Batch statistical process control of a fluid bed granulation process using in-line spatial filter velocimetry and product temperature measurements

Abstract

Fluid bed granulation is a batch process, which is characterized by the processing of raw materials for a predefined period of time, consisting of a fixed spraying phase and a subsequent drying period. The present study shows the multivariate statistical modeling and control of a fluid bed granulation process based on in-line particle size distribution (PSD) measurements (using spatial filter velocimetry) combined with continuous product temperature registration using a partial least squares (PLS) approach. Via the continuous in-line monitoring of the PSD and product temperature during granulation of various reference batches, a statistical batch model was developed allowing the real-time evaluation and acceptance or rejection of future batches. Continuously monitored PSD and product temperature process data of 10 reference batches ( X-data) were used to develop a reference batch PLS model, regressing the X-data versus the batch process time ( Y-data). Two PLS components captured 98.8% of the variation in the X-data block. Score control charts in which the average batch trajectory and upper and lower control limits are displayed were developed. Next, these control charts were used to monitor 4 new test batches in real-time and to immediately detect any deviations from the expected batch trajectory. By real-time evaluation of new batches using the developed control charts and by computation of contribution plots of deviating process behavior at a certain time point, batch losses or reprocessing can be prevented. Immediately after batch completion, all PSD and product temperature information (i.e., a batch progress fingerprint) was used to estimate some granule properties (density and flowability) at an early stage, which can improve batch release time. Individual PLS models relating the computed scores ( X) of the reference PLS model (based on the 10 reference batches) and the density, respectively, flowabililty as Y-matrix, were developed. The scores of the 4 test batches were used to examine the predictive ability of the model.