Human-AI Co-Evolution in Lean Adaptation of Modular Building Production Systems

This presentation will discuss how Artificial Intelligence (AI) agents can learn from production engineers and increasingly help humans identify and resolve time and resource waste in producing modular building products with diverse customizations. Modular building products (e.g., room modules) need customization to ensure occupant satisfaction and energy efficiency in their environments. A building production line can have tens to hundreds of parameters. Small orders of customized building products need frequent adjustments of these parameters. Each new order requires many trials to align many product and machine parameter combinations against the changing machine states, material, and environmental conditions.

Both human and computer algorithms need hours or even days of “trials and errors” to find the optimal alignment between the machine settings and the context formed by materials, machine deteriorations, and production schedules with various transitions. These trials produce unstable qualities of products and machine stoppages, resulting in around 40% of the time and 50% of the material waste. Humans have fluctuating performance in controlling waste. Computer algorithms cannot guarantee safe and efficient transitions between customized production modes, producing unacceptable scraps. 

This presentation will present a series of research studies that enable

1. The interactive sharing of embodied reasoning of humans and production data analytics algorithms in detecting and explaining waste in modular building production systems and

   
   

2. Real-time visual and haptic signal exchange between humans and planning and control algorithms in finding safe and efficient paths towards stable production.

Specific research activities are on

1. Characterization of the human-machine setups in handling small orders to explain how scraps arise through coupling human cognitive behaviors, customization needs, engineering constraints, and machine deterioration;

2. Explainable production process data mining for interpreting how humans detect and prevent scraps in setting up and adjusting production systems for small orders of modular building elements;

3. Co-adaptation in human-machine interaction, where the AI-empowered production line adapts to the human’s evolving minds, skills, and task environments, generating coordinated control and visualization to guide the worker, who, in turn, processes information, provides feedback and collaborate with the machine.

The presentation will explain how using causal rules extracted from human observation and decision behaviors can guide AI agents in adapting machine setup strategies in changing production processes and contexts. The results characterize tasks, workflows, and contexts and capture the best practices of applying the Human+AI collaboration protocols according to these characteristics of tasks, workflows, and contexts.