Transforming cement industry by using AI in Blaine prediction
Sandeep Ramprasad

Transforming cement industry by using AI in Blaine prediction

With the on-going pandemic situation, maintaining quality across the value chain and at the same time reducing the cost of operations remains a key challenge amongst cement companies. ABB explains in this article how advanced technologies like digital APC solutions and AI in Blaine prediction can control, stabilise and optimise various cement processes, helping plant managers achieve profitability and desired targets.

Delivering high-quality products, while keeping production costs low and plant efficiency high, is an ongoing challenge for all process industries. Traditional industries such as cement, face numerous challenges to reduce the cost of operations while maximising the yield and improving quality at the same time. With the global pandemic leading to unprecedented changes around the world, it could not have been more disruptive for the cement industry to embrace digital technologies and harness big data to improve productivity, availability, and quality across the value chain at this time, but for many operators, the results are already showing.

Advanced process control and analytics
Cement is an energy-intensive industry in which the grinding circuits use more electrical energy and account for most of the manufacturing cost. Advanced process control (APC) and related optimisation strategies can help cement manufacturers to reap the real efficiency benefits of digital technology, without sacrificing stability or quality, even as business changes and grows.

Digital APC solutions control, stabilize and optimise various cement processes, helping plant managers achieve profitability and drive towards sustainability targets. These solutions enable manufacturers to optimise coal, raw material, and finished cement grinding by increasing throughput and securing consistent output quality while lowering energy consumption. Digital advanced data analytics offer tremendous opportunities to increase efficiency and further optimise production processes. Now with the emergence of new digital technologies, machine learning models can provide productivity improvements in addition to APC solutions.

Making way for Artificial Intelligence
When planning and implementing a digital strategy, it is important to take a holistic approach. This means moving the process from typically siloed and discrete functions to one in which all processes are connected, via developments in the Internet of Things (IoT) technologies, and then automated. It is then possible to move towards autonomous operations-optimisation and asset management functions happening largely without human intervention, and within a secure environment. The key to successful digitalization is data, collected directly from connected equipment and processes or derived from soft sensor models.

APC and analytics provide the ability to make predictions and estimations about process performance, even in the absence of reliable measurement data, for example, when real-world measurement would be too expensive or to increase the frequency of data input and provide backup for unreliable measurements.

In such cases, analytic models can be deduced from either first principles or process data. Analytic models include graphical (first principles), linear regression, non-linear regression, principal component analysis, artificial neural networks, and support vector machines. Users can test various models and choose the one with either the best fit or performance statistic, thereby leveraging state-of-the-art advanced analytics.

Use of Artificial Intelligence in Blaine prediction
The quality of cement is determined by the Blaine number. The Blaine of cement refers to the measure of the specific surface area or the fineness of the cement. Since process adjustments are made based on this quality measurement, infrequent sampling may result in production loss and inconsistent product quality.

Predictive Quality Analytics makes it possible to accurately forecast cement quality in real-time at any point in the production process, thereby reducing the overspending that is typical in efforts to meet quality targets.

Blaine is measured in a laboratory at a frequency of every one to two hours and is used in the control system to maintain consistent quality and high levels of production. Although the data can be utilised for process control, it does not provide real time insight into the process. This manual approach has limitations and ideally requires a predictive modeling technology that can predict Blaine every few minutes to maintain consistent quality, improve operational stability and reduce variability.


Evolution of the soft sensor
To drive the need to apply predictive quality analytics to Blaine, a soft sensor is developed using data-driven machine learning algorithms to predict Blaine at desired intervals using relevant production parameters such as fresh feed, separator speed, grinding pressure, mill DP, etc.

This can be accomplished by following the steps:

  • Collect historical data from the control system for model training (production parameters and lab data)

  • Data cleansing (e.g. removal of data during mill stoppages, etc.)

  • Create a fullyautomatic regression training model selecting the best fit from the library of models

  • Deploy the model and test the model accuracy using the real-time online data

  • Automatic data pull and retraining of the model if the accuracy is not met

  • Predicted Blaine output is used for further control

As a result, a prediction model transforms cement quality – Blaine from an output process parameter to an input parameter which helps in sustaining the benefits via adaptive re-modeling and tuning. The efficiency of the process can be improved considerably through this approach since Blaine lacks continuous measurement in real-time and can be prone to infrequent sampling. Hence, operators can make more informed decisions using the information available.

Recent developments in advanced analytics have made machine learning models more easily accessible to users. But the true power of a machine learning algorithm can be harnessed only when domain knowledge is applied along with these algorithms. Data cleansing, anomaly removal, analysing the correlation of parameters, result interpretation can be carried out efficiently with expertise in domain knowledge.

Having served the cement industry for more than a century, building up knowledge and know-how of electrification and process control, ABB specialises in increasing plant performance and improving energy efficiency. Using ABB’s proven analytical and process modelling tools, along with our in-depth industry specific knowledge, we can provide a clear path for plants to achieve operational excellence.

Sandeep Ramprasad is the Global Service Product Manager for Cement at ABB Ltd. Actively involved in the fields of engineering, technology management, strategy and product management, he is responsible for driving product and portfolio management, business development and marketing in cement services.

Related Stories

No stories found.