A Manufacturing Exchange for Modular, Composable, and Interoperable Mass Customization

The goal of this project is to enable small businesses and others to manufacture small batches of complex devices at low cost. Our approach is to develop an agile manufacturing exchange (ME) in which suppliers of raw materials, assemblers, transportation companies, banks, etc., participate through standardized protocols to fulfill manufacturing orders. We plan to develop new data-centric and control techniques for designing, dynamically optimizing, and transforming (through adaptation) an agile manufacturing exchange system (MES) that provides the highest level of performance and operational efficiency. These techniques will provide the foundation for a smart software mediation layer (i.e., a "broker"). Such a broker will enable an MES to be self-learning, and adaptive to dynamic/diverse service requests and resource availability, as well as support a large network of service providers and users within a complex information ecosystem The key challenge we propose to address in this project is the design and analysis of a smart and networked MES that can learn from data, as well as accommodate multiple product flows, dynamic and uncertain user demands, and uncertain network components, i.e., links between providers and providers themselves. To achieve this goal, we will focus on the following research thrusts:
1) Policies for admission control: Real-time decision-making capabilities on whether an order can be fulfilled on time with satisfactory quality. The highly dynamic demands and resource availability need to be handled by machine-learning-based predictive analysis.
2) Production Planning and Workflow Discovery: Identify best production plan for incoming new service. Automated discovery and timely updates of the actual production workflow.
3) Distributed Online Optimization: Handle uncertainty in distributed manufacturing network via learning statistics of different network parameters
4) Fault Tolerance: Accurate identification and fast recovery for different types of errors.
We developed an anomaly detector for our watchdog agent deployed in the manufacturing exchange system. It consists of three subcomponents: comparative analyzer which filters a large amount of stable anomalies, threshold analyzer which locks on suspect features, and correlation analyzer which remove redundant and irrelevant anomalies. Experimental results show that the proposed anomaly detector can identify three types of commonly encountered manufacturing errors with high accuracy. We have also developed the Stochastic Accelerated Distributed Augmented Lagrangians (SADAL) method to address the case where convex optimization problems need to be solved distributedly in the presence of uncertainty and noise, which can be applied for distributed control and optimization in manufacturing exchange systems.