Choose Language Hide Translation Bar

Scaling up the Use of Machine Learning in Chemical Process Industries (2023-EU-30MP-1346)

Solvay was the first company to succeed in industrializing the production of soda ash in 1863, a product that only requires salt and limestone as raw materials. However, behind this simple idea is an intricate continuous process able to handle reactant solids, liquids, and gases. Similarly, using data to bring value needs a deep understanding of any manufacturing process behind it. This presentation will showcase how Soda Ash at Solvay is scaling up the use of data-driven techniques in the chemical industry. To succeed, we trained our subject domain experts (process engineers) to use JMP and its predictive analytics capabilities to accelerate daily tasks such as monitoring and root-cause analysis. We will discuss our open-source JMP add-in to connect to industrial historians (Aspentech IP.21 and OSIsoft PI), the current training program, and the lessons learned in this digital transformation journey.

 

 

Hello  all.  I'm  David  Paige,  I'm  the  Digital  Champion  of  the  Global  Business  Unit  of   Soda Ash & Derivatives  at  Solvay .  Together  with  me,  we  have  Carlos  Perez,  who  is  Industrial  Data  Scientist  at  Solvay  at  corporate  level,  who  will  be  co- presenter  of  this  presentation.

This  presentation  is  about  the  scaling  up  of  the  use  of  machine  learning  techniques  in  the  chemical  process  and  concretely  at  Solvay.  Here  in  this  slide,  we  have  the  agenda  of  this  presentation.  First  of  all,  I  will  share  with  you  a  brief  introduction  of  our  multinational  company,  Solvay. T hen  some  words  also  about  our  general  business  unit,   Soda Ash & Derivatives.

Then  here  in  the  point  number  three,  we  will  enter  in  discussion  about  how  machine  learning  techniques  helps  improve  our  production  process.  Then  we  will  go  a  little  bit  deeper  about  the  usage  of  JMP  in  our  GBU.  I  will  explain  to  you  the  awareness  sessions  and  the  training  that  we  provided  to  our  population  of  engineers.  A lso,  we  will  see  a  couple  of  practical  use  cases.

Then  my  colleague  Carlos  will  share  with  you  one  add- in  that  they  developed  internally  at  corporate  level  at  Solvay,  which  is  very  useful  for  us,  for  the  final  users  to  connect  to  our  main  source  of  data,  which  is  the  MES,   manufacturing execution systems.  Finally,  I  will  share  with  you  the  main  challenges  that  we  faced  during  this  journey  and  also  the  lessons  learned.

Brief  introduction  of  our  group,  Solvay.  We  are  a  science  company  founded  in  1863  whose  technologies  bring  benefits  to  many  aspects  of  daily  life.  Our  innovative  solutions  contribute  to  a  safer,  cleaner,  and  more  sustainable  product  found in homes,  food,  and  consumer  goods,  planes,  cars,  batteries,  smart  devices,  health  care  applications,  and  water  and  air  purification  systems.

Very  important,  our  group  seeks  to  create  sustainable  shared  value  for  all.  Notably,  through  its  Solvay  One  Planet  program,  we  have  three  pillars:  protecting  the  climate,  preserving  natural  resources,  and  fostering  better  life.

Here  at  the  bottom  of  the  slides,  you  can  see  some  key  figures  of  the  group  in  2021.  As  you  can  see,  we  have  a  little  bit  more  of  21,000  employees  all  over  the  world.  We  have  presence  in  63  countries  and  we  have  98  industrial  manufacturing  sites.

Now,  if  we  jump  to  our  general  business  unit  of   Soda Ash & Derivatives,  which  is  the  business  unit  that  I  work  for,  coordinating  the  implementation  of  what  we  call  the  digital  transformation  initiatives.  As  you  can  see,  we  have  11  production  sites  distributed  around  the  world.  We  have  six  production  sites  here  in  Europe,  but  also  we  have  two  production  sites  in  North  America  and  one  production  site  in  Asia  out  of  other  locations  around  the  world  like  warehousing  and  buildings.

