2014 AAPS Advanced Process Control and Continuous Processing in Pharmaceutical Manufacturing: What Can We Learn from Other Industries Paul Brodbeck/Control Associates Inc.
Aug 07, 2015
2014 AAPSAdvanced Process Control and Continuous Processing in
Pharmaceutical Manufacturing: What Can We Learn from Other Industries
Paul Brodbeck/Control Associates Inc.
What is Process Control? How? Why? Benefits? Why Advanced Process Control? Advanced Controls
◦ MPC◦ Kalman Filter◦ Neural Networks◦ LP Optimization
Agenda
Controlling process variable to a desired SP.◦ Reactor Temperature◦ Heat Exchanger Flow Rate◦ Boiler Pressure◦ OTC Tablet (API) Concentration.◦ Dryer Moisture Content◦ House Temperature◦ Car Speed◦ Distillation Column Production Rate
Controller
What is Process Control?
How is Process Control done? First Feed Back Controller – Humans Closed Loop Control – Level, Press, Flow, API Concentration (%) Vary output – Valve, Pump, Agitator, Fan
WHY? Manual vs. Automatic Easy Quality – Temperature Variability Temperature Cycling
◦ Poor Quality◦ Inefficient◦ Wear and Tear on Heater and Parts
Auto change SP at day/night Cost Savings Control Improves Quality & Reduces
Costs
House Temperature Control
WHY? Manual vs. Automatic Quality – Constant Speed Speed Cycling
◦ Poor QualityInefficient◦ Wear and Tear on Car and Parts
Get there faster!◦ Set Speed closer to speed limit◦ Less Risk/Less Speeding Tickets
Control Improves Quality & Reduces Costs
Car Cruise Control
Distillation Columns/Refinery
WHY? Manual vs. Automatic Production Yields Profit Reduces Costs
◦ Labor◦ Energy
Safer Lower Risk
Improves Quality Reduces Costs Increases Production
Control Generalization
Reduce Variability!◦ Almost at end in itself.
Edward Deming – Quality Program Founder Japan Post WWWII Better Quality
◦ Autos, Semi-Conductors, Steel…
1980’s American Manufacturing Poor Quality Statistical Process Control Introduced into US
Get Process under Control (Statistically)◦ Control Charts
Reduce Variability Increase Quality
How does it improve Quality?
Variable Parameters
Variable Quality
Controlled Parameters
Fixed Quality
Statistical Process Control
Pharma Poor Manufacturing Techniques
Why Process Control?
Improve Quality Increase Yields Increase Production Reduce off-spec Reduce Bad Batches Reduce Energy Costs Reduce Production
Costs Improve Safety Reduce Risk Increase Profitability
Optimal Control Better Control Control Poor Control Manual Control
Process Control Benefits
Basic vs. Advanced Control
BasicPID
Advanced PID
AdvancedControl
NoControl
OptimalControl
Optimization
Control
Basic Process Control = PID
Tuning Constants:1. Proportional (P)2. Integral (I)3. Derivative (D)
Applications Car Cruise Control Home Heating/AC, Distillation Columns Chemical Reactors Bioreactors Crystallization Chromatography
Basic Control (PID) - Ubiquitous
Industries Chemical Pharmaceutical Petroleum Automotive Robots Aerospace Boilers Missile Guidance
Applications Distillation Columns Robotics Drones Aerospace Robots Missile Guidance Stock Market, Operations Research Economics Scheduling
Advanced Control - Ubiquitous
Industries Chemical Pharmaceutical Petroleum Automotive Robots Aerospace Boilers Missile Guidance
1. Model Predictive Control (MPC)◦ Distillation Columns, Robotics, Drones,
Aerospace… 2. Kalman Filter
◦ Robots, Aerospace, Missile Guidance… 3. Neural Networks (NN)
◦ Pattern Recognition, Stock Market, Genetics… 4. Linear Programming (LP) Optimization
◦ Operations Research, Economics, Scheduling
What are Advanced Controls?
