Human-Robot Team Anomaly Detection in Discrete Manufacturing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional machine learning technologies face challenges in controlling robotic systems due to complex physical laws like Rigid Body Dynamics, where measurements such as velocity and acceleration are often unavailable, and are limited in monitoring and controlling discrete manufacturing processes involving joint human-robot teams, as they struggle to detect anomalies reliably.
Innovation Solution
The development of systems and methods that learn to optimize human-robot collaboration by characterizing normal operations in discrete manufacturing processes, using event relationship tables to detect anomalies in real-time, and adjusting robot assistance based on human worker conditions, such as health and alertness, to improve process efficiency and quality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional machine learning approaches are used for robotic control, then the system can learn from examples, but the system cannot reliably detect anomalies in discrete manufacturing processes because physical variables remain within normal ranges even during anomalies
Solution Approach 1:
The patent replaces conventional physics-based monitoring systems with a machine learning model that learns normal process patterns from historical data. Instead of relying on physical variables like temperature or pressure, the system uses sequence-to-sequence models to detect anomalies in process sequences and human-robot collaboration patterns, substituting mechanical measurement systems with intelligent pattern recognition.
Solution Approach 2:
The patent transforms the monitoring approach by changing from monitoring physical parameters (temperature, pressure) to monitoring sequence patterns and temporal relationships in manufacturing processes. The system encodes process sequences into numerical representations and uses learned models to detect deviations from normal patterns, effectively changing the parameter space from physical to behavioral/temporal.
2Measurement precision
If sensors are mounted on robotic systems to measure physical quantities, then position can be measured, but velocity and acceleration remain unavailable because sensors typically only measure position components
Solution Approach 1:
The patent introduces machine learning models as intermediaries that process position data from sensors and infer velocity and acceleration information. Instead of directly measuring these quantities, the system uses learned relationships from training data to estimate derived parameters, acting as a computational mediator between position sensors and the control system.
Solution Approach 2:
The patent replaces physical sensors that would directly measure velocity and acceleration with a computational approach using machine learning models. The system substitutes mechanical measurement devices with intelligent algorithms that infer these quantities from position data and learned patterns.
3Reliability
If conventional anomaly detection methods are used in discrete manufacturing, then the system can monitor physical variables, but it cannot detect anomalies because no physical variables are out of range during anomalous situations
Solution Approach 1:
The patent replaces conventional threshold-based monitoring systems with sequence-to-sequence machine learning models that detect anomalies based on pattern recognition. Instead of comparing physical variables against fixed thresholds, the system learns normal process sequences and detects deviations from these learned patterns, enabling detection of anomalies that do not manifest as physical parameter excursions.
Solution Approach 2:
The patent changes the detection parameters from physical variables (temperature, pressure, speed) to temporal and sequential patterns in the manufacturing process. By encoding process sequences and using learned models to compare actual sequences against normal patterns, the system detects anomalies in the structure and timing of processes rather than in physical magnitudes.
4Adaptability or versatility
If human workers execute tasks in discrete manufacturing, then flexibility is maintained, but process speed decreases compared to fully automated robotic systems
Solution Approach 1:
The patent merges human workers and robotic systems into collaborative teams where each performs tasks based on their strengths. The system learns to allocate tasks dynamically, having robots handle high-speed repetitive operations and humans perform complex or adaptive tasks, combining the speed of machines with the flexibility of human judgment in a unified manufacturing process.
Solution Approach 2:
The patent introduces dynamic task allocation and process sequencing that adapts to real-time conditions. The machine learning models learn optimal human-robot collaboration patterns and dynamically adjust who performs which task based on current process state, human worker condition, and robot availability, creating a flexible yet efficient hybrid system.
Data Source
AI summary
A system for detection of an anomaly in a discrete manufacturing process (DMP) with human-robot teams executing a task. Receive signals including robot, worker and DMP signals. Predict a sequence of events (SOEs) from DMP signals. Determine whether the predicted SOEs in the DMP signals is inconsistent with a behavior of operation of the DMP described in a DMP model, and if the predicted SOEs from DMP signals is inconsistent with the behavior, then an alarm is to be signaled. Input worker data into a Human Performance (HP) model, to obtain a state of the worker based on previously learned boundaries of human state. The state of the HW is then input into the HRI model and the DMP model to determine a classification of anomaly or no anomaly. Update a Human-Robot Interaction (HRI) model to obtain a control action of a robot or a type of an anomaly alarm.