Also,  we  have  three  R&I  centers  located  in  Brussels,  in   Dombasle,  which  is  our  manufacturing  site  in  France,  and  in  Torrelavega  in  the  north  of  Spain.  Globally,  we  are  3,200  employees  all  over  the  world.

Our  products.  These  are  the  two  main  products  that  we  produce  in  our  general  business  unit.  We  produce   soda ash  and  we  produce  sodium  bicarbonate.  The   soda ash  is  mainly  used  for  the  glass  manufacturing  production,  different  types  of  glass  for  building,  but  also  for  photovoltaics  panels  or  for  containers,  as  you  can  see  here,  the  example  of  the  bottles.

Also, soda ash  is  used  to  produce  detergents  and  very  important  and  very  new  with  the making of the   [inaudible 00:04:44]   of  electrification  that  we  are  seeing  around  the  world.   Soda  ash  is  also  used  for  the  production  of  helium  for  the  batteries.

Bicarbonate.  Our  sodium  bicarbonate,  it's  used  for  different  markets.  First  one,  for  the  exhaust  gas  industry  cleaning,  which  is  our  SOLVAir  market.  A lso  a  very  new  application  for  the  same  purpose,  gas  cleaning,  but  for  the  ships.  The  sodium  bicarbonate,  it's  also  used  for  the  pharmaceutical  industry  and  for  the  food  industry.

In  this  slide,  I  would  like  just  to  show  you  about  the  complexity  of  our  production  processes  to  manufacture  our  final  products.  As  we  said,  our  final  products  can  be   soda ash,  light  or  dense,  and  also  the  refined  bicarbonate.

To  produce  them,  we  need  to  consume  different  raw  materials  such  as  the  limestone,  the  brine,  and  also  we  use  the   coke and anthracite in  our  lime  fields.  As  you  can  see,  the  the  production  process  is  quite  complex  because  we  are  using  different  assets  like  the  absorbers,  like  distillation  in  the  distillation  sector,  dissolvers ,  precipitation  columns,  filters,  compressors.

We  have  a  long  variety  of  assets  used  in  the  manufacturing  process  and  very  complex  chemical  reactions,  mixing  gasses,  liquids,  solids.  We  need  to  take  into  account  thousands  of  parameters  in  terms  of  temperature,  pressure,  flows,  and  so  on.  It's  very  important  the  use  of  advanced  analytics  and  machine  learning  techniques  in  order  to  improve  this  production  process.

Here  we  are  entering  in  the  chapter  for  how  machine  learning  can  help  to  improve  our  production  process.  First  of  all,  to  share  with  you  our  strategy.  Clearly,  in  the   soda ash  and   bicarbonate,  our  strategy  is  to  be  competitive  and  keep  our  leadership  worldwide  position  in  this  global  commodity  market,  which  is  soda  ash,  but  also  in  the  premium  market  as  the  buyer.

What  is  our  objective  to  reach  this  ambition  in  our  strategy?  Our  objective  is  to  reduce  as  much  as  possible  the  variable  and  fixed  cost  in  our  manufacturing  sites  while  ensuring  the  overall  equipment  efficiencies, so  the  OEE,  and  the  quality  of  our  products.

Let  me  put  some  examples  of  how  we  can  impact  in  our  variable  cost  and  fixed  cost.  In  the  variable  cost  side,  clearly,  one  of  the  levers  that  we  can  improve  is  the  yield.  If  we  are  able  to  increase  our  sodium  precipitation  yield  in  our  carbonation  sector,  clearly  what  we  are  going  to  do  is  to  reduce  the  need  of  raw  material  and  energy  in  our  production  process  to  produce  the  same  quantity  of  soda ash.

