Online monitoring of bioprocesses via multivariate sensor prediction within swarm intelligence decision making

• self-organizing sensor network capable of getting over sensor failures
• correlations between raw data from several sensors used for multivariate statistical process control trajectories
• models scored by swarm intelligence (PSO) leading to the optimal sensor/model combination at certain time step
• determination of the fermentation trajectory in combination with sensor output validation
• more robust online monitoring to insure optimal and timely effective processing as well as sensor failure detection

In this work, a methodology combining process knowledge with computational efforts is presented. The aim is to create a self-organizing sensor network capable of getting over sensor failures. First, multivariate linear and non-linear models are utilized creating a search space based on the multi-sensor data pool. Simple correlations between the raw data retrieved from several sensors are used for extracting multivariate statistical process control trajectories. Those different models are scored by swarm intelligence (particle swarm optimization) leading to the optimal sensor/model combination at certain time step aiming at determination of the fermentation trajectory in combination with sensor output validation.The results on online data indicate the possibility of more robust online monitoring using the swarm sensing idea for biotechnological processes to insure optimal and timely effective processing as well as sensor failure detection. Adjustments of the basic algorithms, cost function, accuracy of output as well as the dynamic behaviour are addressed. This methodology is not restricted to the number of sensor inputs as well as the use of specific sensor readings, which makes it beneficial over other approaches.