—INDUSTRY GUIDANCE WORKSHOP - 30-10-2019
A Framework for IIoT AnalyticsEric Harper
—Papermaking has been a human endeavor for 3,000 years
March 13, 2020 Slide 2
—IT Analytics
March 13, 2020 Knowlton, B., “Data and Analytics Strategy Framework for Credit Unions and Banks” (2016)Slide 3
An iterative process
Data and AnalyticsStrategy Framework
Discover 1 Strategize2
Develop3
Deploy4Analyze 5
Iterate 6
—Industrial Internet Consortium
March 13, 2020 http://www.iiconsortium.org/industrial-analytics.htmSlide 4
Things are coming together
INDUSTRIAL INTERNET OF THINGS ANALYTICS FRAMEWORK
We are pleased to announce the Industrial Internet of Things Analytics Framework (Industrial IoT Analytics Framework)for system architects, technology leaders and business leaders looking to successfully deploy industrial analytics systems
Advanced analytics is at the core of the Industrial Internetof Things (IIoT). When analytics are applied to machine and process data, they help optimize decision-making and enable intelligent operations. These new insights and intelligence can be applied across any level of any industry if the appropriate data can be collected and analytics are applied correctly. If data is the new oil, data analytics is the new engine that propels the IIoTtransformation
—Agenda
March 13, 2020 “Analytics Framework”, Industrial Internet Consortium (2017)Slide 5
IIoT Analytics
Business Viewpoint
Usage Viewpoint
Functional Viewpoint
Implementation Viewpoint
Crosscutting Concerns
Questions and Answers
—Creating Business Value
March 13, 2020 “Business Strategy and Innovation Framework”, Industrial Internet Consortium (2016)Slide 6
Who: Business and technology leaders
What: Increase throughput, reduce expenses and inventory
Why: Generate higher margins to create business value
How: Identify performance bottlenecks in overall operations continuously and remove them one-by-one
IIoT Strategy
Market Context
Ideation Preparation Evaluation Initiation
IIoT Business Model Innovation
IIoT Foundations
—Agenda
March 13, 2020 “Analytics Framework”, Industrial Internet Consortium (2017)Slide 7
IIoT Analytics
Business Viewpoint
Usage Viewpoint
Functional Viewpoint
Implementation Viewpoint
Crosscutting Concerns
Questions and Answers
—Papermaking Process
March 13, 2020 Slide 8
Buled puplslushing
Paper mill
Softwood
Refining
Paper mill Pulp mill
Hardwood
Broke/recycleWhite water
Blending
Cleaning
Wet-end Dry-end
From thickener
Pm2 eucalyptus pulp hd-silo
Pm2 eucalyptuspulp chest150m3
RefinersRefiners
Pm2 softwoodpulp chest80m3
2000 m3
4000 m3
4000 m3
80 m3
150 m3
4000 m32000
m3
Pm2 softwoodpulp silo
Bale pulpers2ps
Pm2 dry broke silo Pm2 wet broke silo
Dry broke thickener
Dry broke chest 80m3
Wet broke thickener
1. Stage broke screen
2. Stage broke screen
3. Stage broke screen
To couch pit
Dilution Pm2 eucalyptusrefining chest 80m3
Blend chest
Machine chest
Recovered fibre 30m3
Disc filter
Clear filtrate 80m3
Cloudy filtrate 80m3
Ultraclear filtrate 30m3
5 Bar dilution water bottoms of silo bale pulpers dilution waterWhite
water siloPm2 knock-
off showers
Ret. aid dilution
To blend chest
Broke reject chest 10m3
Broke reject chest 25m3
DeflakerBroke proport. chest 80m3
Wet broke chest 80m3
To/From Pm1
Cleaners reject handling
Cleaners 6 stages
Deculator
Wire silo 180 m3
Headbox dil.
Rej. tank
White water tank 100m3
Couch pit.
