(438c) Multivariate Monitoring of a Continuous Manufacturing Process for API Synthesis: Enhancing the Power of Real Time Data | AIChE

(438c) Multivariate Monitoring of a Continuous Manufacturing Process for API Synthesis: Enhancing the Power of Real Time Data


Multivariate monitoring of a continuous manufacturing
process for API synthesis: enhancing the power of real time data

Melanie Dumarey, Product Development and Supply, R&D,
GSK (melanie.x.dumarey@gsk.com)

Martin Hermanto, SGP Technical Development (TD), Global
Manufacturing Site, GSK (martin.x.hermanto@gsk.com)

The pursuit of efficient manufacturing approaches has
increased the implementation of continuous processing technologies within the
pharmaceutical industry as they can enhance the supply chain flexibility,
reduce the environmental and spatial footprint and deliver more consistent
quality. An additional advantage is the enhanced opportunities for process
monitoring delivering high volumes of data which can be linked by multivariate
tools to assure consistent product quality and optimal process operability.
Traditionally only a few process sensors, e.g. those recording controlled
process parameters, are monitored proactively and fitted with an alarm, whilst
many others are only studied in case of an anticipated problem or as part of a
root cause analysis.

In the presented case study, multivariate modelling tools
were applied to exploit the wealth of process data generated during the
synthesis of an active pharmaceutical ingredient (API) combining five chemical
transformations in a continuous flow process. Not only the controlled process
parameters as flow rate and temperature were monitored, but also not quality
critical sensors such as pump speeds and pressures were included. In total 69
process sensors measuring flow rates, temperatures, reactor pressures, pump
suction and discharge pressures, pump speeds and conductivities were summarized
by multivariate models. The aim was to detect and prevent upcoming process
failures based on the real-time multivariate trends. The main technical
challenge was balancing sensitivity and robustness of the implemented models
ensuring significant process changes were detected and minimizing false alarms
requiring interventions by the plant operators.

Multivariate monitoring was extensively tested during
development campaigns at the R&D facilities, where on several occasions it
successfully detected upcoming process failures (e.g. upcoming pump blockage)
before activation of the univariate quality-limiting alarm. In the example
visualized in Figure 1 an upcoming pump blockage was detected based on the
multivariate trend, which flagged an increase in pump speed approximately one
hour before the flow rate was exceeding the defined limits. As the process
remained within the proposed control ranges before the flow rate excursion, the
product quality was not affected at the time of the multivariate model
excursion and corrective actions could have been taken to prevent process failure.
Secondly, multivariate models of the generated data were used to perform a
retrospective root cause analysis of the recurrent blockage of a reagent
supply. The outcome of this analysis, i.e. the blockages originated from supply
and not from reactor, facilitated further investigations by narrowing down the
possible causes.

Figure 1: Flow chart for the multivariate monitoring
of a primary continuous process displaying symptoms of an upcoming process
failure (pump blockage)

Based on these encouraging results, the informatics platform
and modelling approach were transferred to the manufacturing site, where it
detected other patterns indicative of process drift. In the future, the same
methodology will be implemented as part of the predictive maintenance strategy
for commercial manufacture. More specifically, any significant changes in the
process trend will trigger a well defined action, e.g. a pump switch if standby
pumps available, or a controlled shut down if no corrective action is feasible.
These early interventions aim at increasing manufacturing efficiency by
avoiding divert to waste of the process stream or preventing a forced shut

The application described in this paper supports GSK’s
intent to increase the use of advanced analytics and modelling in order to
enhance quality by design development and improve quality assurance and
efficiency of manufacturing processes. As part of the same intent, online HPLC
capability was developed for the first primary continuous process introduced at
GSK’s commercial manufacturing plant (specifically Singapore), where it is now
used for real time process monitoring, continuous improvement and support of
investigations triggered by the quality management system. Enhanced monitoring
combining online HPLC data and multivariate modelling tools is currently under
investigation. All these new technologies introduced significant improvements
to process understanding, process robustness and quality assurance. GSK’s
longer term aim is to develop reliable and accurate tools to support real time
release of API or even to implement real time control models enabling process
optimization in real time, which would further improve product quality and
manufacturing efficiency.


Peter Shapland, Hannah Robinson, Peter Hamilton and Irene
Areri for keeping the continuous primary process and supporting platforms
running and helping to interpret observed process changes.

Ian Barylski for the informatics support enabling the new

Malcolm Berry for leading the project and supporting


This paper has an Extended Abstract file available; you must purchase the conference proceedings to access it.


Do you already own this?



2017 Annual Meeting
AIChE Members $150.00
AIChE Graduate Student Members Free
AIChE Undergraduate Student Members Free
Non-Members $225.00
Pharmaceutical Discovery, Development and Manufacturing Forum only
AIChE Members $100.00
AIChE Food, Pharmaceutical & Bioengineering Division Members Free
AIChE Graduate Student Members Free
AIChE Undergraduate Student Members Free
Non-Members $150.00