Smart Manufacturing: A Big Data Perspective Andrew Kusiak Iowa City, Iowa USA andrew-kusiak@uiowa.edu https://research.engineering.uiowa.edu/kusiak/ ISPR 2017, Wien, Austria
Outline Introduction Data-driven modeling Pillars of smart manufacturing Hypothesizing the future Data science in manufacturing Optimization in a data-reach environment Conclusion ISPR 2017, Wien, Austria
The Future is Promising In 2001 R. Kurzweil (Director of Engineering at Google) in an essay The Law of Accelerating Returns predicted that the 21st century may experience 20,000 years of progress (at today s rate) D. Butler, Nature, Vol. 530, Feb 2016
Smart Manufacturing Concept Cyberspace System intelligence Data Decisions Interface Standard connectivity Data Decisions Manufacturing equipment Local intelligence A. Kusiak, Smart Manufacturing, IJPR 2017 (published online)
Pillars of Smart Manufacturing Materials Data Manufacturing technology and processes Smart manufacturing Predictive engineering Resource sharing and networking Sustainability A. Kusiak, Smart Manufacturing, IJPR 2017 (published online)
Making Manufacturing Smart with Data Data Science Bottom up modeling No limits on the type and number of parameters High model accuracy Data Mining Decision Making/ Optimization ~½ Solution ~½ Solution
Example: Wind Power Balancing 1 Pa = ρπ RC ( λ, β ) v 2 2 3 p = P Classical science = Data science a = ω T r Pictures courtesy of Danish Wind Energy Association A. Kusiak, Share Data on Wind Energy, Nature, Vol. 529, No. 7584, 2016, pp. 19-21.
Classical Control Known set point Adjustable input P0 Industrial process P Controller Today s manufacturing: Known set point = Production output
Wind Turbine Control Anticipatory Control Unknown set point Non adjustable input P0 Wind Turbine Controller P Tomorrow s manufacturing: Predicted set point = Production output
Intelligent Manufacturing: History 1990
Common Manufacturing Models of the Last Four Decades Flexible manufacturing systems (late 1970s) Computer-integrated manufacturing systems Reconfigurable manufacturing systems Holonic manufacturing systems Bionic manufacturing systems Intelligent manufacturing Smart manufacturing
International Activities in Intelligent Manufacturing IMS Program (Japan, 1995) NGMS, IMS (CAM-I, USA) IMS EU 12
Smart Manufacturing: Contributing Computing Concepts Service-oriented architectures Cloud computing Cyber-physical systems Internet of things (and everything) Sensor networks A. Kusiak, Smart Manufacturing Must Embrace Big Data, Nature, Vol. 544, No. 7648, 2017, pp. 23-25.
New Manufacturing Initiatives Industrie 4.0 (Germany) Factories of the Future (EU) Made in China 2025
Characteristics of Smart Manufacturing (1) Expanded condition monitoring Self-diagnosis Self-correction, repair, self-healing Self-organization Increased adaptation and scalability Variable batch size (from 1 to large) Reduced production ramp-up time Reduced change-over time
Characteristics of Smart Manufacturing (2) Polarization of coupling between manufacturing enterprise and manufacturing assets Corporations with a weak coupling, e.g., sharing and leasing of mfg equipment and facilities Corporations with a strong coupling, e.g., material, product, and process created to serve the same purpose
Smart Factory Primary differentiators: Predictive engineering Seeing the future Sustainability (including energy and transportation) From product conception to the end-of-life
Product End-of-Life Restored 1949 VW Bug Reuse (most preferred) Remanufacture Recycle Disposal (should disappear)
Emerging Priorities New materials, processes, and products Quick path from material design meeting customer needs and production Material-process-product paradigm Engineering biology and bio-products Developments in biology and genetics to benefit manufacturing chemicals, materials, fuel, and cells Integrated manufacturing E.g., integration of manufacturing medication substances and medications into a single integrated process
Bio-based Materials: Examples Petro-based products replaced with bio-based products E.g., rubber from dandelions; Fraunhofer Institute for Molecular Biology, Munster, Germany By 2020 IKEA plans to manufacture all plastic products and toys from renewable/recycled materials Lightweight plastics from agave Ford Motor Corporation 20
Additive Manufacturing: A Game Changer (1) The success hinges on manufacturing of artifacts: having the right properties (e.g., strength, surface quality, material shrinkage) viability in providing unattainable features (e.g., materials of different elasticity in one) by the progress in: component and product design materials, and processes
Additive Manufacturing: A Game Changer (2) Big Area Additive Manufacturing (BAAM) E.g., car chasees, molds for wind turbine blades Small Area Additive Manufacturing (SAAM) E.g., medical implants Material-Process-Product Design Paradigm
New Business Models What these companies have in common? Each is the largest in its category None of them owns or produces any assets it is known for
What Have we Learned from Them? Using customers to design products Innovation
Smart Transportation Sustainable vehicle design Renewable energy Electric vehicle Renewable energy Electric vehicle Non-renewable electricity Traditional vehicle Fossil fuel Vehicle type/ Fuel type Shared Vehicle automation/ Use mode Autonomous Semi-autonomous Connected Traditional
Integration of Manufacturing and Transport Internal and external material handling and transport E.g., wind energy supply chain Globally distributed production Transportation in supply, distribution, and maintenance Meeting changing market needs Transport sharing
The Future of Smart Manufacturing Imagining the future of smart manufacturing Ten conjectures A. Kusiak, Smart Manufacturing, IJPR 2017 (published online)
Conjecture 1 Manufacturing Digitalization Manufacturing will increasingly depend on data Justification Manufacturing could benefit from wind energy and process industry where supervisory control and data acquisition (SCADA) systems have been used to capture, store, and share data A. Kusiak, Smart Manufacturing Must Embrace Big Data, Nature, Vol. 544, No. 7648, 2017, pp. 23-25
