Towards a Domain-Specific Framework for Predictive Analytics in Manufacturing. David Lechevalier Anantha Narayanan Sudarsan Rachuri

Transcription

1 Towards a Framework for Predictive Analytics in Manufacturing David Lechevalier Anantha Narayanan Sudarsan Rachuri

2 Outline 2 1. Motivation 1. Why Big in Manufacturing? 2. What is needed to apply Big in Manufacturing? 2. A Framework for Predictive Analytics in Manufacturing 1. Capabilities of the Framework 2. s of the Framework 3. Interactions of the Framework s 3. Research Challenges to Develop this Framework 4. Summary 5. Future work

3 1.1 Why? A Huge Return on Investment 3

4 1.2 What is needed? Infrastructure Solutions to connect and Infrastructure Processes Resources Analytical models scientists Manufacturers 4 Source: Manyika, James, et Products al. Big data: The next frontier for innovation, competition and productivity. Technical report, McKinsey Global Institute, Computing 2011 Infrastructure

5 2.1 Capabilities of the Framework For Predictive Analytics Be intuitive and easy to use Represent physical components and their behaviors Generate analytical models Handle high volume, velocity and variety of data Make results understandable for manufacturers 5

6 2.2 s of the Framework A ing Interface (DSMI) Schema Repository Analytical Acquisition Visualization 6

7 2.3 Interactions between Framework s Schema Repository Libraries management Acquisition reuse of ing Interface Analytical Energy=f(x1,x2,x3,x4) Visualization x1 x2 x3 x4 7 Translation Visualization

8 3. Research Challenges in the Framework Schema Repository Libraries management Acquisition reuse of ing Interface Analytical Energy=f(x1,x2,x3,x4) Visualization x1 x2 x3 x4 8 Translation Visualization

9 3. Research Challenges for the Domain- Specific ing Interface (DSMI) Multilevel modeling Factory level, machine level, process level ing Interface Multiple viewpoint based on user interest Highlight flows (material, throughput, energy ) Highlight specific machine Highlight specific process 9

10 3. Research Challenges in the Framework Schema Repository Libraries management Acquisition reuse of ing Interface Analytical Energy=f(x1,x2,x3,x4) Visualization x1 x2 x3 x4 10 Translation Visualization

11 3. Research Challenges for the Domain- Specific Schema Repository Schema Repository Representation of: 1. Manufacturing components and their behavior To allow manufacturers to represent their systems 2. Analytical concepts Optimization Analytics: Bayesian Network, Neural Network System states and diagnostic concepts To define a specific problem that needs to be studied

12 3. Schema examples Manufacturing components Analytical concepts System states Combination Schema to combine manufacturing component, analytical Schema concepts to represent and system a Bayesian states Network 12 Schema to represent machine states Schema to represent a factory

13 3. Research Challenges in the Framework Schema Repository Libraries management Acquisition reuse of ing Interface Analytical Energy=f(x1,x2,x3,x4) Visualization x1 x2 x3 x4 13 Translation Visualization

14 3. Research Challenges for the Analytical Analytical Energy=f(x1,x2,x3,x4) Translation: Automatically generate analytical models that: represent manufacturing system and its behavior are appropriate for the user s objective are standard-based to facilitate interoperability 14

15 4. Summary Manufacturers Schema Repository scientists Uses concepts from ing Interface Visualization x1 x2 x3 Get analytical models from Analytical Energy=f(x1,x2,x3,x4) Acquisition 15

16 5. Future work Continue working on the schemas to cover additional concepts in manufacturing and analytics Define model translation to map between descriptive and analytical models Work towards standardized representation of analytical models for manufacturing 16 Implement a case study to illustrate the potential of the framework

17 Questions / Discussion

18 3. Research Challenges in the Framework Schema Repository Libraries management Acquisition reuse of ing Interface Analytical Energy=f(x1,x2,x3,x4) Visualization x1 x2 x3 x4 18 Translation Visualization

19 3. Research Challenges for the Acquisition and Visualization s Handle structured and unstructured data Handle data flows with: Real-time data Big volume of data Handle data with standard-based formats 19