Applied Statistical Consultants 12 Wildwood Drive Springvale, Maine 04083 Contact Us

Electronics Case Study   |   Neural Networks Case Study

Case Study: Neural Networks Application

Background:
During the manufacture of an injectable biopharmaceutical product there is the need to titrate the product with an acetic acid solution.  The titration process is a critical process step in maintaining the product's integrity through the remainder of the process.  Titration of the product with the acetic acid is done in batches costing over $1M on average.

Problem:
The titration process required an iterative process to lower the pH to the desired target.  At the start of the titration process the operators would make an "educated guess" as to how much acid was required to achieve the final pH without adding too much.  A typical scenario is described in the following process flow chart:

Such an iterative process presented two problems:  First, every sample taken exposed the product to contamination.  Second, the iterative process took a considerable amount of time.  An average batch took four to six cycles before the target pH was achieved.  It was not uncommon for this stage of the process to take over an hour of valuable production time.

ASC Solution 

Goal:
Add acid in a single titration cycle by predicting how much acid to titrate to achieve the target pH. 

Objective:
Develop a "user friendly" method of collecting and archiving the information required to make accurate predictions of the amount of acid to titrate.

Results:
Historical data was used to define a neural networks model.  The model was validated using data points removed from the model definition.  Residual error related to the model was minimized using Response Surface Methodology (RSM).  Model was robust to a wide range of cooperating conditions (tank volume, initial pH, etc). 

A database utility was developed to collect and archive the data related to the predictor variables.  In addition to collecting data, the database made the necessary predictions for titrating acid to achieve the target pH. 

The new process does not require an iterative process (see below).  All of the titration is done with a single addition.  Since the process is cGMP, a pH test is still required for verification.  The new process saves about an hour of process time and limits the exposure of the product to contamination.

► Electronics Case Study