eSMART Factory Conference 2019, Day 1
I had the pleasure of attending the eSMART Factory Conference in June in Dearborn, Michigan. This is the second annual conference on the smart factory. SMTA described it as "A technical conference with a focus on electronics manufacturing from software systems/processes to augmented reality and smart inspection." It was held at the rustic Dearborn Inn across the street from the Henry Ford Museum/Village and Ford Development Center.
Trevor Galbraith, the conference chair, opened the conference and introduced the keynote speaker, Irene Petrick, Ph.D. Her keynote was titled, "Building an AI Ready Culture."
Irene Petrick is the director of Intel’s industrial innovations. Her keynote highlighted the overview of the elements of the intelligent factory, including:
- Software-defined industrial equipment and collaborative machines
- Self-organized, flexible production flow
- Self-aware production systems
- Human in the loop-enabled connected worker
- Real-time and reliable computer and connectivity
- Embedded equipment-to-equipment (E2E) security
In her talk, Petrick stated that AI has been a long time coming and may not be ready for everyone yet. But the technology underpinnings for AI are part of the strategic fit of a modern corporation and fit the competitive strategies and architectures for the future. As seen in Figure 1, the smart factory has already started. The underpinnings are currently in place, such as:
- Data collection
- Data aggregation and actions
- Data scoring, analytics, and model updates
- Business intelligence and results sharing
- Feedback to update models and achieve corrective actions
Figure 1: Opportunities for the eSmart Factory involve data collection analytics and resulting corrective actions (Source: Intel Presentation).
AI's contribution to data collection will be hindsight, insight, foresight, forecast, and act/adapt. Further, machine learning is one of the big opportunities for AI progression. The journey from unconnected machines to smart, connected machines leads to intelligent factories. In traditional machine learning, the output is known and has the least workload requirements; meanwhile, with advanced machine learning, the output is unknown and has the highest workload requirements.
Petrick shared that Intel’s own journey started in the ‘80s with wafer fabrication (hands-on) through the '90s until today’s lights-out wafer fabrication (no-hands). Intel conducted extensive interviews with industry professionals (411 participants and over 93 hours of interviews) to discover the stages the industry is today for digital usage . The results included:
- Low usage (25%), medium usage (46%), and high usage (29%)
- The participants had multiple facilities in multiple countries (43%), changeovers or setups very frequent (73%), used no automation or control systems (19%), and work or support predominately discreet manufacturing processes (68%)
- Industries participating included petrochemical (27%), metal fabrication (29%), electrical equipment and components (17%), food and beverage (9%), wood and paper products (7%), and transportation (6%)
- All participants wanted the future intelligent factory
As important as machines are, automation and the smart factory is not going to replace all workers. AI must facilitate the coevolution of workers and operations. It will influence decision-makers and influencers at all levels. Foremost, it will address skill gaps and provide an understanding of gaps. It is the leaders and hardcore doers that will drive transformational changes.
In conclusion, Petrick emphasized leadership and how it’s all about data and trust. She advised the audience to think big, start small, be holistic, and be the partner of choice, which will be a competitive advantage.
The second presentation was by Michael Ford of Aegis Corporation on “How IPC Standards Are Pivotal for Industry 4.0 Achievement.” He provided a summary and report on how IPC-2581 and IPC-CFX (Connected Factory Exchange-2591) together are enabling the smart factory. Ford is chairman of the IPC-1782 Component Traceability Committee and has played a significant role in bringing CFX into fruition. CFX is the only free, open-source, consensus-based Industry 4.0 standard on the market for electronics assembly.
CFX was conceived and implemented because of numerous “standardless” interfaces being created all over the globe for electronics assembly. The dilemma that OEMs faced was a collection of different protocols that would require massive re-engineering to have a coherent operating system. Many were omnidirectional and incompatible, moving data from point to point but not achieving the real goal of Industry 4.0, which is improved productivity, profits, and quality. Thus, an IPC committee was formed to create a consistent, open solution that everyone could use. In a very short time, IPC-2591 was created with over 300 parties participating from over 100 companies around the world to produce a digital manufacturing standard suitable for “plug-and-play” connectivity. No licensing, contracts, or dependencies are required.
The plug-and-play data content is defined by three industry standards:
1. Communication protocol, including advanced message queuing protocol (AMQP) used in banking
- Secure (like a bank ATM)
- Encrypted (as an option)
- Send and forget (host)
- Point to point (direct)
2. Defined language content
- Structured topics and messages across all manufacturing elements (materials, machines, production, resource performance, maintenance, information systems)
- Easily build a model for any type of automation
IPC has made available free, open-source software development kits (SDK) for Windows, .Net, Linux, Java, etc., including:
- Extended-reach Linux SDK
- Small outline integrated circuit (SOIC) CFX client (20 mm x 20 mm for a $9 chip)
- CFX client kernel (
- Examples: Inside a soldering iron, torque wrench, pick-and-place machine, etc.