Machine Learning◦ Computer Science & Statistics ◦ Real World Problem Prediction/Optimization
Search Engines Stock Market Prediction Pattern Recognition (OCR) Robotics Recommender Systems DNA Sequencing Chemometrics
Big Data/Data MiningAPC Subset
Numerical Methods Least Squares Statistics Modeling Analytics Linear Programming Optimization
Machine LearningUnderlying Technology
MPC Neural Networks MVA Tools
MLR PCA PLS
Kalman Filter Multivariate SPC
Optimal Control Slow Processes Large Dead Times Multiple Loops (50x25) Complex Dynamics Strongly Correlated Loops
Model Predictive Control (MPC)
MPC Batch Reactor
Multi-Loop Control
Multi-Loop PID Multi-Loop MPC
Model predictive control (MPC)
22 2
1 1 1 1 1 1
1 1y u un n nP M M
y set u uj j j j j j j j
i j i j i j
J w y k i y k i w u k i w u k i u
y: Controlled variableu: Actuator△u: Predicted adjustment
manipulated variable
deviations
Controlled variable deviations
controller adjustments
Singh, R., Ierapetritou, M., Ramachandran, R. (2013). European Journal of Pharmaceutics and Biopharmaceutics, http://dx.doi.org/10.1016/j.ejpb.2013.02.019.
Tuning parameters1. Output weights (wy
j) 2. Rate weights ( ) 3.Input weight ( ) 4. Prediction horizon5. Control horizon
ujw
ujw
Rutgers Engineering Research CenterContinuous Manufacturing
Single Loop MPC
MPC – API Controller
MPC Model Builder
Actuator Control variable
Linear Model for MPC
Linear Model for MPC
MPC Operator Interface
Actuator
Control variable
MPC Performance - Simulation
MPC Performance – Pilot Plant
Filtered NIR signal CV(API composition)
Actuator Ratio SP
NIR signal
MPC Performance v. RobustnessPenalty on Move/Penalty on Error
3
1
2
Statistically Optimal Estimator Numerous Applications
◦ De facto Standard Robotics◦ Aerospace◦ Missile Guidance◦ Economics ◦ Signal Processing
State Prediction based on:◦ Noisy Data◦ Physical Model (Error increases w/ Time)
Kalman Filter
Takes a statistical average of:◦ Measured Variable◦ Model
Acts Recursively to continuously predict most probable state.
First used by NASA to predict location of rockets◦ Uncertain GPS Signal◦ Physical Model error increases with time
Use measurement signal to correct errors with model.
Use model to validate measured values.
Kalman Filter
Kalman Filter Cannonball
Kalman Filter – Logistic Growth
dF(x)/dx = f(x)*(1-f(x))
E-mail Spam Internet Browser Recommender systems Pattern Recognition
◦ Bar coders◦ Facial identification◦ Robotics
Pharma ◦ Soft Sensors◦ Non-Linear Control
Neural Networks
Non-Linear Data Modeling Combination of Linear Regressions Build up a series of linear models
(regressions) to create a non-linear model
Neural Network Algorithm
Linear Regression
1 2 3 4 5 6 7 8 9 10 110
2
4
6
8
10
12
14
FitRaw Data
1 3 5 7 9 11 13 15 17 19 21 23 250
100
200
300
400
500
600
700
1 3 5 7 9 11 13 15 17 19 21 23 250
100
200
300
400
500
600
700
1
2
3
Combine Multiple Linear FitsRegress Against Y Data
Neural Net ExampleContinuous Digester Degree of delignification indicated by Kappa number.
Neural Net Parameters
Build Neural Net
BUILD MODEL
PREDICTVALUE !
Neural Net Configuration
LP Optimization Linear Programming A mathematical/computer optimization
technique – Simplex Method Solve a system of linear equations Can be used to find the minimum and
maximum states of process control Can be made subject to multiple constraints
Pusher Function Maximize Flowrate subject to constraints
LP Optimization w/MPC
Complexity PID vs. APC
Introduction of Technology◦ PAT◦ Continuous Manufacturing
Introduce Advanced Controls ◦ MPC◦ Kalman Filter◦ Neural Net◦ LP Optimization◦ MultiVariable SPC
Opportunity