Investigating and predicting the occurrence of capping for tablets prepared by melt granulation

Abstract

Tablet capping is a critical manufacturing defect whose early prediction is essential to prevent product failures and process interruptions. This study investigated the processability and capping tendency of fluidized bed melt granules prepared according to a Design of Experiments (DoE), integrating conventional granule characterization, compression analysis, machine vision, and multivariate data analysis. Capping was observed in 6 of 15 formulations, predominantly, but not exclusively, in those processed at lower temperatures with capping onset typically occurring at the higher end of the applied compression pressure range. An OPLS-DA model (Q² = 0.615) discriminated capping from non-capping formulations based on granulation process variables.

Hierarchical cluster analysis on compression-derived parameters identified three distinct groups: the failure-free group exhibited lower yield pressure, higher tensile strength (mean 2.29 MPa vs. 1.44 MPa in the capping group), lower detachment stress (5.90 MPa vs. 7.34 MPa), and reduced ejection forces. Process optimisation identified 16.2% Macrogol and a granulation temperature of 63 °C as the robust setpoint, within a design space characterised by only 1.4% probability of failure. Machine vision, while unsuitable for precise particle sizing, demonstrated good capacity to indirectly estimate flowability through agglomerate detection, yielding predictive models for bulk density and angle of repose (Q² > 0.9). OPLS-DA models on machine vision data discriminated capping from non-capping batches with Q² of 0.99 and 0.98, respectively. A data fusion approach further enabled accurate prediction of tablet breaking force and disintegration time, supporting feed-forward process control in pharmaceutical manufacturing.

Introduction

Although tablets may seem obsolete compared to the plethora of innovations that have emerged in the field of pharmaceutical dosage forms, they remain the easiest to administer, the most widely used, and are obtained by low-cost and robust manufacturing processes. Over the years, the compression process has been studied in detail on countless combinations of active substances and excipients, but manufacturers still face processing difficulties or defects in the tablets. Therefore, research is focused on the in-depth understanding of the processes and products to avoid manufacturing errors and quality issues.

Moreover, efforts are made to streamline processes and increase productivity, while human involvement in tablet manufacturing is expected to gradually decrease in the coming years through the adoption of continuous manufacturing lines (Domokos et al., 2020, Záhonyi et al., 2022). However, their advantages such as easy scale-up, robust and flexible operations and time-to-market reduction, also come with a series of challenges (Domokos et al., 2020, Záhonyi et al., 2022) like a well-thought-out feed-forward control strategy with in-process controls and accurate prediction models.

Among the quality attributes of tablets, the appropriate mechanical strength is essential for proper packaging, transport, manipulation and administration, and contributes to the expected disintegration and dissolution profiles. Capping is a frequent mechanical defect that is characterised by the detachment or fracture of the upper and/or bottom layer of the tablet that occurs during the ejection phase or immediately after the compression process (Xu et al., 2018). It is usually observed after the ejection, especially with biconvex tablets. Capping, like other failure phenomena, represents an imbalance between the strength of a material and the stresses that have arisen in it. The high elastic recovery of the materials, associated with residual die wall pressure during the unloading phase of the compaction process, are among the premises that can lead to such defects (Basim et al., 2019, Meynard et al., 2022a). Anisotropic mechanical properties and high air entrapment were also reported (Mazel et al., 2017; Akseli et al., 2013). Traditionally, capping solutions were sought empirically once the defect appeared in a given product, through troubleshooting options such as reducing compression pressure, increasing precompression, extending dwell time, modifying punch geometry (Basim et al., 2019, Mazel et al., 2017, Xu et al., 2018, Li et al., 2026), or changing the unloading phase profile (Meynard et al., 2022a).

Nonetheless, modern drug development approaches emphasizing process understanding to avoid quality risks require the search for solutions to anticipate the occurrence of defects. The measurement of the shear strength acted as a defect predictor during development studies; more precisely, the ratio between the shear strength and the diametral compression strength gave good correlation with capping phenomena, as measures of anisotropy (Mazel et al., 2017). The observed relationship between tensile strength and capping led many authors to treat it as a capping predictor; subsequent studies, however, showed that tensile strength alone, or indeed any single parameter, cannot reliably predict capping (Imayoshi et al., 2025, Meynard et al., 2022a, Paul and Sun, 2017). Ultrasonic data collected on solid oral dosage forms and analysed by machine learning instruments achieved the prediction of both quality attributes and mechanical defects (Sultan et al., 2024). Compaction simulation also plays an important role in the investigation of tableting and tableting defects as it can reproduce realistic manufacturing conditions and enable the calculation of parameters that further describe capping such as in-die elastic recovery, residual die-wall pressure, Poisson’s ratio, and tensile strength (Paul and Sun, 2017).

Powder blend properties, such as rheology and deformation behaviour are also known as strongly connected with the occurrence of capping events, therefore, various research groups have attempted to use them as capping predictors. For instance, Basim et al. have developed a quantitative relation between the capping score and the API load and powder flow-related properties like air permeability, pressure drop and cohesion (Basim et al., 2019). Dudhat et al. demonstrated through tabletability profiles that capping can arise with powders with low permeability (Dudhat et al., 2017). Using machine learning tools, Meynard et al. created a predictive model for the capping of biconvex tablets, based on the plastic energy per volume, the in-die elastic recovery and the residual die wall pressure measured during the compression of flat faced tablets (Meynard et al., 2022a).

