Manufacturing System Monitoring via Arm Motion Digital Representation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current monitoring solutions for manufacturing systems with multiple stations are impractical due to noise interference from upstream or downstream stations, leading to inaccurate detection of failures and inefficiencies, as they are typically tailored for single stations and require multiple tools, making it cumbersome to assess the entire system effectively.
Innovation Solution
A monitoring process using a convolutional neural network and machine learning model to analyze video stream data from multiple manufacturing stations, generating digital representations of moving arms and computing operation data to detect predefined conditions, such as time series and distance variations, thereby isolating and assessing each station's performance within the system.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If multiple monitoring tools are deployed for each manufacturing station, then monitoring coverage is improved, but device complexity increases
Solution Approach 1:
The patent implements a universal monitoring tool that can monitor multiple manufacturing stations simultaneously. The system uses a single monitoring tool equipped with multiple sensors (cameras, microphones, temperature sensors, vibration sensors) that can capture data from different stations, replacing the need for multiple dedicated monitoring tools for each station.
Solution Approach 2:
The patent combines multiple monitoring functions into a single integrated system. The monitoring tool aggregates data from various sensors and multiple manufacturing stations, processing all information through a centralized machine learning model that identifies anomalies across the entire system, thereby merging multiple monitoring tools into one unified solution.
2Reliability
If monitoring is performed on the entire manufacturing system, then system-wide detection capability is improved, but measurement precision deteriorates due to noise interference
Solution Approach 1:
The patent extracts and isolates specific anomaly indicators from the complex system-wide data. The machine learning model processes aggregated data from multiple stations and sensors, extracting relevant failure patterns while filtering out noise from upstream or downstream station variations, thereby maintaining measurement precision despite system-wide monitoring.
Solution Approach 2:
The patent introduces a machine learning model as an intermediary between raw sensor data and failure detection. This intermediary processes the noisy system-wide data, identifying patterns specific to each manufacturing station while accounting for environmental factors and upstream/downstream variations, thus preserving measurement precision in system-wide monitoring.
3Measurement precision
If monitoring solutions are tailored for single manufacturing stations, then measurement precision is improved, but adaptability deteriorates
Solution Approach 1:
The patent creates a monitoring solution that maintains station-specific precision while adapting to different system configurations. The machine learning model is trained to recognize station-specific failure patterns but can be applied across multiple stations with different configurations, making the solution both precise and adaptable without requiring customization for each station.
Data Source
Figure 1
Figure 2
Figure 3~4
AI summary
The invention provides a monitoring process of a manufacturing system, such as assembly line in the automotive or aircraft domain. The system comprises a plurality of manufacturing stations which are configured for manufacturing a workpiece. The monitoring process comprises the steps of: obtaining (100) video stream data of a moving arm transporting successive workpieces from the first manufacturing station to the second manufacturing station of the manufacturing system; the moving arm exhibiting a first end receiving the workpieces successively; inputting (106) the video stream data in a convolutional neural network algorithm in order to generate an arm digital representation of the moving arm motions; computing (108) operation data, such as time series data, of the manufacturing system by means of the arm digital representation; inputting (110) the time series data in a machine learning model in order to compute manufacturing data; generating (112) a signal if the manufacturing data meet a predefined condition.