The  same  for  the  topic  related  with  energy  efficiency.  In  the  previous  slide,  I  showed  you  the  complexity  of  the  production  process  and  the  energy  that  we  need  to  use  in  the  different  sector  such  as  the  distillation  sector,  calcination  sector,  or  lime  kilns.  If  we  are  able  to  improve  the  main  parameters  on  these  sectors,  we  will  be  able  to  reduce  the  specific  consumption  of  energy  in  our  production  process.

In  terms  of  fixed  costs,  one  of  our  main  fixed  cost  in  our  production  process  is  the  maintenance  cost.   We  have  unplanned  events,  unplanned  mechanical  breakdowns  in  our  industrial  assets,  and  also  we  perform  regular  maintenance  activity  and  cleaning  of  our  assets.  If  with  these  machine  learning  techniques  we  are  able  to  anticipate,  to  predict  these  unplanned  events,  we  could  also  potentially  reduce  our  fixed  costs.

Our  idea,  our  ambition  is  to  combine  the  deep  expertise  that  our  process  and  control  engineers  have  on  the  domain,  the  soda ash  production  process,  together  with  the  IT  and  computer  science  skills  and  math  and  statistics  skills.  This  is  our  ambition.

Traditional  method.  Traditionally,  what  our  engineers  is  doing  is  to  use  the  inputs  of  our  process  using  a  theoretical  model.  For  example,  the  thermodynamics  or  the  chemistry  in  order  to  understand  the  process  and  to  get  an  output.  This  is  the  traditional  method.  But  now  with  the  machine  learning  techniques,  what  we  can  benefit  for  is  about  the  historical  data.

In  our  site,  as  I  explained  before  in  this  very  complex  environment  of  the  production  of  soda  ash,  we  store  thousands  of  different  sensors  data  in  our  MES  systems,  in  the   manufacturing execution systems.  Data  from  temperature,  flows,  pressure  in  different  parts  of  the  process.  We  have  this  historical  of  data,  so  we  can  provide  with  the  machine  learning  algorithms,  inputs  and  outputs  of  our  process.  Creating  machine  learning,  big  data  models  that  could  help  us  to  improve  our  process  and  to  understand  better  our  process  for  the  future  inputs.

Now,  here  in  this  slide,  just  to  share  some  publications,  also  from  Dow,  another  important  multinational  chemical  company,  that  is  sharing  with  us  here  that  a  chemical  company  must  invest  to  create  a  critical  mass  of  chemical  engineers  with  technical  skills  in  statistics,  mathematics,  modeling,  optimization,  process  control,  visualization,  simulation,  and  programming.

But  it's  much  easier  to  train  chemical  engineers  on  data  analytics  topics,  rather  than  to  train  data  scientists  on  chemical  engineering  topics.  We  completely  agree  on  this  statement,  and  this  is  what  we  want  to  do  at  Solvay.

We  have  a  lot  of  very  skilled  chemical  engineers,  and  we  want  to  train  them with  this  advanced  analytics  techniques.  T his  is  the  main  reason  why  we  launched  the  program  of  Machine  Learning  Techniques  with  JMP  in  our  GBU,  our  Global  Business  Unit.

It  was  a  program  that  started  in  2021.  The  target  population  of  this  program  was  47  engineers  in  our  GBU. I t  was  led  by  the  industrial  data  science  team  at  corporate  level.

What  was  the  content  of  this  program?  We  had  one- hour  session,  first  of  all,  one  day,  where  we  explained  why  we  want  to  use  the  machine  learning  techniques  with  JMP  in  order  to  improve  our  production  processes,  as  I  explained  you  before.

Then  during  seven  days,  we  made  an  individual  online  course  for  each  of  the  47  engineers  related  to  a  statistical  thinking  part.  It  was  just  an  introduction  of  a  statistical  thinking  methodology.  Then  we  enter  it  on  the  JMP  introduction  part,  explaining  the  tool,  explaining  the  main  feature,  the  benefits  of  using  JMP,  and  the  main  tips  to  start  creating  some  graphics,  some  statistical  reports,  and  some  basic  things.  We  combine  theoretical  lessons  with  practical  exercises  and  planarizations  of  the  web.

