vertimill predictive analytics - osisoft€¢ inadequate loading of grinding media in the vertimill...
TRANSCRIPT
© Copyright 2017 OSIsoft, LLCUSERS CONFERENCE 2017 #OSIsoftUCosisoft@
Presented by
Vertimill Predictive
Analytics
Scott Schemmel, Jolene Baker and Wyatt
Keller
© Copyright 2017 OSIsoft, LLCUSERS CONFERENCE 2017 #OSIsoftUCosisoft@
Vertimill Predictive Analytics
© Copyright 2017 OSIsoft, LLCUSERS CONFERENCE 2017 #OSIsoftUCosisoft@
Soda Ash 101
– Ciner Wyoming, LLC (pronounced jin-ner) is a natural soda ash producer in the Green
River basin in Wyoming
– We refine soda ash from a natural feedstock of sodium sesquicarbonate (Trona)
– Trona was deposited in the basin over a million year period when 4 million year old
Lake Gosiute(\ˈgōˌshüt\ ) became closed off from a freshwater source and the alkali
concentration increased
– To produce soda ash, Trona is mined and calcined to remove water and CO2 to convert
the Trona ore to sodium carbonate aka Soda Ash
2(Na2CO3NaHCO32H2O) → 3 Na2CO3 + CO2 + 5H2O
– Used in
• Glass – 49%
• Chemicals – 27%
• Soap and detergents – 11%
• Flue gas treatment – 3%
• Pulp and paper – 2%
• Water treatment – 2%
• Misc. – 6%
© Copyright 2017 OSIsoft, LLCUSERS CONFERENCE 2017 #OSIsoftUCosisoft@
Problem Overview
– Continuous Drum Miners mine Trona ore
– Ore Grade varies throughout different areas in the beds
• Low Ore Grade, below 83%, is referred to as “Bad Ore”
• Variances in Ore Grade can lead to process upsets and unplanned downtime
– Lab analysis provides ore grade after the fact - no real time ore analysis
• Process Operators are “blind” to sections of “Bad Ore”
© Copyright 2017 OSIsoft, LLCUSERS CONFERENCE 2017 #OSIsoftUCosisoft@
Problem Overview in Detail
– Trona ore is calcined and then dissolved to separate the desired soda ash from the insoluble impurities
– Insoluble impurities are ground to recover any trapped soda ash and produce a PSD that generates a paste for disposal of the tailings
• The amount and type of insolubles are a direct function of ore rate & grade
• The Vertimill is capable of handling a fixed amount of insolubles
• Variations in ore grade can send too many insolubles to the Vertimill
– The Vertimill can be overloaded when…
• There is too high of a insoluble loading, and/or
• Larger PSD of the insoluble, or
• Inadequate loading of grinding media in the Vertimill
– …reducing grinding effectiveness and ultimately spilling over the top of the Vertimill
When the Vertimill is down 60% of total production is lost
© Copyright 2017 OSIsoft, LLCUSERS CONFERENCE 2017 #OSIsoftUCosisoft@
7C26A
12.48 AMPS
GR.CR.7T30:II7C26A.PNT
7C26B
269.63 TPH
GR.CR.CRUSH_CRUSH_MISC:WI7C26B.PNT
7C26C
47.64 AMPS
GR.CR.7T30:II7C26C.PNT
7C24A
14.32 AMPS
GR.CR.7T30:II7C24A.PNT
4821.72 SCFM
GR.U7.7H50:FI7H01V.PNT
Nat. Gas Burner Header
7CS31
28.01 AMPS
GR.CR.7K01:II7CS31.PNT
190.90 DEGF
GR.CR.7K01:TI7K01E.PNT
Dust to 7ES10
7ES10To 7BL01
7CS10
11.52 AMPS
GR.CR.7ES10:II7CS10F.PNT
7ML01A
75.91 AMPS
GR.CR.7ML01:II7A01A.PNT
0
20
40
60
80
100
7T19
136.03 GPM
GR.CR.7ML01:FI7CH01AA.PNT
5T57 and 7T10
1364.78 GPM
GR.CR.7ML01:FIC7CH01AB.MEAS
7P01A
7P01B
32.45 AMPS
GR.CR.7ML01:II7P01A.PNT
48.95 AMPS
GR.CR.7ML01:II7P01B.PNT
343.00 DEG
GR.CR.7K01_BMS:TI7K01G.PNT
7T01
179.99 DEGF
GR.U7.7ML04:TI7RC01.PNT
856.24 GPM
GR.CR.3TH01:FIC7P01G.MEAS
7P01F
7P01G
783.82 GPM
GR.CR.4TH01:FIC7P01F.MEAS
795.24 GPM
GR.U5.5TH01_THCKNR:FIC7P01E.MEAS
To 3TH01
To 4TH01
To 5TH01
5T57264.74 GPM
GR.CR.7ML04:FIC7RC01C.MEAS
198.10 AMPS
GR.CR.7ML04:II7ML04.PNT
7ML04
31.41 AMPS
GR.CR.7ML04:II7P04.PNT
7P04
7MLC04
7P02A
7P02B
7T02
0
20
40
60
80
100
To 5TH02
587.26 GPM
GR.CR.7ML04:FIC7P02A.MEAS
12.56 gpm
GR.U6.6T01_PRIMARY:FI11P01BC.PNT
From Deca
91.54 %
GR.CR.7ML01:LIC7ML01AA.MEAS
64.23 %
GR.CR.7T01:LIC7T01A.MEAS
80.02 %
GR.CR.7ML04:LIC7T02.MEAS
7RC01
7.50 AMPS
GR.CR.7ML04:II7RC01M1.PNT7ML01B
188.73 AMPS
GR.CR.7ML01:II7A01BB.PNT
First Warning Insol Saturation
Second Warning Insol Saturation
Final Warning Insol Saturation
Calculated Ore Rate310.17
1/17/2017 12:49:32 PM1/17/2017 4:49:32 AM
280
320
340
360
380
400
440
Calculated Ore Grade83%
1/17/2017 12:49:32 PM1/10/2017 12:49:32 PM
82%
83%
84%
85%
86%
Calciner
Spiral
Classifier
Problem Overview (2 hour processing time)
Ore BinBelt Conveyor Screw Conveyor
DissolverDissolver
Vertimill
© Copyright 2017 OSIsoft, LLCUSERS CONFERENCE 2017 #OSIsoftUCosisoft@
Existing Signals
– Existing real-time data showed that the process was affected by Bad Ore or the Vertimill was not properly loaded with grinding media prior to a process upset
• 10 process data streams presented the best pre-upset visibility
– Patterns in the 10 data streams immediately around upset conditions were not consistent enough from upset to upset.