Conjecture 2 Increased Need for Modeling, Optimization, and Simulation Delivery of value from manufacturing data Justification Data flow across different domains (e.g., product, process, and logistics) Dynamic and predictive models Virtual and augmented reality
Conjecture 3 Product-Material-Process Phenomenon Growing instances with the material, process, and product developed simultaneously Justification Design of a part that for which a new material and a 3D printing process have been developed A. Kusiak, Innovation Science, Nature, Vol. 530, No. 7590, February 2016
Conjecture 4 Vertical Separability of the Physical Assets and the Cyberspace The physical and the logistics layers to be designed for ease and speed of connecting and disconnecting Justification The need to reconfigure physical assets driven by the changing product needs
Conjecture 5 Enterprise Dichotomy Two extreme smart enterprise models may emerge, one where the physical and logistics layers are tightly horizontally connected and the other with vertical separability of the two layers Justification The horizontal connectivity and the vertical separabilty models may emerge as the result of Conjecture 3 and Conjecture 4, respectively
Conjecture 6 Horizontal Connectivity and Interoperability Increase of horizontal internal and external connectivity and interoperability Justification The growing volume and flow rate of data across an enterprise will naturally lead to greater horizontal connectivity and interoperability
Conjecture 7 Resource Sharing Sharing manufacturing and transportation resources across manufacturing chains will become a common practice Justification Horizontal connectivity combined with dynamic markets will facilitate sharing manufacturing equipment, transportation, and other resources Expanding globalization and competition form emerging markets may enhance resource sharing
Conjecture 8 Equipment Monitoring, Diagnosis, and Repair Autonomy Diagnosis and prediction of equipment faults will become routine. Autonomous repair will occur. Justification Sensors will provide data to monitor and predict health status of equipment and systems.
Conjecture 9 Cybersecurity and Safety Cybersecurity and safety issues will remain a challenge Justification Increasing degree of automation, system autonomy, and connectivity will raise the importance of cyber protection and human safety
Conjecture 10 Standardization and Collaboration Collaborative development of standards may naturally emerge to meet the emerging needs of integration and interconnectivity Justification Growing reliance on data (Conjecture 1), resource sharing (Conjecture 7), and the need for vertical separabilty (Conjecture 4) and horizontal connectivity and interoperability (Conjecture 6) will drive the need for standardization and collaboration
New Platforms Three practical steps need to be taken to accelerate progress in smart manufacturing A. Kusiak, Smart Manufacturing Must Embrace Big Data, Nature, Vol. 544, No. 7648, April 2017
Establishment of Cyber-platforms for Modeling, Sharing, and Innovation Online or physical spaces are needed enabling interaction among experts and practitioners to develop models and technical solutions Such platforms could mirror maker spaces or innovation hubs Transparency and openness as well as diverse ideas and cultures should be supported Schemes for modelers to access data are needed
Enact Smart Manufacturing Policies Government should fill the gaps lacking ownership or that are too risky to pursue by private companies The 2016 Report by the Information & Technology Innovation Foundation called upon the U.S. Congress to expand federal resources for training and to assist small and medium-size businesses to adopt smart manufacturing technologies
Data-Driven Manufacturing Modeling from data Solving data-derived models
Data Modeling from Data Application Model building Model solving ~½ Solution ~½ Solution
Extreme Learning What is extreme learning? Extreme learning machines involves feedforward neural networks for classification or regression with a single layer of hidden nodes The value of the weights connecting inputs to hidden nodes are randomly assigned and never updated
Extreme Learning Machine Extreme Learning Machine (ELM) Single hidden layer feedforward neural network A three-step learning model Offers favorable generalization and quick learning
Deep Learning What is deep learning? Deep learning involves a class of machine learning algorithms that: Use multiple layers of nonlinear processing units for feature extraction and transformation Learn multiple levels of representations corresponding to different levels of abstraction May be supervised or unsupervised
Algorithms Deep Neural Networks (DNNs) Involve of many hidden layers Suitable for modeling complex non-linear problems Used in both classification and regression
Algorithms Deep Auto-encoder Intended for dimensionality reduction Same number of input and output nodes Unsupervised learning
Algorithms Deep Belief Network (DBN) Involves Restricted Boltzmann Machines (RBMs) where a sub-network hidden layer serves as the visible layer for the next layer Has undirected connections at the top two layers Supports unsupervised and supervised learning
Algorithms Convolutional Neural Network Inspired by the neurobiological model of the visual cortex Well suited for 2D data such as images
Model Solving Algorithms Evolutionary computation Particle swarm optimization Ant colony optimization Artificial immune system 50
Innovation High Market indicator 2 Creation Invention Innovation High risk Low success rate path Market indicator 1 Low High A. Kusiak, Innovation Science, Nature, Vol. 530, No. 7590, Feb 2016
Conclusion Materials, products, and processes are becoming smarter, sustainable, energy aware, and innovation driven Growing importance of data collection, analytics, modeling, and knowledge deployment Co-dependence of materials, manufacturing processes, and products Emergence of new manufacturing domains, e.g., healthcare ISPR 2017