Biomedical Engineering Reference
In-Depth Information
The applications of ANNs in process industries range from multisensor data
interpretation in chemical processes [
65
], through NIR spectra calibration with
RBF [
56
] to enzyme engineering [
32
]. Most of these publications demonstrate the
applicability of ANNs as a non-linear modelling tool in the bioprocess area, but
often also raise issues with model parsimony and the ability to deal with unseen
data. The latest developments thus combine these 'black box' approaches with
first-principles modelling (see Chap. 6). In such hybrid schemes, biological
understanding of the first principles is complemented by MVDA models capturing
the dynamic characteristics of the system to improve the overall performance of
the hybrid modelling framework [
5
,
13
,
55
].
5 Conclusions
This chapter set out to provide a brief overview of MVDA methods used to
advance the interpretation of measurement and monitoring data from a bioprocess.
Methods used for exploratory data analysis, clustering and classification together
with regression methods were described with a brief description of the funda-
mental characteristics of the most frequently used linear and non-linear methods in
each category. Readers are referred to relevant literature for more details on any
particular technique. The bioprocess case studies used in this chapter have been
selected to demonstrate the areas of successful application throughout bioprocess
development and monitoring and control across typical unit operations encoun-
tered in bioprocess manufacturing.
Issues with regards to industrial applicability were highlighted, where relevant,
and it is important to note that significant progress in the industrial implementation
of MVDA techniques within the PAT and QbD frameworks remains to be realised.
However, continuous developments in the robustness and reliability of sensors and
data analysis techniques provide the necessary requirements for much wider take-
up and implementation of such frameworks within the (biopharmaceutical)
industry. Whilst the highly regulated nature of this particular sector may appear to
preclude the application of certain types of MVDA methods, it is important to
emphasise that the regulatory environment has advanced in this respect over the
past decade and fundamental process understanding and risk-based approaches to
process development and operation are actively encouraged. The MVDA methods
described above represent a useful tool for achieving this goal and have already
been proven to enhance bioprocess performance at varying scales of manufac-
turing process. More sophisticated measurement techniques will push further the
boundaries of MVDA method development, which in turn will lead to wider
acceptance of these techniques by the industry.
Search WWH ::
Custom Search