Mobile Robot Image Anomaly Detection With Bounded ML Thresholds
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Industrial facilities require numerous sensors for anomaly detection, which are costly, require extensive wiring, and are prone to failure, leading to inefficiencies and potential malfunctions.
Innovation Solution
Utilize a mobile robot equipped with a camera to capture images of components and process them using a machine learning model, such as a convolutional neural network, to detect anomalies by generating reduced dimensionality output, determining anomalies based on specific probability thresholds to mitigate false positives and negatives.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If numerous fixed sensors are deployed to effectively monitor an industrial facility, then anomaly detection coverage is improved, but device complexity and wiring requirements increase significantly
Solution Approach 1:
The patent transitions from static fixed sensors to a dynamic mobile robot that moves through the facility to perform inspections. The robot dynamically positions itself to capture images of different components, eliminating the need for extensive fixed wiring infrastructure while maintaining comprehensive monitoring coverage.
Solution Approach 2:
The mobile robot serves as a universal inspection platform that can monitor multiple different components and locations throughout the facility. A single robot with a camera can replace numerous specialized fixed sensors, providing multi-functional anomaly detection across various industrial components without requiring dedicated wiring for each sensor location.
2Reliability
If numerous fixed sensors are deployed throughout the facility, then monitoring capability is improved, but the quantity of sensors and associated costs increase
Solution Approach 1:
The patent merges the functions of multiple fixed sensors into a single mobile robot platform. By combining imaging capabilities, mobility, and processing resources into one unit, the system achieves comprehensive monitoring coverage with far fewer individual sensor units, reducing both quantity and associated costs.
Solution Approach 2:
Instead of deploying physical sensors at every monitoring point, the system uses a mobile robot that sequentially visits different locations to capture images. The robot creates temporal copies of the monitoring function across different spatial locations, achieving comprehensive coverage without the need for simultaneous physical presence of multiple sensors.
3Ease of operation
If fixed sensors are deployed in fixed locations, then monitoring is simplified, but ease of maintenance and replacement becomes difficult
Solution Approach 1:
The mobile robot provides dynamic positioning capabilities that allow it to access sensors and components in difficult-to-reach locations. When maintenance or replacement is needed, the robot can navigate to the specific location, perform inspections, and facilitate maintenance operations without requiring human operators to physically access hazardous or confined spaces.
4Measurement precision
If machine learning output values above threshold are used to detect anomalies, then detection sensitivity is improved, but false positive occurrences increase
Solution Approach 1:
The system changes the decision parameter from a simple threshold comparison to a bounded range evaluation. Instead of detecting anomalies when values exceed a single threshold, the system identifies anomalies only when values fall within a specific range between a lower bound and an upper bound. This parameter modification reduces false positives by excluding values that are too high, which may indicate external factors rather than actual anomalies.
Data Source
AI summary
Implementations process an image, using a machine learning (ML) model, to generate a reduced dimensionality ML output. The image captures component(s) of an environment (e.g., an industrial automation facility) and can be captured via a vision component of a mobile robot within the environment. The ML output indicates, for each of N regions of the image, a corresponding anomaly detection probability that indicates whether an anomaly is present in a respective region, of the N regions of the image. Those implementations determine, based on the ML output, a quantity of anomaly detection probabilities that each satisfy a threshold. Those implementations further determine whether the quantity is both greater than a lower bound value and less than an upper bound value and, if so, cause remediating action(s) to be performed, such as causing rendering of an alert that indicates an anomaly is present for the component(s).