Monitoring exhaust air temperature and detecting faults during milk spray drying using data-driven framework

Abstract The advanced data analytics platform bridges the gap between industrial automation technology and new cloud-based technology. The information on the implementation of a data analytics platform to convert huge data into valuable information and use it to serve the scheduled maintenance of the components involved in the food processing industry is rarely reported in the literature. This work reports a data-driven framework for prediction and fault detection in key performance parameters for a milk spray drying process plant. The framework consists of different data analysis methods and it helps to take decisions about the selection of key performance parameters involved in improving the spray drying thermal efficiency. The neural network-based NARX model demonstrates a better performance than the linear models in the prediction of cyclone exit air temperature which is the key performance parameter in spray drying as it governs thermal efficiency. The performance of the predictive model is validated using RMSE. The ML-based classification methods are also used in the present work to classify the different faults and the decisions regarding the maintenance of the components responsible for the faults. The performance of these models was verified using a confusion matrix. It is proposed that the decision tree classifier and random forest classifier are best suitable for fault finding as their accuracy is highest at 99.83%.