(666a) Effect of Signal Processing on Feeding Performance Evaluation

Powder feeding is the start of every continuous pharmaceutical manufacturing line. A constant mass flow into the process is crucial for high quality product. Therefore, feeding performance characterization can aid in detection of feeding issues, even during the early-stage process development phase. Often, many different pre-selected material-equipment pairs with different equipment configurations have to be tested before a suitable setup can be found. Gathering such a large dataset is time and material intensive. A common approach to reduce the experimental in-house effort is to review the data of feeding studies which have been published in literature. Usually, the relative standard deviation (RSD) is used as single measure for the feeding performance comparison. However, signal processing details, like sampling frequency and filter characteristics (cutoff frequency) are rarely reported with the RSD data. However, this additional information is crucial when comparing the results of different feeding studies.

We show in the presentation, how selection of sampling frequency and filter cutoff frequency is affecting the statistical evaluation of feeding performance. Illustrative examples are created to demonstrate aliasing, corrupting the signal dynamics. Furthermore, the effect of sampling frequency on the computed RSD value is highlighted by means of the confidence interval of mean, standard deviation and RSD. Additionally, the RSD is highly affected by the filter characteristics applied to the data. Too conservative filtering might not sufficiently reduce the noise resulting in a too high RSD, whereas too restrictive filtering might remove relevant signal contributions, resulting in a too low RSD. The reliable judgment of feeding performance is not possible in both situations.

We propose, additionally to RSD, to report details about data acquisition and signal processing when evaluating feeding performance based on reference scale data. Furthermore, the confidence interval can give an impression on the statistical significance of the feeding performance characterization, allowing more reliable comparison between the different measurements.