Manufacturing System Simulation Using ML Digital Twins
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current simulation solutions for manufacturing systems in the automotive industry are limited by the need for extensive data collection, which is time-consuming and impractical, and do not adequately consider physical constraints and biases, making it difficult to analyze and optimize manufacturing stations effectively.
Innovation Solution
A simulation process that inputs operation data into a machine learning model to compute manufacturing data, using video stream data and convolutional neural networks to generate digital representations of manufacturing stations, allowing for the simulation of manufacturing systems in different configurations to optimize workpiece processing times and identify potential issues.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If data collection for simulation is performed using traditional methods, then comprehensive manufacturing system data can be obtained, but the process becomes time-consuming and impractical
Solution Approach 1:
The patent creates a digital twin (virtual copy) of the manufacturing system that replicates real-world operations. Instead of collecting extensive real data through time-consuming measurements, the system uses a virtual model that mimics physical system behavior, allowing simulation data to be generated quickly while maintaining reliability through the digital twin's accurate representation of manufacturing processes
Solution Approach 2:
The patent replaces traditional mechanical data collection methods (physical sensors, manual measurements) with machine learning models and digital simulation. The ML-based digital twin substitutes physical data gathering infrastructure with computational models that generate simulation data algorithmically, dramatically reducing data collection time while maintaining data quality
2Productivity
If traditional simulation methods are used without considering physical constraints, then simulation speed increases, but the ability to detect real manufacturing failures decreases
Solution Approach 1:
The patent applies different levels of detail and constraint enforcement to different parts of the simulation. Critical manufacturing parameters and physical constraints are modeled with high precision in the digital twin, while less critical aspects use simplified models. This localized quality approach maintains simulation speed by avoiding unnecessary complexity while ensuring failure detection accuracy for important manufacturing parameters
Solution Approach 2:
The system incorporates feedback loops where simulation results are continuously compared against known manufacturing constraints and physical limits. When the digital twin predicts operations that violate physical constraints or typical failure modes, the system adjusts the simulation to account for these realities, improving failure detection accuracy without significantly impacting simulation speed
3Ease of manufacture
If manufacturing stations are analyzed in isolation, then individual station optimization becomes simpler, but the impact of environmental constraints and upstream/downstream dependencies is ignored
Solution Approach 1:
The patent segments the manufacturing system into individual stations, each represented in the digital twin with its own parameters and performance characteristics. This segmentation allows analysts to focus on optimizing individual stations while the digital twin maintains the contextual relationships between stations, enabling both simple individual analysis and comprehensive system-wide simulation when needed
Solution Approach 2:
The digital twin serves multiple functions: it can analyze individual manufacturing stations in isolation for simple optimization, or it can simulate the entire interconnected system to account for environmental constraints and dependencies. This multi-functionality allows the same tool to provide both localized simplicity and system-wide adaptability depending on the analysis requirements
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The invention provides a simulation process of a manufacturing system comprising a plurality of manufacturing stations which are configured for manufacturing workpieces; the simulation process comprising the steps of: inputting (110) operation data of the manufacturing system in a machine learning model in order to compute manufacturing data of the manufacturing stations; obtaining (112) a test configuration, of the manufacturing system; simulating (120) the manufacturing system manufacturing a workpiece depending on the manufacturing data, wherein the manufacturing system is arranged in the test configuration.