- For dumb machines
- CFX client on Raspberry Pi ($35)
- Multi-digital I/O with tailored software
- Example: Seica “shoebox”
IPC was also able to put together CFX because of an older standard—IPC-2581, the digital product model. CFX includes a complete digital PCB product model all in a single file with design and local bill of materials (BOM) data, process information and variants. Further, it is ready for direct process engineering tasks, is suitable for legacy products, and provides complete machine-readable data for all aspects of PCB manufacturing. Further, CFX is the only standard that provides electronic stackup exchange and is proven to improve product quality and first-pass success, enabling smart factory best practices. For more information on CFX and IPC-2581, visit www.ipc-cfx.org.
The third presentation, "Deployment Excellence: Electronic Production Process Planning," was presented by David Meyers of Siemens PLM. He gave an overview of Siemens’ software products for electronics assembly. The electronics market is one of the world's largest at $2.1 trillion with a steady 3.2% compound annual growth rate (CAGR). Computers and telecommunications are the largest segments, but automotive is the fastest growing at 4.2% CAGR. Top generators of electronic revenues are China, the U.S., Japan, and Germany. PCB assembly and box-build are a $400 million market with 5.1% CAGR. There is estimated to be over 20,000 SMT lines worldwide with 60% being OEM and 40% CEM (but this excludes Foxconn, the largest OEM). Electronics is the most dynamically growing market with innovations adding new products, such as autonomous vehicles, IoT connected everything, 5G, and augmented and virtual reality. Thus, electronics assembly provides the greatest opportunity for the smart factory.
Modern manufacturing also introduces many new challenges:
- Lot sizes are getting smaller, heading toward a custom lot size of one
- Materials, especially components, are sensitive and growing in cost and complexity
- Flexibility in the workforce is essential along with rapid training
- Data collections are growing and must be used effectively
- Planning and optimization become the foundation of profitability
- Top quality is expected even with rapid product evolution
Siemens' suite of product lifecycle management (PLM) products is called Camstar and consists of:
- Enterprise resource planning (ERP), PLM, manufacturing analysis, design verification, manufacturing planning, production execution
- Advanced materials management to reduce inventory, waste, and inspection/repair
- Support for the digital twin of product, production, and performance
- Intelligent data formats of Mentor, Valor, ODB++, open manufacturing language (OML), and OPC unified architecture (OPC UA)
- Seamless new product introduction (NPI) integration with detailed planning and scheduling
- Plug-and-play shop floor connectivity hardware
Meyers concluded with three eSmart Factory case studies, including 1) their own Furth assembly facility in Germany, 2) a high-volume telecommunications, such as the federal telecommunications system (FTS), and 3) a high-volume automotive FTS.
The fourth presentation was by Michael Schuldenfrei of Optimal Plus on "Smart Manufacturing: Shifting the Focus From Machine to Product." He has focused on the rapidly growing automotive electronics market and its need for new levels of quality and reliability for its electronics.
The automotive electronics environment has the following challenges:
- 90% of car innovations and new features are driven by electronics; for example, 7,000 semiconductor devices/car X 1 dppm failure rate X 4,000 cars produced/day = 1 car failure an hour and 50% of failures are “no defect found”
- 22% of warranty costs are related to electronics and semiconductors
- 3X increase in car recalls from 2014–2016 due to electronics
- By 2025, it is estimated that 50% of a car’s cost will be for electronics
These challenges can only be solved by adding a product-centric approach to advanced analytics. Data sources must expand from machine data to product data as well. Analytics have to collect and harmonize data from all parts of manufacturing. And this data must be turned into actions unified across all silos, including product, process, machine, and time.
The product-centric approach has three application characteristics:
- Multiple dimensions handled in a single query (product, equipment, genealogy, measurements)
- Complex filtering, grouping, and aggregating in the database (descriptive statistics, pre-filtering of outliers, aggregations)
- Data schema and query engine must be scalable, flexible, and comprehensive
- Should answer questions like:
- Which configuration parameters in the reflow oven best predict inspection failures?
- Which boards contain chips from a specific batch of wafers?
- Which combination of parts is likely to cause failures?
- The user experience must be simple
- Should do things like:
- Prevent bad parts from shipping
- Reduce scrap
- Drift detection
- Requires end-to-end integration, control, and monitoring across the supply chain
Putting it all together from applications characteristics to data characteristics, the product-centric approach leads to a high-quality index. Schuldenfrei also provided a few automotive case examples to illustrate the new level of quality and reliability challenges to be met that siloed machine data alone could not solve, saying, “A holistic product analytics approach is required to meet zero-defect requirements.”
