ABSTRACT
In a cement plant, the focus of the quality management is on the characteristics of the Portland cement clinker and the performance of the final cements. The quality manager has clear targets regarding the properties of the clinker, including the composition of the cement clinker. The characteristics of the other intermediate products (raw materials, raw meal, hot meal, fuels) and the production conditions are so set up and controlled that the needed quality of the cement clinker can be achieved. With the prevailing standards, controlling and insuring product quality is a must in the cement production as it provides both quality confidence and a trust with always fulfilling the requirements of the different accreditation bodies. The quality control process provides cement plant managers and engineers the confidence through evident figures that the system is still operating satisfactorily, and the results can be accepted. Traditionally, the setup and correction measures are based on the physical-chemical understanding of the production process and the experience.
The laboratories in different cement plants are equipped with a wide range of automated tools. It is thus possible to generate automatically a lot of data describing the process and material characteristics either at the calibration step or the verification of measurement or to control on regularly basis the quality of the production. The development of the analytical method and automation and digitalization enable more and more application of the modern tools including artificial intelligence. Artificial intelligence and machine learning are powerful tools that can operate mainly, but not exclusively, at the level of processing these data, which is a difficult proposition in the traditional methods. This chapter focuses on the main laboratory methods and quality control actions along the process of cement production and associated with the material flow in an integrated cement plant. It describes in short details the potential deviation causes and comments on the possibility to detect it via the application of machine learning. Two examples are given at the end to clearly show the power of artificial intelligence techniques in comparison to the traditional statistical approach.
