.jpg)
In today's automotive engineering, modal shape recognition remains a critical process within NVH (noise, vibration, and harshness) analysis. For decades, this task has required manual intervention by specialists who, relying on their experience, visually interpret vibration modes from simulation or test data. This method, although effective, is costly, slow, and difficult to scale across multiple vehicle programs. Automatic approaches based on heuristics or the MAC (Modal Assurance Criterion) criterion exhibit low robustness when geometries, meshes, or sensor configurations change, limiting their industrial application.
To overcome these limitations, a new generation of techniques has emerged that combines canonical graph representations with deep learning. Instead of working directly on finite element (FE) meshes of each vehicle, heterogeneous models —both simulations and experimental data— are transformed into a common graph whose nodes represent structural regions with engineering meaning. These regions are connected through relationships that reflect domain knowledge, not numerical topology. On this structure, attention mechanisms on graphs and region-aware pooling are applied, achieving interpretable and physically coherent predictions.
The great advantage of this methodology lies in its transferability. By decoupling engineering knowledge from numerical discretization, the model can be applied to different vehicle programs without needing to retrain from scratch. This drastically reduces development time and allows leveraging data from previous projects. Furthermore, the inherent explainability —each prediction is directly associated with structural regions defined by engineers— builds trust in the results and facilitates prototype debugging.
This type of artificial intelligence for businesses solutions fits perfectly into Q2BSTUDIO's offering. Our experience in custom software development allows us to create specialized platforms that integrate graph models and machine learning, adapting to the specific needs of each client. Whether to automate mode classification in commercial vehicles or to implement real-time monitoring systems, our custom applications ensure performance, scalability, and full transparency.
Furthermore, at Q2BSTUDIO we approach the implementation of these systems with a comprehensive vision. We offer AWS and Azure cloud services to deploy models in production environments, ensuring high availability and flexibility. Cybersecurity is another fundamental pillar, protecting clients' intellectual property and sensitive data. For result analysis, we integrate business intelligence services such as Power BI, allowing intuitive visualization of patterns and trends. We even explore the use of AI agents that, based on the knowledge of the canonical graph, can recommend design modifications or alert about anomalies in real time.
This approach is not only applicable to the automotive sector. Any industry that handles heterogeneous simulation and experimentation data —such as aerospace, energy, or electronics— can benefit from a canonical graph representation. At Q2BSTUDIO, we are prepared to help companies adopt these technologies, transforming engineering knowledge into robust and transferable predictive systems. The combination of AI for businesses with a deep understanding of the technical domain allows us to offer solutions that make a difference in competitive environments.
Ultimately, robust and explainable 3D mode recognition with graph neural networks represents a significant advance towards the intelligent automation of engineering processes. By freeing engineers from repetitive and subjective tasks, the development cycle is accelerated and the quality of the final product is improved. At Q2BSTUDIO, we firmly believe in the potential of these technologies and work every day to make them accessible through custom applications and custom software that respond to the real challenges of the industry.