Marybeth Allen of KIC Thermal Systems Inc. gave the fifth presentation about, “Industry 4.0: The Next Industrial Revolution—The Smart Factory.” This was another presentation about the benefits of creating a smart factory. Allen summarized how to:
- Use data to run the factory more effectively
- Use data analytics and optimization
- Acquire sensor-based technologies to obtain data
- Utilize industry connectivity, such as CFX
- Employ process visibility and traceability for sustainable production
Challenges include automating and integrating, which means:
- Factory-wide connection to all equipment (automatic and manual)
- Event data/connected devices
- Search, filter, and analyze
- Web and mobile
- Enterprise Integration
Thus, smart factory integration involves:
- Refined decisions
- Production planning
- Machine-to-machine learning
- Corrective actions
- Scheduled maintenance
- Product tracking and traceability
- Acquisition and processing of machine parameters
But the benefits to be achieved include:
- Immediate corrective actions
- Lower production costs
- More competitive operations
- Consistent quality
- Higher profits
- More happy customers
The final three speakers of the day had subjects on software systems and processes. The first was Dr. Bill Cardoso of Creative Electron who discussed "How AI Is Changing the Way We Make and Inspect Things." He gave an overview of how AI helps in the data collection and interpretation for automated optical inspection (AOI) and X-ray image analysis.
On the modern connected SMT assembly line, defect data is associated with paste application, component placement, and reflow. Statistical process control (SPC) can statistically analyze the data to identify drifts, trends, and other relevant quality issues. But how does data become information? The process still depends on operators catching the problem and requires an experienced line manager to debug the problem and start corrective action.
The replacement for this is data fusion and AI. AI software lowers the barrier to entry by training a fusion system using the AI engine. AOI and automated X-ray inspection (AXI) identify bad ball grid arrays (BGAs) or solder joints and the fusion dataset maps the source of the problem. The AI training progresses from operator warnings to fully automated, self-healing actions.
The end goal is more complete digital twin models that catch all of the problems and provide predictive results.
The second afternoon talk was by Greg Vance of Rockwell Automation on “Automating Detection of Pick-and-Place Nozzle Anomalies.” He recapped their progress by implementing smart factory data analytics—in this case, the analysis of pick-and-place nozzle monitoring. Rockwell Automation has 23 SMT lines operating 24/7 with up to 2,200 pick-and-place nozzles in use placing 5.5M components per day. Nozzle monitoring detects clogging and wear on the pick-and-place nozzles that lead to machine downtime, placement errors, and reduced quality.
The original software approach was Cloud-based software to consolidate and format streaming data, but there was latency and a lack of the single point of failure for the visualization. Using a new AI-based Cloud analytics software with machine learning, they created six pick-and-place nozzle monitor dashboards. The real-time analytics can self-correct or settle in placement performance while providing dynamic optimization and alters, even to their cellphones.
The impact of the new app and alarm brokering will lead to:
- Descriptive notifications to technicians indicating what machine, head, and nozzle has a problem, reducing troubleshooting time by 30 minutes
- Improved SMT line productivity by reducing machine stoppage and waste by up to 15%
- Resolved machine component pick difficulties that may result in an end-of-line defect
Future applications include in-circuit test (ICT) probe characterization, software process improvement (SPI) failure rate, AOI process failure rate, and product performance by work order over time.
The final talk was on "Smart PCBA Manufacturing: How Software-Powered Automation Increases Innovation and Reduces Production Time" by Shashank Samala of Tempo Automation. The startup in San Francisco focuses on developing PCB assembly automation from the IPC-2581 backbone. Their connection of solder paste, pick-and-place, AOI, and AXI for assembly to BOM sourcing software, fabrication simulation, inspection defect detection, and design records is now eclipsed by CFX. They started before CFX was developed and were surprised by the rapid development and implementation of CFX. Fortunately, since CFX is open, they can rapidly adapt their unique brand of software to the CFX format.
The day ended with a panel discussion moderated by Trevor Galbraith of Global SMT and Packaging with panelists Michael Ford, David Meyers, Michael Schuldenfrei, and Greg Vance who discussed the topics covered during the day.
The conference was co-organized by SMTA and Global SMT, and presentations will be available to SMTA members after September in the SMTA Knowledge Base.
Dr. Faith McCreary and Dr. Irene Petrick, “Industry 4.0 Demands the Co-Evolution of Workings and Manufacturing Operations
,” Intel Corporation, Q1 2018.