Page 10 - Industrial Technology February 2020
P. 10


                                             DEEP LEARNING

                                                                       CONDITION MONITORING

                                             MARTIN GADSBY, DIRECTOR AT OPTIMAL INDUSTRIAL TECHNOLOGIES,
                                             LOOKS AT HOW DEEP LEARNING-POWERED PROCESS ANALYTICAL
                                             TECHNOLOGY CAN BOOST MANUFACTURING PROCESSES
                                                    eep  learning  has  the  potential  to  revolutionise  a  relevant  information  from  the  in-line  PAT-led  measurements
                                                    broad  range  of  industries  by  offering  increasingly  on the product’s chemical and physical make-up.
                                                    accurate  predictive  capabilities  with  little  to  no  When  deep  learning  is  applied  to  PAT,  it  is  not  only
                                             Dhuman supervision. These can have a tremendous  possible to simplify the creation of a predictive algorithm, as
                                             impact  on  the  manufacturing  sector  by  supporting  the  no  coding  is  required,  but  the  resulting  model  could  also
                                             application  of  process  analytical  technology  (PAT)  and  its  improve  as  manufacturing  plants  develop  more  and  more
                                             concomitant  increase  in  process  performance  and  product  products. As larger volumes of process and quality data are
                                             quality.                                    generated, the system can use them to determine additional,
                                                Deep  learning  is  a  highly  flexible  and  adaptive  artificial  less obvious connections between data. As a result, industries
                                             intelligence  tool  that,  when  exposed  to  new  datasets,  can  can  build  a  futureproof  processing  unit  that  continuously
                                             increase  its  ability  to  identify  patterns  and  classify  upgrades process efficiency and product quality without the
                                             relationships  between  data.  This  means  that  the  larger  the  need to re-programme the modelling algorithm.
                                             volume of data fed into a deep learning-generated predictive
                                             model, the higher the probability that the system will create  Process orchestrators to stay in control
                                             more  accurate  and  precise  forecasts.  Furthermore,  the  As  this  technology  becomes  established,  PAT  knowledge
                                             evolution  of  the  model  is  automatic,  ie  no  programming  or  management  platforms  take  on  a  more  important  role.  As
                                             other actions from human operators are required.  larger volumes of data are being generated and ANNs analyse
                                                These unique capabilities are enabled by artificial neural  them  without  offering  any  insight  into  their  prediction
                                             network  (ANN)  architectures  that  mimic  the  human  brain.  generation process, it is important for manufacturers to have
                                             ANNs  are  collections  of  interconnected  artificial  neurons  or  a  clear  overview  of  what  is  happening  on  the  factory  floor,
                                             nodes organised in layers. Each neuron receives an input with  what the real-time multi- and uni-variate data looks like, and
                                             data  to  analyse  and  automatically  performs  different  how the ANN models are evolving.
                                             computations  on  it  without  the  need  for  any  rule-based  Therefore, by implementing a PAT knowledge manager, it
                                             programming.  The  resulting  output  is  then  sent  to  another  is possible to monitor and respond quickly to the presence of
                                             node for further processing. Every time an input is fed to the  anomalies  or  when  the  predictive  model  is  ceasing  to
                                             ANN,  the  system  may  be  able  to  notice  new  correlations  represent the input data, ie ANNs have identified correlations
                                             between data and implement them into its predictive model.  that are not relevant or unrealistic.
                                             An  extremely  advanced  ANN  may  even  be  able  to  find  out  One  of  the  most  advanced  PAT  knowledge  management
                                             interdependencies that are not known to human experts, thus  platforms on the market is Optimal’s synTQ, its efficacy having
                                             delivering forecasts with unprecedented accuracy.  been  proven  worldwide  by  several  of  the  world’s  largest
                                                                                         pharmaceutical and life-science organisations. By choosing a
                                             Adopting quality by design strategies       software  solution  like  this,  manufacturers  can  rely  on  a
                                             Deep  learning’s  abilities  make  the  technology  a  potentially  platform that is able to interact with cutting-edge technologies
                                             powerful  ally  for  manufacturing  industries  that  adopted  and  methods,  such  as  deep  learning,  as  soon  as  they  are
                                             Quality  by  Design  (QbD)  strategies  and  PAT.  These  two  available. In addition, synTQ offers a robust and user-friendly
                                             operational  process  methodologies  rely  heavily  on  in-depth  interface  to  keep  both  product  development  and  scale-up
                                             process understanding in order to maximise the efficiency of  production organised at all times. In this way, manufacturers
                                             the overall production process. In fact, knowing how critical  can remain in control as they improve plant efficiency, product
                                             process  parameters  (CPPs)  affect  products’  critical  quality  quality and consistency.
                                             attributes (CQAs) is essential in order to control the different  As  deep  learning  applications  gain  popularity  in  quality
                                             processes in real-time and obtain products that meet elevated  prediction,  offering  a  unique  tool  to  boost  competitiveness,
                                             quality standards.                          PAT  knowledge  management  software  products  like  synTQ
                                                The relationships between CPPs and CQAs are assessed by  can  provide  the  key  to  successfully  implementing  these
                                             means of multivariate analysis (MVA) and chemometrics, ie by  strategies and driving productivity.
                                             using  mathematical  and  statistical  procedures  to  extract  MORE INFORMATION:

                                                                                                   INDUSTRIAL TECHNOLOGY • February 2020
   5   6   7   8   9   10   11   12   13   14   15