Granulation is known for its positive impact on flow and deformation and has shown a reduction in capping tendency (Belič et al., 2009, Bozic et al., 2008). The granule size, morphology, distribution, porosity, density vary upon the way the granulation process is conducted and have an important impact on tablet properties (Munu et al., 2025). It could be regarded as a tableting error reduction strategy, but the extra steps it involves add sources of variability whose influence on compression performance need to be assessed. While the impact of high-shear granulation and fluid bed granulation on capping occurrence was previously reported (Fayed et al., 2017), to the best of the authors’ knowledge, information on the defect incidence of processing fluidized bed melt granules is scarce.

Because capping is a multifactorial phenomenon, correlation modelling requires multivariate data analysis (MVDA) (Casian et al., 2018, Iovanov et al., 2024) or machine learning techniques, applied individually or in combination. For instance, Casian et al. applied MVDA, and artificial neural networks (ANNs) to explore granules downstream processability by fusing NIR spectra and particle size analysis results. Orthogonal projections to latent structures (OPLS) was used to identify the links and differences between NIR and particle size data, then both techniques were employed to build predictive models for tablet tensile strength, ejection stress, and detachment stress. ANNs generally showed better model performance than traditional Partial Least Squares (PLS) methods (Casian et al., 2023). Machine vision, a new analytical technique that allows non-destructive evaluation of pharmaceutical products, achieved the prediction of active pharmaceutical ingredient (API) content (Ficzere et al., 2025), API dissolution profiles from tablets (Meszaros et al., 2024) and the occurrence of defects (Dioszegi et al., 2024). The identification of defects such as chipping, scratches or spots was possible due to the training of Convolutional Neural Networks algorithms to identify and classify them from tablet images, reaching a recognition accuracy of 99.2% (Dioszegi et al., 2024). To understand capping, Paul et al. applied both multivariate data analysis and machine learning tools to identify associated factors, concluding that both compaction parameters and those related to material characteristics determine capping tendency. Notably, they were able to assess quantitatively a predictor of capping occurrence and to compare the ability of available tools (multivariate partial least squares analysis, logistic regression, decision tree and random forest classification tools and neural networks) to correctly identify critical parameters or capping signs (Paul et al., 2021).

Until now, models developed for capping prediction, whether based on ultrasound evaluation, built on parameters calculated during compaction analysis, or derived from tablet images, provide information at a late stage of the technological process, when intervention on the compression blend is difficult. Such approaches are valuable during the development of solid dosage forms, however late in the case of continuous flow processes where real-time intervention in adjusting process parameters is necessary.

The current study aims to explore the processability and capping tendencies associated with fluidized bed melt granules of Clopidogrel hydrogen sulphate as a model API. Design of Experiments (DoE) and multivariate data analysis tools are used to provide a mechanistic understanding of capping occurrence in relation to available in-process controls, granulation and tableting conditions, and parameters derived from single-punch compaction analysis. We further test the ability of a videometric machine-vision system to predict key powder blend properties and discriminate granule populations that lead to capping failures. Finally, prediction of tablet critical quality attributes (CQAs) is evaluated through a data fusion approach combining machine-vision particle size profiles with granule in-process controls and process variables collected prior to the tableting step.

While the currently reported applications of machine vision are focused on the inspection of tablet defects and the prediction of the pharmaceutical properties, this study proposes its use beforehand to prevent quality issues. The novelty therefore lies not in the use of imaging per se, but in (i) the granule-stage sensor placement, (ii) the binary classification of capping tendency from videometric particle-size profiles, and (iii) the integration of these outputs in a data fusion framework with process and IPC variables.

Download the full article as PDF here Investigating and predicting the occurrence of capping for tablets prepared by melt granulation

or continue reading here

Materials

The pharmaceutical composition consisted of Clopidogrel hydrogen sulphate (MSN Laboratories Ltd., Telangana, India) as active substance and selected excipients. Mannitol 35 (Pearlitol 50C, Roquette Frères, Lestrem, France) and Cellulose, microcrystalline type M103D+ (Mingtai Chemical Co., Ltd., Taoyuan City, Taiwan) were used as fillers. Macrogol 8000 (Dow Chemical Company, Hahnville, LA, USA) and low–substituted hydroxypropylcellulose (L–HPC, LH–11, Shin-Etsu Chemical Co., Ltd., Tokyo, Japan) were used as binders. The obtained granules were lubricated with hydrogenated castor oil (Kolliwax HCO, BASF SE, Ludwigshafen, Germany).

Bela Kovács, Sonia Iurian, Alexandru Ioan Pop, Valentin Toma, Rareş Ştiufiuc, Atilla Farkas, Lajos Madarasz, Brigitta Nagy, Erzsebet Orsolya Tőkés, Katalin Zöldi, Éva Katalin Kelemen, Marcela Achim, Răzvan Cherecheş, Zsombor Kristof Nagy, Ioan Tomuţă, Tibor Casian, Investigating and predicting the occurrence of capping for tablets prepared by melt granulation, International Journal of Pharmaceutics, 2026, 127407, ISSN 0378-5173,  ttps://doi.org/10.1016/j.ijpharm.2026.127407.

Enjoy our new free webinar, registration & information here:

Developing Reliable Capsule-Based Dry Powder Inhalers