Outcomes & Report Out for Roundtable 3

Roundtable 3 outcomes

1) Why do we want to share data as an industry (e.g. weather data, cancer data, financial data)?

How do you categorize the specific types of manufacturing challenges or opportunities with respect to greatest benefit and readiness or ability to share and combine data for AI applications; what curated/contextualized data are of greatest value and how do we position for scaled availability and use of curated data?

  • Within industry we can learn about user interface issues, user experience and opportunities for improvement.
  • Industry may want to share data but is concerned about the unintended consequences (e.g. data set can be used to reverse engineer the part). What research can be done to help mitigate?
  • Network building block relationships and interconnections can be important to build motifs for forming a decision.
  • Manufacturers need guidance to start their AI journey and what data to collect, how it can be analyzed and then utilized.
  • If the machine manufacturer can share data with the end-user and vice versa, AI models may be trained faster and predict maintenance and quality issues with more precision.
  • Manufacturing maintenance logs in natural language have been examined by NIST and offer an AI opportunity that may be less likely to experience reverse engineering.
  • How can AI help guide knowledge extraction from subject matter experts. Can it be generalized or is the issue machine specific?  Can an AI assistant help interview people.
  • Gamification of data collection and supplied information usage incentivization for the worker to support AI.
  • User friendly ontologies could be used as a mapping tool.
  • Quality assurance in additive manufacturing could deliver value using AI to optimize the machine material interaction for each layer.
  • Data can facilitate research
 

2) How do we share data?

With respect to the current options, how do they align with scaled use of shared data and AI models in a secure and privacy-preserved manner while providing assurances of intellectual property protection to cultivate trust across the industry?

  • Why is there a disparity in the extent of data sharing across sectors and what can we do to enable manufacturing data sharing to a larger extent?
  • Use cases where manufacturers benefited from other manufacturers could help others understand the benefits and value of data sharing.
  • If the use case or demo is difficult to scale/generalize, its impact will be limited both internal and external to the manufacturer.
  • Protecting data security vs. privacy may require different levels of protection.
  • Common data standards, ontologies are needed for meta data structures to be effectively shared.
  • Legal liability issues need to be understood and/or mechanisms like “Data Trusts” could be used.
  • How do we leverage small size data that may be complex with large data with high signal to noise ratio and low dimensionality?
  • Data generation needs to consider the complexity of the source and not lose the richness.
  • Would federated learning help alleviate intellectual property issues? Should data from USG grants be broadly available?
  • Receiver of data needs to be able to trust the data (source, maintenance and communication).  Standards and methods to maintain trust in the data chain need to be developed.
  • Smaller, focused data cooperatives may be a good way for SMMs to store and share data.
  • Data curation is required.
 

3) What drives/motivates the sharing of data?

What are the business models for data sharing adoption to start and accelerate scaled AI implementation in the SMM supply chain so all manufacturers can participate in the ecosystem of data and models?

  • If the use case is understood and valued, there is much more interest in supporting with data.
  • SMM adoption is often driven by the large OEMs who have developed successful use cases and cascaded the requirements.
  • SMMs are concerned that they will not have enough resources to support multiple unique OEM processes.
  • Certification and validation company relationships may offer a view of how to handle proprietary data.
  • SMMs will be motivated by a return on investment and the ability to connect with their supply-chain.
  • Non-certified/regulated industries/processes may offer a white-space for data sharing.
  • Digital twin of a manufacturing process could be a strong motivator for the sharing of data.
  • Clearly restricting or limiting the use of the data is a must.
  • Requiring government funded research to provided curated data could provide a foundational trusted data set.
  • SMMs are concerned they will be priced out of the market as AI applications become more prevalent.
  • Bottom-up: recognize SMM needs, develop the right incentives, build track-record of success.
  • Sustainability requirements may require more data sharing.