Introduction:
With ever-growing demand for infrastructure and building materials, concrete manufacturers have been expanding its production facilities more than ever. With every ton of expansion, there is a challenge observed in optimization of resources, fuel and inventory. Cement manufacturers hold the processes which are one of the most energy intensive and emission generating plants. Being one of the early adopters of digital transformation, this sector has always been finding the opportunities to optimize their processes hence improving the overall energy efficiency. Manufacturing of optimal grade of cement demands for calculative consumption of fossil fuels, raw mix and alternative fuels. Several unforeseen influences go unfactored even behind the traditional techniques of process control and inferential. With the ongoing rapid developments in the world of AI has enabled the applications with opportunities to improve the operations from the current baseline. Previous intricacies including – unmeasured disturbances, complex process dynamics and multivariate nature of the processes, which were left unaccounted, has now figured out the ways to be integrated with the advanced AI-driven models.
In this article we will be looking at a practical example leveraging AI based closed loop control solution to improve the Kiln efficiency. For kiln operations, manufacturers desire to minimize fuel consumption, maximize the feed rate without compensating for the clinker quality in the output. To achieve so, several other parameters must be accounted for, which includes more than 50-70 parameters in total. Some of these parameter’s manipulative, controlled, and disturbances by nature. Below is the schematic which represents some of the critical parameters which needs to be captured for building the closed loop model.
Fig. Source: Imubit – the world’s first reinforcement learning-based model
Reinforcement learning for process optimization:
AI has made life easy for realizing complex relationships among several parameters which would have not been possible just by using the 1st principles. As mentioned above, factors including – unmeasured disturbances, complex process dynamics and multivariate nature of the processes and others could have been never accounted for using barely 1st principles. 1st principles can take care of thermodynamics, mass balances very well, but does not account for the factors which were never considered in it. AI on the other hand has shown the potential to integrate and extract complex relationships from any data which can be quantified. Several stages of development have been made recently from supervised, and unsupervised to now reinforcement learning techniques allowing the developers to continuously improve their modeling strategies.
We are going to showcase 1 such case study which leverages reinforcement learning for model development. With the developed model, the operators can leverage the what-if capabilities to simulate multiple scenarios including influences of quality in feed raw mix quality variations, alternative fuels (AFR), and fossil fuels on the target clinker production, energy efficiency and overall cost of operations in real-time. With the simulated scenarios being validated in open-loop, the model finally gets to sit on top of controllers in closed-loop mode. In closed loop mode, the model runs in real-time (every minute) to optimize the processes by making the necessary changes in the set points to minimize the fuel consumption and maximize the feed. Such implementations have potentially shown several benefits in the KPIs such as a 2-4% improvement in the yield, a 15-30% reduction in fuel usage, and over 2kMT reduction in CO2 emissions yearly.
Fig. Deep learning (reinforcement) based process control
To know more about Imubit’s DLPC (deep learning-based process control) kindly reach out to us.