Cons. control
Press section
Press pulper
Shower water tank wire showers
Warm water tank
Chem. purified water
To ret. aid dilTo clear and u. clear tank level control Sym-sizer
Size-press pulper
(Soft calender)
Pope
Calender pulper
Showers
—Analytics Framework
March 13, 2020 “Analytics Framework”, Industrial Internet Consortium (2017)Slide 9
Getting started
Descriptive Analytics
Gain insight from historical or current data streams including for status and usage monitoring, reporting, anomaly detection and diagnosis, model building or training
Predictive Analytics
Identify expected behaviors or outcomes based on predictive modeling using statistical and machine-learning techniques, e.g. capacity demand and usage prediction, material and energy consumption prediction, and component and system wear and fault predictions
Prescriptive Analytics
Uses the results from predictive analytics as guidance to recommend operating changes to optimize processes and to avoid failures and the associated downtime. An example of prescriptive analytics is on-demand production from a solid geometric assembly model to find the optimal set of manufacturing processes to achieve the final product
—Agenda
March 13, 2020 “Analytics Framework”, Industrial Internet Consortium (2017)Slide 10
IIoT Analytics
Business Viewpoint
Usage Viewpoint
Functional Viewpoint
Implementation Viewpoint
Crosscutting Concerns
Questions and Answers
—Pulp and Paper Plant
March 13, 2020 Slide 11
Data sources and systems
1
1. Energy Management
2. Order Management
3. In Line Measurement
4. Paper Controls & Optimization
5. Lab Measurement System
6. Quality Controls System
7. PM Drives System
8. Web Inspection System
9. Production Planning & Measurement
10. Pulp Mill Controls & Optimization
2
3
4
5
6
7
8
10
9
—Analytics Architecture
March 13, 2020 “Analytics Framework”, Industrial Internet Consortium (2017)Slide 12
Functional Domains
Business
Application
Streaming Analytics
Queries
Data ServicesHistorical Current
Information
Statistical CEP ModelQueries Statistical Machine Learning
Batch Analytics
Sense Actuation
Control
Edge Analytics
Industrial ControlHMI
Mac
hin
e Ti
me
Ho
rizo
n
Operations
Pla
nn
ing,
Pro
cess
Co
ntr
ol,
Engi
ne
eri
ng
De
sign
OperationalConditions & Events
Physical Systems
Models, Rules, Operational ParametersPlanning Time Horizon
—Agenda
March 13, 2020 “Analytics Framework”, Industrial Internet Consortium (2017)Slide 14
IIoT Analytics
Business Viewpoint
Usage Viewpoint
Functional Viewpoint
Implementation Viewpoint
Crosscutting Concerns
Questions and Answers
—Analytics Deployment Considerations
March 13, 2020 “Analytics Framework”, Industrial Internet Consortium (2017)Slide 15
Evaluation Criteria Analytics Location
Plant Enterprise Cloud
Analysis Scope
Single Site
Multi-Site
Multi-Customer
Response Time
Control Loop
Human Decision
Planning Horizon
…
—Analytics Design and Implementation Process
March 13, 2020 Slide 16
Business Case
– Customer Profile Analysis– Value Proposition Design– Environment and Data Catalog
Stakeholder Engagement
– Subject Matter Expert Review– Customer Story
Data Explorationand Preparation
Modelingand Analytics
Automation
– Data Integration– Application Integration
ContinuousImprovement
—Analytics Design and Implementation Process
March 13, 2020 Slide 17
Business Case
Business Case
– Customer Profile Analysis– Value Proposition Design– Environment and Data Catalog
Stakeholder Engagement
– Subject Matter Expert Review– Customer Story
Data Explorationand Preparation
Modelingand Analytics
Automation
– Data Integration– Application Integration
ContinuousImprovement
—Challenges in Papermaking Process
March 13, 2020 Slide 18
Machine Length
Water/Fiber
200
100
10
0
Consistency %
0.5 20 30 36 40 92
Measure, Control and Optimize
Papermaking needs to measure many process variables and product qualities for monitoring production operations and control automations
Papermaking consists of many complex processes that require various control solutions and techniques to produce high quality products
Papermaking is an energy intensive production process. There are many opportunities to optimize consumption of raw materials, utilities, and energy
—Analytics Design and Implementation Process
March 13, 2020 Slide 19
Stakeholder Engagement
Business Case
– Customer Profile Analysis– Value Proposition Design– Environment and Data Catalog
Stakeholder Engagement
– Subject Matter Expert Review– Customer Story
Data Explorationand Preparation
Modelingand Analytics
Automation
– Data Integration– Application Integration
ContinuousImprovement
—Data Science vs. Model-Based Analytics
March 13, 2020Harper, E., Zheng, J., Jacobs, S., Dagnino, A., Jansen, A., Goldschmidt, T., Marinakis, A., “Industrial Analytics Pipelines”,IEEE BigDataService (2015)
Slide 20
Prescription
Optimization
Prediction
Confirmation
Discovery
Description
Extract knowledge from data, using scientific discipline
Collect and clean raw data, explore relationships, develop models and algorithms, uncover patterns and predict outcomes
Effective with large amounts of data
Hindsight Insight Foresight
What am I missing?