Then  during  15  days,  we  enter  it  in  more  details  about  what  we  can  do  related  with  machine  learning  techniques,  which  we  did  the  same  individual  online  course  and  also  practical  exercises  and  plenary  session.

All  of  this  program  training  last,  let's  say,  around  one  month.  But  then  the  most  important  part  was  the  selection  of  the  real  cases  to  solve  in  the  different  manufacturing  sites  for  the  different  participants  of  the  training.  We  made  this  selection  and  we  provide  a  license  of  JMP,  of  course,  and  a  regular  support  with  weekly,  linear  meetings  and  individual  coaching.

Let  me  put  two  examples  of  practical  use  cases  about  this  selection  that  we  did  afterwards  this  training  session.  The  first  one  is  about  increasing  the  sodium   precipitation  yield  in  the  production  processes  of  Rheinberg,  our  manufacturing  site  in  Germany,  and   Torrelavega,  our  site  in  Spain.

How  we  use  the  JMP  on  this  project?  First  of  all,  to  screen  the  multiple  variables  that  we  think  that  can  impact  our  main  target  variable,  which  is  the  sodium  precipitation  yield,  in  order  to  explain  the  variability  of  this  target.

The  goal  was  to  investigate  what  are  the  variables  that  can  explain  better  the  variability  of  our  target.  For  this,  we  use  one  of  the  tools  that  we're  learning  during  this  one  month  course,  which  is  the  predictor  screen.

This  is  very  important,  because,  as  I  explained  to  you  before,  we  have  hundreds  of  variables  impacting  this  output,  which  is  the  yield  in  the  process, s o  it's  very  difficult  to  analyze  one  by  one.   This  tool  allow  us  to,  in  a  very  quick  way,  in  a  very  fast  way  and  intuitive  way,  to  understand  what  are  the  main  contributors  explaining  the  variability  that  we  have  in  our  tool.

Also,  we  need  to  say  that  JMP  is  a  very  intuitive  and  code- free  advanced  analytics  tool,  and  this  is  very  important  because  the  production  engineers,  not  all  of  them  have  the  knowledge  to  use  these  programming  code  tools.  Also,  important  to  say  that  to  visualize  the  long- term  variability  of  the  target,  but  also  its  relationship  between  the  most  important  variables  is  a  very  important  feature  that  JMP  has.

Finally,  also,  we  use  JMP  to  elaborate  the  statistical  reports  about  the  performance  of  the  different  approaches,  different  trials  that  we  perform  along  the  project.  This  is  the  first  example  where  we  use  JMP  in  order  to  understand  better  our  process  and  improve  our  yield  in  both  of  our  management  designs.

Second  one,  in  this  case,  we  are  talking  about  finding  the  root  causes  for  the  variability  of  one  important  parameter  of  the  final  product,  which  is  the  carbonate  content  in  the  sodium   bicarbonate.

Here  we  use  JMP  similarly,  like  in  the  project  that  I  explained  before,  we  screen  the  multiple  variables  and  select  the  most  important  ones  to  explain  the  variability  of  this  target.  For  this,  of  course,  we  use  to  use  it  again,  the  predictor  is  giving  input,  as  you  can  see  here  in  the  right- hand  side  of  the  slide.

Also  here,  it  was  very,  very  important,  the  visualization  in  a  graphical  way,  the  interaction  of  the  main  variables  that  we  identified  thanks  to  the   predictor screening.  You  need  to  understand  that  on  this  type  of  projects,  we  need  to  collaborate  with  different  stakeholders.  The  production  engineers  cannot  solve  this  type  of  very  complex  projects  alone.

