What s new in The Unscrambler X version 10.4 Lars Gidskehaug
Who am I? Born Bergen, Norway 2002: MSc, chemometrics, NTNU Trondheim 2007: PhD, bioinformatics, NTNU Trondheim 2007-2011: genomics, CIGENE, Aas Since 2011: CAMO Currently Product Manager for Unscrambler products
Agenda Integration of Design-Expert software version 10 General changes Sample Alignment Piecewise Direct Standardization Process control charting Next versions
Design-Expert integration CAMO and StatEase partnership (www.statease.com) World leading Design of Experiments (DOE) software Unsc. 10.4 includes Stat-Ease Design-Expert software version 10 Replaces the old DOE module First integration is basic Large potential for tighter integration
General changes Unique headers Split text variable New UI for PCA, PCR, PLSR New model structure Sample grouping Remember previous view
Unique headers Both row and column headers have to be Existing Unique Improved quality control when mapping different data tables applying models for online prediction Non-valid header names will be automatically changed
Split text variable Replicate headers can be split into original sample names replicate index
New UI for PCA, PCR, PLSR Modified panes X/Y Weights Algorithm Removed panes Warning Limits Set Alarms Autopretreatments New Outlier Limits pane
Algorithm pane changes Can set max iterations for NIPALS Changed convergence criterion Faster convergence Reporting convergence statistics Exact Q-residual limits Allow missing values in prediction Model warnings
New Outlier Limits pane Merges previous functionality from Warning Limits and Set Alarms panes Warning and Alarm limits can be applied for Marking suspect samples or variables in calibration models Detecting outliers in prediction Online prediction tools Optional
New model structure Category Variables added to Raw data node New warning nodes Sample Outliers Variable Outliers Model Warnings Column headers for entire preprocessing range included (hidden)
Sample grouping More options, less mouse clicks External variable option matches samples based on (unique) name in both matrices
Remember previous view Current/last plot settings always given in tabs in Unscrambler window New: click on model node to show previous view in open project Works after closing tabs Not remembered after save/open
Sample Alignment
Background Most statistical methods require that the data are aligned Users often gather data from different sources Time-consuming to align them manually Prone for errors with manual work X Variables Objects X
New Sample Alignment (SA) transform Align data as a transform before building model Apply as auto-pretreatment during prediction Supported in Unsc. and PP (not yet in engines)
Different alignment approaches Implementation handles different types of alignment Sample ID Event Polling Sample ID with additional time stamp Only numeric data can be aligned Multiple options for the various alignment types Nearest Mean, Median Interpolation Alignment window that allows for time offsets
Ex 1: Sample ID alignment Samples are different raspberry jams Sensory measurements Laboratory instruments Aligned based on sample ID
Ex. 2: Sample ID alignment Similar example but replicate sensory measurements Different options for how to handle replicates Mean/Median/Last row/all rows
Piecewise Direct Standardization
Piecewise Direct Standardization (PDS) Calibration transfer method applied on spectra Separate transfer model created for each «slave» instrument Transformed to look like collected on «master»instrument
Process control charting Statistical Process Control (SPC) Moving Block Methods (MBM)
Process control In process industries there is a demand to monitor and control process parameters at a specific level Be notified when a process reaches steady state Two added methods for process control charting Statistical Process Control (SPC) Moving Block Methods (MBM)
Statistical Process Control (SPC) Univariate monitoring of steady state process parameters Includes charts Individual/Moving range (I-MR) Average/Range (Xbar-R) Average/Standard Deviation (Xbar-S) Output Basic statistics Tests for normality Capability analysis
Moving Block Methods (MBM) Uni- or multivariate monitoring of changes Includes Individual block mean and std (IBM, IBSD) Moving block mean and std (MBM, MBSD) %RSD (=MBSD/MBM*100)
MBM: Trend charts Moving Block Mean 0.004 0.003 0.002 Moving Block Mean, Region 1 Right-click charts to set limits* 0.001 Sample #44 Sample #13 Sample #22 Sample #30 Sample #39 Sample #48 Sample #58 Sample #66 Sample #75 Blocks 0.0006 Moving Block Standard Deviation, Region 1 0.0005 Standard Deviations 0.0004 0.0003 0.0002 0.0001 Sample #44 Sample #13 Sample #22 Sample #30 Sample #39 Sample #48 Sample #58 Sample #71 Sample #80 Blocks *) F-residual limits scheduled for maintenance upgrade v10.4.1 (2016)
Next versions
Upcoming versions 10.4.1 (2016): Maintenance, Contributions, MBM with F-test 10.5 (2017): Performance, Dimension Reduction 11 (2018): New UI, plots Workflow scripts Settings manager
THANK YOU! Lars Gidskehaug lg@camo.com