Why might this be happening?
What is happening now?
What has just happened?Complexity
How do I make it happen?
What is the best that can happen?
What will happen if I takethis action?
What will happen next?
Val
ue
Recommendation
Maximization
Modeling
Forecasting
Data Mining
Correlation
Alerting
Situational Awareness
System models using subject matter expertise, physics, mechanics and dynamics of component interactions
Approximations using linear simplification of non-linear behavior
Effective with small amounts of data
Statistical function
Condition function
Health index assessment for a single asset
Statistical data
Utilization data
Condition data
RL, Statistical
RL, Utilization
RL, Condition
Degradation function WeighingHealth index
Value, numberClassification (G, F, P)ObservationExpert opinion
colour
inte
nsi
ty
—Analytics Design and Implementation Process
March 13, 2020 Slide 21
Data Exploration and Modeling
Business Case
– Customer Profile Analysis– Value Proposition Design– Environment and Data Catalog
Stakeholder Engagement
– Subject Matter Expert Review– Customer Story
Data Explorationand Preparation
Modelingand Analytics
Automation
– Data Integration– Application Integration
ContinuousImprovement
Extend Value Proposition
Revise Algorithms
—Simulation of Water Papering Process
March 13, 2020 HPC4MfG paper manufacturing project yields first results, https://phys.org/news/2017-05-hpc4mfg-paper-yields-results.html (2017)Slide 22
Multi-scale, Multi-physics Modeling
HPC for Manufacturing
Leverage advanced simulation capabilities, high performance computing resources and industry paper press data to help develop integrated models to accurately simulate the water papering process
Researchers used a computer simulation framework, developedat LLNL, that integrates mechanical deformation and two-phase flow models, and a full-scale microscale flow model, developedat Berkeley Lab, to model the complex pore structures in the press felts
Save paper manufacturers up to 20 percent of the energy usedin the drying stage – up to 80 trillion BTUs (thermal energy units) per year –and as much as $400 million for the industry annually
0.00 0.05 0.100.15
0.200.00 0.05 0.10 0.15 0.20 0.25
0.00
—Model Building Process
March 13, 2020 “Analytics Framework”, Industrial Internet Consortium (2017)Slide 23
Prepared Data(Selected Features)
Apply Learning Algorithm to Data
Candidate Model
Iterate to find best model
Machine Learning Algorithms
—AI / Machine Learning
March 13, 2020 Nazre, A., Garg, R., “A Deep Dive in the Venture Landscape of Artificial Intelligence and Machine Learning” (2015)Slide 24
Algorithms
KnowledgeRepresentation
PlanningDeduction, Reasoning,Problem Solving
Perception:Computer Vision
Artificial Intelligence
Supervised Learning
Decision Tree Learning Association Rule Learning
Support Vector MachinesInductive Logic Programming
Unsupervised Learning
Genetic AlgorithmsSparse Dictionary Learning
Clustering Similarity and Metric Learning
Reinforcement Learning
Manifold LearningDeep Learning
Bayesian Networks Neural Networks
Machine Learning Robotics: Motionand Manipulation
Natural LanguageProcessing
Social Intelligence
—Analytics Design and Implementation Process
March 13, 2020 Slide 25
Automation
Business Case
– Customer Profile Analysis– Value Proposition Design– Environment and Data Catalog
Stakeholder Engagement
– Subject Matter Expert Review– Customer Story
Data Explorationand Preparation
Modelingand Analytics
Automation
– Data Integration– Application Integration
ContinuousImprovement
—Streaming and Batch Integration
March 13, 2020 “Analytics Framework”, Industrial Internet Consortium (2017)Slide 26
Ind
ivid
ual
Pro
cess
es
Ap
plic
atio
nsData source/
Message queue
Speed (Real-time) processing
Machine learning model
