Inadequate mixing quality shows up either directly in the product or indirectly through unstable process signals. For reliable inline monitoring, direct, product-near measurement methods such as NIR or Raman spectroscopy are combined with indirect process data such as torque, power consumption, temperature and flow. A single sensor signal is generally not sufficient to reliably assess the homogeneity of an entire batch; what matters is a validated measurement concept with a representative sensor position, suitable reference analytics and clearly defined acceptance criteria.

Signs in the product

An obvious sign is spatial differences in concentration of individual components. In powders, these can appear, for example, as uneven active ingredient, additive, colour or moisture content. Visible indications include colour differences, streaks, striping, specks, agglomerates, dry or over-wetted pockets, and unevenly distributed liquid fractions. In suspensions or pastes, sediment, uneven particle distribution, phase separation, undissolved agglomerates or deviating consistency can occur.

Further indications often only emerge in subsequent process steps. These include fluctuating bulk density, deviating flow behaviour, unstable dosing, segregation during discharge, uneven tablet weights, fluctuating tablet hardness, changing extrusion pressures, varying coating quality or uneven reaction control. In solid mixtures with strongly differing particle sizes, bulk densities or shapes, segregation can occur even when the mixture was initially homogeneous in the mixer. Assessment should therefore not be carried out only in the mixing chamber, but where appropriate also during discharge and at critical points in the subsequent process chain.

Process signals

Indirect process data are well suited to detecting deviations and monitoring reproducibility. These include torque, power consumption, rotational speed, product temperature, pressure, flow and, where applicable, vibration or acoustic signals. In powder and paste mixers, for example, a sudden increase in torque can indicate agglomeration, increasing cohesion, uneven wetting or a faulty raw material addition. If torque falls unexpectedly, the product structure, fill level or moisture may have changed. A stable torque value alone, however, does not yet prove chemical or material homogeneity, because different product states can produce the same mechanical signal.

Temperature profiles can provide additional indications in temperature-controlled, reactive or friction-heated processes. Local temperature differences, unusual heating rates or unstable temperature profiles can indicate insufficient mass transfer, uneven liquid distribution or unwanted reactions. In continuous plants, pressure and flow fluctuations can be used to detect dosing deviations, blockages, bridging or unstable conveying performance. Such signals are particularly valuable when assessed within a statistical process control framework using target values, trend limits and alarm strategies.

Direct inline measurement

For powder mixtures, near-infrared spectroscopy, or NIR for short, is one of the most important methods for inline or online monitoring. An NIR probe measures reflected light at a defined point in the mixer, on a process line or at the discharge. From the spectra – once a chemometric model has been built and validated – concentrations of individual components, moisture content and the time course of homogenisation can be derived. Spectral variation typically decreases during mixing and, ideally, reaches a stable level. NIR has been successfully used for real-time monitoring of the homogeneity of pharmaceutical powder mixtures; one study showed that the mixture was already homogeneous well before the usual blanket mixing time had elapsed.

Raman spectroscopy is particularly suitable when the components to be monitored have distinctive Raman signals and can be sufficiently distinguished spectrally. It can be used for the quantitative determination of individual components, such as an active ingredient, and for assessing homogeneity in real time. Studies show that Raman spectroscopy, with suitable chemometric models, can determine the mixing endpoint of powder mixtures inline. In continuous pharmaceutical processes, Raman has also been used for real-time measurement of active ingredient content and homogeneity, and for feedback control of dosing.

For liquids, suspensions, pastes and emulsions, conductivity measurement, pH value, inline viscosity, torque, ultrasound, turbidity, refractometry or optical image analysis are additionally suitable. Conductivity and pH are particularly suited to the distribution of dissolved ionic components. Turbidity, scattered light and ultrasound can support the dispersion of solids or the homogeneity of an emulsion. Inline viscosity measurement is useful for liquids, suspensions and pastes, but not as a characteristic value for the flowability of dry powders. For dry powders, cohesiveness, bulk density, flow function, moisture and particle size distribution are the more suitable product-related characteristic values.

Defining the mixing endpoint

The mixing endpoint should not be defined as a mere point in time, but as a validated process window. With spectroscopic methods, it is often checked whether the spectra or the predicted concentrations remain stable within defined limits over a set period. A common approach is the moving standard deviation of consecutive spectra, known as the moving block standard deviation. If it falls below a validated threshold and remains stable, the process can be assessed as steady. For actual homogeneity, it must additionally be demonstrated that the measured concentration corresponds to the target value.

With low-dose mixtures, the inline method must be expressly suitable for the low-concentration critical component. Otherwise, an apparently stable NIR signal may merely confirm the even distribution of the main components while the low-dose active ingredient remains unevenly distributed. The FDA points out that endpoint criteria for low-concentration components must specifically reflect their even distribution. The spectral mixing endpoint should therefore be confirmed with an independent reference method, for example a validated blend uniformity analysis.

For a robust practical solution, a graduated concept is recommended: first, several mixing times are tested as part of development trials and samples are analysed from different positions as well as across the discharge. Inline spectra and process data are recorded in parallel. A chemometric model, for example using partial least squares or principal component analysis, is then calibrated against the reference analytics. This produces clear criteria for the mixing endpoint, permissible signal variance, concentration range, alarm limits and the required hold time in the stable state. The method should also work reliably across different raw material batches, fill levels, moisture contents and operating conditions.