Neural Screen Discharge Model

Application

Minealytics offers a range of neural networks to model screen discharge in a screenhouse that can understand and predict the behaviour of screened ore fractions based on various inputs. Our neural networks, with their ability to learn complex patterns from data, can be particularly effective in environments with multiple complex, cross-correlated variables, such as a screenhouse.

Our screen models use the following process variables to model discharge rates and split ratios:

1. Level change rate above screen: this serves as a primary indicator of the input flow to each screen. A higher rate of level change would typically indicate a higher flow rate of input material.
2. Cumulative fractions: the total output from all screens, measured accumulatively, helps in understanding the overall performance and efficiency of the screening process.
3. Screen current: the electrical current drawn by each screen can indicate its load; higher currents may suggest more material is being processed or difficulties in processing (possibly due to screen wear or the characteristics of the material).
4. Feeder speed: this controls the rate at which ore is fed into the screens and can significantly affect the screening efficiency and the load on each screen.

Our neural model is trained using historical data from the screenhouse operation. During training, the network learns the complex relationships between the inputs (level change rate, screen current, feeder speed) and the desired outputs (individual screen flow rates and split ratios). The neural network, through its layers and interconnected neurons, learns the non-linear patterns and interactions between different variables. For example, it might learn how changes in feeder speed affect the screen’s discharge rates in conjunction with the current load on the screens. Once trained, the neural network can predict the individual screen discharge rates and split ratios based on real-time data. This predictive capability allows for proactive adjustments to be made to optimise the screening process.

Further process or operator inputs can be incorporated into the Minealytics Screen Models natively or in the format of hybrid models. These inputs are usually:

• Screen mesh size and condition: the size and condition of the screen mesh can affect the screening efficiency and the split ratio. Including data on mesh size and monitoring for wear can enhance model accuracy. Mesh size can either be input by the operator or fed in the format of image data using cameras above the screens.
• Ore characteristics: properties such as moisture content, particle size distribution, and hardness can significantly affect screening performance. Incorporating real-time or periodic assessments of these properties can refine predictions. Image data is often utilised to infer these parameters.
• Environmental factors: temperature and humidity can also affect screening efficiency and are often added to the screen rate prediction models.

Benefits

• The Minealytics Screen Discharge Rate Neural Model can be used to continuously optimise the screening process by predicting discharge rates and, in closed loop, making real-time decisions and adjustments to the feeder speed.
• By understanding the load and performance of each screen, maintenance can be scheduled proactively, reducing downtime.
• The model can adapt to changes in ore characteristics or operational goals, providing a flexible tool for process optimisation.
• By leveraging our neural networks for screen discharge modelling, mining operations can achieve higher efficiency, better predict and manage screen wear and maintenance, and optimise the entire screening process based on real-time data and predictive insights.

Technology

The screen discharge model is evaluated against measured individual streams and split ratios. Inputs can include levels, screen current, feeder speed, mesh condition and ore properties. The implementation may use a time-series neural model or a hybrid process model. The product is the validated discharge estimate and the control response around it, not a fixed network architecture.