Events, anomalies
Batch processing
Diagnostics, predictions
Hot, temporary storage
Cool, permanent storage
Queries
Responses
Queries
Responses
—Analytics Technology Choices
March 13, 2020Harper, E., Zheng, J., Jacobs, S., Dagnino, A., Jansen, A., Goldschmidt, T., Marinakis, A., “Industrial Analytics Pipelines”,IEEE BigDataService (2015)
Slide 27
Type Rationale
ExampleTechnology
Development Motivation Constraints
Hadoop Hive/Pig SQL Large static capacity and fault tolerance through data replication Computations are disk bound, high latency and response execution times
Indexed Solr/Lucene Indexed Query Real-time, scalable search with support for almost any type of data and file format Latency to create and maintain indexes
RDBMS JDBC SQLHigh productivity, legacy data store use requires no retraining. Support for transactions
Scalability, limited to structured data
Key-Value Pair
NoSQL CQL Simplicity of design, reliability (fault tolerance), and scalability Unstructured and schema-less data
Time Series Time SeriesSummary, Aggregates
Efficient storage and processing for high frequency dataData access as single columns, robustness is of importance given susceptibility to error due to missing data
Streaming Storm Java, Topology Quick insight from streaming and real-time data Slow recovery from faults
In-Memory Interactive SQL, Script Scalable dynamic capacity, low latency, and low (quick) overall response time Large main memory (RAM) requirements
Single Node R / PythonScript and Packages
High productivity, quick prototyping and proof of concept; rich data science libraries Limited in terms of scalability
GraphGraphX / GraphLab
Spark, TensorFlow
Intuitive and visual representations of computational problems, represent arbitrary data and systems as nodes and connections
Not all algorithms can be represented as graphs
Custom Custom Java High level of data and algorithm flexibility Custom programming, lower productivity
—Analytics Design and Implementation Process
March 13, 2020 Slide 28
Continuous Improvement
Business Case
– Customer Profile Analysis– Value Proposition Design– Environment and Data Catalog
Stakeholder Engagement
– Subject Matter Expert Review– Customer Story
Modelingand Analytics
Automation
– Data Integration– Application Integration
ContinuousImprovement
Propose New Products and
Services
Collect More Data
Data Explorationand Preparation
Identify New Tools
Identify New Technologies
—Agenda
March 13, 2020 “Analytics Framework”, Industrial Internet Consortium (2017)Slide 29
IIoT Analytics
Business Viewpoint
Usage Viewpoint
Functional Viewpoint
Implementation Viewpoint
Crosscutting Concerns
Questions and Answers
—System Characteristics Related to Analytics
March 13, 2020 “Analytics Framework”, Industrial Internet Consortium (2017)Slide 30
Industrial analytics requires many services from IIoT
Safety
Design industrial analytics processes and computations to prevent unintended operation and independently validate that the resulting actions do not harm lifeor property
Security
Provide defense in depth so that if a malicious or un-intended action compromises one security or accountability measure then another measure still guards the assets
Data Management
Common across tiers and accessible using a federated information model that supports search, classification and markup to enable rapid industrial analytics application development
Connectivity
Distributed architecture requires connectivity between components, not only between collocated processes but also across wide-area and global networks
—Agenda
March 13, 2020 “Analytics Framework”, Industrial Internet Consortium (2017)Slide 31
IIoT Analytics
Business Viewpoint
Usage Viewpoint
Functional Viewpoint
Implementation Viewpoint
Crosscutting Concerns
Questions and Answers