Swarm intelligence based self-monitoring of bioprocesses

• Applicant: Professor Dr.-Ing. Thomas Becker
• Subject Area: Biological Process Engineering
• Term: from 2017 to 2023
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 383534822

This project addresses a novel methodic concept in the area of bioprocess monitoring. Not a single sensor is in focus but a network of sensors. This sensor network will acquire intelligence by communication among the sensors. The project goal is the realization of a self-monitoring sensor network based on swarm intelligence. Particle Swarm Optimization (PSO) is the used algorithm. The intelligent sensor network will be capable of identifying faulty sensors and signals by means of multivariate linear and non-linear modeling approaches in comparison with other measurements of the sensor network and process information. Additionally, it will be able to replace these faulty sensor values. By this auto-compensation process control can be carried on independent of sensor failures and batches that would be lost otherwise can be finished successfully. By using the swarm intelligence idea, the functionality of each physically built-in sensor (temperature, pressure, pH, pO2, offgas CO2/O2, methanol, turbidity) can automatically be examined. This allows continuous monitoring and evaluation of process data, as well as the creation of a learning, autonomous, intelligent system with memory.Intended research results:- Establishment of a network of process-relevant sensors and information technology networking of the sensors (AP 1)- Setting up of a fully automated reactor system (AP 2), which is necessary for the application of the swarm intelligence concept- Construction of the individuals of a swarm: For each individual, a defined partial model is formulated that maps the different inputs (measured values of the sensors) to one another (AP 3)- Analysis of the suitability of the submodels used in a swarm (AP 3)- Development of a functionality that uses historical data for the automatic parameterization of a submodel (online parameter adjustment) (AP 3)- Online prediction of the trajectory of each sensor value via the whole sensor network (AP 4)- Evaluation of the measured values of each process-relevant sensor in comparison with the sensor network (AP 4)- Identification of faulty sensors and assignment of the correct error category (systemic error, drift, noise, partial / complete failure of one or more sensors) (AP 4)- Reconstruction of faulty sensor values (AP 5)- Validation of the functionality of the network in the present process by varying the sensor inputs and process conditions (AP 6 and AP 7)- Assessment of information redundancy within the sensor network (AP 8)- Analysis of the transferability of the system to other fermentations (AP 8)

• DFG Programme: Research Grants