Here  we  need  to  align  and  speak  and  generate  debates  with  the  production  operators  on  the  field,  with  production  operators  in  the  control  room,  with  site  managers,  other  engineers  in  other  plans,  experts  at  corporate  level,  and  so  on.  It's  very  important  to  translate  what  we  analyzed  in  analytical  way,  in  a  graphical  way  to  make  and  generate  these  debates.

Finally,  also  very  important  to  help  on  the  decision- making  process  in  order  to,  at  the  end,  of  course,  taking  decision  on  these  main  variables  that  we  demonstrated  in  an  objective  way  to  the  people  that  at  the  end  decides  to  make  a  modification  in  the  process  to  take  these  decisions.

This  is  what  we  have  done  also  in  this  project.  At  the  end,  it's  about  make  a  modification,  make  a  small  investment  to  modify  a  part  of  our  installation  in  order  to  reduce  the  variability  of  this  carbon  content  of  the  final  product.

That's  all  for  my  side  for  the  moment.  Now,  I  will  give  the  floor  to  my  colleague,  Carlos,  data  scientist  at  the  corporate  level,  who  will  explain  to  you  an  add- in  that  we  developed  internally at  Solvay  that  allow  us  to  connect  the  data  from  our  MES,  which  is  the  player,  as  I  explained  before,  where  we  store  all  the  data  into  JMP.

Thank  you,  David.  I  will  go  ahead  and  share  my  screen. Can you all  see?

Yes.

Okay,  I  want  to  get  started.  Do  you  see  this  ribbon  or  it's  only  me?

Yes,  it's visible .

Yes,  thank  you,  David.  In  this  section  of  the  presentation,  I  will  show  you  one  tool,  one  add- in  to  demonstrate,  an  open  source   add-in  that  we  have  created  in  the  team  of   industrial  data  scientists  at  corporate  level  in  Solvay. T his  is  a  team  that  supports  all  of  the  global  business  units.  That  means  that  we  have  to  provide  for  solutions  for  all  of  the  different  MES  that  exists  in  Solvay.

We  did  the  automation  of  this  task  because  we  saw  the  situation  that  was  happening  before  where  we  had  to  download  the  data  in  a  spreadsheet  sometimes  without  a  lot  of  advanced  capabilities.  Then  we  had  to  import  this  data,  treat  it,  and  then  finally  be  able  to  use  it  without...

Sometimes  it  was  not  even  clearly  identified  because  it  was  only  the  name  of  the  sensor,  but  sometimes  the  name  of  the  sensor  is  not  very  clear,  because  it's  not  very  well  standardized,  the  notation.

To  leverage  the  power  of  data,  we  say,  "Okay,  let's  make  the  process  of  extracting  the  data  as  automated  as  possible  so  that  all  the  process  engineers  can  use  it."  We  have  leveraged  this  power  in  JMP,  in  GBU  soda  ash,  and  also  in  other  GBUs  with  an   add-in  that  is  able  to  connect  to  the  two  most  common  data  bases  in  Solvay  that  are  the  MES  historians,  IP21  and  PI  from  Aztec  and  from  AVEVA,  respectively.

This  add-in  connects  directly  to  the  databases  if  we  are  in  the  local  network  and  is  able  to  fetch  with  a  query  whatever  information  is  stored  there.   We  have  automated  the  task  of  connecting  to  the  server,  selecting  the  query  parameter,  downloading  the  sensor  data  table  in  a regular  format  with  the  description  and  units.  A lso,  we  have  integrated  other  functions.

It's  also  worth  mentioning  that  in  this  case,  we  are  dealing  with  a  lot  of  sites.   The  sites  of  soda a sh  are  among  these,  of  course,  where  we  use,  as  I  mentioned,  two  main  databases  for  historian MES,  and  where  this  is  more  or  less  the  range  of  sensors  that  we  have  to  take  into  account.

This  is  how  it  looks  today  the  add-in  which  is  available  from  the  menu  add-ins.  It  was   developed  from  an  application  menu,  and  it  has  also  some  script  on  the  background.  I  will  show  a  clear  demo  with  this,  so  bear  with  me.

