Wet granulation is a frequent process in the pharmaceutical industry. As a starting point for numerous dosage forms, the quality of the granulation not only affects subsequent production steps but also impacts the quality of the final product. It is thus crucial and economical to monitor this operation thoroughly. Here, we report on identifying different phases of a granulation process using a machine learning approach. The phases reflect the water content which, in turn, influences the processability and quality of the granule mass. We used two kinds of microphones and an acceleration sensor to capture acoustic emissions and vibrations. We trained convolutional neural networks (CNNs) to classify the different phases using transformed sound recordings as the input. We achieved a classification accuracy of up to 90% using vibrational data and an accuracy of up to 97% using the audible microphone data. Our results indicate the suitability of using audible sound and machine learning to monitor pharmaceutical processes. Moreover, since recording acoustic emissions is contactless, it readily complies with legal regulations and presents Good Manufacturing Practices.

Pellet production is a multi-step manufacturing process comprising granulation, extrusion and spheronisation. The first step represents a critical control point, since the quality of the granule mass highly influences subsequent process steps and, consequently, the quality of final pellets. The most important parameter of wet granulation is the liquid requirement, which can often only be quantitatively evaluated after further process steps. To identify an alternative for optimal liquid requirements, experiments were conducted with a formulation based on lactose and microcrystalline cellulose. Granules were analyzed with a Powder Vertical Shear Rig. We identified the compression density (ρ_press) as the said alternative, linking information from the powder material and the moisture content (R² = 0.995). We used ρ_press to successfully predict liquid requirements for unknown formulation compositions. By means of this prediction, pellets with high quality, regarding shape and size distribution, were produced by carrying out a multi-step manufacturing process. Furthermore, the applicability of ρ_press as an alternative quality parameter to other placebo formulations and to formulations containing active pharmaceutical ingredients (APIs) was demonstrated.

For process monitoring, an adequate data preprocessing is crucial to link accessible inline process data with offline measured target variables. Literature, however, does not provide systematic preprocessing strategies. The effects of five different preprocessing strategies on data from a Dielectric Spectroscopy system applied to the Viable Cell Density (VCD) of a mammalian cell cultivation were thus evaluated. Single-frequency measurements are typically used to model the VCD over the growth phase using linear regression or the Cole-Cole model and served as a reference. As multi-frequency measurement is promising to model the VCD beyond the growth phase using Partial Least Squares Regression (PLSR), we further aimed to determine, whether replacing linear regression by PLSR shows comparable modeling performance. All five preprocessing strategies led to comparable results. Exemplary, when using capacitance values at a frequency of 3347 kHz, linear regression resulted in a $\mathrm{R}^{2}$ of 0.90 and a standard deviation of 0.4 % on average. Both normalization techniques had the same positive effect on the results of PLSR. The order of smoothing and normalization was irrelevant for both regression methods. Comparing the results of linear regression and PLSR, the latter obtained on average 9 % better results. Therefore, we concluded that PLSR is preferable over linear regression and is potentially suitable to model the VCD beyond the growth phase, which is suggested to be investigated based on more data sets.

Wet granulation is a frequent process in the pharmaceutical industry. As a starting point for numerous dosage forms, the quality of the granulation not only affects subsequent production steps but also impacts the quality of the final product. It is thus crucial and economical to monitor this operation thoroughly. Here, we report on identifying different phases of a granulation process using a machine learning approach. The phases reflect the water content which in turn influences the processability and quality of the granule mass. We used two kinds of microphones and an acceleration sensor to capture acoustic emissions and vibrations. We trained convolutional neural networks (CNNs) to classify the different phases using transformed sound recordings as the input. We achieved a classification accuracy of up to 90 % using vibrational data and an accuracy of up to 97 % using the audible microphone data. Our results indicate the suitability of using audible sound and machine learning to monitor pharmaceutical processes. Moreover, since recording acoustic emissions is contactless, it readily complies with legal regulations and present Good Manufacturing Practices.

We present a comprehensive, time-resolved dataset of Chinese Hamster Ovary (CHO) cultivation dynamics to support data-driven bioprocess engineering and process analytical technology development. The dataset includes 24 controlled bioreactor cultivations, representing 12 experiments performed in duplicate in a parallel bioreactor system with a working volume of 0.55 – 0.8 L and cultivation durations of 161.5 – 328.5 h. A total of 48 variables are provided, comprising 38 continuous inline variables and 10 offline variables sampled up to twice daily. In addition to standard process measurements, the dataset includes multimodal inline monitoring using turbidity, permittivity, and dissolved CO₂ sensors. The experiments systematically vary cultivation medium, agitation rate, aeration rate, and process mode (batch and fed-batch), providing diverse operating conditions for sensor evaluation, process modeling, and bioprocess analysis. Furthermore, the inoculation cell density, glutamine concentration, and impeller position were varied to further expand the range of investigated process conditions. The dataset preserves realistic measurement characteristics, including noise, and missing values, enabling robust benchmarking of data-driven methods for process monitoring, modeling, and control.

Determining cell density and viability with (semi‐)automated methods enables rapid cell cultivation analysis. However, scientific validation is required, typically by comparing results to manual counting. To address this, we present a method to evaluate and compare two cell counting methods exemplified by a manual and a semi‐automated method (Countstar BioTech). Experiments followed validation parameters aligned with the International Council for Harmonization (ICH) Q2(R1) guideline and a dilution series design based on ISO 20391‐2:2019. Both the semi‐automated and manual methods showed comparable specificity and linearity for Chinese hamster ovary (CHO)‐K1 cells. The semi‐automated method exhibited superior repeatability for total cell density, whereas cell viability results showed no significant difference.