SMART MANUFACTURING: THE NEXT INDUSTRIAL REVOLUTION Caroline Hargrove & Rob Bowyer Cambridge, 3 rd February 2017 Presented at Consor,um for the 4th Revolu,on | Execu,ve Briefing Day (#C4IR) Cambridge, UK 2-3 February 2017 | www.cir-strategy.com/events
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
SMART MANUFACTURING: THE NEXT INDUSTRIAL REVOLUTION Caroline Hargrove & Rob Bowyer Cambridge, 3rd February 2017
The world of manufacturing is evolving: • Rapid technological development • Ever increasing computational power have sent
ripples of digital disruption throughout the industry • Rise of the Industrial Internet of Things (IIoT) is
paving the way for the 4th industrial revolution
BACKGROUND
4
The IIoT combines real and virtual production environments into a single view of the factory Smart manufacturing describes the application of these systems to real production lines
– using embedded systems, – communication platforms and – process intelligence
Decision support tools enable manufacturers to understand the impact of individual line productivity on the overall factory performance
BACKGROUND
5
The motivation to harness the IIoT is usually to achieve improvements across a range of manufacturing KPIs including: • Increased yield • Increased throughput • Increased productivity • Reduced costs • Reduced downtime
This can be difficult to achieve due to unique characteristics of each production process
CONDITION INSIGHT
PRODUCTIVITY MIRAGE
7
Typically, manufacturing plants prioritise their efforts on reducing downtime, of which maintenance programmes are a key driver
Condition-based monitoring (CBM) or Pre-emptive maintenance (PEM) are seen as the natural solutions but • It is difficult to develop robust solutions • Ignores link to quality - usually separate
functions within the business
Hence the ‘mirage’ effect – one never really gets the expected benefit from CBM/PEM
MAT IIoT FOCUS
8
Our approach is different – rather than starting from • Condition Monitoring / PEM focus: fault
diagnostics – Challenging problem; may not be unique solution – Needs cloud to join sources of data – Needs ground-truth from NLP platforms etc. – Long term validation required
… we start with a • Quality focus: anomaly detection
– Needs high rate sensing at edge of network – Appropriate choice of sensors – Bespoke key features – Rapid deployment – Informs CBM
SYMPTOMS & LEAD TIMES
9
Time
Conditions start to change
Vibrations Noise
Heat
Smoke
Emergency Stop
3 month 2 weeks 2 days
10 min
Mac
hine
con
ditio
n Anomaly detection works early on (lead indicator) Traditional CBM/PEM
<10ms
CLOUD vs EDGE of NETWORK
Developing solutions for the edge of the network requires application specific engineering for the end-to-end solution to work including • Domain knowledge • Agile feature development • Secure over-the-air download • Configuration management
Our integrated data acquisition, workflow & algorithm development environment inspired from F1 racing allows us to do this efficiently
10
GENERIC SYSTEM OVERVIEW
11
Smart Sensor
Analytics Engine
PLC
Analytics Engine Execution of local analytics for: − Real-time decisions for
machine operation. − Reducing data volume for
upstream analysis through various strategies.
Gateway
Data storage
ATLAS / MIDAS
Market App
Market App
Data & Analytics Platform
EDGE LAN
Smart Sensor Feature extraction from very high frequency data.
CLOUD
Market Application Dashboards containing performance KPIs, and decision support.
Data Analysis & Model Design Tools Workflow & tools to undertake diagnosis of issues, and to design & improve models for execution line-side.
CASE STUDY: MDI CRIMPING
CRIMP OPERATION - PHYSICAL
13
CRIMP OPERATION SEQUENCE
14
FEATURE EXTRACTION
15
Reasonably consistent force traces: • Most variation between
heads
• Stochastic noise evident on ‘shoulder’
FEATURE EXTRACTION
16
Extracting key features from within a force trace: • Contact force • Max force • Number of Stiction/friction
peaks • Contact force multiplet • Crimping force multiplet • Force duration
Use features in thresholding/clustering for identifying anomalies
NUMBER OF OUTLIERS
17
For the contact force 0.0425% of cans considered outliers (in bins with very low numbers of samples <1000) out of 1.34 million cans For 75,000 cans this would be ~30 cans Example: Contact force – force at initial contact of crimp head and can
Cloud platform
HADOOP
Real-time gateway
SYSTEM OVERVIEW - PHYSICAL
18
Analytics Engine PLC
Analytics Engine Feature detection and thresholding
HMI
LINE-SIDE FACTORY CLOUD
Data Store
SPARK
DECISION INSIGHT
DECISION INSIGHT Production is not only about uptime and quality but also operating the facility with competing KPI’s (cost, quality, schedule, shelf-life, inventory) • Real-time decision support tools help operators
avoid ‘local’ optimisations and miss the bigger picture
• Best results come from being forewarned => predictive modelling supported by what-if scenario planning
20
DECISION INSIGHT
21
Decision Insight – a framework for prescriptive analytics
DECISION INSIGHT
22
We use simulation to aid understanding in a complex world • Performance (KPIs,