Using Multi-Sensor Data Fusion for Process Analysis and … · 2019-06-28 · Setup Unscrambler...
Transcript of Using Multi-Sensor Data Fusion for Process Analysis and … · 2019-06-28 · Setup Unscrambler...
6/19/2019
1
Using Multi-Sensor Data Fusion for Process Analysis and Control
Geir Rune Flåten
• Combination or fusion, of multiple data types can provide an improved description of a system, such as a production process. One can for instance use traditional process parameters to describe the physical properties of a process and complement this with chemical properties from a spectroscopic method. The two main challenges when combining multiple data sources in process analysis are i) synchronization and ii) scaling of the measurements. In the Camo Industrial Analytics platform, the data fusion challenges are resolved by Sample Alignment for synchronization and either block-scaling or dimension reduction for the scaling challenge. The challenges and opportunities for multi-sensor data fusion are discussed and it is shown how to successfully implement data fusion for process analysis and control
Abstract
6/19/2019
2
Agenda1
2
3
4
5
Introduce data fusion
Multi‐sensor approaches
Real‐time applications
Conclusions
Examples
Data fusion is the process of integrating multiple data sources to produce more consistent, accurate, and useful information than that provided by any individual data source.
wikipedia.org
6/19/2019
3
Data fusion
processInput Output
Raw Material Data
Process Data
Sensor Data
Quality DataIPC Data
Multi sensor data fusion
processInput Output
Raw Material Data
Quality DataProcess Data
Sensor Data
IPC Data
6/19/2019
4
Multi operation
https://www.pharmamanufacturing.com/articles/2017/true‐continuous‐manufacturing/
process process process process
Multi operation data fusion
https://www.pharmamanufacturing.com/articles/2017/true‐continuous‐manufacturing/
process process process process
Data 1 Data 2 Data 3 Data 4
6/19/2019
5
Contact Lenses
Multi operation data fusion example
H2O MEK H2O PAA H2O H2O H2O
ContactLense
carrier
ContactLense
carrier
Process Data
Position Data
Fixated lenses – finished productFixation baths – process
Visual system
Quality controlMoulded lenses– raw materials
Synchronisation and scaling
Data Sources Data Preparation Data Combination Data Analysis
6/19/2019
6
Block weighting
YD X1 X2
ProcessRaw
materialsQuality
Block weighting
Dimension reduction
Scores from each
YResponse variables; quality
X1 X2 X3Process steps
“Superscores”
Combine scores from individual steps
Dimension Reduction
6/19/2019
7
Process NIR, other Process NIR, other
PollingEventSample ID
Sample Alignment (synchronisation)
Sample Alignment
Decision 1:
- Alignment criteria- Polling- Event- Sample ID
Decision 2:
• Alignment Basis
‐
+‐
+‐
+
• Nearest• Mean• Median• Interpolation
6/19/2019
8
Sample Alignment-BTW
Decision 1:
- Alignment criteria- Polling- Event- Sample ID
Decision 2:
• Alignment Basis
‐
+
• Nearest• Mean• Median• Interpolation
‐
+
Real time
Data Sources Data Preparation Data Combination Data Analysis
6/19/2019
9
Real time
1. Create alignment/weighting pattern in Unscrambler
2. Upload to Process Pulse
3. Use in realtime
Data Sources Data Preparation Data Combination Data Analysis
Real time-MPC
Data 1ReadAlignScale
Run model
External Output
Data 2
Control System
Process
6/19/2019
10
Real time-IPC
Data 1ReadAlignScale
Run model
IPCManual Sampling
LIMS
IPC samples
6/19/2019
11
Multiple IPC samples
Examples
Unscrambler – historical data
Process Pulse – real‐time data
6/19/2019
12
Setup Unscrambler
Setup PP
6/19/2019
13
Example 1 –Fluidized Bed Dryer
Example 1 -results
NIR
NIR + Process
6/19/2019
14
Example 1:Karl Fisher
Develop an Integrated Real-Time environmental monitoring system
Challenge
Multiple sensors were required to provide input to the data model in order to provide real-time data monitoring.
Various types of sensors with data sampled at different frequencies
Solution
The empirical models provided real-time insight to subsea environmental conditions.
Situation
6/19/2019
15
The LoVe Ocean Observatory
• LoVe: Acronym for Lofoten-Vesterålen
• Lander with multiple sensors located at 258 m depth 20 km off the coast
• Cabled
• Data from the sensors (~100 individual variables) submitted online more or less continuously
• Real-time environmental monitoring
Ingvar Eide, Frank Westad, Automated multivariate analysis of multi-sensor data submitted online: Real-time environmental monitoring. PLoS One, 13, 1, 2018
29
The LoVe Ocean observatory
https://love.statoil.com/ 30
6/19/2019
16
Online sensors at LoVe
• A total of ~100 individual variables are collected every 5th or ten minutes: Chlorophyll (2 sensors), conductivity, depth, temperature (3 sensors), turbidity and Total Suspended Matter, salinity biomass at three different depths, and current speed
• Current speed is measured in two directions (N and E) using two sensors covering different depths, however, with overlap: • Aquadopp 3-21 m, 2 m resolution• Continental 6-146 m, 5 m resolution
• RGB Camera
31
The modelling approach
32
6/19/2019
17
The LoVe case – on-line monitoring
Map of variables (model) Drill down
Higher temperature in August33
Conclusions
• Sensor data fusion can be useful• Tools for sensor data fusion is
available (and easy to use)• Online use of alignment patterns is
straightforward