, service providers, and equipment manufacturers from across the mining value chain have leveraged AI to provide step changes in process safety, performance and efficiency.
The term AI was first coined in the 1950s as “the capability of a machine to imitate intelligent human behaviour,” but since then its capability has advanced so rapidly machines can now easily exceed human performance.
The most common family of techniques used for AI is machine learning, which is often used either to assist humans or to fully automate repetitive tasks.
For example, Caterpillar integrates AI to support machine vision in haul trucks, which improves safety by automatically identifying obstacles, and increases productivity by supporting autonomous operations.
The process of training a truck to identify objects in images requires thousands of labelled images that are used to tune the AI, which in this case, is based on a deep learning algorithm. This deep learning algorithm mimics the human brain to identify objects based on its experience.
Newcrest Mining has integrated AI to assist operation of a mineral processing flotation plant by providing real-time insight into plant efficiency and the onset of failures.
This insight is generated using a model, which in some cases is referred to as a digital twin, of the plant.
Here, productivity improvements are achieved as AI processes large amounts of data and compares it to past good performance, alerting the operators when divergence occurs.
In real-time, the model provides insight into equipment behaviour that is difficult for humans to identify and interpret. This in-time intervention allows the operators to handle complex plant operations over long shifts in sometimes harsh environments.
Integrating AI can provide a step change in operational performance, so companies want to understand how they can get started.
To get started and gain business support, it is important to choose a problem that offers large production value if it can be fully or partially automated.
A secondary consideration is to ensure that production data is available, as this will allow the project to gain momentum quickly. Examples of production data could be existing camera monitoring systems or existing production plant data.
Lastly, ensure that a reliable and dependable deployment platform integrated with the existing production systems is available. Most AI projects fail at this final stage as they are unreliable and cannot be integrated into a production environment.
MATLAB provides an AI development platform that supports the project from concept to production deployment:
Data cleaning and exploration – Data quality is a recurring challenge in mining. In the case of Newcrest Mining, the input from the metallurgists and operators provided valuable contributions in the task of separating data issues from true plant anomalies.
This domain, or first principles, knowledge was leveraged in the creation of derived or engineered features from the raw data. With Caterpillar, the challenge was the collection and labelling of large amounts of images.
Model development – When building a model, it can be challenging to identify the correct model structure.
MATLAB supports high-level workflows for evaluating and optimising the numerous different types of models that support new and experienced users. Typically mining process can be partially described by first principles and this combined with machine learning yields excellent results.
In the case of Newcrest, the domain experts helped identify the structure of the plant and process flow. This knowledge was used to represent the larger plant as a series of smaller models. The high-level workflow streamlined the creation and evaluation of many candidate models for each unit process.
Production deployment – Deployment requirements vary greatly for each application. MATLAB provides flexible multi-platform deployment that can target both production IT/OT environments and the machine hardware through the generation of standalone C/C++, structured text and CUDA code.
Newcrest deployed the models into the IT environment that interfaced the plant data through plant historians. In this case the ease of deployment and updating the model were important requirements.
On the other hand, Caterpillar targets high powered embedded hardware running on the haul truck using code generation technology. In this case, speed, small memory footprint, and dependability requirements were imperative.
AI and its ability to assist in task automation offers great benefit to the mining industry. These benefits come in the form of increased safety and improved operating performance.
To successfully integrate AI into production processes, it is important to understand the production integration and integrity requirements from the beginning of the project.
To further reduce project risk, it is important to consult subject matter experts, and partner with experienced vendors.