• Varied in frequency, consistency, and magnitude
– Conventional analytics were not good as preventative warning
• Time consuming application of statistical analytics to filter and refine the data did not work
• Some other method or tool was needed…
© Copyright 2017 OSIsoft, LLCUSERS CONFERENCE 2017 #OSIsoftUCosisoft@
Enlightenment
• By chance, we met Crick Waters from Falkonry at a regional OSIsoft event
– Falkonry provides a Pattern Recognition software designed for use by front
line process experts or Subject Matter Experts (SME)
– Trial run POV
• 2 months to repeatable insightful patterns for all 10 data streams
– Falkonry “crunched” our data streams
• Similar operating conditions were grouped and color coded
• No context…yet
© Copyright 2017 OSIsoft, LLCUSERS CONFERENCE 2017 #OSIsoftUCosisoft@
Existing signals – New Tools & How They Work
– Adding context - time periods defined for Good and Bad Ore events or inadequate grinding media
charge
• Software found similar patterns to create Bad Ore or Media Charge prediction model
• Ran multiple iterations and tests on the models to confirm validity
– Post validation, applied the model to real-time data flow
• Ever improving predictive model
• Able to be adjusted anytime there are new events
© Copyright 2017 OSIsoft, LLCUSERS CONFERENCE 2017 #OSIsoftUCosisoft@
Combining Learned Models
– Best Solution? All 10 data streams in one model (pipeline)…
• Expected to see Bad Ore move through the system in an hour or two…not the case…
– Broke 10 data streams into 3 like process pipelines
• Dry-Burner/Calcining, Wet-Dissolving, and Grinding/Milling-Vertimill
• Detectable Variations
– Insolubles from Bad Ore actually buildup hours before Vertimill affected
• Patterns show buildup of Bad Ore cascading from calcining to dissolving to the Vertimill
• Plenty of time for corrective response versus reactive response
Stage 1 trigger
Stage 2 trigger
Stage 3 trigger
© Copyright 2017 OSIsoft, LLCUSERS CONFERENCE 2017 #OSIsoftUCosisoft@
Operationalize
DCS Alarm
Alarm 3 Stage Alarm Real time Ore
Grade
Prediction
PI Coresight™ Dashboard
© Copyright 2017 OSIsoft, LLCUSERS CONFERENCE 2017 #OSIsoftUCosisoft@
Benefits
• Operations
– Alarms provide visibility for operators where they were “blind” before
– Confidence to make decisions regarding tonnage flow to run at optimal state
• Business
– Reduce lost tons of production
– Benefits are measurable and significant!
• Technical
– No time spent teaching outside parties process details
• Subject Matter Expert (SME) is directly involved
• “Data science in software” significantly reduces time spent performing data analysis
– Visual Pattern Recognition is relatable and easily interpreted
• Models are easily modified to meet current conditions
– Reduced development and deployment time leads to quicker realization of Revenue Growth and Cost Savings
© Copyright 2017 OSIsoft, LLCUSERS CONFERENCE 2017 #OSIsoftUCosisoft@
Lessons Learned
• More than one problem may be revealed
• Iterative process requiring input from many areas of the
process
• Opportunity is knocking…
© Copyright 2017 OSIsoft, LLCUSERS CONFERENCE 2017 #OSIsoftUCosisoft@
Picture / Image
RESULTSCHALLENGE SOLUTION
COMPANY and GOAL
Company
Logo
Predictions with the PI System and
Falkonry's Pattern Recognition
Ciner Resources is a leading natural soda ash
producer, and wanted to predict and reduce process downtime.
Difficult to find patterns to use for alarms when combining multiple data sources.
Required a more advanced
pattern recognition
solution.
More detailed insight into
current operation
conditions.
• Recognized upset using PI Coresight
• Unable to capture all instances using tools in PI Asset Framework
• Leveraged Falkonry with PI to identify meaningful patterns
• Detect Bad Ore Grade, Mechanical
issues and Process Anomalies
• Generated Calculators using PI AF
Analytics based on new insights
• Justified hypothesis around abnormal
events
© Copyright 2017 OSIsoft, LLCUSERS CONFERENCE 2017 #OSIsoftUCosisoft@
Contact Information
Jolene Baker
SMART Plant Lead
Ciner Wyoming LLC
Wyatt Keller
Process Engineer
Ciner Wyoming LLC
15
Scott Schemmel
Head of IT (CIO)
Ciner Resources
© Copyright 2017 OSIsoft, LLCUSERS CONFERENCE 2017 #OSIsoftUCosisoft@ 16
Questions
Please wait for the
microphone before asking
your questions
Please remember to…
Complete the Online Survey
for this session
State your
name & company
http://bit.ly/uc2017-app
© Copyright 2017 OSIsoft, LLCUSERS CONFERENCE 2017 #OSIsoftUCosisoft@
Thank You