(301a) The Use of Linear Regression and R2 in Continuous Manufacturing to Dictate When Loss-in-Weight Data Is Acceptable to Control Feeding Following Feeder Refill

Continuous manufacturing is becoming an ever more popular approach to pharmaceutical manufacturing because it provides many advantages compared to conventional batch manufacturing. Continuous manufacturing utilizes feeders to constantly add material, at the appropriate concentration, to the system through rotating screws. This fed material will end up downstream in tablet or capsule form. One of the major advantages of continuous manufacturing is the ability to continuously monitor and control the amount of each material that is being fed into the system. This is accomplished by monitoring the weight loss in the feeders over time. The loss-in-weight data provides the feed rate (kg/hr) of the material. Maintaining the target feed rate of each material throughout production is critical to meeting the critical quality attributes of the product.

By monitoring the real time weight data the control system can make updates to the screw speed to ensure the target feed rate is being maintained. However, the feeder hoppers have a limited capacity so material needs to be added at regular intervals to refill the hoppers throughout the batch. While the refill is occurring, the system loses the ability to control and monitor the feed rate. Hence, it is imperative for the feeder to return to loss-in-weight feeding mode using real time weight data as quickly as possible.

A criteria is required to determine at what point the weight data coming into the system can be relied upon to control the feeders through loss-in-weight. The continuous manufacturing line at Thermo Fisher Scientific continuously monitors the weight loss using linear regression to determine the feed rate and how well the actual data matches the predicted data to provide continuously updated coefficient of determination (R²) throughout the feeding process. While feeding in loss-in-weight mode the R² value remains close to 1. However, when in refill the value decreases to approximately zero. Once the refill has been added the slope of the weight loss begins to stabilize and the R² begins to increase. Thermo Fisher Scientific has implemented an approach that relies on the R² value obtained through linear regression of the weight loss to dictate when the loss-in-weight data coming into the system is once again acceptable to control the feed rate. Once, the R² value reaches its designated set point the system will change from volumetric to loss-in-weight feeding.

Studies will be presented examining different materials, feeder sizes, and refill methods. Feed rate, weight, screw speed, and R² data will be presented before, during, and after refills to show the behavior of the feeders throughout the entire process. The data will show how using linear regression with an R² set point standardizes the criteria to return to the desired state of loss-in-weight feeding following feeder refill while removing material and refill method variability inherent to the process.