• Accumulation
• PHA
• Polyhydroxyalkanoates
• Process
• Sensor
• Turbdidity

Polyhydroxyalkanoates (PHAs) are bacterial polymers that are gaining interest in recent years because of their versatile properties and their biodegradable nature. Efforts amongst researchers and companies have been focused on the optimization of the production process in order to reduce the price of the final polymer, which currently is not competitive with fossil-based polyolefins with similar properties such as polyethylene (PE) and polypropylene (PP).
Within the Wetsus facility, the production of PHA from activated sludge is carried out at pilot scale. Key process parameters and further optimization of the accumulation and extraction steps are evaluated to facilitate the upscaling. Regarding the accumulation, a well-established fed-batch process based on previous investigations is already employed. However, the current practice has still a bottleneck with respect to choosing the duration of the accumulation phase.
To date, the process is carried out for 20 hours. Such method does not consider differences in biomass activity from batch to batch and can lead to overextended accumulations. When the run is carried out for too long, the process is less efficient because of the higher feed and energy consumption. The polymer quality is also affected since longer runs produce a PHA with a lower molecular weight. Thus, the focus of this work was to establish a new approach to online monitor the process and choose the right time to stop the accumulation. To accomplish this, a soft sensor was developed based on the data collected from an online probe. Soft sensors are tools that focus on the process of estimation of any system variable or product quality by using mathematical models, substituting some hardware sensors, and using data acquired from other available ones.
As a solid foundation for the experiments, a preliminary test on non PHA-rich biomass was performed to establish the main parameters affecting the turbidity readout. It was found out that out of all factors considered (position of the probe, presence of light and speed of mixing), the speed of mixing was the only one that has a detectable influence, hence a particular attention was focused on this parameter in further experiments. The probe was then used to follow several accumulation runs in order to collect data on the turbidity response overtime.
The turbidity signal from the probe was found to be related to three main parameters of the PHA accumulation run: biomass, polymer, and inorganics concentrations. From the experimental data collected, a multivariable regression equation was created to predict the turbidity trend during the process. The model was then retrofitted on a set of 16 old accumulation runs with no turbidity data in order to create a historical dataset. This step enabled the creation of a data-driven soft sensor, a tool that helps process monitoring and control by mathematical estimation of process variables. A non-linear regression was applied on the historical data in order to model the increase in polymer content as a function of the online turbidity signal.
With the development of an online estimator for PHA production is then possible to predict the PHA content based on the turbidity signal and to choose the endpoint of the accumulation when a predefined polymer content is reached. Further steps are required to acquire more data in order to strengthen and fine tune the output model. However, the insights presented in this work can be considered as a valuable tool that improves process efficiency and facilitates the pathway towards industrial scale application.