It's  a  focus  on  process  engineers.  As  I  mentioned,  it  integrates  a  description  and  engineering  unit,  which  is  very  useful  for  identifying  what  data  you  are  using  in  your  analysis.

A t  the  end  of  the  day  you  get  two  data  tables.  One  of  them,  so  what  I  see,  what  you  have  with  typical  statistics,  and  the  other  one,  which  is  a  time  series  with  all  the  details  of  every  sensor  according  to  your  extraction.  But  above  that,  the  functions  that  I  mentioned.

I  will  go  out  of  this  script  to  be  able  to  show  you  the  demo  of  this  tool.  Just  bear  with  me.   This  is  JMP,  as  you  know  it.   Here  we  have  an   add-in  that  when  it  is  connected  to  the  database,  you  are  able  to  see  all  the  list  of  servers  here  in  the  list.

It  is  connected  to  a  server  list  that  is  maintained  by  another  team,  which  is  an  IT  team.  In  this  server  list,  it  is  also  possible  to  modify  the  details  in  case  one  of  the  servers  is  not  available.  For  example,  one  can  put  the  IP  address  and  domain  here  to  add  a  new  server  that  is  connected  to  the  internal  network  of  Solvay.

After  that,  after  selecting  your  server,  the  next  step  is  to  go  ahead  and  filter  your  sensor  by  name  or  description.  This  is  important  because  as  we  mentioned,  we  have  in  the  order  of  thousands  of  sensors,  which  means  that  if  you  are going  to  go  ahead  and  try  to  see  everything  that  is  available,  it  might  take  a  long  time  or  the  server  might  crash.

For  that  reason,  we  have  this  filter  so  that  you  can  see  what's  relevant  to  you  in  case  you  want  to  see  flow  sensors,  temperature  sensors,  pressure  sensor,  or  you  want  to  look  the  three  sensors  by  description  or  both.

After  you  are  done  with  this  filter,  what  you  need  to  do  is  select  the  relevant  tags  from  this  other  list  which  are  given  in the  presentation,  just  an  example.  But  in  this  case,  I'm  not  connected  to  a  local  network,  but  you  can  see  an  example  here.

You  will  see  here  all  the  list  of  available  sensors,  and  what  you  need  to  do  is  to  add  them  to  right -hand  side,  which  is  in  this  way.   The  right- hand  side  list  means  that  these  sensors  are  ready  to  be  extracted  for  you.  Now  what  you  need  to  choose  is  the  start  time  and  end  time  for  your  extraction.

You  select  that,  it  perhaps  is  one  day,  one  month,  one  year,  I  don't  know.  Then  you  have  to  choose  what  type  of  method  you  want.  The  most  common  is  interpolated  because  it  means  that  you  will  have  evenly  spaced  data  by  minute,  by  second,  by  hour  or  by  day.  Also,  we  offer  an  aggregation  that  is  in  this  case  is  the  average.  A lso,  we  offer  to  extract  the  actual  data  as  it  was  recorded  by  the  sensor.

One  more  thing,  if  for  some  reason  you  already  know  the  list  of  sensors  that  you  want  to  download,  and  you  don't  want  to  browse  by  name  or  description,  you  can  directly  paste  also  this  list  from  a  CSV  format  that  you  have  available.

When  you  have  all  these  parameters  ready,  what  you  have  to  do  is   heat  from  extraction.  T his  will  take  the  time  it  takes  the  SQL  query  to  go  to  IP 21  or  PI.  W hen  it  finishes,  you  will  get  two  tables  with  what  I  mentioned  before.  One  with  summary,  which  will  allow  you  to  understand  the  typical  statistically  values  for  each  sensor,  row  by  row.

In  this  case,  you  have  the  name  of  the  sensor,  description  units,  and  also  the  mean,  standard  deviation,  max,  mean  range  of  the  sensor.  In  this  way,  you  can  understand  if  the  sensor  is  perhaps  not  working  or  something  odd  is  going  on  so  that  you  don't  need  to  extract.

