Tracking end of line manufacturing issues to their source can be a daunting task. Boston Scientific, in partnership with GraphAware, has used the Neo4j platform to build a manufacturing quality tool that offers dramatic improvements to the time, quality, and quantity of investigations. In this talk we will review a manufacturing value stream in a graph and discuss the analysis methods available, which result in striking increases in business efficiencies, for this unique application. We will also present how the system was implemented within the existing data architecture and then scaled from a laptop investigational tool to an enterprise-grade solution with Neo4j Server.
*Talk at GraphConnect NYC 2018*
3. ‣ Introduction
‣ Project Themes
‣ Business Problem
‣ Graphs Add Business Value
‣ Graph Model Evolution
‣ Extracting Insights From Graphs
‣ Additional Use Cases
‣ Questions
Outline
GraphAware®
4. ‣ Data science team lead
‣ Engineering background
‣ High-volume manufacturing & development
Boston Scientific
GraphAware®
Eric Wespi
eric.wespi@bsci.com
11. ‣ Business problem
‣ Discover graphs
‣ Exploration, vetting of technology
‣ Proof of concept (PoC) / Minimum Viable Product (MVP)
‣ Success & business value demonstrated
‣ Evaluate additional business opportunities
‣ Repeat
Boston Scientific:
A Case Study in Graph Adoption
GraphAware®
12. What caused a failure?
‣ Vertically integrated
‣ Batch processing
‣ Multiple teams
‣ Nonstandard analysis methods
‣ Lots of spreadsheet manipulation
Business Problem & Use Case
GraphAware®
+
15. ‣ Query times decreased from ~ 2+ minutes to ~ 10-55 seconds
‣ Streamlines the process
‣ Enhances overall efficiency
Graphs Add Business Value:
Performance
GraphAware®
16. Largest value added from non-functional areas:
‣ Simplicity
‣ Explainability
‣ Whiteboard friendly
‣ Data Accessibility
Graphs Add Business Value
GraphAware®
17. Estimates:
‣ 2.5 quintillion bytes of data created daily, accelerating
‣ 90% of the world’s data generated in the last 2 years
Graphs Add Business Value:
Data Accessibility
GraphAware®
18. Everyone has data
The competitive advantage
is going from data to
wisdom to action
Graphs Add Business Value:
Data Accessibility
GraphAware®
19. ‣ Shortest path
‣ Variable length queries
Graphs Add Business Value:
New Capabilities
GraphAware®
23. ‣ Every batch (node) gets a “score”
‣ Scores can be analyzed in a number of way
Extract and Analyze Graph Data
GraphAware®
Date
Score
Score
Process Data
24. ‣ Prepare data
Python
‣ Build and test database
py2neo/cypher
‣ Augment properties
Weights and scores
Extract Insights From the Graph
GraphAware®
Product
Qty: 100
Part A
Part B
Weight:
0.75
Issued: 75
Weight:
0.25
Issued: 25
Failure
…then develop a production-worthy pipeline through Hadoop:
26. Many start by transfer existing data model to the graph 1:1
‣ This approach is ok when getting started
‣ Follows crawl, walk, run
‣ May not take advantage of graph strengths
Graphs are highly flexible, easy to change
‣ Many accustomed to high barriers and costs of changes in non-graphs
‣ Leads to the “Mentality of initial perfection”
‣ Avoid this mentality and revise, evolve model easily, as needed
Graph Adoption Lessons
GraphAware®
27. ‣ TopAssembly (TA) has a “product” text property
‣ Next: extract “product” to dedicated node, relationship
Boston Scientific:
Graph Model Evolution
GraphAware®
28. match (ta:TA) with ta
create (:Product {name: ta.product})-[:HAS_PRODUCT]->(ta)
remove ta.product
Graph Model Evolution
GraphAware®
29. Data analysis expansion
add supplier and supplier facility data
Boston Scientific:
Graph Model Evolution
GraphAware®
30. ‣ Apply findings from existing products to new ones
‣ Alert other internal users of suspicious raw material batches more quickly
‣ Improve sensitivity to weak signals
Connecting Different Products
GraphAware®
Part
Part
Product
Failure
Raw Material
New Product
Failure
31. • Import raw text describing inspection failures
• Extract and correlate topics for better root cause
investigation
Future NLP Use Cases
GraphAware®
Event 1
Pertains to
Topic 2
Topic 1
Event 2
Pertains to
Pertains to
Part
Date/Time
Topic
1
Frequency