A sensor-based endpoint determination replaces a fixed mixing time with a demonstrated criterion for the process state achieved. It only works if the sensor signal has been shown to be meaningful for the specific product, the mixer and the relevant quality requirement. A stable torque or spectral signal alone does not automatically prove homogeneity or releasability.

Defining the measurement task

First, it must be clearly established what the endpoint means. Depending on the process, this can be an adequate distribution of active ingredient, moisture, concentration, viscosity, particle structure, wetting or chemical conversion. This target variable is referred to as a critical quality attribute. It is then established which process variables influence the endpoint, for example mixing time, rotational speed, fill level, dosing sequence or temperature.

For dry powder mixtures, near-infrared spectroscopy is often a suitable method for assessing concentration, moisture or – with sufficient calibration – homogeneity. Raman spectroscopy can capture chemical or crystalline characteristics for certain substances. Torque, motor current or power consumption provide more indirect indications of product behaviour, agglomeration or wetting. They are only suitable as an endpoint criterion if their relationship with the target quality has been demonstrated in the respective process.

Conductivity, pH value, turbidity or viscosity are relevant above all for liquids, pastes, suspensions or solutions. They are generally not suitable primary signals for dry powders. Acoustics, ultrasound and image processing can be helpful in individual specialist applications, but are usually considerably more demanding to develop and validate for a robust mixing endpoint.

Integrating the sensor correctly

The measuring point must supply representative process information. A probe at a locally preferentially mixed point can create too optimistic a picture; a probe in a dead zone can cause unnecessarily long mixing times. The installation location should therefore be chosen on the basis of an understanding of flow, preliminary trials and comparative measurements. With powders it must also be checked whether windows become soiled, whether the product adheres to the probe, or whether changing fill levels affect the measurement signal.

Sensors must be designed to be pressure-, temperature- and abrasion-resistant and, where applicable, suitable for explosion protection. They require a hygienic, cleanable integration, suitable seals and a clear plan for cleaning, calibration, functional testing and replacement. Cleaning in Place means cleaning in the installed state without extensive dismantling; the common abbreviation is CIP. Sterilization in Place means sterilisation in the installed state; the abbreviation is SIP. If a probe cannot be reliably cleaned or tested, it is only of limited suitability for hygienically or regulatorily critical processes.

Data transmission to the programmable logic controller, process control system or Manufacturing Execution System must occur with correct timing and be traceable. A programmable logic controller is an industrial computer for controlling machines and processes. A Manufacturing Execution System supports production control and batch documentation. The sampling rate should be derived from the signal dynamics. Higher data rates are not automatically better; they generate more data and noise without necessarily improving the endpoint decision.

Developing the model and limit values

The endpoint is usually developed through series of trials with the original product. During these trials, sensor data are recorded together with suitable laboratory references, for example assay determination, moisture measurement, particle analysis or samples for homogeneity testing. High Performance Liquid Chromatography, or HPLC, is, for example, a laboratory method for the quantitative determination of specific chemical substances. Karl Fischer titration is an established method for determining water content.

A simple, transparent endpoint criterion or a chemometric model can be developed from the data. Chemometrics combines chemical measurement data with statistical methods. These include, for example, Principal Component Analysis, or PCA, for representing patterns in complex data, and Partial Least Squares, or PLS, for calibrating a relationship between spectra and reference values. A complex model is not automatically better. For routine operation it should be robust, traceable, maintainable and sufficiently stable against typical raw-material and process fluctuations.

An endpoint criterion could, for example, be that a calibrated concentration or moisture model lies within a defined target range and no longer changes materially over a defined time window. A moving standard deviation can also be used if it correlates with actual quality data. A mere approach of the first or second signal difference to zero is not, by itself, sufficient proof of quality.

The limit values must be set in such a way that measurement uncertainty, raw-material variation, sensor soiling, measurement point and time delay are taken into account. There should also be a clear approach to invalid or missing signals. If the measurement fails, the control system must not release a batch or end the process in an uncontrolled manner. Suitable options are defined fallback strategies, for example a conservative maximum permissible mixing time, an alarm and a qualified decision by operating personnel or quality assurance.

Automating and safeguarding

Once successfully developed, the endpoint criterion can be integrated into the control system. The mixer can then stop when the criterion is met, switch to discharge or generate a message for review. In quality-critical applications, a semi-automated logic is often more sensible than fully autonomous release: the system recognises the probable endpoint, documents data and deviations and, where necessary, requires confirmation from qualified personnel.

The integration must be qualified and, depending on industry and risk, validated. This includes Design Qualification, Installation Qualification, Operational Qualification and Performance Qualification. Design Qualification documents the suitability of the concept. Installation Qualification confirms correct installation. Operational Qualification checks functions within the intended range. Performance Qualification demonstrates performance in routine operation. In addition, the sensor, data processing, model versions, alarm limits, user rights, audit trails, backup and change management must be controlled.

Process Analytical Technology, or PAT, is described by the FDA as a system for designing, analysing and controlling manufacturing through timely measurement of critical quality and performance attributes of raw materials, in-process materials and processes. The FDA regards PAT as a way to improve process understanding and product quality, not as an obligation or an automatic release strategy.

Real Time Release Testing, or RTRT, can be supported by validated process data. However, it does not mean that mixing sensor technology alone replaces batch release. Under ICH Q8, batch release remains an independent decision based on test results, manufacturing records, GMP conformity and the quality system. RTRT can supplement or replace individual end-product tests where this is regulatorily justified and approved; it does not necessarily replace all end-product testing.