Robot Work Quality Determination With RNN-Based Anomaly Response
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing work determination apparatuses face a time lag between anomalous occurrences and quality determination due to the need for storing and processing time-series data, which affects determination accuracy and speed.
Innovation Solution
A work determination apparatus utilizing a sensor data output unit that provides state data at intervals corresponding to the control period of an industrial machine's motion, combined with a determination unit employing a recurrent neural network (RNN) for real-time inference to assess the quality of work.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of time
If the duration of time for data processing is shortened to minimize time lag, then the responsiveness to anomalies is improved, but the determination accuracy is lowered
Solution Approach 1:
The patent segments the time-series data processing into multiple feature quantities with different durations of time. Each feature quantity is calculated from data of a specific duration, allowing the system to evaluate anomalies at multiple time scales simultaneously. This segmentation enables the system to use shorter-duration features for rapid anomaly detection while using longer-duration features for accurate quality determination, thus resolving the contradiction between speed and accuracy.
Solution Approach 2:
The patent applies partial action by selectively using different feature quantities based on the specific evaluation needs. For rapid anomaly detection, it uses feature quantities calculated from shorter time periods, while for comprehensive quality assessment, it incorporates feature quantities from longer time periods. This partial application of different processing durations allows the system to optimize both responsiveness and accuracy without requiring full processing of all possible time scales for every decision.
2Measurement precision
If multiple feature quantities with different durations are extracted to achieve high determination accuracy, then the accuracy is improved, but the time lag from anomalous occurrence to determination is still caused due to data storage requirements
Solution Approach 1:
The patent implements preliminary action by continuously calculating and maintaining multiple feature quantities from time-series data as the data flows in. Rather than storing all raw data and processing it when needed, the system pre-calculates feature quantities with different durations in real-time. This preliminary computation of multiple time-scale features eliminates the need to reprocess stored data, thereby reducing the time lag while maintaining the ability to perform accurate multi-scale analysis.
3Measurement precision
If a certain time period worth of data sets are stored for calculating feature quantities, then the determination accuracy is maintained, but the time lag between anomalous occurrence and determination is caused
Solution Approach 1:
The patent extracts only the essential feature quantities needed for determination from the time-series data, rather than storing and processing all raw data. By extracting pre-computed feature quantities with different durations directly from the data stream, the system eliminates the need to store large volumes of raw time-series data. This extraction approach maintains determination accuracy while significantly reducing the time lag, as the required features are readily available without needing to retrieve and process stored data.
Data Source
AI summary
It is an object of the present invention to obtain a work determination apparatus capable of preventing or reducing a time lag between the occurrence and determination of an anomaly in work. A work determination apparatus includes: a sensor data output unit that outputs sensor data with an output interval of an integral multiple of a control period with which a motion of a robot is controlled, the sensor data representing a state of work of the robot; and a determination unit that determines a quality of the work of the robot by inference using a recurrent neural network, based on the sensor data outputted from the sensor data output unit.