Optimisation in the US

The optimisation of cement operations has become a goal for many producers around the world, not only focusing on throughput maximisation or energy savings, but also on the reduction of emissions and CO2 footprint. There are a wide range of technologies able to deliver Advanced Process Control (APC) solutions, but not many of them use Artificial Intelligence (AI). OPTIMITIVE’S AI system, OPTIBAT, learns from process data and continuously adjusts optimal setpoints in closed-loop mode. The key factor is the capability to adapt to changing and uncontrollable process conditions. This approach enhances the capability of finding optimal states.

Problem description

In the cement industry, many decisions affecting energy, throughput, quality or profit are based on human experience and judgement. Plant operators usually stay in their comfort zone, focusing on keeping process stability. Improving operation performance or getting the most from their plant assets’ operation is difficult achieve consistently. Traditional control technologies are capable of improving performance, but they do not focus on optimising at each moment. In order to go that extra mile, the OPTIBAT tool goes beyond regular control and adapts each setpoint to the best possible configuration. OPTIBAT guides the decision-making process to the optimum with a demonstrated impact on the business. This software is used in the optimisation of complex processes, such as the operation of a cement kiln.

The solution

OPTIMITIVE has experience deploying AI Real Time Optimisation (RTO) solutions. The company offers advanced analytics, as well as process monitoring features, integrated into the current operation system. OPTIBAT sits on top of the APC or directly over the Distributed Control System (DCS) providing optimised setpoints for controllable variables that must be reached by the underlying controls. Data driven optimisation can be adjusted to producer requirements, adapting it to the plant operation goal. In terms of optimisation, OPTIBAT has achieved great results by either maximising throughput or minimising plant specific energy consumption, whilst meeting process requirements (quality and stability). It has also helped cement producers to lower CO and NOx emissions.

OPTIMITIVE provides a solution based on AI models that can learn non-linear relations, which are fast to set up and are only dependent on the availability of process data. On top of that, AI models remain continuously updated, thanks to machine learning technology.

Optimising a cement plant

OPTIMITIVE’s last deployment in the US successfully optimised a whole cement plant, including a vertical raw mill (VRM), a kiln, and three ball mills. This project was delivered in 2020, resulting in an improvement in plant performance. Throughput was incremented 10% in the ball mills, 6.8% in the kiln and 6% in the vertical raw mill during the test period. The optimisation deployment also allowed for the kiln to operate at its full capacity for 97% of the testing period, while keeping stable the free lime values. Additionally, the ball mills and the vertical raw mill reached 4 – 5% improvement on specific energy consumption. At the kiln, the savings of specific thermal energy consumption totalled 3.4%. All five assets are maintained and currently used in closed-loop mode without human intervention, while the process is in a steady state.

In the VRM, the solution provided uses six setpoints for throughput optimisation in closed-loop. There were difficulties predicting vibrations, a key indicator of process stability. Vibrations have a direct relation to the bed height and the mill differential pressure. The relationship between these two inputs were not as simple as expected. Depending on the mill under or over-loading, the relationship between these two variables changes. This modulation effect needed to be identified.

When vibration levels are under control, the system maximises throughput while keeping the main motor and fan consumption within range, as well as the mill differential pressure stable. Above this, residue quality measurement was also controlled.

Regarding kiln and cooler installation, the optimisation target is defined as multi-objective optimisation, which includes throughput maximisation and specific thermal energy minimisation (kiln and calciner), while keeping free lime within range. Once the optimisation target is defined, OPTIBAT keeps the maximum allowed feed, 400 tph, at its maximum 99% of the time during the commissioning test with a decrease of 3.4% in specific thermal energy. Also, the solution kept the free lime value within the desired ranges for 97% of the testing period. In the current implementation, OPTIBAT manipulates eleven setpoints that enhance both kiln and cooler performance to higher optimisation results in closed-loop mode.

During the commissioning phase, the need for an accurate free lime prediction to keep the kiln and product stable was identified. The free lime was measured every 2 – 4 hours and it was essential to predict the trend of the free lime between samples. For this reason, OPTIMITIVE developed a ‘soft sensor’ predictive model which uses any measured signal that can help identify the current free lime value, and from there, operate the kiln always fulfilling the free lime requirements. This free lime soft sensor considers more than 25 variables from the system and calculates the FCaO current value. In the cement world, there may be certain factors which have an impact on the free lime that is not even measured, such as the precise raw material composition, and the variability of specific caloric value of fuels, among others.

In this case, OPTIMITIVE was able to find an accurate model able to capture overall trends of the free lime, which allowed the system to optimise the process without compromising the product quality. This model also learns from new situations and adapts to them, so the free lime prediction stays updated with changing conditions.

Currently, OPTIBAT recommends optimum setpoints for the controllable variables meeting the optimisation target. Optimising a kiln process does not only involve the free lime control, but it involves many other variables that must be controlled within operative ranges, emissions, torque, secondary air and more. OPTIBAT is able to find the optimum setpoints that allows the system to run efficiently at a minimum cost. AI is definitely beneficial, considering the complex relationships between different variables.

Regarding the ball mill process, OPTIMITIVE implemented a throughput maximisation optimisation in three different ball mills with a capacity of 100 to 140 tph using four setpoints. This throughput optimisation translates into an improvement of 5% in specific energy consumption. During the project definition phase, the importance of process stability, and avoiding mill underloading and overloading was identified. With that in mind, the total feed (fresh feed and the rejects) variable was predicted and defined as a key factor for project success. Additionally, the quality measurements are kept within the allowed range.

One of the main issues identified while developing the solution was avoiding the inertia of the process. When maximising the throughput, there is a risk of overloading the mill as the rejects levels are also increased due to quality product requirement fulfilment. When the commissioning phase started, the first approach was not able to handle the inertias on the total feed and the mill was reaching unacceptable overloading levels. However, the company discovered a way of incorporating the inertia effects into the AI models, so the system was able to predict these effects in advance and minimise them at acceptable levels. Later the mill was able to run smoothly in closed-loop mode.

Once the stability issues were tackled, the next step was to reduce the variability of Blaine and passing quality measurements. Added to the sampling frequency of 2 – 3 hours and raw material variability, which complicates the prediction, the specific sampling method may affect the result. Depending on the sampling method, the approach for the AI model can be different. In this case, the team chose to develop a corrective AI model. This approach provides the current quality measure towards the quality target, applying recommendations based on the manipulated variables. This approach helped the system to reduce the quality variability up to 38%.

Conclusion

Availability of data is a key factor in AI-RTO technology. It allows the system to continuously adapt and learn from new conditions, so operations reach higher levels of optimisation. Implementation of AI RTO in closed-loop has been successfully deployed in a whole cement plant with a great impact on the cement operation and business.

Authors

Raúl Rivero, Jesús Ortiz, Alejandro Llaves, Iñigo Llopis, Igantzi Rodríguez, Bhavyashree B., Javier A. García Sedano.

Read the article online at: https://www.worldcement.com/special-reports/07122021/optimisation-in-the-us/