Since the early 80s, AI development has had an impact on the application of Advanced Process Control in mineral processing plants. During the first wave of AI development, the so-called symbolic AI era, fuzzy-based and expert models were put into practice to support a wide range of operations.
The challenge lies in expert system development, which takes time and heavily depends on specialists since it lacks support for industrial applications.
Ultimately, this problem was solved by advancing the technology and vendors providing user-friendly packages. Still, the majority of processing plants are using qualitative models. While qualitative models are accepted in control systems due to the simplicity and transparency of their logic, they don't offer high efficiency.
Second Age of AI Development
The second era of AI development started with the rediscovery of the algorithm and neural networks in the mid-90s. Simultaneously, the application of mathematical control modeling, including physical and black box models, was reported in the processing plants.
Physical models stem from the physical laws of energy, mass, or momentum balance, and their parameters describe real quantities. But they have limited applications for complex processes such as grinding circuits where the physical models are not well defined and understood.
Conversely, black box frameworks pinpoint the relationship between output and input variables. In fact, their parameters rarely have any physical interpretation. With the lack of transparency of the black box model and the high degree of maintenance, some operations are skeptical about deploying these models for Advanced Process Control Systems, especially for Model Predictive Control (MPC).
By the end of the 90s and early 2000, the lack of computational power, infrastructure, software environment, data, and acceptance of computer-generated advice prevented the expansion of the first generation of machine learning model applications.
New Wave of AI Applications
The latest wave of AI applications in processing started in about 2005. The new idea is to leverage advanced computing power, data availability, and software to develop a highly robust, transparent, and accurate system. But one of the main challenges of the new approach is how to put prior knowledge into the first generation of machine learning and combine Symbolic AI with Numeric AI. To develop such a new system, data scientists with basic knowledge can put the right numeric tools into practice.
One of the applications of the next-gen machine learning models is developing advanced soft senor. While it is helpful to support plant operators with real-time information, the real benefits of soft sensors are realized when combined with MPC.
It is particularly suitable for this application as the response of the soft sensor is modeled against the system inputs, allowing the quality value to be tightly controlled. Now, you can seamlessly integrate these soft sensors into the MPC scheme.
Another application is developing hybrid models with historical data and using mechanistic casual understanding to select a proper structure and variables. It is not a complex task for a data scientist. With advanced technology and proper training, it is feasible within reasonable time and effort.
What is the Obstacle?
Despite mining and metals investment in digital infrastructure, they are 30% to 40% less mature digitally than other similar industries based on the BCG Digital Acceleration index. It supports different aspects of digital applications, like deploying third-generation AI models in APC systems.
In addition, the performance of most implemented conventional APC systems continues to decline over time due to ore variation, asset performance degradation, and lack of in-house knowledge to maintain the system in optimal performance.
As a result, mining is not eager to practice an Advanced MPC system which is not well known in the industry regardless of its benefits.
What is the Opportunity?
The new AI applications are less expert-oriented and easier for young professionals to pick up quickly. Using advanced technologies also motivates the young generation to develop their carries in the mining sector and practice the same knowledge as other sectors.
Many mining companies have invested in digital infrastructure that reduces the cost of implementing new AI solutions and ensures compatibility. Using new AI generation solutions, you can address common causes of failure in deploying older generations of AI.
For example, control models based on historical data and hybrid structure are more robust than other models when operational conditions change and require less maintenance. That leads to a more sustainable solution and provides more benefits.
Higher computing power and powerful algorithms allow the development of more complex and comprehensive AI models for processing plants. You can also develop an integrated model for the entire plant despite the number of process components. It can include all re-circulating and unit interactions to work in tandem with each other simultaneously.
The hybrid structure presents more transparency, and metallurgists can infer from the models which align with their prior knowledge. As a result, users are more likely to trust the system and follow the recommendations.
Another advantage of implementing the third generation of AI is lower risk. You can develop and test the model stepwise without the usual operation interfering. But to develop other models, it is crucial to interrupt the operation to generate the required data. For example, step testing for system identification can interrupt production for days and reduce production rates.
Comments