Anggaran kualiti klinker dalam kilang simen memainkan peranan penting untuk menghasilkan simen yang berkualiti dan mengekalkan operasi kilang yang ekonomi. Kos buruh dan kos bahan-bahan boleh dikurangkan dengan menukar parameter operasi kilang mengikut kualiti anggaran klinker. Secara tradisinya, kualiti dan komposisi klinker ditentukan dengan menjalankan ujian fizikal dan kimia di makmal. Ini memerlukan lebih banyak tenaga kerja, masa dan sisa bahan klinker. Kualiti klinker yang rendah masih memerlukan proses lagi. Dengan model seperti rangkaian neural, masalah-masalah ini dapat dikurangkan dan meningkatkan operasi kilang. Dalam kajian ini, model rangkaian neural telah dibangunkan untuk meramalkan kualiti klinker seperti C2S, C3S, C3A dan C4AF komposisi di Pahang Cement Sdn Bhd (PCSB). Komposisi ini adalah penting di dalam kekuatan, kehalusan dan combinasi dalam tindak balas klinker. Tiga model rangkaian neural telah dibangunkan, masing-masing dengan masukan, keluaran dan seni bina rangkaian yang berbeza. Di antara tiga model, model yang terbaik adalah Model 2, dengan seni bina PP [2-8-1]. Model ini mempunyai MSE agak rendah 0.0011 and nilai r 0.985 dalam set latihan. Ini seni bina optimum dapat meramalkan dengan tepat pengeluaran set pengesahan, memberikan MSE daripada 0.00082 dan nilai r 0.988. Kesimpulannya, model rangkaian neural dalam anggaran klinker membuktikan kemampuan teknik ini dalam kilang simen. Ia mempunyai potensi yang besar untuk mengatasi masalah di kilang simen dan menggantikan kaedah konvensional mengawal kualiti klinker.
_______________________________________________________________________________________________________
The estimation of clinker quality in cement plant plays a significant role in order to produce good quality cement and to maintain economic plant operation. Labor cost and materials cost can be reduced by changing plant operation parameters according to the estimated clinker quality. Traditionally, the clinker quality and composition are determined by conducting physical and chemical tests in the laboratory. This will require more manpower and time, even the waste of materials because clinker with unsatisfied quality produces lower quality cement which needs to be process again. With modeling such as the artificial neural network, these problems can be minimized and result in an optimized plant operation. In this study, artificial neural network models were developed to predict the clinker quality such as the C2S, C3S, C3A and C4AF composition in Pahang Cement Sdn Bhd (PCSB). These composition are important in the clinker strength, fineness and combinability in reaction. Three neural network models were developed, each with different inputs, outputs and network architectures. Among three models, the best performed model is Model 2, with architecture of PP [2-8-1]. This model has relatively low MSE of 0.0011and r-value of 0.985 in training set. This optimum architecture is able to accurately predict the output of validation set, resulted in MSE of 0.00082 and r-value of 0.988. In conclusion, the neural network modelling in clinker estimation proves the feasibility of the technique. It has a great potential to overcome the current limitations in cement plant and replace the conventional method of controlling clinker quality.