Furthermore,  you  will  also  get  the  time  series  data,  which  in  this  case  looks  like  this.   You  get  a  column  with  the  time  stamp,  then  also  one  column  by  sensor  with  the  proper  format.  For  example,  this  is  a  continuous  amount,  this  is  a  discrete  amount,  and  everything  is  properly  formatted.  On  top  of  that,  you  get  the  description  of  the  sensor  as  I  mentioned,  and  the  units,  which  is  very  useful  for  processing  units.  This  allows  you  already  to  apply  all  the  methods  from  JMP.

Here  you  have  an  automated  version  of  the   add-in  that  allows  you  to  extract  data  directly  to  JMP.  It's  also  open  source,  so  if  you  are  interested  in  contributing,  you  can  go  to  the  community  in  JMP  or  in  GitHub  and  contribute  your  own  developments.

On  top  of  that,  we  also  offer  three  functionalities,  which  are  the  update  table.  The  update  table  will  make  sure  that  when  you  are  done  extracting  your  data  and  you  perform  one  analysis,  you  can  keep  updating  the  same  analysis  the  next  day.

For  example,  let's  say,  yesterday  I  downloaded  this  data  and  I  created  one  column  for  calculating  some  value.  L et's  say  that  today  I  want  to  also  see  how  this  calculated  value  is.   That  means  I  just  have  to  hit  this  button  and  this  data  will  be  updated  with  the  newest  data  from  yesterday  to  today.

Also,  we  offer  a  refresh  functionality  in  which  it's  meant  to  work  as  some  dashboard.  This  means  that  it  will  fix  time  window  and  you  will  be  able  to  see  your  analysis  with  respect  to  the  current  time.  That  means  that  if  I  perform  an  analysis  yesterday  and  I  have  a  new  column  with  a  new  formula,  I  can  hit  this  button  and  only  see  the  relevant  data  table  for  the  actual  period  for  getting  the  past.

That  means  that  as  I  said,  some  fix  window  is  fixed,  and  then  you  can  see  the  same  analysis  with  the  current  time  instead  of  with  the  old  time.   One  of  them  will  only  see  one  time  window  and  the  other  one  will  update  the  full  time  window.

Furthermore,  you  also  have  the  add  new  tags,  which  means  that  for  some  reason  you  forgot  to  add  a  tag  and  you  are  remembering  that  it's  very  important,  you  can  add  a  new  functionality.

With  all  this  said,  I  will  go  to  the  next  slide.   That  means  that  by  this  point  you  have  already  a  nice  data  table  in  JMP,  which  with  all  these  functionalities  that  we  mentioned,  update  table,  refresh  table  and  add  new  tags.  This  allows  you  already  to  use  the  typical  methods  for  advanced  analytics  in  JMP.  For  example,  this  one  I  am  showing  here  both  the  JMP  and  JMP  Pro  version,  but  this  is  up  to  you.

We  also  empower  the  user  to  use  another  add-in  that  we  also  developed  that  is  called  Predictor  Explainer,  which  will  be  presented  in  another  Discovery  talk.   We  also  have  other  types  of  analysis.  This  will  allow  us  to  perform  the  typical  task  in  data  analytics,  which  could  be  root  cause  analysis,  anomaly  detection,  process  optimization,  and  others.

With  this,  I  will  let  David  to  conclude  on  the  presentation.

Yes,  thank  you  very  much,  Carlos.  I  don't  know  if  you  are  seeing  now  my  screen.  If  you  stop  sharing,  maybe.

Yes,  stop share .

Okay.  Good.  Let  me  reshare  the  screen.

Yeah.

Okay.  Could  you  see  now  my  screen?

Yeah.

Perfect.  Thanks,  Carlos.  Thanks  for  your  support  that  you  provided  to  our  GBU,  not  only  developing  this   add-in,  but  also  coaching  our  production  process  engineers  on   JMP, too .

