Issue link: https://www.e-digitaleditions.com/i/1420709
46 Pharmaceutical Technology APIs, EXCIPIENTS, AND MANUFACTURING eBOOK 2021 P h a r mTe c h . c o m The toolbox can be integrated with sensor de- vices and real-time monitoring software to col- lect data from the plant. The prediction and the diverge signal can be sent to control platforms to communicate with the reject gate or a direct com- munication with the plant can be established. For RTD model fitting, a measured pulse or step re- sponse of certain inlet concentrations is needed as inputs. Fitting will be done by the program auto- matically with RTD model parameters as outputs. Real-time prediction of tablet content uniformity An RTD model is used to predict the outlet con- centration of the tablets from a continuous phar- maceutical manufacturing plant based Equation 1: !"#$%&'()*+) ! !"# " # $ %&! $% " # $ '("#$!! ! !"#$%&'% !"#$%&'(),+) ( " #)# & $ % '#(# ! ) "#$ ' %*+ ) ,- % " . " * ' #"&'(' ! ) % ) &! !"#$%('% !"#$%&'()*+) ! !"# " # $ %&! $% " # $ '("#$!! ! !"#$%&'% !"#$%&'(),+) ( " #)# & $ % '#(# ! ) "#$ ' %*+ ) ,- % " . " * ' #"&'(' ! ) % ) &! !"#$%('% [Eq. 1] where E(t) is the pulse response of an RTD model, C in is the real time input concentration, and C out is the outlet concentration. This RTD-based control toolbox keeps computing convolution of the inlet concentration with the pre-fitted RTD model. The inlet concentration should be a real-time updated .csv file, which can be generated by a real- time prediction tool (e.g., Process Pulse II [CAMO]). The real-time prediction tool should receive inlet raw data (spectra) from near infrared sensors and update the input file frequently, less than one data point per second. Higher frequency is supported, but modification of the code is needed. Once the direc- tion and filename are selected, the RTD-based control toolbox will monitor the .csv file and read new inlet concentration data points from the file. Similar to the inlet concentration, the RTD model parameters should be stored in a .csv file. Parameters fitted in an RTD model fmodule can be read as well as other parameters generated from different sources. After reading all inputs, the RTD-based control toolbox convolutes the data point with the RTD model. According to the properties of convolution, adding the convolution results of each input data point in time series will lead to the convolution of the whole input signal. Each time the prediction of outlet concentrations will be compared with the preset upper and lower limits, before running the program. If the prediction falls within the limitation, the reject signal of tablets will be 0, meaning that the tablets produced are qualified. Otherwise, the reject signal will be 1, and the rejection gate will reject the produced tablets. RTD modeling and experimental data The tank-in-series modeling approach has been used to develop this toolbox. The module is used to fit an RTD-model based on Equation 2: !"#$%&'()*+) ! !"# " # $ %&! $% " # $ '("#$!! ! !"#$%&'% !"#$%&'(),+) ( " #)# & $ % '#(# ! ) "#$ ' %*+ ) ,- % " . " * ' #"&'(' ! ) % ) &! !"#$%('% ) !"#$%&'()*+) ! !"# " # $ %&! $% " # $ '("#$!! ! !"#$%&'% !"#$%&'(),+) ( " #)# & $ % '#(# ! ) "#$ ' %*+ ) ,- % " . " * ' #"&'(' ! ) % ) &! !"#$%('% ) [Eq. 2] Here, E(t+t d ) is the response signal, n is the number of tanks that need to be determined, τ and t d are fur- ther parameters which need to be determined. Among them, τ is the mean residence time and t d stands for the delay time. To determine RTD parameters, an experi- ment was performed to generate the outlet concentra- tion response of a pulse or step input concentration. Manufacturing The toolbox can output a diversion signal for the rejection gate, to divert tablets containing components outside of the preset limits.