Last  slide  to  share  with  you  what  are  the  main  challenges  that  we  faced  during  this  journey  of  scaling  up  the  usage  of  JMP  in  our  GBU,  and  also  the  lessons  learned  and  the  next  steps.

Today,  let's  say,  that  around  20 %  of  the  target  population  that  two  years  ago  we  started  with  this  program,  is  continue  using  JMP  today  in  a  routine  basis.  The  main  blocking  points  that  we  found  are,  of  course,  resistance  of  change.  Some  people  are  more  comfortable  using  another  tools  like  Minitab  or  only  Excel  files.  In  any  project  or  initiative  that  requires  a  change  in  a  tool,  there  is  always  this  resistance  of  change  that  requires  time  and  efforts  to  change.

But  another  reason  is  also  the  lack  of  time.  Lack  of  time  that  is  linked  also  to  priorities.  The  priorities  of  the  role  of  production  and  process  engineers  is  not  always  fully  oriented  on  process  optimization  only,  because  sometimes  there  are  too  much  reporting  to  do  and  other  topics  to  cover  in  their  role.

The  main  points  to  keep  during  this  process  are  this  type  of  awareness  that  we  did  with  the  practical  industry  success  for  example.  This  is  very,  very  important.  In  order  to  convince  the  people  and  to  show  the  value  to  use  machine  learning  techniques  to  improve  our  process  and  to  reach  this  competitive  level  that  we  want  as  a  company,  as  a  business,  we  need  to  use  these  practical  industry  successful, for  examples.

Because  this  is  related  with  a  population  of  chemical  engineers  that  they  will  not  understand  if  we  start  to  talk  about  different  examples  in  marketing,  in  finance,  to  improve  all  of  these  other  areas  with  which  we  need  to  show  them  clear  and  concrete  examples  related  to  process  industry.

Then  also  an  important  point  about  the  importance  of  this   predictor screening   tool  as  a  kit  tool  for  us  for  the  variability  sourcing.  The  main  problem  that  we  have,  as  I  explained  it  before,  is  the  variability  that  we  have  is  certain  parameters  of  our  process  that  we  need  to  reduce.

If  we  are  able  to  reduce  this  variability  of  the  key  parameters,  we  are  going  to  really  reduce  our  variable  and  fixed  cost  in  our  production  manufacturing  sites.  This  tool  is  very  important  for  our  production  engineers  to  find  the  root  causes  of  this  variability  and  act  on  them.

Also  an  important  thing  is  this  combination  that  we  did  between  planarization,  so  all  together  sharing  thoughts  and  experiences,  but  also  with  individual  practice.  P rovide  time  to  the  people  to  practice  by  their  own  and  then  exchange  in  a  common  call.

Finally,  the  points  that  we  identified  to  reinforce  and  to  implement  in  the  near  future  are,  first  of  all,  of  course,  in  order  to  tackle  this  problem  of  resistance  of  change,  we  need  to  convince  the  site  management  about  the  importance  of  analytics  for  the  production  and  process  engineers.  We  need  to  launch  a  series  of  awareness  sessions  dedicated  for  them.  This  is  in  the item  we  are  going  to  do  a  lot  of long  this  year.

Also  very  important  for  us,  we  identified  this  strong  individual  coaching  for  the  production  and  process  engineers  when  they  start  to  use  JMP  in  the  real  cases,  in  the  real  projects.  Because  JMP  requires  time,  the  different  tools  as  per  the  tool  screening  and  other  tools  requires  time.  It's  very  important  for  the  very  first  projects that  one  engineer  developed  using  JMP  to  have  a  good  coach,  a  good  trainer  to  have  a  company  during  the  process.

That's  all  from  our  side.  Thanks  a  lot  for  your  attention.  If  you  have  any  questions  for  me  or  for  Carlos,  we  are  available.  Thanks  a  lot.